Doing Liberal Arts Education pp 31–43 Cite as
Interdisciplinary Curriculum and Leadership Education: The Case of FLAME University, India
- Santosh Kumar Kudtarkar 3
- First Online: 11 December 2018
Part of the Education Innovation Series book series (EDIN)
FLAME University can truly be said to be a pioneer of liberal arts education in India. This chapter provides a historical perspective to the current Indian higher education system and its deficiencies. It explains why and how FLAME came to depart from conventional university education and nurture the intellectual and personal development of the individual rather than provide a narrowly academic and vocational course of study. FLAME’s model of liberal arts education is based on an inter- and multidisciplinary approach to cultivating thoughtful, sensitive, tolerant, ethical, and well-informed citizens who can occupy leadership positions in all walks of life. But as shown, it also has its roots in Indian culture and philosophy, as shown in the Discover India Program which also serves as an example of the intersection of interdisciplinary learning and personal development. The chapter concludes with the lessons learned and challenges that remain.
This is a preview of subscription content, log in via an institution .
- Available as PDF
- Read on any device
- Instant download
- Own it forever
- Available as EPUB and PDF
- Durable hardcover edition
- Dispatched in 3 to 5 business days
- Free shipping worldwide - see info
Tax calculation will be finalised at checkout
Purchases are for personal use only
Béteille, A. (2010). Viable universities – set conflicting goals, centres of higher learning cannot prosper. The Telegraph , Calcutta, April 22, 2010. Retrieved 6 Mar 2017, from https://www.telegraphindia.com/1100422/jsp/opinion/story_12356150.jsp
Capelli, P., Singh, H., Singh, J. V., & Useem, M. (2010). Leadership lessons from India. Harvard Business Review. March 20, 2010. Retrieved 6 Mar 2017. From https://hbr.org/2010/03/leadership-lessons-from-india
FLAME University. (2016). Discover India program . Retrieved 6 Mar 2017, from http://www.flame.edu.in/academics/undergraduate/program-structure/experiential-learning/dip/projects
International Baccalaureate: India. (n.d.). International Baccalaureate . Retrieved 6 Mar 2017, from http://www.ibo.org/about-the-ib/the-ib-by-country/i/india/
Majumdar, S. (2016). A brief history of the modern Indian university. THE. Retrieved 6 Mar 2017, from https://www.timeshighereducation.com/features/a-brief-history-of-the-modern-indian-university
Ministry of Human Resources Development. (2016). Educational statistics – at a glance. Retrieved 6 Mar 2017, from http://mhrd.gov.in/sites/upload_files/mhrd/files/statistics/ESG2016_0.pdf .
National Knowledge Commission. (2009). National knowledge commission report to the nation 2006–2009 . New Delhi, India: National Knowledge Commission. Retrieved 6 Mar 2017, from http://knowledgecommissionarchive.nic.in/downloads/report2009/eng/report09.pdf
Pal, Y. (2009). Report of “The committee to advise on renovation and rejuvenation of higher education” . New Delhi, India: Ministry of Human Resource Development Retrieved 6 Mar 2017, from http://mhrd.gov.in/sites/upload_files/mhrd/files/document-reports/YPC
Times of India. (2015). International baccalaureate schools in India post 10-fold growth in 10 years. Times of India, May 19, 2015. Retrieved 6 Mar 2017, from http://timesofindia.indiatimes.com/home/education/news/International-Baccalaureate-schools-in-India-post-10-fold-growth-in-10-years/articleshow/47349322.cms
Zakaria, F. (2015). In defense of a liberal education . New York: W. W. Norton & Company.
Authors and affiliations.
FLAME University, Pune, India
Santosh Kumar Kudtarkar
You can also search for this author in PubMed Google Scholar
Correspondence to Santosh Kumar Kudtarkar .
Editors and affiliations.
International Christian University, Mitaka-shi, Tokyo, Japan
Mikiko Nishimura & Toshiaki Sasao &
Rights and permissions
Reprints and permissions
© 2019 Springer Nature Singapore Pte Ltd.
About this chapter
Cite this chapter.
Kudtarkar, S.K. (2019). Interdisciplinary Curriculum and Leadership Education: The Case of FLAME University, India. In: Nishimura, M., Sasao, T. (eds) Doing Liberal Arts Education. Education Innovation Series. Springer, Singapore. https://doi.org/10.1007/978-981-13-2877-0_4
DOI : https://doi.org/10.1007/978-981-13-2877-0_4
Published : 11 December 2018
Publisher Name : Springer, Singapore
Print ISBN : 978-981-13-2876-3
Online ISBN : 978-981-13-2877-0
eBook Packages : Education Education (R0)
Share this chapter
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
- Publish with us
Policies and ethics
- Find a journal
- Track your research
FLAME University offers smooth, personalised student engagement with Salesforce
A single platform for all student engagement allows FLAME University to create a personalised learning environment, and offer a smooth student experience.
FLAME University’s philosophy is rooted in the concept of liberal education. The university encourages students to interact with various disciplines to address world challenges so they can evolve into critical thinkers and lifelong learners. FLAME strives to create a personalised, participatory environment that provides a transformational student experience.
As part of this vision, FLAME invests heavily in future-ready technology to build highly customised student experiences.
Wow your Customers with Customer 360. Create unforgettable moments with the world’s #1 CRM.
State of the Connected Customer: Insights from 15,000+ global consumers and business buyers on a new era of customer engagement
Future-proofing the university
The Information Systems team at FLAME was earlier limited by an underlying system that was neither easily customisable nor integrable. Also, it was slow to respond – resulting in frequent downtime.
“Conventional, out-of-the-box, enterprise systems do not give you the capability to build and customise on the fly,” observes Mukesh Joshi, Director Technology - Information Systems, CTO (Integrator), FLAME University.
So, FLAME began looking for a highly customisable, reliable solution. To fulfil the vision of personalisation, it would have to be cloud-based, and offer scalability, efficiency, and integrability.
Salesforce emerged as the clear choice.
downtime of the solution built on Salesforce
Technology enables accessibility and interconnectedness.
With Salesforce, FLAME is now able to offer its stakeholders a frictionless experience. “We no longer worry about maintenance, increased user load, and downtime. We use the freed-up time to build customisations quickly and easily, helped by Salesforce’s click-not-code functionality. So, we can provide our end-users with the best possible experiences,” says Joshi.
The first step in this journey was to improve collaboration and student engagement.
To this end, FLAME prioritised the building of a student community portal on Salesforce Experience Cloud . This portal has become the first touchpoint for prospective students looking to engage with the university, and current students seeking to connect with one another.
Prospective students can explore the programs offered and submit their applications here. The admissions team receives these and takes the process forward for smooth onboarding.
Current students, on the other hand, use the portal to collaborate on studies and projects. Students use Chatter to form focused groups, enabling speedy learning and project completion. Faculty members can also be invited to these groups as mentors, making the learning experience fruitful.
Streamlining the admissions funnel with Salesforce
A unified platform boosts service efficiency across the flame ecosystem.
For a university that values stakeholder well-being, a seamless service experience is key. Students, faculty members, and non-faculty support teams such as Information Systems and administration can all raise cases using email, forms, and the community portal.
These cases all flow into Salesforce Service Cloud and are assigned to the relevant departments for resolution. For example, the admissions office resolves cases raised by prospects, while students’ queries around subject and course choices are routed to the program office. With the solution functioning as a single service console for all departments and operations, cases to non-academic departments, such as the infirmary, facility maintenance, and travel desk, are also raised here.
“With one collaborative service system, we are able to efficiently manage and resolve all stakeholder enquiries and student expectations,” says Joshi.
Leveraging the Salesforce ecosystem for process improvement and optimisation
FLAME is also thinking of ways to continuously optimise its processes and systems to boost efficiency and the stakeholder experience. “We did not want to spend effort on these optimisations and felt that it would be ideal to find out-of-the-box solutions that could be deployed quickly and seamlessly within the Salesforce environment,” explains Joshi.
Sure enough, the team found apps to address its different needs on Salesforce AppExchange . Using Formstack, the team consolidated data from forms across the ecosystem; and streamlined workflows on Work-Relay. Booker25 helped automate the booking of classrooms and conference rooms. All these apps could be easily integrated with Salesforce, making it the single source of truth.
“We wanted to have one system, one place for all interconnected data. And Salesforce delivered,” says Joshi.
360-degree engagement for contextual learner experience
FLAME now has a unified platform for 360-degree engagement with each student, which captures the student’s entire journey with FLAME - from their first interaction to their latest grades, awards, and disciplinary concerns. Whenever a prospect or a student reaches out via any channel, the stakeholder uses the up-to-the-minute 360 view to respond in a sensitive, relevant manner.
”Student satisfaction is the core of our DNA. Salesforce gave us the platform and tools to easily capture student sentiment and satisfaction,” observes Joshi.
A recent survey showed a high 90% student satisfaction rate with the technology available to them to engage with the university.
student satisfaction rate with the technology
On the path to a best-in-class student system.
The university’s belief in leveraging top-notch technology has reaped dividends. With access to real-time reports and dashboards on Salesforce, senior leadership can now make speedy, data-backed decisions. FLAME is now adopting Tableau for data visualisation and analytics, to further underpin strategic thinking.
The university will also deploy Salesforce Marketing Cloud to improve lead acquisition and engagement. The insights will be used for better strategy and designing effective marketing campaigns on the right channels.
Eventually, FLAME plans to deploy Salesforce Education Cloud for a state-of-art Student Information System. “All our engagements and interactions will be on this one system. That is the dream,” concludes Joshi.
Deliver personalised student experiences and support from the very first touch point.
Explore the products that help our customers grow..
Want to check out everything Salesforce offers?
See how College for America launched a degree program in just 16 weeks.
ISBF is blazing a trail for year on year growth
KLAY Schools looks to hit milestones early with Salesforce
Get monthly updates and fresh ideas delivered to your inbox.
- planning ecologically
- cradle to cradle
- timber construction
- blocher partners
- blocher partners sens
- blocher partners sustain
- blocher partners india
- blocher partners general planning
Flame University Pune
Expansion of the campus — Gat No. 1270, Lavale, Off. Pune Bangalore Highway, Pune, Maharashtra 412115, India — 85.500 m² — 2026 — Flame University, Pune
Expansion of the Flame University
Near the Indian city Pune lies the campus of Flame University, a private institute of higher learning with a strong focus on classical education. Its student numbers have risen greatly since its founding in 2003, so that its expansion will be conducted by the planning office blocher partners.
The master plan incorporates the character of the existing arrangement of buildings and expands it to include the future faculties and additional functions. The architects will add new buildings such as apartment buildings, auditorium and sport facility which are distributed in an open layout on the site, taking environmental aspects into account. The environmental design and orientation of the campus, conducted by the planning office, will have a lasting impact on the decision-makers of the future.
State of the construction site
Castle of Mannheim
Quarter Pankower Tor Berlin
Health Care Center Berg
Outletcity Metzingen – New centre
Kita Landteilstraße Mannheim
Q 6 Q 7 Mannheim
Campaign Management, Email Marketing, PPC Management, SEO Services, Social Media
Digital strategy for FLAME University – Pioneers of Liberal ed in India [UPDATED 2022]
FLAME University is the first private university to introduce liberal arts education in its true essence and structure to India back in 2007. Ever since, it has established its position as one of the most sought after liberal education programs in the country with a world-class campus and enviable roster of reputed academicians and professors from all around the world.
We have worked with the institution for 5+ years helping on social media marketing, SEO and performance marketing so a lot of our work has been through hundreds of incremental changes that have compounded over time to produce direct value and bottomline results. And we are only getting started.
Increase in organic traffic
Liberal education is a relatively new concept in the Indian educational ecosystem. This presented a 2-pronged challenge of both identifying the right audience and also striking the right balance of educational and intent based marketing.
Additionally, a university student application process has multiple touch points and steps. In a complex performance campaign, it was crucial to get the analytics and attribution piece correct to be able to maximize results – both at the top of the funnel, with the ad platforms and at the post-lead part of the funnel, with the university CRM – Salesforce.
Bonoboz has been a trusted digital partner for FLAME University for many years. We have relied on their expertise with social media, design, video and especially paid marketing to take 10x the impact of our digital presence. Their methodical approach, keen insights and transparency set them apart.
Member, Academic Council, FLAME University
What we did we did
We adopted a very organized and structured process to the campaign. There were 2 primary channels that were used to drive 95%+ leads – Google & Facebook.
We started by defining a campaign design language that could extend to all visual assets. We know we would have a lot of assets used at different stages of the campaign and a consistent visual style would help build recall. Further, we created a system to being able to experiment headlines, primary text, colours and layouts through small experiments. In the long run, these small optimization helped us run a lean an d efficient campaign.
For search, we created a neatly divided campaign and ad set structure that helped us perform on keywords that we were already strong with at low cost, and divert more spends to keywords and areas where we knew there was scope of improvement – both in terms of visibility and conversions. Another crucial element that we had to implement was to avoid having our campaigns for FLAME (different programs) compete against each other. This hurt us in the short run in terms of attribution but it was the right move to make sure that every client Re. spent on media gets the max impact that’s possible.
Full funnel approach with a tight feedback loop from the panel as well as the super responsive team at FLAME admissions that provide constant feedback on the quality of leads. What conversations they were having, where they were getting stuck, what kind of leads were not useful and so on.
As for attribution, we made sure that pixel installation was done properly with all channels and tested thoroughly both for getting accurate feedback and also improving the input signal for the automated ad platforms directly improving results. With a comprehensive GTM setup, we were able to track events and page views across all landing pages and map the customer journey well.
This admissions year 2019-20 was a strong success for us not only in that we improved numbers but also in the backdrop of the Covid-19 pandemic with challenges of remote work, communication and team safety.
- Generated over 1000+ application leads from across India
- 260+ completed applications
- 75+ shortlisted candidates – chosen from these leads after the interview process
Building upon this, we continued to grow applications up from the previous year to another 25%++ in terms of shortlisted candidates while keeping the media spend budget within +5% of the original amount.
Other case studies case studies
Expert growth for Expert Education Australia & India
Campaign Management, PPC Management, SEO Services, Website Development
Building a luxury real-estate brand [Case-Study]
Campaign Management, PPC Management, Social Media
Restart 2021 – Hubilo Virtual Event for North America
Campaign Management, Social Media
Making the Baker’s Dozen India’s bread
Write in to us! us!
I practice what I teach and I teach what I practice.
Case Study on MSTC
Twice a year, I go to Flame Investment Lab at Flame University to teach a program called “Case Studies in Business and Investment Analysis.” These case studies are based on real businesses and sometimes the coverage is also in real time. In June 2023, I presented several case studies, including one on MSTC.
In this post, I am reproducing what was presented in the class. Many aspects of this case were not covered in the presentation such as potential competition from other private players like Metal Junction, the risks of investing in government-controlled companies called PSUs (public sector enterprises) and whether there is a way to reduce the chances of those risks materialising by being very choosy, and also whether investing in PSU makes sense at some cheap valuation.
The presentation was followed by an excellent debate. Participants had different views about investing in PSUs. What fascinated me were the strong views people had against investing in such businesses regardless of valuation and regardless of the quality of the business ignoring ownership structure. The prejudice was quite extreme, in my view. This, I think, is also reflection of the prejudice outside the classroom, which is so widespread that it’s probably a major cause for a few PSU stocks with excellent business models becoming extraordinarily cheap and attractive as investments.
There were many other PSU businesses that were also covered in the program and the same prejudice was witnessed by me when those cases were presented as well.
While these topics are not covered below, they were discussed in detail in the class subsequent to the presentation of the case study.
As of this writing, I and my firm’s clients are invested in this stock. Nonetheless, this is not a recommendation to own this stock, nor was it during the presentation of the case. The platform business model of this company is unique. I presented it to the participants so they could learn about it.
“MSTC” stands for Metal Scrap Trading Corporation. The name is somewhat boring, but the economics of this business is anything but.
MSTC as an Auctioneer
The jewel in MSTC’s crown is the “e-commerce” vertical, which conducts electronic auctions (“e-auctions”) primarily on behalf of the government of India and related parties like government-owned companies. The government of India has another entity called GeM (”Government e-Marketplace”), which it uses for procurement . For sales , however, the government relies to a very large extent, on MSTC. There is no significant overlap between these two entities.
In 2022, the government of India auctioned 5G telecom spectrum with the help of MSTC. This complex auction done over 40 rounds, was conducted in a fair and transparent manner, and generated more than USD 18 billion for the government.
The two key characteristics of auctions, fairness and transparency , are essential not only for attracting genuine buyers and sellers to a marketplace, but also to help the government in another way. The government aims to avoid any perception of favoritism when it comes to the sale of natural resources such as iron ore mines, coal, minerals, sand blocks, and resources extracted by government-owned companies like iron ore and natural gas. This principle also extends to the ongoing sale of scrap, surplus stores, old plant and machinery, e-waste, and obsolete items belonging to different branches of both the state and central government across India.
Indeed, after a huge coal mines allocation scam , numerous judgments issued by Indian courts now require government departments to sell natural resources only through transparent e-auctions and MSTC is often the only company that is trusted by the government to assist in this process. As it is owned by the government, this brings monopolistic rights to MSTC.
For example in FY22, MSTC was given the exclusive rights to conduct e-auctions of 7,665 liquor shops in 34 districts in the state of Rajasthan on behalf of the state’s excise department.
Public sector banks holding assets such as land, buildings, and apartments seized from defaulting borrowers, must sell them to recover loans that have turned into non-performing assets (NPAs). In FY22, MSTC helped Indian public sector banks sell 6,134 properties for about USD 650 million. This is a tiny number as of now but will grow significantly over time as MSTC helps these banks convert their NPAs into cash. Indeed, MSTC has been hired by Indian Bank Association to develop a platform for online auction of assets attached by its members.
NPAs are not the only thing that needs to be converted into cash. In India many temples receive enormous quantities of gold and jewellery from devotees as offerings to the gods. These valuable assets too need to be converted into cash. Temple trusts in the south Indian states of Karnataka and Tamil Nadu have hired MSTC to help them in this matter.
These are just a few examples to illustrate the diversity of auctions conducted by MSTC. Some of these auctions are lumpy in nature — for example sale of telecom spectrum does not happen every year. On the other hand, some auctions, such as sale of scrap metal by government-controlled steel companies produces a significant and regular income stream for MSTC. This diversity of earnings from conducting thousands of auctions every year means that the company is not overly dependent on just one or two types of auctions. Indeed, the company is very focused on increasing the diversity in its business as was described by the company’s Chairman recently when he said:
We are entering into the new areas, new frontiers and new kind of clientele. We are not limiting ourselves to the scrap sale. We are entering into new kinds of things, new kinds of business.
I like the growth mindset of MSTC’s management. Indeed, the company is also working towards helping private sector clients sell unwanted but valuable stuff via e-auctions. In the company’s annual report for FY22, the company wrote:
MSTC is casting more focus on the untapped e-commerce business from the private sector and in this stride MSTC has signed big ticket agreement with Reliance Industries, Indus Tower, Tata Power, L&T, Jindal Group, Vedanta etc to name a few.
Online auctions are a form of e-commerce that takes advantage of the digital platform’s capacity to breach geographical limitations, furnish real-time data, and minimise transaction costs. This results in huge convenience benefits to both bidders and sellers transacting business on the platform. The platform, of course, becomes more valuable to a seller if it already attracts more buyers. In turn, buyers go where the sellers already are, which makes the site still more attractive to sellers. These “network effects” are playing out in MSTC as can be seen from the table below.
As gross merchandise value (GMV) has grown, so have the company’s revenues. But operating costs have not risen in tandem — just what one would expect from a profitable, and growing platform business. As a result pre-tax margins in this business have soared from 29% in FY18 to 69% in FY23.
Notice that MSTC’s revenues in FY23 were just 0.12% of the GMV generated by the company in that year. This tiny percentage on GMV represents a significant entry barrier for new entrants and add to the competitive advantage of MSTC derived from monopolistic rights granted by the government as well as the network effects which attracts a large number of buyers and sellers to its platform.
The government of India and its related parties will always have the need to sell off things in a transparent manner. I expect the GMV to soar in the next decade or more. And as GMV grows, even a tiny spread of 12 basis points will produce rapidly rising revenues. And thanks to the significant operating leverage in the business, earnings should grow even faster.
One aspects that I really like about MSTC’s e-commerce business is that it requires no incremental capital to run and grow. Indeed, surplus cash of INR 4 billion matches the book net worth of the e-commerce business. This surplus cash can be taken out of the business without affecting its ability to grow.
The attraction of a wonderful business which can grow without requiring incremental capital has been mentioned by none other than Warren Buffett in the past. For example, in the 1994 annual meeting of Berkshire Hathaway, Mr. Buffett said:
There’s a huge difference between the business that grows and requires lots of capital to do so and the business that grows and doesn’t require capital. And generally, financial analysts don’t apply adequate weight to the difference between those. In fact, it’s amazing how little attention is paid to that. Believe me, if you’re investing, you should pay a lot of attention to that.
And in the annual report for 2007, Mr. Buffett wrote this about See’s candy — a wonderful business owned by Berkshire Hathaway:
There aren’t many See’s in Corporate America. Typically, companies that increase their earnings from $5 million to $82 million require, say, $400 million or so of capital investment to finance their growth. That’s because growing businesses have both working capital needs that increase in proportion to sales growth and significant requirements for fixed asset investments.
A company that needs large increases in capital to engender its growth may well prove to be a satisfactory investment. There is, to follow through on our example, nothing shabby about earning $82 million pre-tax on $400 million of net tangible assets. But that equation for the owner is vastly different from the See’s situation. It’s far better to have an ever-increasing stream of earnings with virtually no major capital requirements. Ask Microsoft or Google.
While MSTC is neither Microsoft, nor Google in terms of global dominance, it does enjoy their ability to grow earnings without requiring incremental capital.
One big reason why MSTC’s growth requires no incremental capital from shareholders is the presence of float. To deter non-serious bidders from ruining the sanctity of the auctions, every bidder has to give a refundable deposit to MSTC before any auction. This deposit is refunded on the completion of the auction but if a successful bidder fails to pay the final bid price, it is forfeited. The table below shows that MSTC has enjoyed the possession of significant interest-free float for the last several years.
While it has temporary possession of this float, MSTC invests the cash in short-term, high-quality, fixed-income securities which adds to the operating profitability of this business.
Key point discussed in class: Are earnings from deploying this float money in fixed income instruments operating or non operating earnings? In my view they are operating earnings because they come from investing float which is an integral part of the operations. But that is not how the company’s financial statements as well as databases treat MSTC’s treasury income. They treat it as other income which is automatically considered to be a non operating item which is not the correct treatment. So we have to make the necessary adjustment and treat it as operating income. By the way, this is what Graham used to prescribe as well. He used to tell his students to make the necessary adjustments to analyse the business properly. In accounting one of the key lessons is to focus on substance and not form and this is what we are doing here.
Another key point discussed in class: MSTC has a large amount of cash on its balance sheet, but a significant part of it is represented by float money which is other peoples’ money. Therefore, for calculating the return on net operating assets and also for estimating the market value of the operating business net of excess cash, the float money should not be counted as surplus.
MSTC as an IT Consultancy Business
The company is also helping clients set up their own auctions and, in effect, is becoming an IT consultancy business. Over the years, the company has delivered e-bidding packages for various ministries of government of India. For example:
- During FY22, MSTC helped in making improvements in platforms for e-bidding solutions like DEEP (Discovery of efficient electricity price) , TBCB (tariff-based competitive bidding for electricity), eBKray (online auction platform for sale of assets attached by banks)..
- MSTC developed and operates the portal for Ministry of Coal to auction coal mines and 25 mines were successfully auctioned during FY22
- Developed the EXIM Portal for Petroleum Industry — the online bidding platform for Export & Import of petroleum products for Indian Oil Corporation.
- In FY22, MSTC launched e-RaKAM (e-Rashtriya Kisan Agri Mandi) a nationwide electronic portal for trading in food grains, vegetables, fruits, spices and all agriculture-related commodities.
- Directorate General of Hydrocarbons has hired MSTC as one of the service providers for conduction of e-bidding events for natural gas exploration.
- Ministry of Power is using MSTC’s help for Tariff-based competitive bidding for electricity transmission services including solar power and wind power projects.
- National Thermal Power Corporation is using a procurement portal developed by MSTC for the establishment of FGD (Flue gas desulfurisation) at its various sewage treatment plants.
A Poor-Quality Legacy Business Camouflaged Earnings
Another interesting aspect about MSTC is just how much of the wonderful performance of its e-commerce business was camouflaged by the performance of a poor-quality legacy business, which produced losses, but is now an insignificant piece in the larger picture. Moreover, the causes for those losses have also been fixed and this business is not expected to burn cash going forward.
For many decades, well before the e-auctions business started, MSTC acted as a facilitator for procurement of industrial raw materials like heavy melting scrap, low ash metallurgical coke, HR Coils, crude oil, naphtha, coking coal, steam coal etc. for supply to industrial customers. This was essentially a trading business in which, on orders received from customers, the company would procure the raw material, hold it as its own inventory and then sell it to them at a small markup over cost. Unlike the e-commerce business, this trading operation was hugely capital intensive not just because of inventories but also because of receivables thanks to sales on credit.
Given large capital requirements in this business, the small markup over cost meant that the returns on capital were pathetically low. On top of all this, many of MSTC’s privately-owned customers (but not government-owned ones), defaulted on their obligations to pay the company for material supplied to them. This resulted in huge bad debts which brought down the reported earnings of the company. Over the ten years ended on March 31, 2023, aggregate provisioning for bad debts reduced pre-tax earnings of the company by a whopping 54%.
Thankfully, this trading operation has now been fixed. The company no longer supplies material to its customers until it has in possession a bank guarantee for 110% of the value of the supply. So, there will be no further bad debts. Moreover, this trading operation which now breaks even, represented just about 3% of MSTC’s FY23 revenue.
MSTC is Also Into Scrapping and Recycling of End of Life Vehicles
Yet another interesting aspect about MSTC is its 50:50 joint venture with the Mahindra group called Mahindra MSTC Recycling Private Limited (MMRPL). This company is an authorised RVSF (“Registered Vehicle Scrapping Facility”) for vehicles reaching end of their life. These old vehicles are purchased for de-polluting, dismantling and converting the metallic parts in an environmental friendly manner.
Since the Mahindra group is in the business of manufacturing automobiles, this joint venture with MSTC gives it access to steel scrap which is cheaper that buying newly minted steel. This also helps Mahindra group achieve the objective of reducing its carbon footprint.
But why is MSTC into this business at all? One key reason is that MSTC owns the platform through which the end of life vehicles are sold via e-auctions. Notably, in December 2022, the Government of India issued a directive regarding its Scrappage Policy, a government-funded initiative which seeks to phase out old passenger and commercial vehicles, thereby reducing urban air pollution, increasing passenger and road safety, and stimulating vehicle sales. According to the directive:
It has been decided that Government vehicles which are older than 15 years and owned by Government of India and its Ministries/ Departments, State/ UT Governments and their Departments, Local Government institutions, State Transport Undertakings, Public Sector Undertakings, and Autonomous Bodies with the Government of India and State Governments shall be scrapped immediately in order to achieve policy objectives… All condemned vehicles (including vehicles which have been prematurely condemned) are required to be scrapped through RVSFs.
In order to facilitate seamless scrapping of such vehicles, it is proposed that the e-auction platform developed by Metal Scrap Trade Corporation Limited (MSTC) … be used to conduct e-auction of such vehicles. RVSFS which have been commissioned as per provisions of MoRTH Notification GSR 653 (E) dated 23rd September 2021 shall only be allowed to participate in the auction. This would support operations of existing RVSFs by providing them with a base volume of end-of-life vehicles and would also encourage private investment in establishing new RVSFs.
This means that all government owned vehicles will be auctioned off on MSTC’s e-commerce platform when they reach end of life. Over time, volumes will be huge. In FY23, only 2,084 vehicles were auctioned by MSTC but the company estimates that 1.5 million vehicles owned by the government and related parties are approaching end of life. So this is going to be a big opportunity for MSTC and its e-commerce platform will enjoy monopoly rights too.
Even more interesting is this — the commission earned by MSTC as auctioneer will be between 2.5% and 3% of GMV and also that this commission will be earned by MSTC and not the joint venture company. Recall that in FY23, the company’s revenues were just 0.12% of GMV. So, this scrapping opportunity is not just going to be big over time, it will also be very profitable.
And of course, if the JV company is the successful bidder in the vehicle auctions, then for any vehicle it acquires, it will earn, according to company estimates, a return of anywhere between 10% to 50% over the cost of the vehicle by removing and selling re-usable things like the mirrors and stereo, and also from the sale of scrap steel, aluminium and rubber (from tyres) — which again could get sold on MSTC’s auction platform creating another stream of earnings for MSTC. Clearly, there are significant synergies between MSTC’s e-commerce business and that of this JV.
At current valuation, one is not pay anything for this opportunity because the earnings from the e-commerce activities alone are more than sufficient to justify its current market valuation.
Ferro Scrap Nigam Limited (FSNL)
There is one last bit on MSTC before I talk about valuation. The company owns a 100% subsidiary called FSNL which provides steel mill services such as processing steel mills slag for recovery of iron scrap and other metallics. It also provides custodian service for warehouse management to the clients of MSTC and valuation services for plant and machinery, scrap, movable and immovable material and properties.
Interestingly, this business, which is profitable (see table below) is up for sale and the government of India has started the process of divesting it completely. The bids have already been invited and as per DIPAM (Department of Investment and Public Asset Management) the planned divestment may happen in the near future.
In FY23, FSNL delivered pre-tax earnings of INR 512 million. While I don’t know if and when the sale will happen and at what price, I do know that should the sale go through, all the sale proceeds will come to MSTC and it’s likely that the company will either distribute part or all of it as a special dividend and/or invest it in the recycling JV, which is a profitable but capital-intensive operation. Either of these two outcomes will be fine. In any case, at current valuation, one is not paying anything for FSNL because the earnings from the e-commerce activities alone are more than sufficient to justify the company’s current market value.
At INR 339 per share in June 2023, the stock is quoting at less than eight times FY23 pre-tax earnings, implying a pre-tax earnings yield of about 14% which compares favorably with pre-tax AAA bond yields of about 8% a year. Equally important is the fact that the core e-commerce business is growing rapidly (pre-tax earnings grew by an annual average rate of 38% over five years ended on March 31, 2023) but requires no incremental capital investment. In FY23 this exceptional business delivered pre-tax margins of 69%. Dividend payout ratio was more than 40% and dividend yield alone is more than 5%. And on top of all this, one would also get to own a promising recycling business for nothing.
Interestingly, MSTC’s stock has no significant institutional ownership and has no significant coverage in the investment analyst community. Perhaps the camouflaging of an emerging and attractive platform business by the poor operating performance of a legacy business has played a role in the relative lack of interest from the street.
* * * * * * * * * *
June 28, 2023
Published by fundooprofessor
I am Distinguished Adjunct Professor at Flame University, Pune. Previosuly, I was Adjunct Professor at MDI, Gurgaon, where I taught MBA students two popular courses titled “Behavioral Finance & Business Valuation"(BFBV) and Forensic Accounting and Corporate Governance (FACG) for many years. I am also Managing Partner at ValueQuest Capital LLP— an investment advisory boutique which helps its clients invest their capital in good quality businesses. View all posts by fundooprofessor
2 thoughts on “Case Study on MSTC”
This is the best stock write-up I have read this entire year. Thank you from Toronto, Canada.
I was looking for such insightful research report. Just to learn how to value and research stocks. Thank you so much for this. @f
Comments are closed.
- Already have a WordPress.com account? Log in now.
- Subscribe Subscribed
- Copy shortlink
- Report this content
- View post in Reader
- Manage subscriptions
- Collapse this bar
You must be logged in to post a comment.
Search research topics, case studies, institutions and more:
- Log in/sign up
[excerpt from University website]
FLAME University has been established as a state private university in Maharashtra. It is a fully philanthropic initiative with an aim contribute to society in a noble manner. FLAME University is the pioneer of liberal education in India. The University's main objective is to offer an inter-disciplinary platform of education that provides both breadth and depth in diverse areas of knowledge. The liberal education ethos at FLAME University inculcates leadership and risk-taking abilities in students and focuses on discovering one's passion. The student-centric learning is delivered by esteemed faculty scholars in a world-class campus setting. FLAME University currently offers undergraduate, postgraduate and doctoral programs across various disciplines all anchored in the core ethos of liberal education. The university currently consists of – FLAME School of Liberal Education, FLAME School of Business, FLAME School of Communication, FLAME School of Computing and Data Sciences and FLAME School of Design, Art & Performance.
Other FLAME University links
Connect with flame university, want to contribute or amend this page.
Found an inaccuracy, something out of date, or want to add something you think is important to understanding FLAME University?
Add your knowledge by contacting us and requesting a page update.
How India and Australia are navigating online learning
A recording of a joint Australia-India online seminar exploring 'The Future of the University' is now available.
- Higher education
- online learning
An official website of the United States government
The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.
The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.
- Account settings
- Advanced Search
- Journal List
- Elsevier - PMC COVID-19 Collection
User experiences using FLAME: A Case study modelling conflict in large enterprise system implementations
The complexity of systems now under consideration (be they biological, physical, chemical, social, etc), together with the technicalities of experimentation in the real-world and the non-linear nature of system dynamics, means that computational modelling is indispensible in the pursuit of furthering our understanding of complex systems. Agent-based modelling and simulation is rapidly increasing in its popularity, in part due to the increased appreciation of the paradigm by the non-computer science community, but also due to the increase in the usability, sophistication and number of modelling frameworks that use the approach. The Flexible Large-scale Agent-based Modelling Environment (FLAME) is a relatively recent addition to the list. FLAME was designed and developed from the outset to deal with massive simulations, and to ensure that the simulation code is portable across different scales of computing and across different operating systems. In this study, we report our experiences when using FLAME to model the development and propagation of conflict within large multi-partner enterprise system implementations, which acts as an example of a complex dynamical social system. We believe FLAME is an excellent choice for experienced modellers, who will be able to fully harness the capabilities that it has to offer, and also be competent in diagnosing and solving any limitations that are encountered. Conversely, because FLAME requires considerable development of instrumentation tools, along with development of statistical analysis scripts, we believe that it is not suitable for the novice modeller, who may be better suited to using a graphical user interface driven framework until their experience with modelling and competence in programming increases.
Over the past two decades, the use of computational approaches to investigate social systems, including organizational and behavioural research has progressed from mere quantitative data analysis to one that complements traditional social science techniques through the use of advanced techniques in modelling and simulation. The techniques used within computational social science have evolved from mathematical approaches that utilize deterministic or stochastic equations, through to computational modelling approaches such as agent-based modelling and simulation  ,  . An advantage of these computational modelling approaches is the ability to fully control the underlying mechanistic component interactions of the model, thus once a computational model is established, quick studies of the effects of changes to the elements and associated parameter values can be seen through equation-solving or simulation. As such, simulation-based experimental results from a well engineered model are directly related to the level of abstraction and assumptions made during model design and development. This provides a firm basis for testing hypotheses on the mechanisms behind social behaviours and dynamics than traditional approaches such as participant observations or action research, where the unseen variables/factors can introduce additional complexity and uncertainty, which limits our ability to translate interpretations of observational/qualitative results to the real-world system.
Real-world systems are complex, with a range of system behaviours, properties and characteristics that emerge from the low-level interactions of a highly connected set of system components that function through temporal and spatial dimensions. One of the main strengths of the systems approach to investigating complex systems, is that it focuses on three key properties when modelling the systems: 1) system structures, 2) system dynamics, and 3) system control  . Indeed, previous discussions around Agent-Based Modelling and Simulation (ABMS) advise that it can be used to test working theories of the underlying mechanics of component interactions within systems and their resulting dynamics  ,  . As such, we believe that ABMS is the preferred approach to facilitate the micro-to-macro mapping of complex dynamical systems, in particular, social systems  . This is performed through altering the way in which agent interactions occur, or relaxing of assumptions at the individual agent-level, in order to investigate the emergence of system-level behaviours at the system-level. We therefore believe that ABMS provides a low-cost computational approach for developing hypotheses of the real-world system. With ABMS having recently become accepted as a scientific instrument for investigation  , it is now imperative that agent-based models (ABMs) adhere to the scientific method through a principled approach to their design, development and use, to ensure they are fit-for-purpose  .
Technology advances in hardware architecture, data storage and processing speed over the past twenty years has encouraged the development of new simulation frameworks that can handle increasingly complex computational models. With respect to ABMS, this has resulted in the development of a number of frameworks aimed at either subject matter experts or novice modellers, which use graphical user interfaces, or experienced computational modellers that are comfortable with command line terminals, a range of basic programming languages and shell scripting. A relatively new ABMS framework that falls in the latter category is the Flexible Large-scale Agent-based Modelling Environment (FLAME). Within this study we have used FLAME to model the development and propagation of conflict within a large multi-partner enterprise system implementation.
This manuscript provides an account of our experiences in using FLAME, with particular reference to a number of it’s strengths and weaknesses with respect to: ABM design; ABM development; constraints due to its underlying technical architecture; and performance. In particular, we highlight: the ease of model design for experienced computational modellers who are familiar with object-oriented programming and the Unified Modelling Language (UML); and the ease of model development for experienced computational modellers who are comfortable with the logic behind UML State Transition diagrams. In addition, we discuss a number of significant constraints of FLAME due to the underlying architecture, such as: issues when setting the Pseudo-Random Number Generator seed value; the inability to send messages between simulation time-steps; the inability to use global mutable parameters; performance challenges due to the I/O rate-limiting characteristics and the need for significant fast storage allocation to accommodate realistic simulations of social systems. The manuscript is structured as follows: Section 2 provides an overview of the major concepts that contribute to the theory behind our study; Section 3 discusses the case study; Section 4 provides an account of our user experiences when using FLAME to model the case study; Section 5 is the discussion section; and Section 6 provides our conclusions.
2. Related work
This section provides an overview of the major concepts that contribute to the theory behind our study.
2.1. Agent-Based modelling and simulation of social systems
Social networks and the various dynamics and emergent behaviours that develop from interactions between individual human agents are complex, meaning that they often display non-linear dynamics. Computational social scientists often model these social dynamics through equation-based mathematical approaches (e.g., system dynamics or differential equations) or ABMS. These computational modelling approaches facilitate the analysis and investigation of complex social systems that would otherwise be intractable. This may be due to direct experimentation of the system being either unethical or impossible across the scale and hierarchies of the social system; difficulties in observing system dynamics due to extremely short or indeed large timescales; difficulties in observing the whole population due to the magnitude of the system, or to the location or complexity of the system (e.g., virtual team).
Importantly, it is frequently apparent that a significant amount of data generated through observations, participant interviews or surveys in the social sciences, accumulates and is collected in a manner comparable to the object-oriented paradigm for design of computational systems. Historically, computational modellers have taken inspiration from the natural and physical sciences and utilized the reductionist approach, which advocates that systems should be investigated through their decomposition to their smallest indivisible unit, which we believe is analogous to looking for ‘objects’ within nature. As such, we believe that an object-oriented approach to computational modelling, through the use of a bottom-up approach for design and development of the models is a useful formalism for modellers in general, and those interested in Computational Social Systems in particular. ABMS intuitively expands upon the object-oriented approach to model design and development, and significantly benefits through the principal enhancement that an agent is active instead of passive, and secondarily, that ABMS is not bound by a central computational mechanism of control, but instead has multiple threads of control. Macal and North  provide a tutorial on ABMS, which describes this by stating that the “fundamental feature of an agent is the capacity of the component to make independent decisions” . In addition, Jennings  further extends the notion of an agent by advising that they are asynchronous components of a computational model that interact with the environment. As such, agents have the capability to sense their environment, and through processing the input(s) that they receive from the environment, may act upon it.
The origins of ABMS are in the modelling and simulation of human societies, systems, and networks and the associated dynamics and relationships within these hierarchical structures  , with particular connection to the fields of: Complex Systems, Computer Science, and Management Science  . In fact, the agent-based approach was developed to allow investigation of complex social systems at the level of individual actors/people, so borrowed the reductionist perspective from the Natural Sciences. The belief here, which was also borrowed from Engineering, is that complex systems are built from the ground up, which was a marked contrast to the beliefs of Systems Thinkers who took a top-down view. Since its inception, the agent-based approach has gained the attention of researchers from fields far away from its origins within the social sciences. Examples of the use of ABMS to model complex dynamical social systems include: the seemingly inocuous renting and sales patterns of houseowners/tennants within a reimplementation of Shelling’s Bounded Neighbourhood Model  ; the most recent banking crisis associated with the British banking sector  ; the way that virtual project team performance is affected by communication technologies  ; and the recent Covid-19 pandemic  .
The behaviour of individual agents is dictated by rules for their behaviour(s) and interactions. The interactions between compatible agents gives rise to behaviour(s) and dynamics at the system-level. This population-level behaviour, develops through the aggregated dynamics of individual agents, and is consequently deemed an emergent behaviour because it is not defined in the technical design of the computational model. Epstein  advises that this study of the emergent system-level behaviours is one of the key advantages of the ABM approach, and the ability to investigate the effects of the heterogeneous and autonomous individual agents has led to its increased adoption over the past decade  ,  . Moreover, an agent-based approach allows the modeller to design and develop the model by explicitly taking the individual agent’s location into consideration, which results in the model more accurately reflecting the spatial aspects of the complex social system, and in particular the network of social interactions within the system.
The engineering of viable computational models of social networks is not a straightforward process. This is due to the real-world system containing multiple feedback signals and non-linear dynamics, along with the model requiring the use of numerous parameters whose values have a degree of uncertainty (or may even be unknown) from observations/participant interviews. This uncertainty is further compounded through the introduction of randomness at key decision points/system events to provide the heterogeneity in behaviours that are seen across a population of human actors within the system  . ABMS equips the model developer with a set of techniques to integrate multiple data types gained from empirical data collection and knowledge themes gained from secondary data sources. Furthermore, one of the key strengths of the agent-based approach is that it allows us to model the complex non-deterministic and heterogeneous behaviours, which allows us to investigate the role of differences in chracteristics (e.g., demographic or behavioural) of individuals within social systems and develop testable hypotheses for new empirical research. See Williams  for lessons learned on development and application of ABMs of complex dynamical systems.
The work of Nikolai and Madey  has reviewed the main agent-based modelling frameworks available, and discussed their different computational underpinnings; examples include MASON  , Repast  , and NetLogo  . The commonality between all of these frameworks, is that they represent the individual agent as having a specific state at any particular point in time. Consequently, representations of each individual agent may be considered as a state machine , where the combination of current and previous inputs (previous states), along with the logic/functions associated with the agent-type, determines the output (next state) of the agent.
Increasing progress in hardware architecture, storage and speed over the past twenty years has encouraged the development of new simulation frameworks that can handle increasingly complex computational models, which simulate at magnitudes and dimensions that are much more closely aligned to real-world societal populations. For example, the Flexible Large-scale Agent-based Modelling Environment (FLAME 1 ) was developed  to provide an intuitive development framework for large-scale ABMs. FLAME was developed for use over the full range of computer hardware, which allows for the early coding and unit/system testing on a personal computer, before being moved to high-performance computing infrastructure for full-scale testing and simulation. FLAME has been used by researchers across a variety of domains, ranging from Biology  , to Economics  , to Human Resources  , to transport and logistics  .
2.2. The FLAME simulation framework
The FLAME simulation framework comprises a collection of Application Programming Interfaces (APIs), templates to define the agent-based model, and compilation and parsing routines to create the C code for running simulations. Due to FLAME’s conceptual architecture being based on communicating stream X-Machines, templates are used to define the rule-based functions (written in C code) of agents, with the agent types being defined through XML templates, and the instantiation of the initial iteration of a simulation run using another XML template that defines the attributes and internal agent states for the simulation run. FLAME requires installation and configuration of the Message Passing Interface (MPI) to develop messages that communicate interactions between specific agents and changes within the system’s environment. These agent interactions are simulated through transition functions that specify the rule-based logic of the ABM, with all agents that transition to a new state having their internal memory updated accordingly. Following specification of the model in the appropriate templates (e.g., C and XML), the xparser tool is used to generate simulation code (see Fig. 1 ). The xparser is written in C with the use of standard libraries, and is actually a collection of template files and header files that are stored in the desired directory of the computer (i.e., laptop, desktop, or HPC) and compiled using the freely available GCC compiler. Furthermore, through the use of the MPI communication framework, the generated simulation code is also portable to HPC platforms that use parallel architectures, thus enabling efficient communication between agents that may be across different compute nodes, and therefore remain in sync  .
FLAME relies on the definition of models using the XML and C templates, and their parsing and subsequent compilation into simulation code. This simulation code is then run, through linking the main executable file to an initial starting parameters file (0.xml). Due to FLAME utilizing the discrete-event approach, individual simulation runs generate separate output files for each time-step increment within the simulation, with each output file acting as the input file for the subsequent simulation time-step (after  ).
An X-machine is a formalized specification developed by Eilenberg  that has the capability to model both the system’s data and the specification method ( function ) for controlling the system. They were originally introduced in 1974, but did not receive widespread acknowledgement in the modelling community until 1988  when the approach was advocated as the basis for a formal model specification language. Subsequently, this was built upon in 1992 through the concept of stream X-Machines, which utilized sets of input and output symbols that were read through in a stream. From a conceptual modelling perspective, X-Machines employ both a diagrammatic approach and formal notation to model the system, where the X-Machine contains infinite internal memory, a set of functions, and a current state. The current state of control (the function defined in the specification method) and the current state of the memory, is processed alongside an input symbol from the input stream to determine the next state of the X-Machine, update it’s memory state, and calculate the output symbol, which becomes part of the output stream that is used for communicating to other X-Machines. This can be summarized in that the system’s new state is a product of it’s current state (using memory and reference to a list of input symbols) and the relevant function  .
A communicating stream X-Machine model is a formalized specification that builds upon that of X-Machines to introduce additional modelling capabilities. In particular, communicating stream X-Machines can be used to compute the functional behaviour of the system at a lower-level of granularity (i.e., individual agents), whose individual dynamics may be aggregated up to the overall system level in order to generate the emergent behaviour of the system as a whole. The formal notation of the communicating stream X-Machine specification utilizes a 10-tuple notation, with C i X representing the i th communicating stream X-Machine component, which is defined by Stamatopoulou  as:
C i X = (Σ i , Γ i , Q i , M i , Φ i , F i , q 0 i , m 0 i , IΦ i , OΦ i ) where:
An X-Machine is defined by Holcombe  as a system that has internal memory and an internal computational state, which dependent on environmental input and their current internal state, can transition to another computational state (see Fig. 2 ). The additional functionality incorporated into communicating stream X-Machines means that they are able to model in a single process specification, the functional behaviours and dynamics of an agent, as well as the intrinsic system data that it is modelling. A communication matrix using MPI facilitates communication (and thus interactions) between individual communicating stream X-Machines. This communication matrix is essentially a simulation-level message board that facilitates the reading and writing of information between every communicating stream X-Machine.
Diagrammatic representation of a generic communicating stream X-Machine. Relevant input messages that are held within the communication matrix are read by an agent, and potentially processed to initiate state transitions and updating of the agent’s current memory and current state  .
2.3. Enterprise system implementations
The Gartner Group devised the term Enterprise Resource Planning (ERP) in 1990  to define the software that provided functionality for organizations to use to manage their core administrative (back office) functions such as Finance, Payroll and Human Resources. A number of leading software vendors (e.g., Oracle, JD Edwards, PeopleSoft and SAP) offered ERP software packages to store the organizational data and provide assistance to facilitate compliance to a standardized set of business processes  . With the expansion of Information Systems (IS) and Information Technology (IT) throughout the late twentieth century, the individual software modules that focused on the different organizational business functions were integrated together. These larger software systems, which were termed Enterprise Systems, not only integrated the individual ERP modules, but also provided additional functionality, such as business intelligence, advanced planning, and automatically processing data from external supplier/customer relationships. Enterprise systems are now commonly used to help organizations (of all sizes) manage their human, financial and physical resources (e.g., staff, money, and products) along with key external relationships (e.g., customers and suppliers) more effectively  .
Enterprise systems require considerable time and resource commitments in order to be implemented properly within an organization. These implementation projects within large organizations are usually implemented by third-party service providers, with the largest implementations using the consultancy services of both the vendor (e.g., SAP or Oracle) and professional services firms (e.g., Accenture, IBM, CapGemini, etc.). The various customer, vendor, and professional services personnel are usually structured into project teams that relate to the functional modules within the enterprise system, such as Finanicals, Payroll and Human Resources, which gives rise to a wider implementation programme.
Whitty  has postulated that the increasing complexity and scale of reach into the business by these enterprise system implementation programmes, contributes to them exhibiting similar behaviours and traits to those of complex systems. We have previously discussed that the emergent behaviour generated within the social network of these large implementation programmes, may, to a large degree, be due to the complexity stemming from the large number of team members, and the growing trend of using multiple third-party providers to implement the individual projects that combine to form the programme. In addition, we discovered that the individual project teams may have competing priorities and objectives, and that these may lead to various forms of conflict propagating throughout the wider programme, which we view as a social network of formal workplace relationships  .
2.4. Conflict within enterprise system implementations
Conflict within group situations, such as the project teams that implement functional modules of enterprise system software, has been defined as interpersonal incompatibilities or the holding of divergent views/perspectives amidst the members/participants, which may be individuals within a single group (intragroup conflict) or between different groups (intergroup conflict)  . Conflict has been reasoned to be an intrinsic part of group  and project team dynamics, which can propagate throughout the network of group members as a shared affect  . Conflict develops during a variety of circumstances relating to the implementation of group- and team-based activities, and in three main forms: task, process, and relationship conflict  .
Enterprise system implementations at large organizations, often require the personnel, expertise and resources of multiple third-party organizations, which may have different, and often incompatible, business objectives and commercial drivers. As discussed above, the programme-wide implementation of the enterprise system is routinely divided into discrete project teams that map on to the corresponding functional modules within the enterprise system software (e.g., Finance, Payroll, HR). Our recent study  showed that within large multi-partner enterprise system implementations, conflict (be that task, process or relationship) can develop between members of a particular project team, or between members of different project teams, and once developed, can propagate throughout the social network of the multi-partner enterprise system implementation. Furthermore, this study conceptualized the propagation of conflict within large multi-partner enterprise system implementations as conflict propagating between team members of an individual project team, and also between team members that are situated within different project teams (see Fig. 3 ). In addition, we built upon the work of Gamero et al.  who discovered that conflict is dynamic and can transition over time, by hypothesizing that task or process conflict can transition to relationship conflict, and that relationship conflict (being directly related to shared affect), is the most common form of conflict to propagate through the social network of enterprise system implementations. We therefore hypothesize that the conflict, which initially developed between a subset of team members from two different project teams, has the potential to propagate throughout the social network of the wider enterprise system structure, and may negatively affect implementation of the enterprise system as a whole.
Rich picture: Observable phenomena that emerge from the interactions between individual team members. It is hypothesized that conflict can develop through a number of mechanisms, and that once formed, it may propagate throughout the social network of the wider programme. If any of these occur, they may be observed as deviations away from the agreed scope, time, budget and quality of the enterprise system implementation. Reproduced from  under Creative Commons Attribution 4.0 License.
3. Case study
The case study relates to a large UK-based organization that was implementing an enterprise system along with associated IT hardware as part of a major strategic modernization initiative. This IS/IT change programme was underpinned by the need to integrate the enterprise system with multiple, existing, third-party legacy systems, to facilitate significant cost savings and promote more efficient business processes. The client selected a large Enterprise Applications vendor to implement the installation and configuration of the enterprise system, and to lead the design and assist in the development of the technical architecture that the enterprise system was embedded within. In addition, two different Professional Service Providers were also contracted to act as Domain Expert Consultants for the configuration and extension of functional modules within the enterprise system, along with a third Professional Service Provider who consulted on the hardware and middleware requirements for hosting the enterprise system.
Such an environment is best described as a multi-partner enterprise system programme implementation. The resulting social network was structured into project teams that aligned to the functional modules within the enterprise system, along with a Programme Management Office (PMO) and project teams that focused on developing technical extensions (e.g., custom forms, reports, and database triggers), training material, and installing and configuring the technical architecture that hosted the enterprise system. Overall, there were 159 resources from these five organizations, with 972 undirected workplace relationships between them, which formed the basis of the social network map (see Fig. 4 ). The Conceptual Framework presented in  , acted as our conceptual model for the ABM, by providing the phenomenological behaviours that are the basis of the emergent behaviours generated from the ABM (see Fig. 3 ), alongside the detailed person-level data and social network map that formed the basis for the individual agents within our computational model, and the rules that dictate which agents can interact with each other.
Social network of the case. The programme structure of the multi-partner enterprise system case study is represented by a social network map that defines how resources from the various organizations are assigned to the individual project teams. It can be seen that the individual project teams are structured around the functional modules within the enterprise system software (e.g., Financials, HR, etc.) and that they do not work in isolation, but instead require interactions with other project teams in order to implement their specific project objectives, and in turn facilitate implementation of the wider programme. Reproduced from  under Creative Commons Attribution 4.0 License.
The case therefore constitutes a large multi-partner enterprise system implementation, which utilizes IS/IT consulting personnel from four external organizations. These external resources were combined with client resources, to form a number of project teams that align to the underlying functional modules of the enterprise system. The resources worked together within their respective project team, with a number of them also acting as bridgers to interact with other project teams, which is diagrammatically represented in Fig. 4 as the programme-wide interfirm social network. This multi-partner social network, introduces a risk that conflict may develop due to the differing backgrounds of project team personnel (e.g., different professional identities, cultural backgrounds, education, and normalized behaviours), along with their constrasting drivers and motives for being at their chosen employer (and occupation), and thus on the implementation programme. It was discussed in our previous work  , that conflict is to all intents and purposes inherent within large multi-partner enterprise system implementations, and as such is almost always going to occur at some point during the lifecycle of the programme-wide implementation, or within one or more of its constituent project teams. Our ABM has been designed to model this case. It allows us to simulate the development and propagation of conflict throughout the social network of the workplace relationships within the individual project teams and the wider programme as a whole.
4. User experiences of FLAME when modelling the case study
The ABM  was designed and developed to reflect findings from our previous qualitative studies  ,  . In the subsections below, we discuss our experiences using FLAME following the design, development, and simulation-based experimentation of an ABM of conflict development and propagation within our case study. Our discussion focuses on four main areas: the ease of model design having made the decision to use FLAME; the ease of model development using FLAME; constraints due to the underlying architecture of FLAME; and performance of FLAME in running simulations.
4.1. Ease of model design
As discussed above, the FLAME simulation framework utilizes the concept of communicating stream X-Machines to define the logical entities and the rule-based logic of their interactions within the agent-based model. The conceptual framework  ,  , acted as the functional specification for our ABM, which in turn was used as the basis for developing our technical specification. When developing the technical specification, we were cognizant of the underlying architecture of FLAME, and made explicit reference to: the C programming language and the XML markup language, which were used for coding agent interaction functionality and defining agents; the communicating stream X-Machine architecture for defining agents and their associated behaviours and dynamics; along with the use of the centralized message board to facilitate communication between agents.
Briefly, our technical specification consisted of a two-dimensional, undirected network of 159 agents that represented the individual team members from the case study. Communication between individual members followed the rules specified in the adjacency matrix and resulting social network map ( Fig. 4 ), and dummyTXdummy-(this communication within the ABM is modelled as a process with no velocity (i.e., link between two agents, but no speed). Specification of the agents was straightforward because the communicating stream X-Machine architecture facilitates easy compliance to an object-oriented approach to design. In addition, FLAME is flexible with respect to defining the physical boundary of the model, which allowed us to design a 2-Dimensional physical environment for our ABM, and the use of Cartesian co-ordinates to position individual agents within the environment. This allowed us to group agents together in 2-Dimensional space to represent the clustering of resources into their respective project team, and to disperse individual teams within an open-plan office, another office in the building, or potentially in another location.
System-level behaviours corresponded to those depicted in our Rich Picture ( Fig. 3 ), with the main agent attributes that facilitate simulation of conflict development and propagation relating to: project team (e.g., Programme, HR, Financials, etc.); Organization (e.g., Client, Vendor, Consulting Firm); Role Type (e.g., Management, Functional, Technical, Training); Conflict Quotient for each type of conflict; Stress Quotient for each type of conflict; and Formal Authority Quotient, which indicates how much formal authority/power the agent has over others. With respect to communication between agents, although FLAME utilizes a centralized message board, we were able to design restrictions into the communication mechanism to ensure agent communication corresponded to the workplace social interactions, as defined within the adjacency matrix from our case study (i.e., the 972 workplace relationships between agents). This was facilitated through adding an attribute to the agent definition that explicitly defined the other agents that it is able to communicate and interact with.
One point to note is that when diagrammatically modelling an X-Machine as part of the technical specification, the X-Machine state is associated with the computational system state transitions and does not correlate to the social (domain specific) states. These internal X-Machine state transitions are achieved through the use of transition functions, which encode the logic that changes the agents memory, and communicates with other agents through the use of messages sent to (using output port) and received from (using input port) the centralized communication matrix. For those familiar with UML, an ABM developed using X-Machines can be specified (from a system behavioural perspective) as a set of connected state transition diagrams.
4.2. Ease of model development
For those comfortable with integrated development environments (to write code), the command line terminal (to run simulations and submit batch jobs to HPC architecture), and the use of shell scripts (to submit jobs to a server, link to scripts for post-processing of output files, and link to analysis scripts [such as R or Matlab] to generate descriptive statistics and graphs), the FLAME simulation framework is an intuitive and very friendly tool for developing agent-based models using XML and C. Unfortunately, the converse of this holds for the novice modeller, and we conjecture that a more user friendly agent-based modelling framework that uses a graphical user interface, such as NetLogo, would be suitable as a first step to developing competence in model development and simulation-based experimentation.
FLAME provides a simulation engine that parses the ABM specification files (XML agent definition files and C functions files), manages the execution of simulations (on either a single node or over parallel architecture), and manages potential interactions between individual X-Machine agents. Within FLAME simulation runs, the notion of time is modelled as discrete time-steps . Each time-step therefore involves the individual X-Machines to iterate through their internal computational states, which may result in updates to their real-world social states following interactions with either the environment or other X-Machine agents. Furthermore, for those familiar with object-oriented design and development, FLAME, being underpinned by the concept of communicating stream X-Machines, allows the various functions within the ABM to be constructed in a modular fashion. Consequently, FLAME ensures that each type of X-Machine agent can be designed in an object-oriented way, which allows their construction and modification to be performed without the risk of affecting the functionality of other agent types as long as a standard interface is used for communication and interactions between agents.
The underlying process related to the technical architecture of FLAME is the requirement for synchronization, through a simulation-level time-step after all agents within the simulation run have completed one internal computational transition function. The effect of this technical processing constraint, is that the end and start of computational transition functions, incorporate the ability to synchronize communication and internal computational state transitions. With this in mind, care is required to ensure that agent behaviours do not contain any loops, which would incur the risk of undecideability problems, and that the interactions between agents (through use of the centralized message board) does not result in causal loops, which could impact upon the step-by-step updating of the agent internal states.
Coakley openly acknowledged during his design and development of the FLAME simulation framework that “when communicating X-Machines are used to represent agents in an [ABM], communication is [often] restricted to interactions with neighbouring agents that are [located] close to one another”  . Within our social system model, there are two broad categories of messages used within the simulation runs to communicate an X-Machine agents’ current location and their potential to initiate or be part of a social interaction. The location message is sent from individual X-Machines to the communication matrix to broadcast their current location so that other X-Machine agents can calculate whether they are within the appropriate spatial range for initiating an interaction, or alternatively to determine whether they are connected within the social network. With specific reference to the latter, the agents will elucidate whether they are connected (as defined in the adjacency matrix), and once [a] suitable agent(s) is/are identified, messaging to the centralized communication matrix would then be used to initiate interactions between the relevant agents.
Finally, one point to be aware of, is that although development of an ABM is one of the principal essential activities, in isolation, it will not enable simulation-based experimentation. In order to run simulations and analyze the resultant data, a number of additional computational tools need to be developed (termed instruments ). For our research, we developed the following instrumentation tools alongside the development of the ABM itself: Unix shell (specifically, BASH) and Ruby scripts to facilitate a semi-automated submission of simulation runs to the HPC resources; Python scripts to facilitate a semi-automated mechanism to generate the initial starting parameters files (using XML templates) for initiating simulation runs; scripts for the processing and transformation of XML output files to CSV files (Python and Matlab scripts); scripts to automatically generate graphs (in Matlab and R) for analyzing the data; a visualization tool to provide a graphical and animated view of the simulation run’s system dynamics over time (separate package provided by the FLAME development team, that can be configured to the ABM); and various statistical analysis scripts (developed in Matlab and Python) to analyze the simulation output data for statistical significance using the Kolmogorov-Smirnov test  , and effect magnitude using the A-Test  . These instrumentation tools (apart from the FLAME visualizer) were developed using the appropriate scripting/programming language, so in order to ensure that ABMs developed in FLAME can be robustly tested and analyzed, the developer requires a working knowledge of various statistical techniques along with competence in various scripting/programming languages. With that in mind, we do not believe that FLAME is appropriate for the novice modeller or subject matter expert.
4.3. Constraints due to underlying architecture
With FLAME being designed and developed from the outset to utilize parallel processing hardware architectures, the notion of a global mutable parameter unfortunately does not exist within the simulation framework. This is because such functionality would introduce issues around concurrency control and potentially lead to dead-locking. For instance, an individual compute node within the HPC architecture may be processing the simulation data for a particular X-Machine agent that is dependent on the current value of a global parameter, but the global parameter is currently being processed by another compute node. To mitigate risks around using outdated parameter values, the compute nodes would be required to wait until they confirmed that the global parameter was not being processed. When amplified up to the level of the overall simulation run, the cumulative waiting would incur a significant overhead regarding the Wall-Clock time of simulation runs, which would significantly offset the benefits of parallel processing. Consequently, we were unable to utilize a global counter to keep track of system-level quantities or to set system-level chracteristics, such as a pseudo-random number generator seed value.
4.3.1. Setting the pseudo-Random number generator seed
Pseudo-Random Number Generators (PRNGs) allow us to introduce probabilistic behaviour into our computational models, such as the interactions between X-Machine agents, or between the agents and the system environment. During the calibration exercise, and later more formally during the verification process, we ensure that the computational model is not overtuned through running multiple replicate simulations that utilize different PRNG seed values and performing statistical analysis to ascertain the variance in simulation output data. The principle behind explicitly setting the seed value for a PRNG is that you set once, and use multiple times  . A significant constraint that we discovered with FLAME is that it does not include a mechanism for you to do this very easily, because mutable parameter values can only be defined and set within the definition of an agents’ logic (the C functions file). As discussed in the previous paragraph, the design decisions taken when FLAME was first developed means there is no opportunity to set a global constant within the simulation environment setup of a simulation run due to the inefficiencies involved with it propagating across the multiple nodes within the HPC architecture. With that in mind, not only are global variables, such as total counts, unable to be updated, but the PRNG seed value cannot be set at the global level. This introduces the constraint of either not being able to set your own PRNG seed value, and therefore being forced to use the system clock to derive the seed value, or to develop a technical workaround where the PRNG seed value is explicitly set within an individual agent’s definition and logic (see Fig. 5 ).
The levels within a simulation run at which the PRNG seed value can be set when using FLAME. Although the principle of using PRNG seed values is that you set once per simulation and use multiple times, FLAME’s explicit exclusion of global mutable parameters means that this is not straightforward. Due to FLAME requiring all functionality to be written at the level of agents, this provides us with the ability to experiment to see whether any of the levels allow us to develop a technical workaround, for instance through setting at the level of agent-type, where any function would be called once per agent-type per iteration, or at the level of individual agents, where any function would be called once for each agent and for each iteration.
When setting the PRNG seed value at the level of agent-type, you unfortunately encounter the issue of the seed value being reset during each simulation time-step, and thus do not gain the full benefit of using a PRNG to simulate stochastic behaviours within the computaional model. Through explicitly setting the PRNG seed value a list of random numbers is generated, albeit in a determinstic manner so that you can reproduce simulation runs by using the same seed value (hence the use of the prefix pseudo ), to be used within probabilistic functions  . The resetting of the PRNG seed with the same value and at the level of agent-type, results in the same deterministic list of pseudo-random numbers being generated as per the previous simulation time-step, which reduces the benefits of using a PRNG. This can be succinctly illustrated within our case study: if you set the seed value at the level of an individual agent within our ABM that contains 159 agents (e.g., separate programme team members) over a 2000 time-step simulation, you in effect repeat the resetting of the seed value 318,000 times. Likewise, if you set the seed value at the level of agent-type, you in effect repeat the resetting of the seed value 2000 times (assuming a single high-level agent type of team member and 2000 time-step simulation run). Although the overall effects at this level are orders of magnitude smaller (e.g., far fewer resets of the PRNG seed value), it still significantly affects our ability to incorporate stochasticity into the ABM and associated simulation runs.
One approach for incorporating stochasticity into ABMs developed using FLAME, and therefore gaining variance between replicate simulations is to run simulations using the built-in Production mode within FLAME, which derives the seed value from the system clock. Unfortunately, this approach does not enable exact reproduction of individual simulation runs because the system clock is continuously updating with the progression of time, which seriously impacts our ability to repeat simulation-based experiments that have interesting dynamics, or to help resolve issues during the debugging activities in model development. Fortunately, a technical workaround was identified that resolved this issue, which involved the creation of a dummy agent whose sole purpose is to set the PRNG seed value in the first iteration of an individual simulation run. This utilized an agent-level counter (to count the simulation time-step), and logic to only set the PRNG seed value when the counter equals zero. In addition, and to keep the simulator tidy of computational artefacts, the logic also ensures that this dummy agent is removed from the simulation once the simulation time-step counter reaches a pre-defined number.
4.3.2. Unable to send messages between simulation time-Steps and need to explicitly set memory allocation
Additional constraints, although minor in relation to the above, are that a single simulation time-step is taken as a standalone run of a simulation, and that memory allocation for the agents and messages requires a continuous block size of memory. With respect to the single simulation time-step, this means that upon completion of all functions in that time-step, the global message board is emptied of all messages generated by agents during that time-step. Simulation modellers therefore have to take this constraint into account when designing and developing the ABM, because messages cannot be sent between time-steps, so pertinent information (e.g., temporary parameter values or communication from specific neighbours) will need to be stored within the agent’s memory (i.e., within an agent attribute). Conversely, with respect to the block size, this is important for parallelization when using MPI, because the simulator needs to know how to package up data that is sent across nodes, so the need to explicitly define the size of the agent memory and messages (both in bytes) is crucial.
During the calibration process, initial simulations were run using a desktop PC that had quad-core processor, 8GB RAM and used the Windows operating system. This meant that three separate simulation runs could comfortably be performed simultaneously in distributed mode across three out of the four cores in the processor. We found the performance of FLAME to be very promising during initial evaluation when running a single simulation on a single compute node across multiple hardware platforms. Both CPU time and Wall-Clock time were found to be linear following increases in the number of agents within a simulation, and also linear when the simulation length increased up to 50,000 time-steps (the maximum needed in our simulation), which both indicate that FLAME scales very well. However, further investigation identified that the rate-limiting processes and tasks associated with simulation runs was the Input/Output ([I/O], e.g., Read and Write) speed for writing and reading the large number of XML files that are created as part of the simulation output. The cause for this relates to the underlying design principles of FLAME, with its conceptual architecture being based on communicating stream X-Machines, which require the generation of individual XML files for each simulation time-step (these XML files contain parameter values and states for every X-Machine agent that is instantiated). Due to the output XML file for a specific simulation time-step, becoming the input XML file for the next simulation time-step, a performance bottleneck is encountered with respect to the speed of reading from and writing to, the storage disk.
For example, our ABM contained 159 separate agents that corresponded to the individual team members in the case study, and resulted in each XML output file being approximately 50kb in size. A 50,000 iteration simulation run, would therefore generate 50,000 individual simulation time-step output files, that contain the parameter values and states associated with each of the 159 agents at the respective simulation time-step, which in our case approximated to 2.5Gb of output data per simulation run. Furthermore, following aleatory uncertainty analysis, we discovered that 75 replicates of each simulation run (using different PRNG seed values) are required to stabilize the median averages of system dynamics  , which results in approximately 190Gb of simulation output being generated for each set of simulation runs. As such, complex ABMs that incorporate a large number of agents, can quickly generate significant volumes of simulation output data, which could significantly impact the ability to perform simulation-based experimentation if large capacity storage is not available on the computing hardware.
Our relatively simple ABM took approximately 25min to run on the Windows PC, along with an extra 10min to process and transform the XML simulation output files (using Python scripts) to CSV files, and then subsequently transforming these into a single CSV file that contained the median average dynamics of each agent over the duration of the simulation. Through using a pipeline of scripts in this way, we were able to ensure that the large volume of output data spent the minimum of time on the storage facility.
To confirm I/O rate-limiting characteristics of FLAME, we performed an identical set of simulation runs using the same Windows PC, but for this experiment, the output files were written to an external USB hard drive. We found the overall Wall-Clock time increased due to the flow of data across the USB port being slower than that to the onboard hard disk. We built upon these initial findings through using another desktop setup, which this time encompassed an Apple Mac Mini with 8GB RAM and Solid State Disk (SSD), along with a High-Performance Computing facility run by the Northern Eight Universities Consortium (N8) (based within the North of England). The Mac Mini with SSD was 28% faster (with respect to Wall-Clock time) than the Windows PC with internal disk, and 38% faster than that setup when using an external USB hard drive, resulting in the overall Wall-Clock time of a 50,000 time-step simulation run being reduced down to 18min. Likewise, the N8 HPC was 85% faster than the setup using Mac Mini with SSD, bringing the Wall-Clock time for simulation runs down to just under 3min (see Fig. 6 ). Access to fast storage (of suitable large capacity) is therefore crucial for any moderately complicated ABM that requires replicate simulations to be run in order to account for the aleatory uncertainty within the model.
The performance of FLAME was found to be linear with respect to the number of iterations within a simulation run and the wall clock time. FLAME was found to be Input-Output rate-limiting by running a simulation on various hardware setups. For instance, we ran the simulation on: a Windows PC with internal HDD and on an external HDD, a Mac Mini with SSD, and a high-performance cluster with dedicated fast lustre storage.
An additional observation on performance relates to the communication between agents over parallel architectures, and the resultant sequence of messages being sent and read. Such communication dependencies between agents requires a synchronization block between the cores (on a laptop/desktop) or nodes (on a HPC) to ensure that messages arrive in time to meet the dependency, i.e., functions being performed in the specific time-step of the simulation do not try to read a message before the node/core has received it from another node/core. These synchronization blocks are a considerable time bottle neck, which introduce performance losses into simulations. As such, the fewer synchronization blocks introduced into a simulation, the greater the performance of the overall simulation run.
Real-world social systems are complex, with sets of behaviours, characteristics and dynamics that emerge through the individual relationships that function through time and space. One of the major benefits of the agent-based modelling paradigm, is the emphasis on three principal characteristics of complex systems: 1) system structures, 2) system dynamics, and 3) system control  . One of the advantages of the agent-based approach is that the simulations aim to replicate the dynamics of the real-world system, in order to ensure the validity of the underlying assumptions behind the computational model can be tested. However, in order for these ABMs to be successful in performing their role as scientific instruments that act as credible abstractions of complex social systems, it is crucial that the preferred modelling and simulation framework is able: to realistically represent system structure and dynamics; is modular, so that the ABM can be incrementally updated with new functionality, without the need to re-engineer the entire model; can expand with reference to the hierarchical-scale of the real-world system, e.g., individual team members, to discrete project teams, to the programme-wide network as a whole; and is amenable to a thorough validation and verification process, including stringent statistical analysis of simulation output data.
Within this study, we used FLAME to develop an ABM that was calibrated to the quantitative and qualitative data from our case study around the development and propagation of conflict within multi-partner enterprise system implementations. As discussed above, the FLAME modelling and simulation framework utilizes the conceptual and technical architectures associated with communicating stream X-Machines to facilitate agent-based models in a discrete-event manner (the simulation time-steps are the discrete events). It has been reported to provide very significant improvements in performance over more traditional ABM frameworks  . FLAME’s overarching purpose is to deal with massive simulations, allowing for modelling abstraction levels that cater for large scope with respect to the real-world system of interest, and hundreds of thousands, to millions, of X-Machine agents. Through being designed and developed to comply with the MPI communication framework, FLAME code is also deployable on to parallel hardware platforms.
Our study has identified a number of strengths and weaknesses for using FLAME to model complex social systems. Firstly, with respect to the strengths, and for modellers who are familiar with the object-oriented approach to design of computational models, we have identified that FLAME is an intuitive framework for designing ABMs of complex social systems due to its template-driven approach. Similarly, for modellers who are comfortable with command line terminal, basic programming languages, and shell scripts, we have identified that FLAME is an intuitive framework for developing ABMs of complex social systems. Conversely, we have shown that FLAME suffers with a number of constraints that introduced technical issues into the development process for our ABM. The four major constraints were: the lack of built-in functionality to set the PRNG seed value at the simulation level; the inability to utilize functionality associated with global mutable parameters; the inability to communicate between simulation time-steps/iterations; and the lack of instrumentation to analyze simulation output data, thus requiring the development of various statistical and data transformation scripts, which is fine for experienced modellers, but is a real constraint for novice modellers or subject matter experts.
These constraints are unfortunately a direct consequence of the design decisions taken during the initial creation of FLAME, because the underlying premise was that it would harness the power of parallel processing to run individual simulations across HPC architectures. A number of technical challenges were introduced by these constraints, but fortunately, we were able to develop workarounds to resolve them. As such, we want to inform future users of the FLAME simulation framework that they need to be cognizant of these constraints, especially due to the fact that it has been proposed as suitable for subject matter experts within the real-world domain of interest, who may have limited experience and/or competence in computational modelling or programming. With this in mind, we advocate that this message be moderated, because subject matter experts may not have the technical competence or modelling experience to fully investigate, analyze, diagnose and resolve the full variety of problems and constraints that we have experienced during this study.
Furthermore, and of more significance is that we discovered a major limitation when using FLAME, is that models that incorporate large numbers of individual agents are computationally expensive with respect to Wall-Clock time, the I/O load, the need for HPC architecture to run multiple replications in distributed mode, and the consequent very large size of output simulation data. This computational expense is compounded when the ABM is increased in scale to reproduce results consistent with system-level dynamics from the real-world domain. Due to FLAME’s communicating stream X-Machine architecture, which relies on input and output streams of messages to facilitate communication, we unexpectedly discovered that comparatively small simulations (with respect to total X-Machine agent numbers, the complexity of interaction logic, and the total time-steps within the simulation) could generate a considerable number of XML data files, which directly correlates to the volume of simulation output data produced. The Wall-Clock time for single simulation runs was found during diagnostic tests, to grow linearly in accordance with the number of time-steps within simulation runs. Diagnostic tests also indicated the Input-Output rate-limited nature of FLAME (as opposed to being rate-limited through the Central Processing Unit), which is due to separate XML output files being generated for each time-step within simulation runs.
To put this into further context, the communicating stream X-Machine nature of FLAME requires that individual XML output files are generated at the end of each time-step to record the parameter values for internal computational and simulated real-world states along with internal memory values for each X-Machine agent; these are then used as the XML input file at the beginning of the next time-step to set the corresponding starting parameter values, states, and memory values for the next time-step of the simulation run. A bottleneck is therefore experienced regarding computational performance, which results from the speed of reading from and writing to, the onboard or peripheral storage disk. The results from aleatory uncertainty analysis further compounded this performance issue by identifying that a minimum of 75 simulation replicates were required to achieve stable median dynamics, which resulted in approximately 3,750,000 total XML output files, which as disucssed, amounts to 190GB for our relatively small ABM.
There can often exist a delicate balance with respect to computational efficiency and performance during aleatory uncertainty analysis, which is introduced from the desire to achieve stable simulation results by using a high number of replicates for calculating the median average simulation results, whilst being cognizant of the computational resources required (e.g., Wall-Clock time, number of CPU cores on desktop or nodes on HPC, and access to fast storage disk). The number of simulation replicates chosen to calculate the median average dynamics is therefore usually a compromise between minimizing the impact of aleatory uncertainty versus the acceptable costs in computational resources for the project. Our findings therefore indicate that average, everyday desktop computing resources (such as a Windows PC or Apple Mac Mini) are unsuitable for running simulations that require large numbers of replicates to achieve stable average dynamics to act as a baseline for simulation-based experimentation. We therefore conjecture that access to very fast multi-core desktops with significant storage capacity (such as fast SSD raid capabilities), or to HPC architecture is critical for ensuring that our ABM of conflict development and propagation within a complex social network can be used for simulation-based experimentation.
The above discussion has focused on our contributions to the computational modelling community through out user experiences when using FLAME to model a complex dynamical social system. Specifically, the discussion has focused on the strengths and weaknesses that we identified when using FLAME to model the development and propagation of conflict within the social network of large multi-partner enterprise system implementations. We found the design of our model to be straightforward due to FLAME’s underlying architecture using communicating stream X-Machines. The object-oriented nature of communicating stream X-Machines, which are in effect, computational instantiations of UML state transition diagrams, makes the design and subsequent development relatively intuitive. In addition, the use of XML to define agents and C to code the agent communication and interaction rules, allows for a truly modular approach to ABM development. Conversely, however, the limitations of FLAME are also due to the underlying conceptual and technical architecture of FLAME. We discovered four main constraints relating to the lack of in-built functionality to set PRNG seed values, the inability to use global mutable parameters, the inability to communicate between time-steps, and the lack of instrumentation to analyze simulation output data. In addition, we identified a significant limitation with respect to performance, which is a direct consequence of the communicating stream X-Machine architecture, in that the output file for a simulation time-step acts as the input file for the subsequent time-step. Our analysis indicated that this not only leads FLAME to be I/O rate-limiting, but also means that large simulations (with respect to either total number of agents or total number of simulation time-steps) quickly generate prohibitively large amounts of simulation output data. We were lucky that our ABM of the case study is relatively simple and had only 159 agents within the social network, but even still, following aleatory uncertainty analysis we were required to run 75 replicates of simulations in order to mitigate the effects of aleatory uncertainty, and produced 190GB. As such, we conjecture that for large, complex, ABMs, the large volumes of simulation output data generated by FLAME may mean that it is rejected as a potential simulation framework if access to very large capacity and high-speed storage capabilities cannot be achieved.
To conclude, we believe FLAME is an excellent choice for experienced modellers, who will be able to fully harness the capabilities that it has to offer, and also be competent in diagnosing and solving any limitations that are encountered. Conversely, because FLAME requires considerable development of instrumentation tools, along with development of statistical analysis scripts, we believe that it is not suitable for the novice modeller, who may be better suited to using a graphical user interface driven framework until their experience with modelling and competence in programming increases. In our opinion, FLAME’s major strength is its flexibility, in that once the model definition (along with any technical workarounds) has been developed using the XML and C templates (e.g., the social network topology, communication rules, rules for conflict development, and functionality to explicitly set the PRNG seed value), the augmentation of the ABM with new functionality is comparatively easy due to the template-driven nature of FLAME facilitating its modular approach to design and development. In addition, we have found FLAME to be excellent for quickly developing ABMs of complex dynamical systems in both our case study presented here, and indeed in previous work into complex dynamical biological systems  . We do however believe that it requires significant computational modelling skills in order to develop workarounds to some of the constraints that have been imposed due to the underlying technical architecture, and to develop a pipeline of scripts (e.g., Ruby, Unix Shell, MS DOS, and Python) to semi-automate the submission of jobs to a HPC and to transform and analyze simulation output data to reduce the storage load when large numbers of replicate simulations are required to account for the aleatory uncertainty within the model. Finally, we are aware that a version of FLAME has been developed that harnesses the parallel processing power of Graphics Processing Units (GPU)  , which has recently been extended  ), and could prove beneficial, because a number of the limitations that we discovered (e.g., no global mutable parameter and the I/O resource intensiveness) may be resolved through the difference in processing architecture. One final point to highlight on this however, is that FLAME GPU, being based on CUDA, requires NVIDIA graphics cards, which could prove problematic for those developers who are using Apple Mac computers because of the current issues with drivers and Mac Mojave/Catalina operability, and the need to use eGPU and various scripts to get the thunderbolt connectivity working with older Macs - we are aware that newer Macs have Thunderbolt 3 connectivity and that Mac OS High Sierra supports NVIDIA drivers. To this end, we are in the process of trialling FLAME GPU on such a setup.
This work was in part funded by a Lancaster University Early Career Small Grant and a Management and Business Development Fellowship awarded jointly by the Society for the Advancement of Management Studies, the United Kingdom Commission for Employment and Skills, and the Economic and Social Research Council (SAMS-UKCES-ESRC) with Grant No. ES/L002612/1. The funders had no involvement in study design; development of the underlying ABM, its analysis and interpretation of results; in the writing of this manuscript, or the decision to submit this manuscript for publication.
- INVITE FLAME TO YOUR SCHOOL
- VISIT FLAME - BOOK A CAMPUS TOUR
- DOWNLOAD BROCHURE
- REQUEST INFO
- VOICE OF FLAME
Students are integral to research at FLAME University. Students at all levels are encouraged to participate in research and appropriate funding is made available. University Research Committee facilitates creation and development of strategies and options that students might adopt in order to address their knowledge and skill gaps and build capacity for more effective research and innovation. In its efforts to support enhancement of knowledge and intellectual growth of its students, University supports various projects that the students take up under the guidance of their faculty. In keeping with the underlying ethos of liberal education, research at FLAME University is essentially inter-disciplinary with students led by faculty across disciplines, working together on numerous subjects of mutual interest, appeal and requirement.
Applied Math Student Research Projects
- Managing the Zambezi river: The Kariba Dam, Alyssa O’Donohue, Calvin Gentry, Surbhi Lodha, Tirtha Patel, Mathematical Modelling course, April 2018
- Eradicating Ebola, Basudhara Choudhuri, Louis Kopp, Zheng Qu, Alec Nelson, Mathematical Modelling course, April 2018
- Garbage Patches in Subtropical Gyres: Tracking Particles from Source to Sink, Harshada Balasubramanian, Raghav Chadha, Salil Shinde, Mathematical Modelling course, 2019
- Is it Sustainable? A Mathematical Model, Kyra Manjunath, Aditi Pasumarthy, Mathematical Modelling course, 2019
Environmental Studies Student Research Projects
- Tirtha Patel (Class of 2018)- Where Tourism meets Conservation: Translating Emotions into Actions https://www.currentconservation.org/issues/where-tourism-meets-conservation-translating-emotions-into-actions/
- Smriti Jalihal (Class of 2020)- An annotated bibliography and summary of sea turtle satellite telemetry studies in the Indian Ocean and Southeast Asia https://www.iotn.org/iotn29-06-an-annotated-bibliography-and-summary-of-satellite-telemetry-studies-conducted-throughout-the-indian-ocean-and-south-east-asia/
- Sanjana Singh. (Class of 2015). Nipah Virus: Effects of Urbanization and Climate Change. In Md. A. Rahman & R. Ahmadi (Eds.), International Institute of Chemical, Biological and Environmental Engineering. Paper presented at 3rd International Conference on Biological, Chemical and Environmental Sciences (BCES 2015), Kuala Lumpur, Malaysia, 21-22 Sep, 2015 (pp. 64-68). IICBEE. Adjudged the best paper of the session. https://iicbe.org/upload/7575C0915051.pdf
International Relations Student Research Projects
- 2019 – Published research paper titled “Need for Space Law in India: Analyzing the Space Activities Bill (2017)” with Journal of Science Policy and Governance
- 2019 – Published research paper titled “Radical fundamentalism impeding successful policy intervention: Polio a case in point” with IIS Journal of Social Sciences
- 2019 – Authored an article titled “Beef Down: Implications of Beef Ban on Indian Communities” and published it with International Policy Digest
- 2019 – Co-authored an article titled “Never Again War? Tracing German Military Identity” and published it with International Policy Digest
- 2018 – Authored a research paper titled “Radical fundamentalism impeding successful policy intervention: Polio a case in point” and presented it at The Fourth Annual Seminar on Research in Social Sciences held in Mumbai, India
Current Papers in progress with regards to IR (checking if these are done)
- Developing geostrategy for Sri Lanka (sent as a chapter for an edited book)
- Costly signalling and the US-Iran War
Psychology Student Research Projects
Gender Differences in Personality, Motivation and the Tendency to Follow Social Networking Trends among Indian Millennials
Aishwarya Sastry B.A. (Psychology) FLAME University
Dr. Sairaj Patki Assistant Professor (Psychology), FLAME University
ABSTRACT With the availability of affordable and high-speed internet access on phones, social networking has reached far and wide. While some of the trends that emerge via these platforms have helped steer public movements and supported social causes, some lead to severe detrimental consequences like harm to psychological or physical wellbeing. The present study investigated personality traits and sources of motivation as factors related to the tendency to follow social networking trends on Facebook, Instagram and Snapchat among millennials. The sample included 133 respondents aged between 18-25 years. The findings suggested existence of some gender differences. While tendency to follow social networking trends was found to be related only with extraversion among females, it was found to be related to both extraversion and agreeableness among males. Also, tendency to follow social networking trends was found to be related with intrinsic motivation among females but with extrinsic motivation among males.
Economics Student Research Projects
The Role of Emotions and Mindfulness in Investment Decision Making Medha Arora- 2016-17
Abstract Traditional economic theories have always assumed that human beings make rational choices. However, reality may dictate otherwise. There are many factors that influence how one makes decisions, especially in the context of risk and uncertainty. Emotions are one such factor, and have proven to strongly affect risk attitudes. Mindfulness has also been associated with decision making and emotional regulation, although data for this finding is limited. Hence, this study aimed to investigate the role of emotions and mindfulness in investment decision making through the experimental method. Participants were randomly allocated into four emotional manipulation groups - positive affect, negative affect, neutral affect, and mindfulness. This was followed by an investment task, the results of which were analysed and compared based on the treatment condition. It was hypothesised that diverse emotional states would affect investment decisions differently, and that the mindfulness state would lead to the most rational investment behaviour. Results found that while investment choices between the four groups were not significantly different, higher overall positive affect significantly predicted lesser risk aversion compared to a negative state. The role of mindfulness as a potential anxiety buffer was also evident in a high global risk situation. Time and gender were also identified as important features impacting decision making.
Farmer Suicides in India- A Macro Perspective Navneeth JK-2017-18
Abstract Agriculture is the soul of the Indian economy. Although the contribution of agriculture towards the GDP has been consistently declining, this sector still employs half of the India’s workforce. However, the agricultural sector in India is heavily distressed as evident from the large number of farmer suicides that have plagued the country over the last decade. So, what are the factors that lead farmers to commit this act? Is it the type of crops that they grow? Is it the high cost of inputs? Or is it a lack of infrastructure for farmers to sell their produce at a fair price? This paper aims to answer these questions. This paper uses a state level panel dataset from 2001-2013 to investigate the factors that affect farmer suicide in India. Preliminary results indicate irrigation and credit ratio negatively impacting farmer suicide while sown area affects farmer suicide positively. Moreover, we also find that consumer price positively affects farmer suicide in India. The results thus call for creating a stronger agricultural market in India.
Learn more about FLAME
- Invite Us To Your School
- Download Brochure
- Virtual Tour
Subscribe to Our Quarterly Newsletter
Read what our clients have to say about our writing essay services!
Research papers can be complex, so best to give our essay writing service a bit more time on this one. Luckily, a longer paper means you get a bigger discount!