Researchers use AI agents to test service designs before building them — and it's cheaper than hiring consultants

Researchers built a way to test how people will respond to service design choices by simulating their behavior using large language models instead of running real pilots or hiring human consultants. This means companies can now explore dozens of design options quickly and cheaply, including some they might not have thought to try, before committing money to implementation.

Service businesses have always had a problem: you can't know how customers will actually respond to a design change until you try it, and trying it is expensive, slow, and risky. This work suggests a much cheaper pre-flight check exists. The practical shift is straightforward — a supply chain manager or operations team can now test a new workflow or customer interaction through simulation before running a real pilot, cutting both cost and time-to-decision. What's less obvious is that the simulation sometimes finds good designs that human designers missed, which means the tool isn't just validating existing ideas — it's occasionally discovering new ones.