AI Agent Consumer Simulations

One of the biggest buzzwords in tech these days is “AI Agents”. Mark Benioff just said he’s “never been more excited about anything at Salesforce” as he is about Agentforce, its tool to build and customize AI agents. Sam Altman recently said, “Helpful agents are poised to become AI’s killer function”. One recent market estimate suggests AI agents will be a $47.1B industry by 2030.

I’m not big on hyperbole without specifics. Things aren’t real to me until I can picture the nuts and bolts of what they mean. I also tend to think about things other people aren’t thinking about… yet. So I decided to document one specific area where AI agents could help marketers, with some specific examples. It’s based on my extensive background in market research and marketing. I'm talking about consumer simulation.

By training AI agents on the background, behavior, and/or perspectives of specific consumers, marketers could generate great and very cost effective use cases from simulations played out by those agents.

How would we train these consumer simulation AI agents?

The concept is each agent represents a specific consumer. Training data for a given agent can come from:

• Purchase or interaction history of a specific consumer
• Demographics, segmentation, and psychographics of a specific consumer, with all PII removed of course
• Quantitative and (especially) qualitative market research responses from a specific panelist. Respondents could be asked to rank priorities when making purchase decisions, walk through the process of how they move through an online or physical store, or describe some of their favorite products or concerns about products they dislike.
• Computer vision patterns of specific consumers moving through a shopping environment

How would we use these agents?

• Perhaps the most obvious and scalable use case is ad creative testing: We could track the reactions of these agents to different marketing materials, including the critical aspect of having them think out loud. It’s like the best of A/B testing and panel research combined, but with essentially zero marginal cost.
• Development of sales scripts: We could have these agents interact with human customer service or sales people to test out and track the effectiveness of different scripting and document FAQ. We could even have them interact with customer service or sales AI agents for true scalability! These sales scripts could also be used as vector databases for customer-facing agents.
• Simulations of shopping experiences: We could have these agents simulate the process of going through a shopping experience, thinking out loud all the way, explaining how they’re deciding what to pay attention to, what to consider, and how they decide. This could be great for shopper marketing, shelving, or product launches.
• Word-of-mouth simulations: We could have these agents talk to each other, tracking to see if information passed from one to another becomes useful training data in making the other change their decision on a given product. By tracking when decisions change, we could document convincing arguments that could be used in ads, marketing collateral, and by customer-facing agents and chatbots.

Researching this area of potential use for AI agents was extremely useful for me in solidifying what agents are, why they’re potentially valuable, and how they relate to the world I know. I hope this was helpful to you too. I’d love your thoughts and feedback!