07 Nov Agent-Based Modeling – The new quiver – Individuals v. segments
Agent-Based Modeling (ABM) isn’t looking directly at individual behavior. Instead it looks at a group of individuals overall – a segment – as the key element against which the model is calibrated.
I often get asked the question as to whether ABM requires a longitudinal study of the individual behavior of specific consumers over time. This is probably the most asked and most mis-understood concept when considering agent-based modeling. Hopefully I can clarify it here.
Our MarketSim implementation of ABM uses our understanding of how consumers act in the marketplace to build a framework of rules. Once this framework is put together it needs to be calibrated (fit) against what can be observed in the real world. What ABM does is it simulates virtual consumers acting within this framework, following the prescribed rules. During the calibration process the parameters of the rules are adjusted such that the actions in the aggregate can be calibrated against what we observe.
For example, let’s assume that our tracking study shows top of mind awareness at the beginning of a campaign is 55% and at the end of a campaign it reaches 78%. The calibration process takes this number and the parameter for the rule for advertising (in a simplistic model) and adjusts it such that after the campaign top of mind awareness has grown 23 points from 55% to 78%. This means that 78% of the virtual consumers – agents – have become aware.
During this process, individual agents have become aware, but the individual behaviors aren’t calibrated to individual consumers. Only the aggregate is calibrated against what we have observed, in this case in the aggregate at the segment or category level through our tracking study. ABM doesn’t require knowledge about the individual consumers but instead it is simply calibrated (fit) against the tracking study data such that 78% of the agents become aware at the end of the campaign.
I hope this helps. If you haven’t downloaded our white paper yet, please go here for the complete background on this discussion.
Look forward to hearing from you.