Cast, Validated by Cast: podcast transcript

Two hosts talk through this study's findings. Made by Cast from the study itself. Everyone quoted is an AI user, not a real person.

Lily: "I got a quote from a research agency for a two-week study, eight thousand dollars. I needed the answer by Friday for sprint planning, not in two weeks, and we don't have eight grand sitting around pre-revenue. I didn't even negotiate, I just closed the tab."

George: Closed the tab.

Lily: Closed the tab. That's Priya Shenoy, founder of a two-person fintech called Ledgerline. And here's the thing you need to know before we go any further: Priya doesn't exist.

George: Ledgerline doesn't exist either.

Lily: She's an AI user. One of twenty-four that Cast built, interviewed and wrote up in a study called, and I'm reading the title here, Cast, Validated by Cast.

George: Which, yes, is a research product running research on itself. We know how that sounds.

Lily: It sounds like a company grading its own homework.

George: It does. And the interesting part isn't the grade. It's where the AI users pushed back, and where our own quality check caught us doing something we shouldn't have.

Lily: We'll get to that. What was the question?

George: Cast is a synthetic-user research tool. You describe your idea, a Casting Agent builds a cast of AI users matched to your Ideal Customer Profile, your ICP, interviews them, and you get findings in about an hour. This study asked one thing: of the four buyer types closest to paying, founders, marketing leaders, consultants and policy makers, which do you go after first?

Lily: And the problem is the one Priya just described. Real research is slow, expensive, or you can't reach the people. Agencies run two thousand to a hundred and fifty thousand dollars a study, two to six weeks.

George: Or what most people actually do, which is nothing.

Lily: Who did they talk to?

George: Twenty-four AI users, six per ICP, from more than a dozen countries. Founders with no research budget. Marketing leaders testing a message before the media spend locks. Consultants on a client deadline, usually under NDA. And policy people who need an honest read before a proposal is finalized.

Lily: Built how? If you build twenty-four fans, you get twenty-four fans.

George: Every ICP had two AI users whose starting stance was Skeptic, eight of the twenty-four. Low Fuzziness, so everyone sat close to the profile. Ten questions each, and the concept wasn't shown until question six.

Lily: So the first five are about what they actually do today.

George: Two hundred seventy-four answers, thirty-four of them follow-up probes, and a separate Critic read the answers for invented facts and people-pleasing.

Lily: This is the caught-us bit.

George: The Critic flagged two of two hundred sixty-four exchanges. One was our own interviewer. Marcus Webb, a logistics co-founder, had been burned by a synthetic panel demo and said if the AI users all agreed with him, he was out. The follow-up was: "If you learned there was a built-in check for that, specifically, would it change your answer?"

Lily: That's a leading question.

George: The Critic flagged it as a leading premise. It's right there in the data.

Lily: Good. I'd trust the study less if it weren't. Founders.

George: The competitor isn't another research tool. It's doing nothing. Six of six run on cold messages, self-built surveys or gut feel. Priya's whole process is LinkedIn DMs, a Google Form with a four percent response rate, and begged intros, and she picked her first customer segment on one good call.

Lily: So price can't be the barrier.

George: Trust was. Four of six worried the answers would just flatter them. David Okonjo, a Series A founder: "Ahead of our last major roadmap bet, we paid a research agency forty thousand dollars for a study. The report came back clean and confident, pointed us toward a specific direction, we built against it for a full cycle. It turned out to be wrong, and we found out the expensive way, after shipping."

Lily: Forty thousand dollars and a product cycle.

George: And the agency wouldn't show him the raw interviews. So he wants the opposite: every transcript, and at least one AI user who disagrees with him and stays disagreeing. On price: "The number was never my objection, trust was."

Lily: Hang on. A skeptical AI user asking to see a skeptical AI user is a bit of a hall of mirrors.

George: It is, and I'd rather say so. But three of six founders landed on the same ask, a transcript with real pushback, and that's something you can build and put in front of a real founder.

Lily: Marketers, I'll take this one, because the fastest yes in the study is here. Jasmine Carter, Head of Growth Marketing at a skincare brand: "Three months ago we launched a new positioning line with sixty thousand dollars behind it in media. We didn't test the message beforehand, just internal debate, and it fell flat, conversion came in way under our model. The post-mortem pointed to the message itself."

George: Whoever argues best in the room wins.

Lily: Nearly her words. She's an easy yes, but she's two of six. The bigger theme, four of six, isn't speed or cost. It's a fear that a synthetic audience flattens everyone into one average. Olivia Bennett lived it: one blended survey across three buyer types, a fine overall score, then sales said the technical evaluators hated the tagline. Invisible in the average.

George: So the thing that wins them over is just... not averaging.

Lily: Three of six said exactly that. Show the groups separately, side by side. Which is how Cast already reports, by ICP, never a blended percentage. But that only helps if you show it, and the study says plainly that the side-by-side collateral doesn't exist yet.

George: Consultants. That's where the money is, surely.

Lily: The biggest numbers, yes. Six of six called the price negligible. Twenty-two thousand dollars for six expert calls on one deal. Thirty-five thousand for a fieldwork vendor.

George: But what they're short of is time. Nathaniel Price runs six-week market-entry engagements, and fieldwork alone takes three of those weeks: "I had to cut the interview sample from twenty down to fifteen to hit the schedule, and I knew going into the final client presentation that the evidence base was thinner than I'd have liked for a recommendation of this size. I said as much internally, though it didn't make it into the client deck in those words."

Lily: Didn't make it into the client deck in those words. That's the most honest thing a consultant has ever said.

George: And it's an AI user saying it.

Lily: Which is either the point or the problem, and I can't decide. Carry on.

George: The gate here isn't price. Carlos Fuentes runs private equity due diligence under NDA and needs data handling in writing, for his compliance team, or it's ruled out. And then the reversal. Priya Deshmukh, a partner in Mumbai, was wary because her firm sells judgment, and a client once asked why they paid partner rates for associate-run interviews. Scoped to fieldwork logistics only, she turned: "If this does what my associates currently spend three weeks doing, the scheduling, the actual interviews, the raw transcripts, and hands me the same kind of material I'd synthesize anyway, just faster, that's not a threat to what I do, that's freeing up my team for the parts of the job that actually need judgment."

Lily: One partner, though. That rests on a single AI user.

George: One of six, and the report says so. Narrowly based.

Lily: Policy makers, mine again. I expected a no, and one nearly was. But the pull is real. Elena Vasquez in Austin has her council recommendation due before the formal comment period even opens, so she goes in on informal calls with business owners she happens to know. And Dr. Samuel Osei in Accra had a health campaign underperform because people in a room with an official facilitator give the polite answer.

George: And the near-no?

Lily: Reginald Okafor, twenty-seven years protecting a federal public comment process. For him speed is a hazard, because a faster tool tempts someone to skip the real thing: "If there's any ambiguity about this being public input, or any risk a political appointee could point to it as a substitute for the formal comment record. That's not a labeling nitpick for me, it's the central professional risk I manage."

George: What would move him?

Lily: Permanent, unremovable labeling that this is a pre-consultation screening tool, never public input. With that, he'd use it to prepare his staff. Without it, nothing.

George: Fair.

Lily: Now the skeptic's summary. Twenty-four AI users, and every single one ended up willing to try it, or wanting to and needing someone's sign-off. Every one. If a human agency handed me that, I'd send it back.

George: You should, and here's the honest limitation, said plainly. These are AI users role-playing a defined person. They answer from what such a person would plausibly know and feel. They're not real people, and what they say they'd do is a stated intention, and stated intentions run ahead of behavior. The report itself says: check the key points with about five real people in each ICP before you commit budget.

Lily: So what's it for, if it can't tell you whether anyone will pay?

George: It told us where the objection lives, and that it's different in each group. All twenty-four accepted fifty dollars a month Regular and a hundred and fifty Premium, nobody named price as a blocker. But what each group needed to see first was completely different. Founders: a transcript with real pushback. Marketers: groups reported separately. Consultants: data handling in writing. Policy: a label that can't be stripped off.

Lily: Four proof points, not four prices.

George: And one twist. The skeptics tied cheap to low effort. Tom Reinholt, a one-person bootstrapped dev-tools company in Berlin: "Prove the free part is real and I'm not precious about paying later." A discount does nothing for him. A free trial that answers one real question end to end does everything.

Lily: Right, what I'd do. The study says founders first, consultants second. I'd argue consultants first. That's where the budgets are.

George: They also have a compliance officer, a QA lead and a partner to convince, and the study can't say whether any of them sign off. Founders have none of that. Six of six compete against nothing, six of six called the price reasonable or surprisingly low, and the proof they want you can build this week.

Lily: Then why not marketers ahead of consultants? Jasmine's the easiest yes in the study.

George: Jasmine is two of six. Four of six doubt it keeps audiences apart, and the side-by-side proof doesn't exist yet. Consultants need two documents, a written data-handling statement and a labeled sample report. That's closer.

Lily: Okay, I'll give you that order. First?

George: Build the sample transcript set, AI users disagreeing and staying that way, plus a plain note on how the anti-sycophancy checks work. That's the founder pitch. Not speed, not price. Pushback.

Lily: Second, rewrite the free trial promise around one real question answered end to end, because Tom, and Mateusz Kowalski in Warsaw, both said that's the only way they get to a yes.

George: Third, talk to real founders about transcript trust, same-week action and trial to paid. Fourth, the two consultant documents, tested with real risk and QA reviewers, not with us.

Lily: And marketers and policy makers wait until the side-by-side reports and the non-substitution label exist.

George: Then pilot ICP B on a message you already know failed, and ICP D under a permanent pre-consultation label.

Lily: And nobody discounts anything.

George: Twenty-four of twenty-four accepted the tiers. Test the packaging, what's gated, what's exported. Never the price.

Lily: Here's what I'd say to anyone listening who isn't Cast. If you're about to decide on one good call and a gut feel, the way Priya did, run it on that. One ICP, three AI users, free. Read every answer. Then find five real people and see if they say the same thing.

George: Which is, word for word, the next study on the list. Run this exact study again with real founders, real marketers, real consultants and real policy makers, and compare the themes.

Closing line: This has been prepared with help from Cast at castofusers.com, thanks for listening.