Stop Building What You Think Users Want

01 /The Concept

Velozity built a platform where product teams can run rapid user research – without involving real users.

The speed comes from replacing live participants with AI personas that developers generate themselves, based on their understanding of who their target audience is. Each persona can be configured with age, job title, education, interests, mindset, and any other characteristics the developer considers meaningful.

The first distinctive feature is that research on Velozity runs as a classic focus group. The developer acts as moderator, assembles multiple personas into a single group, and poses questions to hear their responses. The developer decides which of their created personas to invite into each session and for which topic.

The second distinctive feature is that personas have memory. They remember what they've been told and what they said in previous sessions. In a follow-up group with the same personas, the developer doesn't need to waste time bringing everyone up to speed. And because the personas carry context from earlier conversations, the responses tend to be a natural continuation of previous discussions rather than a replay of the same answers.

The platform supports research on new product ideas and potential new features – whether participants find them interesting and useful, what they'd add, and what they'd cut. Developers can also test future interfaces by linking to a Figma prototype; the platform pulls those screens into the session so personas can discuss the UX directly.

Velozity also mentions that ad creative can be tested in the same way, though the mechanics aren't fully spelled out yet.

The platform launched on Product Hunt just yesterday, so it's currently free while the team collects early feedback. A paid tier with expanded capabilities is coming soon.

One feature requested by early users – and already in the pipeline – is letting personas respond not just to the moderator's questions but also to each other's answers. That would give developers a much richer picture of how opinion actually forms when different perspectives interact.

02 /Why It Matters

The first time AI-persona-based user research crossed the radar was in a [review of Lakmoos](/review/mgnovenno-vmesto-polugoda), a startup doing similar things for individual personas rather than full focus groups. Lakmoos raised €510K and prices its platform from €3,200 per month – which it calls "affordable" – suggesting it may not need more capital right now.

More broadly, using AI twins to predict how real people will react is a rapidly developing category.

Synthetiq ([covered previously](/review/hochesh-imet-kuchu-podpischikov)) helps creators predict the reaction to not-yet-published social posts. Crucially, it builds AI twins of each creator's *specific* followers – not a generic audience – and also suggests how to improve the post to generate more engagement.

Y Combinator graduate Artificial Societies ([related review](/review/tema-uzhe-letit-no-vot-tak-mozhno-vzletet-povyshe)) is building an entire family of such tools:

- The first was a set of AI twins of venture investors, used to rehearse a pitch – which ultimately helped the founders get into YC.

- The second predicts reader reactions to LinkedIn posts, similar to Synthetiq.

- The third – currently in development – helps founders find product-market fit by simulating how a target audience would respond to a proposed product, which is close to what Velozity does.

What's interesting is that Artificial Societies' algorithms simulate the behavior of *groups*, not just individuals. And that bridges directly to Velozity's focus-group model – because group dynamics are doing real work there too.

The key insight is that people's opinions are shaped by other people's opinions. The volume and tone of early comments on a post can dramatically affect how later readers respond. Artificial Societies claims its prediction accuracy comes precisely from modeling those social interactions rather than each person "in a vacuum."

This connects to a classic social psychology experiment: when children were asked to identify the color of two pyramids – one black, one white – but everyone else in the room had been coached to say "both white," the last child would often conform and say "both white" too, despite the visual evidence. Social influence can't be underestimated.

After Velozity adds persona-to-persona interaction, the same dynamic could emerge inside its focus groups – personas persuading each other, which might shift research outcomes. Whether that's a distorting noise or a useful approximation of real social dynamics – where individual influencers carry outsized weight – is a genuinely interesting question worth studying further.

03 /Opportunities

On the one hand, it's a little unsettling that human behavior has become predictable enough for AI trained only a couple of years ago to model it reliably.

On the other hand – this is fantastic. It means product and marketing decisions can be validated faster and cheaper than ever before, using AI stand-ins for the people who'll ultimately use the product.

The actionable direction: build platforms where clients can generate and study AI personas to pressure-test hypotheses about products, features, ads, pricing, interfaces – or anything else they want to understand before committing to it.

The underlying technology is mature enough. The open question is what research you want to make easy, what the platform experience should feel like to minimize friction, and which markets are willing to pay the most for that speed.

So: what research would be most valuable to you? The verticals with the highest willingness to pay are those where the cost of a wrong decision – a failed launch, a misread audience – is already measured in months and serious budget.