If your sales-tech vendor bills you by the token, you are carrying a risk that is not yours to carry, and you should stop signing that contract renewal until the vendor moves the meter.
That is not a hypothetical. A recent Andreessen Horowitz analysis of AI pricing, written by Tugce Erten, a partner on the firm’s Growth team focused on pricing and packaging, and Sarah Wang, a general partner on the same team, lays out the argument directly: “we believe companies should price at the highest layer of value that they can reliably measure, attribute, and defend.” Their framework sorts AI pricing into three tiers, tokens for raw model access, credits for a recognizable unit of work, and outcomes for a clear, attributable business result, and it found that of 50 technical AI buyers surveyed, 27 preferred credits tied to real work over just 14 who preferred raw token pricing. Buyers, in other words, already know which meter they want. Most sales-tech vendors have not caught up.
Why this is a RevOps problem, not a pricing footnote
Token pricing does something specific to a buyer’s forecast: it imports the vendor’s cost structure directly into your budget line. When the underlying model gets cheaper, faster, or is swapped for a different one behind the scenes, your bill moves for reasons that have nothing to do with the value you got from the tool that week. A RevOps leader building a forecast on top of an AI sales-tech subscription priced this way is not forecasting the tool’s value to the business, they are forecasting a third party’s infrastructure costs, which they cannot see and did not negotiate.
Advertisement
A separate Madrona Venture Capital enterprise AI report adds the buyer-side half of this argument: enterprises are reevaluating their AI vendors roughly every six months, not locking into the multi-year commitments that used to define enterprise software. If vendors already face that short a leash, the “we have to price by token for cost transparency” defense gets weaker, not stronger. A vendor who expects to be re-litigated in six months has every incentive to show you a clean, defensible outcome metric now, while they still have your business.
The counter-argument, and why it does not hold
The strongest objection a vendor will raise is that outcome-based and credit-based pricing is operationally harder to administer at scale, and that token metering is at least transparent, you can see exactly what you consumed. That is a real operational cost, and it is fair to name it. But it does not answer the buyer’s actual complaint. Erten and Wang’s own framework addresses this directly: “transparency does not require the billing meter and the underlying cost meter to be the same.” A vendor can show a customer full usage detail behind the scenes while still billing against a credit or an outcome the customer actually cares about. The administrative argument is a reason pricing teams have to do more work, not a reason buyers should accept a bill they cannot forecast against.
What it means for the sales leader and RevOps buyer
Practically, this changes how a sales or RevOps leader should run the next AI sales-tech renewal conversation. First, ask the vendor to name the recognizable unit of work behind their token count, a qualified lead scored, a call summarized, a forecast updated, and price against that unit instead. If they cannot name one, that is itself informative about how mature the product actually is. Second, use the six-month reevaluation cycle Madrona describes as real leverage in the negotiation, not just a churn risk to manage defensively; a vendor who knows you can credibly walk in two quarters has much less room to hide behind a token count they control unilaterally. Third, build forecasts on committed credits or outcomes, never on a raw token estimate, so a model-cost change on the vendor’s end cannot silently move your number.
None of this requires AI sales-tech vendors to give up revenue. It requires them to price the thing the buyer is actually purchasing. Buyers who keep accepting token bills because switching feels like more work than it is worth are subsidizing a pricing model built for the vendor’s convenience, not their own forecast.
Related on SalesTech: Sales Software Pricing Is Fracturing, Not Ending and Stop Pricing AI Agents by the Seat They Do Not Fill.
Source: Andreessen Horowitz
