Last Wednesday night, I was fixing a client’s quote sheet in Hangzhou

By the third version, I was staring at the screen thinking: AI-written work keeps getting better, but how does it judge whether one line of advice is reliable? What I was missing wasn’t another model. I needed real expert answers it could reference. That was when I saw Mercor’s news: a new $750M funding round at a $20B valuation. The numbers were huge, but my first question was simpler: can a one- or two-person team like mine tap into this momentum?

Mercor isn’t selling a tool—it’s selling people who’ve gotten it right before

Mercor organizes profiles of doctors, lawyers, engineers, and other specialists into data that AI companies can use as references. The company says its platform has 5 million experts. By its published figures, annualized revenue reached $1B in February 2026, $2B in June, and about $3B now.

My friend Lin Yue runs industry training in Guangzhou. On the evening of September 13, she received an invitation from an unfamiliar company account asking her to review an AI-generated output for the legal industry. She didn’t develop a system. She recorded a work brief with conditions attached, then pointed out omissions in the answers one by one. It wasn’t recruitment, and it wasn’t quite a traditional survey either. It was more like packaging “how an experienced person makes a judgment” for companies to buy.

What we can replicate today isn’t the $750M round

Don’t get carried away by the valuation. To test this idea, the cost can be $0 and 30 minutes. The technical barrier is whether you can write consistent evaluation criteria; don’t hire five experts right away. Money: start at $0. Time: spend 30 minutes preparing and collect feedback over seven days. Barrier: knowing how to send a survey and build a spreadsheet is enough. First step: open the online survey tool you already use, add three questions asking clients, “Which three pieces of information matter most when you make this decision?” and include permission for anonymous use. If you already have one or two clients, ask them to review two old proposals. If you don’t, run the process on an example from your own industry. If you’re not ready to test now, that’s fine too; confirming the need clearly comes first.

If you’re at a different stage

If I were just starting, I’d first talk to three potential clients about “which judgments often go wrong,” create free samples only, and avoid building a platform. If I already had one or two clients, I’d split the quote into “one expert-feedback package” and “ongoing updates,” then sell the one-time service first to confirm that people will pay. If I were scaling up, I’d record each feedback item’s source, version, and error rate before deciding whether to work with an existing data service or hire part-time editors for review. Once volume grows, the easiest place to burn money isn’t software—it’s answers nobody verifies.