Whalen describes how his firm presents synthetic research results: a note on the simulation's limitations first, then the answer, then verbatim quotes from real people. Framed as a possibility to explore, he says, the results make executives ask for more human testing instead of betting a product on them.
Leah, you were kind of talking about building skills. And I think in a previous interview, you were talking about like helping your students over there at the University of Michigan start thinking about this sort of thing. What's your take on that?
…that expertise and experience and point of view. And a lot of people are saying taste. And I think taste actually is some of it because it's like, well, what do I like and not like? And that's okay to say, right? And how do we teach someone to do that without getting it wrong for 10 to 15 years?
…we're going to launch this next $60 million product on this because my chat GPD told me. And really, I think there is, with the right presentation of this is an indication or a possibility or an avenue to explore, they actually get much more interested in coming back and saying, if this is really true, this would be really, really interesting. We need to do more human testing as a result. So we've actually been really pleasantly surprised that ironically, we had to get all the way to synthetic users to get our leaders super interested in real human customer data. But it has evolved into a loop there that sort of gives them feedback that they get interested in hearing from the real consumers. So I appreciate that part of it. So yeah, to your point, we try to say, here are some of the guardrails. Here's the limitations of this. Now go ahead and ask it questions. And here's what we heard from our verbatim people in real life, just to try to ground it. So…
Yeah, the podcast format too. Like, oh my goodness, like the, I know I've been doing that for myself, like creating podcasts that are like summaries of things. And it's just one way that those of us who are busy can like walk and learn at the same time or, you know, like multitask. And, you know…
Well, it is interesting, right? So in Leah's thought here, so far we've been saying that you're only directly interacting with a synthetic user. There's no reason why we couldn't have some sort of expert that says, I think you're asking about this. Let's ask about that. And then have the synthetic…
But you're talking about building separate agents, an expert agent, AI agent, and your synthetic audience agent, and maybe some sort of other agent, and they're all kind of checking each other and guiding it as you go along. Yeah, so