BUILDING AGENT-BASED DECISION SUPPORT SYSTEMS FOR WORD-OF-MOUTH PROGRAMS. A FREEMIUM APPLICATION
Published in the Journal of Marketing Research, 2016 [IF 3.109, Q1 19/120 BUSINESS]
Marketers have to constantly make decisions on how to implement word-of-mouth (WOM) programs and a well-developed decision support system (DSS) can provide them with valuable assistance. The authors propose an agent-based framework that aggregates social network-level individual interactions to guide the construction of a successful DSS for WOM.
The framework presents a set of guidelines and recommendations to:
(1) involve stakeholders,
(2) follow a data-driven iterative modeling approach,
(3) increase validity through automated calibration,
and (4) understand the DSS behavior.
This framework is applied to build a DSS for a freemium app, where premium users discuss the product with their social network and promote the viral adoption. After its validation, the agent-based DSS forecasts the aggregate number of premium sales over time and the most likely users to become premium in a near future. The experiments show how the DSS can help managers by forecasting premium conversions and increasing the number of premiums via targeting and reward policies.
Keywords: Word-of-mouth, Marketing Decision Support Systems, Agent-based Modeling, Targeting and Referrals, Freemium Business Model
We were approached by the firm to help answer the question how can they incentivize additional conversions? We quickly determined that word-of-mouth seemed to have a large impact on conversions, but the firm did not know how to use that knowledge, so we hypothesized that it would be possible to create an agent-based model that used social network data to assist them in targeting users who would have a large impact on overall conversions.
We first constructed an agent-based model that attempted to predict the rate of conversions in a massive-online game, called Animal Jam. The main components of this model were a social network and a model of social influence propagation. Once we had a model that we had shown was accurate at predicting out-of-sample conversions, we then constructed a tool that enabled the examination of the effects of different marketing policies on the social network. Using this model we were able to explore the role of targeting different users with different incentives:
We created a set of guidelines that can be generally used to construct a decision support system for understanding word-of-mouth marketing in a wide variety of domains. In the Animal Jam case, we determined that targeting users who already had a large number of premium friends but who had not converted themselves was likely to have the largest effect on conversion rates, both in terms of the targeted users, but also in terms of spillovers to network peers.
These findings illustrate how we can build analytic tools to help create more data driven decisions in a wide variety of contexts. Managers should shift away from the more intuition -based design of campaigns and instead refocus on data-driven, simulation-supported decision support tools. In particular, our application has implications for most freemium apps that are interested in encouraging a larger conversion rate.
The figure below illustrates the effect of incentivizing 2,000 non-premium users in terms of the number of additional conversions that will happen within the following month. As can be seen, the largest increase happens if the most likely users (>= 6 premium friends already) are targeted:
Presentations about this work:
– Talk at the INFORMS Marketing Science Conference, Baltimore 2015 (PPTX)
– Keynote talk at the Marketing Series by Catedra Ramón Areces, Oviedo 2015 (PDF in Spanish)