BT Technol J Vol 18 No 1 January 2000 33 Agent-based modelling — treating customers as individuals not numbers D Collings, T Reeder, I Adjali, P Crocker and M Lyons Developments in computer hardware and software over the last ten years have made modelling a cheap and powerful support tool in the business world. Routinely, simple models are used to understand how changes in the cost base of a product will affect profits and returns on investments. Used in this way models allow decision makers to investigate the sensitivities between variables to make more informed decisions. 1. Introduction We are now familiar with instore cards which are used to promote loyalty but they also generate valuable information about individual customer purchasing patterns. Similarly, field studies of communications usage are able to generate detailed information about the preferences and motivations for telephony usage. With such detailed information available a technique is needed which will make better use of the information than simply aggregating. Agent-based models in which the properties of the agents are preserved represents the state of the art of business modelling, and are naturally constructed from this type of data. 2. Agent-based models I n an agent-based model the agents represent the real world entities. The beliefs, knowledge, and objectives are captured in the agent so as to mimic the real-life object. In the simplest form these behavioural patterns can be represented by rules which describe what an agent will do under certain circumstances. For example a rule could be: ‘If the price of Internet connection falls below £5, then buy’. Interactions between agents are also captured, as these are central to the communication of information through the system and can lead to modification of the behaviour of the individuals. For example, if word of mouth is important for a customer undertaking a decision to purchase a product then a rule may be: ‘If two or more of my friends recommend a product, then buy’. One of the outcomes of the agent-based model is the emergent behaviour of the system. When a system displays a behaviour which is not immediately obvious from the microscopic rules of the system, the behaviour is described as emergent. A graphic example of emergent behaviour is the ‘boids’, — a computer screen saver. This is a simulation of birds in which they have been given three simple rules to control how they interact — essentially collision avoidance rules. The birds are given random speeds and directions to start the simulation. After a few moments the result is that the birds flock in the same manner that birds are seen to flock in the real world. This result is surprising since this overall behaviour was not explicitly built into the model. Conventional, top-down, modelling techniques would assume the macroscopic behaviour such as this. The bottom-up, agent-based approach does not make these sorts of assumptions and hence the model can give greater insight, moving away from the problem of just ‘getting out what you put in’. 3. Summary R epresenting a population of customers using agent-based modelling is intuitively a more realistic approach for modelling the diffusion of an innovation within the popula- tion. It allows details of the cognitive processes experienced by the individuals and the configuration of the social net- works to be incorporated explicitly. It can overcome many of the shortcomings of conventional diffusion modelling. M ost modelling takes a top-down approach in which key interactions are observed in the real world and then reconstructed in a model. Using this approach a modeller would observe the effects of, say, a price change on the number of consumers who purchased a product at an aggregated level. This would provide the basis for quantifying the strength of interaction in the model. The modeller would not be interested in how the behaviour of individuals gives rise to the level of interaction. This type of modelling is used in a wide variety of applications through its simplicity.