Chapter 31
DYNAMOD—An Agent Based Modeling Framework: Applications to Online Social Networks Aneesh Zutshi, Antonio Grilo and Ricardo Jardim-Gonçalves
Abstract This paper presents an Agent Based Modeling Framework that seeks to provide a generic model that can be used to simulate any internet based business. The model captures the unique characteristics that define how online users interact, share information, and take product adoption decisions. This model can be used to simulate business performance, make business forecasts, and test business strategies. To demonstrate the applicability of the model, we choose two social networks with opposing scales: Facebook and a startup educational social network—weduc.pt. Keywords Digital business models · Agent based modeling · Business forecasting · Business simulation
31.1 Introduction Digital Businesses are the most dynamic of industry segments, and have expanded dramatically over the last decade. Digital businesses encompass the entire gamut of ventures, from online sale of products and services to social collaboration platforms, and have become a vital engine for the new economy. According to a recent McKinsey Report, the Internet economy contributed to 21 % of the overall GDP growth in the past 5 years in mature economies and now accounts for 3.4 % of the GDP [19]. However, understanding this new economy has been a challenge for companies that were tuned to the traditional ways of doing business and were armed with traditional A. Zutshi (B) · A. Grilo UNIDEMI, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, 2825-344 Monte de Caparica, Portugal e-mail:
[email protected] R. Jardim-Gonçalves UNINOVA, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, 2825-344 Monte de Caparica, Portugal
J. Xu et al. (eds.), Proceedings of the Eighth International Conference on Management Science and Engineering Management, Advances in Intelligent Systems and Computing 280, DOI: 10.1007/978-3-642-55182-6_31, © Springer-Verlag Berlin Heidelberg 2014
349
350
A. Zutshi et al.
product development and marketing philosophies. Most companies that rushed into the dot-com bubble in the early part of the last decade perished [15]. However, we also saw the emergence of new business giants such as Google and Amazon, who did the right things at the right time to emerge as winners. There has been limited academic effort trying to understand what makes a digital business a success or a failure. Business Model (BM) has been an active area of research over the last decade, and provides us with a structured approach to represent and comprehend the value delivering mechanism of a business. Walter and Back [28] define BM in the following way: “A business model describes ways of creating value for customers and the way business turns market opportunities into profit through sets of actors, activities and collaborations.” A Business Model is a set of coordinated strategies that ultimately aim to increase the short-term and long-term profitability of an enterprise. In this paper, we first make a review of the characteristics that make a Digital Business different from traditional businesses. Through a Business Model approach we develop a generic framework that can be used to represent a Digital Business. This framework, a Dynamic Agent Based Modeling Framework (DYNAMOD), further incorporates Agent Based Modeling techniques to model real life business scenarios. DYNAMOD provides a framework to assist business managers develop and test their models in a variety of market scenarios. DYNAMOD will enable the simulation of current market conditions, and enable the visualization of market adoption and growth of a particular business offering, thus enabling the development of a sound Business Model with a much greater chance of success. Finally, we demonstrate some of the capabilities of this model through its application to two social networking sitesFacebook and a startup educational social network—weduc.pt.
31.2 Digital Business Characteristics Digital businesses refer to the provisioning of products or services through the Internet. They have been the most dynamic entrepreneurial area, requiring agile and innovative business models [8]. Digital businesses have flourished and failed at a scale and pace never seen before by any other traditional business sector [26]. Despite numerous failures, the phenomenal success of new ventures like Google and Facebook have shown that disruptive business models have the potential to exponentially gain market share and expand virally. Here we discuss some of the key characteristics that form the defining features of digital businesses.
31.2.1 Free and Freemium Models Due to Low Cost per User Digital businesses have a unique feature. While costs of infrastructure and servicing of clients increases over time for traditional businesses, thesecosts have been falling rapidly in the case of digital businesses. The key infrastructure
31 DYNAMOD—An Agent Based Modeling Framework
351
costs—Storage, Processing and Bandwidth, have continued to follow Moore’s Law and have steadily plummeted [2]. As companies have embraced economies of scale, the servicing cost per client have been tending to zero [27]. This has led to experimentation with new business models that reflect these low incremental per-user costs. Very low per-user costs have led to the growth of Free and Freemium business models, where the end user receives tangible products or services without a cost. While Free services are mostly supported by advertising, in case of Freemium, a small percentage of people pay for premium services, or generate revenue through indirect sources [27].
31.2.2 Viral Marketing Viral Marketing refers to the fast and infectious spread of a product adoption across the market. It involves strategies for rapid uptake of electronic peer to peer referrals [5]. It also involves an effective utilization of word of mouth or established networks of clients, through an excellent value proposition that is low cost or free. In electronic business, automated viral marketing strategies are sometimes used to induce viral marketing with little intervention from the users, or sometimes even without the knowledge of users. A commonly cited example is Hotmail, which became very popular due to automatically adding its web-link at the end of each email sent by a Hotmail user [21]. However excessive viral marketing might create a perception of the product as spam and lead to negative opinions about the product in the minds of potential adopters [18]. Hence, although it is an effective tool, viral marketing must be carefully exploited.
31.2.3 Intense Competition Another feature of digital business is intense competition. Since entry barriers to markets tend to be low and national boundaries are not significant hurdles, the entry of new competitors takes place at a phenomenal speed [20]. In traditional businesses, a large number of competitors would have meant a highly fragmented market share. A unique feature of digital business is that often a dominant player assumes an extremely large proportion of the market share. This is because of the low cost structure (often zero or near zero) of the digital business, where pricing cannot be used as a differentiating parameter. The network effects associated with well connected users mean that a well designed product, usually the first mover, would virally expand due to positive customer feedback [10]. Marginal deficiencies in the product or services offered can lead to customer dissatisfactions that can propagate through the network of the potential user base, with a considerable effect on future
352
A. Zutshi et al.
product adoption. This is especially true since forums and product reviews are easily accessible to customers today.
31.2.4 Product and Marketing Integration Digital markets often do not provide room for drawing board development of business strategies that can be slowly tested in the market and tweaked as time goes on. For effective marketing in the digital world, marketing strategies should not be separate from product, but rather marketing should be built into the product itself [16]. Market opportunities must be quickly exploited, before a competitor catches up. This integrated approach to the digital business strategy, which incorporates product development, marketing and pricing forms the underlying principle of a digital business model.
31.2.5 Network Effects Network effects exist when consumers derive utility from a product based on the number of other users [14]. These effects are especially relevant for several online businesses, especially since various online products and services exhibit some form of network effects, such as social networking sites and online marketplaces. Goldenberg et al. [14] also predicts a chilling effect of network externalities. They propose that a product with a network externality has a slower initial adoption compared to a product that does not have any network externalities. The higher growth rate due to the network effects occurs only after the product has crossed a certain adoption critical threshold.
31.2.6 Pricing of Information Goods/Services Pricing of digital products often involves splitting the product into different sub-categories and re-bundling them. The unit of charge for digital products must change. Smaller units of charge, focusing on pay per use or per month subscription charges have met with success in the digital world [8].
31.3 Agent Based Modeling (ABM) Approaches to Simulate Business Environments Traditional management guidelines are ill-equipped to help understand the digital business models and predict success and failure. New tools and techniques are necessary to help model the complex nature of online products and services. Hence we
31 DYNAMOD—An Agent Based Modeling Framework
353
have developed a generic framework for a customizable simulation environment that can capture the dynamics of an online market, and provide Business Managers with tools to simulate and forecast, thus acting as a tool to perfect their Business Model. Online markets can be represented as a network of interconnected online users which share positive and negative feedbacks and respond to different online products and services. If the behavior of individual agents can be sufficiently well modeled, then a natural candidate for representation is Multi-Agent Based Modeling Techniques. ABM is build on proven, very successful techniques such as discrete event simulation and object oriented programming [22]. Discrete-event simulation provides a mechanism for coordinating the interactions of individual components or “agents” within a simulation. Object-oriented programming provides well-tested frameworks for organizing agents based on their behaviours. Simulation enables converting detailed process experience into knowledge about complete systems. ABM enables agents who represent actors, or objects, or processes in a system to behave based on the rules of interaction with the modelled system as defined based on detailed process experience. Advances in computer technology and modelling techniques make simulation of millions of such agents possible, which can be analysed to make analytical conclusions. Tesfatsion introduced Agent-Based Computational Economics as the computational study of dynamic economic systems modeled as virtual worlds of interacting agents. Aliprantis et al. [1] have applied ACE to retail and wholesale energy tradings in the Power Markets. In this paper we extend the concept of Agent-Based Computational Economics, to develop DYNAMOD-An Agent Based Modeling Framework for online Digital Business Models.
31.4 Theoretical Basis for the DYNAMOD Model The development of the DYNAMOD model, is based upon other previous research works in diverse areas. Some key concepts that have been applied in the model are discussed below.
31.4.1 Diffusion of Innovations Diffusion of Innovations has been an active research area and reflects adoption decisions made by individual consumers. These decisions are made in a complex, adaptive system and result from the interactions among an individual’s personal characteristics, perceived characteristics of the innovation, and social influence [23]. There are two major approaches to modeling diffusion: econometric and explanatory. The concept of Econometric Modeling was first introduced by Bass [3]. Econometric approaches forecast market uptake by modeling the timing of first-purchases of the innovation by consumers and are more applicable when market growth rate and market size are of primary interest. Explanatory approaches, as first proposed by Gatignon [11] establish that the diffusion of a product in a defined market is
354
A. Zutshi et al.
equivalent to the aggregation of individual consumer adoption decisions. The adoption decisions are dependent on: Personal characteristics, perceived product characteristics, and social influence. In earlier works, diffusion of innovation has been approached with mathematical modeling [12–14]. However as computational powers increased, relatively recent attempts have been made to complement these classic approaches with Agent Based Modeling tools [6, 7, 24]. We have used these works as the basis for developing the DYNAMOD model with the application of specific characteristics to differentiate online businesses.
31.4.2 Word of Mouth Literature has assumed word of mouth (WOM) to be the influence of neighbors over an individual [9]. This is a relevant assumption for offline word of mouth since such communication is mostly limited by geographical location. Word of mouth communication is more effective when the transmitter and recipient of information share a relationship based on homophily (tendency to associate with similar persons), trust and credibility. Brown et al. [4] conducted research on online word of mouth and report that online homophily is almost entirely independent of interpersonal factors, such as an evaluation of individual age and socio-economic class, traditionally associated with homophily. The idea of individual-to-individual social ties is less important in an online environment than in an offline one. Individuals tend to use websites as proxies for individuals. Thus, tie strength was developed between an information seeker and an information source as offline theory suggests, but the information “source” is a Web site, not an individual.
31.4.3 Network Structure Cellular Automata, is a form of lattice network and has been used by numerous authors to mathematically model word of mouth. They represent users as cells in a cellular grid like network, with each a cell getting influenced by static neighbouring cells surrounding them. Goldenberg et al. [12] used cellular automata and introduce the concept of strong ties and weak ties while discussing word of mouth. However the static nature of the network makes it unsuitable to represent the dynamic nature of online user networks. Another form of a network is a random network where the cell distance is randomly distributed. Another common network methodology is the small-world network which starts with a random network randomly rewiring some of the edges [25]. In the case of DYNAMOD, we shall be using a dynamic random network where user agents start being randomly distributed over a flat world, and get influenced by agents in a fixed radius. However the agents themselves slowly make random walks,
31 DYNAMOD—An Agent Based Modeling Framework
355
Fig. 31.1 Conceptual representation of the DYNAMOD framework
Fig. 31.2 DYNAMOD component architecture
Pricing Analysis
Competitor Analysis
Network Effects / Viral Marketing Effects
Model Core Multi-Agent
Market Segmentation / Regionbased modeling
based
model
and thus the agents within their sphere of influence keep changing. This represents an online world where users constantly meet the opinion of new users through online posts and forums.
31.5 The DYNAMOD Framework The DYNAMOD Framework has been developed based on the academic literature collected. Its purpose is to provide researchers and companies engaged in online businesses with a tool for quickly developing Computational Modeling Systems that can represent their Business Models and their Business Environment, in order to perform advanced simulations for predicting business growth dynamics. DYNAMOD is based on Agent Based Modeling, which enables dynamic representation of the online marketplace. Every online user that could be a potential customer for a product or service is represented as an Agent in DYNAMOD (see Fig. 31.1). These agents interact with each other and share information about new products and services. At the same time, they are influenced by Advertising and Social sites. The model captures these influences, and simulates their impacts in order to predict future scenarios. The model is customizable and extendible to implement a diverse set of Business Model components, and to make a variety of simulations. Figure 31.2 shows a conceptual relationship of the various components the DYNAMOD Framework. The model core consists of many interacting agents that represent a market. The model
356
A. Zutshi et al.
includes standard variables and logics for implementing influence and satisfaction scores for each agent. This core component handles the simulation and interaction, and defines what constants are needed to initialize the key features of the model. Other features are added to the model in the form of modules, as and when necessary, for different case scenarios. In the current scope of the model, four additional modules have been envisaged, namely Competitor Analysis, Pricing Analysis, Network Effects/Viral Marketing Effects, and Market Segmentation/Region Based Modeling, but additional components can always be added for modeling other scenarios. Competitor Analysis involves introduction of competitors who can have competing influences on consumers, and then monitoring the switching behavior of consumers. Pricing Analysis involves introduction of various charging units, and their impacts on consumer adoption. It also involves the introduction of Freemium Business Models into the model, and simulates the adoption of Free and Paid components of the Businesses. This module is not needed in case of Free Business Models. Businesses that have an inherent Network Effect or are based on Viral Marketing need to add additional logics that change the rate of product adoption. The Market Based Segmentation or Region Based Modeling changes the dispersion of agents in the model space, to represent different clusters of agents. This can represent different classes of customers with varying purchasing powers, or can represent customers on different continents.
31.6 Case Studies of 2 Contrasting Social Networks One of the most powerful forms of digital business is a Social Network. It harnesses the power of the web to easily connect communities and promote social interactions. The largest social network site in the world is Facebook in terms of number of active users. However several niche social networking based digital businesses have been launched ranging from product reviews to forums and blogs. We have selected two case studies: Facebook and an online Educational startupWeDuc. While both these cases are examples of Online Social Networks they are two contrasting examples. While Facebook is a highly mature company and due to the positive network effect due to its large number of users, would find it difficult to be replaced by competitors. In contrast WeDuc is a startup and is still trying to establish its Business Model.
31.6.1 Facebook Forecast Using DYNAMOD Model The Model was initialized based on active monthly users historical data from June 2009 untill June 2011, and based on questionnaires collected from 112 respondents out of which 102 were Facebook users. The questionaires gauged their satisfaction
31 DYNAMOD—An Agent Based Modeling Framework
357
Fig. 31.3 DYNAMOD forecasts for Facebook
levels and measures the degree of word of mouth and were used to initialize the model. After initialization, the model was used for making forecasts of future user adoption until June 2015. The actual data after June 2011 are used only to verify the predictions of the model. To test the accuracy of forecast by the DYNAMOD model, we also used another forecasting tool-ARIMA (Fig. 31.3). • Total number of customers at any time Ct according to the Dynamod model = Aiclient = 1 . • Number of customers according to data at any month t = C A t. • Number of customers at any month t according to the ARIMA Forecast = Rt . The Root Mean Square Error for the initialization data was: Jun 11
Jun 09 (C t
− C A t)2
Count − Records (Jan 09–Jul 11)
= 3.18.
The actual growth of Facebook active users slows down over the validation period. Hence the regression line moves away from the actual growth line. However the DYNAMOD model is able to predict the slowdown and closely follows the actual growth line. For the Validation Period between Sep 2011 and Sep 2013, RMS Error during Validation phase for DYNAMOD prediction was:
Sep 13 Sep 11 (C t
− C A t)2
Count − Records (Sep 11–Sep 13)
= 3.91.
For the same Validation phase, the ARIMA Error was: Sep 13
Sep 11 (Rt
− C A t)2
Count − Records (Sep 11–Sep 13)
= 19.08.
358
A. Zutshi et al.
This validates that the DYNAMOD model forecast is closer to real data than the Regression Model. This demonstrates that DYNAMOD can be leveraged for business growth simulations, in this case, a very mature business which has already reached a majority of global users. While the main aim of the Model is not to be used as a pure forecasting tool, a good forecast ensures that the model is efficiently configured and ready for making further analysis.
31.6.2 Case Study of an Educational Social Network Startup-WeDuc Weduc.pt is a Portuguese startup educational social network with the aim of providing a secure environment for Parents, Teachers and Students especially aimed at Primary and Secondary Schools. The site allows for a user-friendly interface where parents can keep track of the progress of their children at schools. Teachers can communicate individually with students and their parents, post homework, evaluations and also conduct online tests. Finally the students can interact with their peers, have discussions on topics, and experience social networking under the supervision of Teachers and Parents to make sure that they are not exposed to any unsuitable content. Despite being a social network, it does not exhibit a network effect because in effect it consists of individual islands of social networks for each school. The key decision for adoption is taken by the school administrators. The website started conducting its first trials in September 2010 and when this research was conducted in July 2013, had registered 300,000 users across 2,000 schools. However the existing Freemium business model was only established in 2013.
31.6.2.1 Business Model Currently weduc.pt is using a freemium business model where basic subscription is offered free to schools and while paid subscriptions cost 5 Euro per child per month. The paid option includes access to high resolution images, videos, file uploads, customised school theme, telephonic support and training of teachers. The school management takes the decision to either go for the free subscription or the paid one. Consequently, most of the schools opting for the paid price plan are the private schools. As on the date of the study, out of the 2,000 schools registered in Portugal, only 40 schools had opted for the paid tariff plan. Ninety nine percent of the total revenues are generated through subscriptions while only 1 % is through online advertisements. While most subscriptions are the result of direct marketing to school administrators, other sources of marketing that have been used include online advertisements, radio ads and advertising on Facebook. 20 % of
31 DYNAMOD—An Agent Based Modeling Framework Fig. 31.4 Willingness to pay by parents of children in schools with free tariff plan
€5 per child per annum 13%
€10 per child per annum 11%
359 €1per child per annum 6% €30 per child per annum 3% €2 per child per annum 2%
€20 per child per annum 3% Other 4%
€3 per child per annum 1%
€0.5 per child per annum 1%
Free 60%
overall budget have been spent on Marketing and 70 % of the budget has been spent on Product Development.
31.6.2.2 Implementation of DYNAMOD Model Since weduc.pt did not benefit from such a vast historically recorded user data after the implementation of the existing freemium pricing model, it was difficult to make accurate growth forecasts using DYNAMOD model. Rather the focus of DYNAMOD in such early stage startups is to test the impact of changes in the current Business Model. Model Objective: Currently the school management takes the decision between choosing a free subscription and a paid subscription and pays for all the students in the school. This has led to only 40 out of 2,000 schools taking the paid service. The objective of this simulation was to see the impact if instead of the School Management making the decision, individual parents were able to choose between paid and free subscription for their kids. Methodology: To develop the DYNAMOD model the schools were selected as the main agents within the Agent Based Model. Since the aim of this model was not to simulate the adoption by new schools but rather the analysis of shift in the business model to allowing parents of free schools to upgrade to a paid version. So a macro model was used where schools were the agents with a range of students from 50 to 500 per school. A subsequent mini model was created to represent a typical school with 300 students to evaluate their parents? willingness to pay. The initialization of this model was based on a survey conducted from July 2013 to November 2013 with 288 respondents. Results: Fig. 31.4 shows the response of the parents of children in schools choosing the free version when asked about their willingness to pay for a paid version. While 60 % of the respondents chose to not pay, 11 and 13 % of the respondents chose to pay annually e10 and e5 respectively. 30 % of the respondents chose to pay at least e5. Hence there was a compelling case for monetization of revenues from a sizable percentage of the new users. However the only drawback was any negative impact
360
A. Zutshi et al.
from different children in the same school having different features. This could reduce the satisfaction level of students whose parents don’t subscribe to the paid version thus impacting the overall satisfaction. According to the simulation result, offering this option could enable weduc to increase their revenue share by 23 %.
31.7 Conclusion The DYNAMOD model demonstrates a systematic approach to representing the key characteristics of digital businesses and developing an Agent Based Model that can act as a forecasting and key decision support tool to study the impact of changes in the Business Model. It is a tool that can be used for startups as well as mature entreprises. We are in the process of developing further application scenarios and case studies to further validate the approach and the results. Acknowledgments The authors also would like to thank Fundação da Ciência e Tecnologia for supporting the research center UNIDEMI through the grant PEst-OE/EME/UI0667/2011. Also, the authors are grateful to ISOFIN and VORTALWAY projects (QREN) for funding Aneesh Zutshi’s research work.
References 1. Aliprantis D, Tesfatsion L, Zhao H (2010) An agent-based test bed for the integrated study of retail and wholesale power system operations. Agent Technol Energy Syst ATES 2010:6 2. Anderson C (2009) Free: the future of a radical price. Hyperion Books, New York 3. Bass FM (1969) A new product growth for model consumer durables. Manage Sci 15(5): 215–227 4. Brown J, Broderick AJ, Lee N (2007) Word of mouth communication within online communities: conceptualizing the online social network. J Interact Mark 21(3):2–20 5. Bruyn AD, Lilien GL (2008) A multi-stage model of word-of-mouth influence through viral marketing. Int J Res Mark 25(3):151–163 6. Delre SA, Jager W et al (2007) Targeting and timing promotional activities: an agent-based model for the takeoff of new products. J Bus Res 60(8):826–835 7. Diao J, Zhu K, Gao Y (2011) Agent-based simulation of durables dynamic pricing. Syst Eng Procedia 2:205–212 8. Docters R, Tilstone L et al (2011) Pricing in the digital world. J Bus Strategy 32(4):4–11 9. Feng J, Papatla P (2011) Advertising: stimulant or suppressant of online word of mouth. J Interact Mark 25(2):75–84 10. Gallagher S, West J (2009) Reconceptualizing and expanding the positive feedback network effects model: a case study. J Eng Technol Manage 26(3):131–147 11. Gatignon H (1985) A propositional inventory for new diffusion research. J Consum Res, pp 849– 867 12. Goldenberg J, Libai B, Muller E (2001) Talk of the network: a complex systems look at the underlying process of word-of-mouth. Mark Lett 12(3):211–223 13. Goldenberg J, Libai B et al (2007) The NPV of bad news. Int J Res Mark 24(3):186–200 14. Goldenberg J, Libai B, Muller E (2010) The chilling effects of network externalities. Int J Res Mark 27(1):4–15
31 DYNAMOD—An Agent Based Modeling Framework
361
15. Goodnight GT, Green S (2010) Rhetoric, risk, and markets: the dot-com bubble. Q J Speech 96(2):115–140 16. Grewal D, Janakiraman R et al (2010) Strategic online and offline retail pricing: a review and research agenda. J Interact Mark 24(2):138–154 17. Herr PH, Kardes FR, Kim J (1991) Effects of word-of-mouth and product-attribute information on persuasion: an accessibility-diagnosticity perspective. J Consum Res 17:454–462 18. Kalyanam K, McIntyre S, Masonis JT (2007) Adaptive experimentation in interactive marketing: the case of viral marketing at Plaxo. J Interact Mark 21(3):72–85 19. Manyika J, Roxburgh C (2011) The great transformer: the impact of the internet on economic growth and prosperity. McKinsey Glob Inst, pp 1–10 20. McGrath RG (2010) Business models: a discovery driven approach. Long Range Plan 43(2):247–261 21. Montgomery A (2001) Applying quantitative marketing techniques to the internet. Interfaces (Providence) 31:90–108 22. North MJ, Macal CM (2007) Managing business complexity: discovering strategic solutions with agent-based modeling and simulation. Oxford University Press, Oxford 23. Schramm ME, Trainor KJ et al (2010) An agent-based diffusion model with consumer and brand agents. Decis Support Syst 50(1):234–242 24. Stonedahl F, Rand W, Wilensky U (2008) Multi-agent learning with a distributed genetic algorithm. In: AAMAS: ALAMAS + ALAg Workshop, Citeseer 25. Stonedahl F, Rand W, Wilensky U (2010) Evolving viral marketing strategies. In: Proceedings of the 12th annual conference on genetic and evolutionary computation, pp 1195–1202 26. Swatman PMC, Krueger C, Beek KVD (2006) The changing digital content landscape: an evaluation of e-business model development in European online news and music. Internet Res 16(1):53–80 27. Teece DJ (2010) Business models, business strategy and innovation. Long Range Plan 43(2):172–194 28. Walter TP, Back A (2010) Crowdsourcing as a business model: an exploration of emergent textbooks harnessing the wisdom of crowdsourcing as business model innovation. Business, pp 555–568
