A customer with 2,000+ sales reps on your product, every contract flowing through your system, sits down at a post-Dreamforce dinner you're hosting — and tells you the restaurant's price tag is evidence you're overcharging him. That happened to Jason Lemkin's EchoSign team with Groupon, then a Top 5 account running EchoSign for Salesforce, and the line has outlasted almost every other memory from those years.

The exact quote, delivered mid-dinner by their lead champion at Groupon: "This is a pretty fancy dinner. That probably means I'm overpaying."

It sounded like a joke. It wasn't. He was smiling, but he meant it. His job was to own a vendor relationship with a real line item attached, and he was reading every available signal about whether that number was fair. An expensive dinner read as one signal: this vendor has margin to spare, and that margin is coming from me.

Lemkin's takeaway cuts against how most vendors operate. Teams treat value as a two-moment question — close and renewal. Customers don't. They run a quiet, continuous audit of whether they're getting their money's worth and update it at every interaction. The absolute price barely matters: a $20,000-a-year customer and a $2 million-a-year customer both want to feel they got a good deal. The bigger customer is often more sensitive, not less, because someone above them will ask why the number is so high.

The signals that actually cost renewals

The fancy dinner is the mild version. The behaviors Lemkin identifies as renewal killers:

Mediocre support. A customer paying serious money waits two days for an answer, or gets a canned response to a real problem, and does the math immediately: we pay this much and this is what we get?

Upcharges that shouldn't exist. Extra fees for SSO, for an API that should be standard, or for admin features required to make the core product usable. Each one tells the customer the pricing was designed to extract, not deliver — and they remember at renewal.

A CS team that exists to upsell. Customers can tell within one or two calls whether their CSM is there to make them successful or to find expansion. If every QBR ends with a pitch, they stop taking the calls, and the vendor loses visibility exactly when it's needed most.

An AI agent that's too hard to train and deploy. Lemkin calls this the 2026 version of all of the above. If a customer buys your agent and spends weeks of their own team's time getting it to work, they're paying twice — once in dollars, again in hours — and they feel that cost every week, coloring how they see the price tag.

None of these are pricing problems on paper. The price might be completely fair. But customers don't experience price in isolation; they experience it next to everything else the vendor does.

Why AI makes the audit faster

With seat-based software, the value audit was slow. You bought 2,000 seats, reps used the product, and the question was mostly "are they using it?" With AI agents and usage-based pricing, the audit gets faster and more granular: customers can see what each task, resolution, and outbound touch costs, and compare it directly against a human or a cheaper tool. Every bad output, every retry, every hour their team spends cleaning up after the agent goes straight into the "am I overpaying?" column. The gap between "a fair price" and "feels like a fair price" now shows up in renewal and expansion numbers much faster than it did with seats.

What EchoSign changed

Three adjustments followed that dinner. The team checked whether its biggest customers were getting its best support, not just its best dinners — a fast, senior answer to a real problem beats any restaurant for a Top 5 account. They got more careful about add-on charges, treating anything that felt like a tax on normal product use as a problem. And they ensured the people talking to big customers were there to make them successful first, with expansion following from that rather than the reverse.

The dinner itself was fine, Lemkin notes. Customers like being taken care of. But a nice dinner can't be the main way a customer feels valued — it has to be the least important way.

The postscript: Groupon was getting a great deal. 2,000+ reps, deeply integrated into Salesforce, at a price that was very fair for what they received. It didn't matter what Lemkin knew. What mattered was what his champion felt at that table, and for a moment the dinner told him the opposite of the truth. As vendors layer usage-based AI pricing on top of seat-based contracts, the number of signals a customer can misread — or read correctly — is about to multiply.