Five9 buys Level AI. Your call center is your best market research

Here's a familiar picture. Marketing pays for surveys, focus groups and social listening. Two floors away, customer service handles hundreds of conversations a day in which customers volunteer exactly what annoys them. Nobody is listening systematically.

That gap just got very expensive.

What Five9 bought

On November 5, Five9, which sells cloud contact-center software, announced it will acquire Level AI. The key points from the press release:

• Price: about $325 million in cash from its balance sheet. Expected to close in 2026.
• Product: AI agents, agent assist, quality management, coaching and voice of the customer, all on a layer that analyzes conversations. Level AI runs on its own model family, Latitude, which it says is cheaper and faster than frontier models.
• Scale: more than one billion customer interactions a year (company figure), with customers including Gusto, Chime and VistaPrint.
• Go-to-market: Level AI stays available standalone, including for companies that don't use Five9, and becomes part of Five9's platform with more than 4,000 customers.

Reports put Level AI's annual recurring revenue at around $20 million at the end of October. Just days earlier, Salesforce announced it would buy Cresta, also voice AI for customer conversations.

They're not buying the bot. They're buying the listening

What I find interesting: everyone talks about AI agents taking calls. The real asset is something else. Whoever analyzes every conversation owns the most honest customer data in the company. Not a survey, not a sample, but what customers say when they actually have a problem.

For marketing that means the best insights for positioning and messaging will soon live in the service stack, not the research budget. The words customers use. The objection that comes right before a cancellation. The ad promise that sets expectations the product doesn't meet. That's copy material no agency can invent.

Second point: Level AI bets on its own specialized models instead of the big generalists. I think we'll see that pattern a lot. For a narrowly defined job at huge volume, the smaller, cheaper model beats the big one, because the math has to work per conversation.

The catch: who owns the insights?

Two deals in a week also means fewer independent vendors. Pick a conversation analytics tool today and you may be buying a piece of someone's platform tomorrow, like it or not. And the insights land in the service team's dashboard, not automatically with marketing. That's not a tech problem. It's an org question.

What I'd do now

• Tap your service data. Ask your service team this week for the top 10 reasons for calls and complaints over the last 90 days. Hold them up against your ad messaging.
• Collect customer language. Take five verbatim customer phrases (anonymized) and test them as headlines or subject lines against your current versions.
• Settle the exit before you buy. If you're choosing a conversation analytics tool, put the right to export your analyses in the contract, in case the vendor gets acquired.

Bottom line: The most expensive market research you're not using is running through your service hotline right now.