Agent Modeling in Marketing
A curious article in Forbes about using a SimCity-like simulation as an inspiration for the next Coca-Cola or Unilever marketing campaign:
William Bean, a research director at Pepsi-Cola North America, is fascinated by vending machine use in a mythical suburban office. There 35 sales, technology and administrative employees work, gossip and quaff lots of soda.
These thirsty workers are characters in a simulated test market on Bean’s desktop computer in Purchase, N.Y., and they are supposed to give Bean, a former biologist, insight into soda consumption. The employees, or “agents,” as they are called among computer-simulation experts, are programmed to act like office staffers and consumers. They are directed to move and interact in response to simple rules (drink when thirsty, for example) but with an element of chance injected. It’s a toss-up whether an agent, confronted with a vending machine in an office lobby, will buy a soda.
SimCity for marketing managers. Thus does Pepsi aim to study how customers react and adapt when it changes the number and locations of vending machines in the office. It hopes to take some of the
guesswork out of placing the machines.
Ok why not but what is suprising is this, isn’t it too emphatic????
It is like a real test market, only faster and cheaper. (…) The analytical technique, which is like a videogame without controlling players, helps companies forecast consumers’ individual and collective response to new product offerings, price changes, media buys or marketing pitches. It also tracks how the agents influence one another, a big selling point as
marketers seek to understand and sway targeted groups of consumers by using specialized appeals, such as word-of-mouth campaigns.
November 17th, 2005 at 4:55 pm
Many people don’t seem to realise that simulations are only as good as the assumptions or models underlying them. People don’t ‘drink when thirsty’ - there are much more complex rules than that. (Cost? Concern about sugar intake? Choice between drinks? Social interaction - ie drinking when others do, not drinking when others don’t? Level of work - ie too busy, or so bored you drink for something to do?)
Simple models run the risk of just learning how that particular model works - with little or no real link to actual consumer behaviour. I’ve no idea what Mr Bean’s model is like, but I hope it’s better than the description in Forbes.