In late April, GitHub Copilot ended the flat-rate era of enterprise AI. The most-deployed AI tool in enterprise software moved every customer to a credits model, billed against direct token usage. Microsoft, after years of absorbing the variable cost of every developer’s experiment, handed that cost back to the customer. Cost spikes on heavy agentic workflows ran 60x overnight.
That’s the signal. If the largest deployment in the category can’t make flat-rate pricing work at scale, no one can. And from the conversations I’m having, most vendors aren’t asking if they’ll follow — they’re deciding when.
Which means CROs and CCOs are about to learn the hard way that credits are far more than a packaging decision. Comp plans break. Renewal forecasts drift. And under every closed deal, a “zombie liability” is accumulating: commissions paid out against revenue that exists on paper and nowhere else.
Credits are everywhere now. More than 30% of AI-native companies had adopted credits as a core monetization mechanism by early 2026, up from a small minority a year prior. Adobe, Salesforce, ServiceNow, Workday, Hubspot, OpenAI, and Microsoft are all in, all within 12 months. The customer experience is intentionally clean: one wallet, draw down as you go, top up when needed.
Your revenue team’s experience is the opposite. Every design choice that makes the wallet feel frictionless — pooled balances, rollovers, weighted rate cards, floating exchange rates — creates a new accounting problem in the back office.
Under a credit model, the question of when a rep gets paid stops being obvious. A rep closes a $5M credit commitment and gets paid on the full $5M. The customer uses 20% year one. The remaining $4M of unused balance remains a liability until the customer uses it or it expires — and the board is now staring at a gap between the CRO’s bookings number and the CFO’s recognized revenue.
That commission paid out against revenue that hasn’t and may never move is the zombie liability, and companies running consumption models the longest have already built around this. Snowflake pays reps on a hybrid of committed capacity and platform usage milestones. Databricks ties commissions directly to month-over-month consumption rather than the signature. AWS and Google Cloud pay a baseline of 60% to 80% on signed contract value, with the remainder unlocked over 12 months of verified usage. Each model rewards consumption velocity over paper bookings.
Starbucks figured this out the hard way. Customers load cash onto a digital wallet, and a predictable percentage of it is never spent. Starbucks now recognizes hundreds of millions a year in breakage revenue from those unspent balances. It took years of auditor negotiations to get there. Every credit-based AI company is arriving at the same question faster, with less runway to figure out the answer.
When credits inevitably expire, the short-term financial win of breakage revenue comes at a real cost to long-term customer success and NRR. Finance wants to book the breakage, while Customer Success wants to roll the balance forward. Both reflexes are rational — both are protecting something real — but both produce a different answer on the same customer call. And if there’s no policy, every conversation becomes a one-off, negotiated by whoever’s in the room.
Set the policy before the call happens: align Finance, CS, and Sales around a single customer answer, or you’ll be negotiating this every time.
Imagine an arcade where game prices are entirely hidden, and you discover at checkout that a single round of Skee-Ball costs five times what you anticipated. After that experience, most players will leave the arcade.
That’s how enterprise AI vendors price today. The per-action cost stays hidden until the monthly invoice. Engineers running agentic workflows burn hundreds of credits per action. Marketing runs a handful per week. Sixty days in, the whole company is locked out because they hit the annual credit cap, and the CRO who sold the deal inherits the dispute call.
Only 26% of companies have a comprehensive view of their AI costs, per a KPMG study reported in the Wall Street Journal this month. The rest are flying blind — and some are landing hard. Life360 discovered that a single user action in its app was triggering 50 to 200 backend calls to frontier models. The budget evaporated in 90 days. The fix is in engineering — caching, batching, rearchitecting the agent flow — but only if the customer can see the problem coming.
The visibility has to come from the vendor. Hide the rate table, and you’re not just making the invoice a surprise, you’re building churn into the point of sale.
Twelve months from now, there will be two kinds of revenue organizations: those that got ahead of the credits model, treating credits as a revenue architecture redesign, and those that spent the year explaining to the board why bookings and recognized revenue aren’t the same number. The difference between them is more about sequencing than resources.
Align comp with consumption. Pay reps on hybrid, milestone-gated structures — Snowflake, Databricks, and the hyperscalers are solid examples of this model.
Set the credit policy upstream. Resolve the Finance and CS tension on expiring credits now, so the answer to every expiry question is already decided before anyone picks up the phone.
Publish the rate table. Lock unit economics and put the per-action rate in writing. Customers who can plan are far less likely to churn.
Unify the ledger. Anchor the customer wallet to a shared system of record. When Finance, RevOps, and CS read different consumption data, every renewal is a negotiation over which number is real.
The revenue leader who architects this now sets the terms for how the category competes for the foreseeable future. The one who waits will spend that time explaining the zombie liability to the board, while their counterpart is closing renewals at 120% NRR.
Amy Konary runs the Subscribed Institute at Zuora, a think tank focused on how companies design and evolve their revenue models. She came up as an industry analyst at the dawn of SaaS and hasn’t stopped paying attention since. Amy loves finding the story in the data — and the data in the story.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.