Vayu launches a customer-by-customer margin view for AI vendors' finance teams

TNW introduces the launch with a scenario. An AI vendor's finance team reaches quarter-end and finds that its best customer on the revenue chart was a loss on the margin chart ^[8]. The customer had used the product heavily. Each use cost the vendor tokens, compute and third-party model calls, while the invoice went out at the agreed price ^[8]. The cost sat in a cloud bill nobody had matched to that customer ^[8].

Product teams tell themselves that heavy use means the product is working, and heavy use is what every vendor wants ^[8]. In practice, a heavy user of an AI product consumes something with every action: a model call, a GPU minute, an embedding, a third-party API ^[17]. On a fixed price, that user's cost keeps climbing and the revenue stays where it is. "A SaaS seat used to cost almost nothing to serve. Today a single heavy customer can wipe out the margin on an entire contract, and most finance teams find out a quarter later," said Erez Agmon, Vayu's chief executive and co-founder ^[6]. He runs the company selling the remedy. Vendors have responded by moving AI pricing to usage, credits and outcomes so that revenue follows consumption ^[16].

The pitch, in TNW's description, is a CFO typing "which customers are below margin this month" and getting the answer the same day instead of after a six-week reconstruction ^[15]. The actual job is joining data. What was promised, what was delivered, what can be billed this month and what it cost to deliver are four different numbers kept in four different places ^[9]. Vayu does the join in its metering layer and contract agents, and the Hub sits on top of them ^[2][11][13]. Anyone evaluating the Hub is mostly evaluating whether the layer underneath can tie cost to a single customer. "Pricing and cost now move together, every day, and the revenue engine has to see both," Agmon said ^[7].

Most of the evidence for the problem comes from Vayu. The survey is one Vayu co-produced ^[3], and its headline figure works out to about 58 of the roughly 100 respondents ^[1]. The launch account does not include a price for the Hub or margin results from any of the six customers Vayu names ^[1][2].

The Hub is built for a finance team whose costs rise with each customer action and whose pricing is spread across several systems ^[2][5]. For everyone else, two questions settle it: whether cost-to-serve rises per action, and whether price does. The old seat business has flat cost and flat price, and a quarterly spreadsheet still answers its margin question. A product with flat cost and usage pricing has a billing problem; metering matters there, and a margin layer adds little. With per-action cost and usage pricing, revenue moves with consumption, but margin depends on each tier's unit price staying above its unit cost. Vayu says the Hub flags tiers where cost-to-serve is eroding margin ^[14]. Per-action cost on a flat price is the case in Agmon's quote. There the first fix is the price itself, and customer-level margin shows which contracts to change at renewal.

The forcing function is a timed request. Finance is asked for last month's below-margin customers, and someone records how long the list takes. An answer within the week means the current stack copes. If the answer takes weeks and has to be assembled from three or more systems, the survey's majority case ^[5], then the buying decision is about the integration layer, and the chat interface on top matters less.