Agent-based Models

Agent-based Models (ABM)

The purchasing and decision-making behavior of consumers can be simulated using agent-based models. This allows the complexity of markets to be recorded and understood, enabling the knowledge to be used to enhance marketing. 

Why does a consumer choose product A rather than product B? Where did he or she gather information before making the purchase? What influence did advertising, friends, or websites have on his or her decision? Marketing executives must know and understand buyers’ behavior so that they can plan their measures reliably and successfully. But the complexity of the markets, and in particular communications, has increased enormously over the past few years. So far, marketing has focused on the communication flow from the company to the customer – usually viewing it as a linear process. However, the growing exchange of information between consumers is also important. And this functions more like a circulatory system with non-linear interrelations and a network-like structure.

Agents Simulate Purchasing Behavior

Previous marketing models have only been able to roughly replicate these network-like structures, as regression models are better suited to linear structures. Agent-based models, on the other hand, assume a network structure. This is promising. An agent represents individual buyers or households within a certain environment which act and make decisions, interact with their environment, and change over time. Agents are assigned various roles, for instance. Some are opinion leaders, others are trendsetters, and yet others prefer to follow the masses.

However, agent-based models have mainly been used for small-scale academic research projects to date. And the few applications in companies have focused more on explaining fractures within a structure. The exciting question which cannot yet answer but want to investigate is: Can such models also simulate entire markets, much like the way the Nuremberg Institute for Market Decisions (formerly GfK Verein) and GfK SE already did for daily consumer goods with the BrandSimulator? Can it perhaps deliver even better results?

In order to answer these questions, we first need a software solution that can run these simulation models. Unfortunately, existing software is not flexible enough and/or unsuitable for large datasets, such as GfK’s 30,000-household panel. That is why we are starting by developing a prototype software solution that can subsequently help us to look into these questions.

The software will analyze the behavioral patterns of the agents. The focus here will be on so-called research online, purchase offline (ROPO) and touchpoint analyses, evaluating the customer journey, assessing the impact of new product launches, and modeling FMCG markets. Our understanding of complex markets is growing, which can be used to derive specific recommendations for marketing. It will also be possible to clearly show which developments and circumstances are behind these recommendations.