Boston Startup CollectivIQ Bets on Multi-Model AI to Fix Hallucination Problem — No Long-Term Contracts

John Davie had a problem that's become familiar to executives across corporate America: his employees were feeding sensitive company information into AI chatbots, potentially training his competitors' models. When he looked for secure enterprise alternatives, he found expensive long-term contracts that still produced hallucinations and flat-out wrong answers that ended up in PowerPoint presentations. So he built his own solution.

CollectivIQ, a Boston-based startup incubated inside Davie's hospitality procurement firm Buyers Edge Platform, launched publicly this month with a pitch that sounds almost too simple: instead of betting on one large language model, query them all at once. The software simultaneously pulls responses from ChatGPT, Claude, Gemini, Grok, and up to 10 other models, then searches for overlapping and differing information to produce what it calls a "fused answer" — theoretically more accurate than any single LLM could deliver on its own.

The wake-up call came about a year ago, according to TechCrunch, when Davie realized the scope of the data leakage problem. "We could be essentially edging our competitor," he said, describing the risk of employees using consumer AI tools with company data. When he investigated enterprise contracts, he hit another wall: "We hated having to decide which employees deserved AI." Worse, employees complained about biased and hallucinated answers that made their way into client presentations — the kind of mistake that can crater a deal or damage a reputation.

Davie challenged his chief technology officer to build something better, and CollectivIQ is the result. The company started rolling out the software internally at the beginning of 2026, and the initial response was strong enough that Davie decided to release it to the public after learning that many of Buyers Edge Platform's customers were dealing with the same confusion and hesitation around AI adoption.

The technical approach is straightforward but potentially powerful: CollectivIQ uses enterprise APIs from multiple AI providers, pays for the token costs itself, and charges customers only by usage. That pay-per-use model is a direct shot at the enterprise AI market's status quo, where companies often lock themselves into expensive multi-year contracts with a single provider. "I'm hoping that this is a breath of fresh air for companies that see that they are not going to have to be committed," Davie told TechCrunch. "They're only going to pay for the value they get out of it."

All data involved in CollectivIQ prompts is encrypted, addressing the security concern that started this entire project. The company is fully funded by Davie, who plans to seek outside capital later this year — a timeline that suggests he's prioritizing product-market fit over rapid scaling.

The multi-model approach isn't entirely new. Researchers have long known that ensemble methods — combining predictions from multiple models — can improve accuracy and reduce errors. What CollectivIQ is betting on is that enterprises will pay for a productized version of this technique, especially if it means avoiding the vendor lock-in and hallucination problems that have plagued early AI deployments.

There's a certain irony in the timing. CollectivIQ is launching just as the enterprise AI market enters what might charitably be called a period of disillusionment. The initial hype around ChatGPT and its competitors has given way to a more sober reckoning with the technology's limitations — hallucinations, bias, data security risks, and the difficulty of integrating AI into existing workflows. Davie's bet is that the answer isn't to pick the "best" model, but to hedge across all of them.

For Davie, who founded Buyers Edge Platform 28 years ago, the experience of building a startup again has been energizing. "It does feel like way back in the day and we are doing it all over again and being scrappy and being very in the weeds on LLMs and post training and all sorts of things I was not trained in," he said. "It's fun and exciting. I go sit hand and hand with the software developers building the product, that's how I got my main company, it's a lot of fun."

The real test will be whether CollectivIQ's fused answers are actually more accurate than single-model outputs, and whether the pay-per-use economics work at scale. If the company can prove both, it might have found a genuine opening in a market that's rapidly commoditizing. The alternative — that enterprises will simply pick the cheapest or most convenient single-model provider and live with the hallucinations — is less appealing but entirely possible.

What's clear is that the enterprise AI market is still wide open for innovation. The first wave of products was about access — getting ChatGPT or Claude into the hands of employees. The second wave, which CollectivIQ represents, is about reliability, security, and economics. The companies that solve those problems will own the next decade of enterprise software.