Strategy & Technology Analysis  ·  shashi.co  ·  February 28, 2026

Who Pays for the AI?

The pricing debate every enterprise buyer will face in 2026 — and why the answer is harder than vendors make it look.

Bessemer Venture Partners recently published a primer on AI pricing. It is worth reading if you are buying, building, or advising on AI products. The core argument: the software industry is shifting from selling access to selling outcomes. Every inference costs real compute money. Unlike traditional software, where one more user costs almost nothing to serve, artificial intelligence has a real cost of goods sold baked into every interaction.

That is all true. Reading it as a buyer rather than a founder, though, a different set of questions comes into focus.

What Enterprises Actually Want

Most enterprises want fixed pricing. A number on a purchase order. A budget line that does not surprise the Chief Financial Officer mid-year. The appeal of Software as a Service was never just the cloud — it was the predictability. Twelve months. One number. Done.

Artificial intelligence breaks that deal almost immediately.

The problem is genuine unknowns. A company deploying an AI agent for customer support does not know in January how many tickets it will handle in October. A legal team using AI to draft contracts does not know how complex the cases will get. A finance team running AI over invoices does not know if volumes will double after an acquisition. Usage fluctuates. And because AI inference has real compute costs attached to each query, vendors cannot absorb that variability the way a traditional Software as a Service company absorbs an extra user.

So what emerges, almost inevitably, is the hybrid model: a base subscription for predictability, plus a usage tier for the variability neither side can forecast. It covers the vendor's risk. It gives the buyer a floor. But it creates a ceiling no one can see in advance.

Fixed plus usage is the pricing model that makes sense on a whiteboard and drives Chief Financial Officers mad in practice.

The usage overage invoice is not a technical problem. It is a budgeting and trust problem. Enterprises run on annual budgets approved months before the technology is actually used. An overage landing in the third quarter is not just an accounting headache — it is a conversation with a Chief Financial Officer who approved a number that is no longer the number. The Bessemer report notes that Chief Information Officers frequently ask vendors to push overage invoices into the following year's budget. That is not a rounding error. That is a structural mismatch between how AI is priced and how enterprises plan.

Three Charge Metrics — and Where the Real Tension Lives

Bessemer outlines three charge metrics emerging across the industry, each representing a different trade-off between cost predictability and value alignment.

⚡ Consumption-based Per token, per API call, per inference. Clean economics for the vendor. Opaque and anxiety-inducing for every buyer outside the technical team. Works for developers only

⚙️ Workflow-based Per completed task — a meeting booked, a contract drafted. Clearer value, but one complex task can cost ten times a simple one. Middle ground

🎯 Outcome-based Per resolved ticket, per letter generated, per case closed. Tightest alignment with value — but only when buyer and vendor agree on what "resolved" means. Best alignment, highest risk

Outcome-based pricing gets the most attention, and rightly so. Intercom charges $0.99 per ticket its AI agent resolves. EvenUp charges per demand letter generated. The logic is appealing: you only pay when value is delivered. Chief Financial Officers can model it. Return on investment is visible.

But this model carries a hidden dependency that pricing playbooks tend to gloss over.

The Agreement Problem Nobody Talks About

Outcome-based pricing only works when both parties agree, in advance and in writing, on what an outcome actually is.

That sounds obvious. It is not simple. What counts as a "resolved" support ticket — one the customer closed, or one they did not reopen within 48 hours, or one where the AI gave a substantive answer regardless of whether the customer was satisfied? What counts as a completed legal document — a draft, a reviewed draft, a signed document? What counts as a successful sales lead — a contact added, a meeting booked, a deal created in the system?

These are not philosophical questions. They are the exact disputes that will surface in contract reviews and renewal negotiations. Vendors have an incentive to define outcomes broadly. Buyers have an incentive to define them narrowly. Without precise upfront agreement, outcome-based pricing simply transfers the conflict from the invoice to the relationship.

The 2026 Renewal Cliff Bessemer flags something worth sitting with: much of 2025 ran on soft Return on Investment — buyers adopted AI on promise, not proof, with minimal price scrutiny. Those pilots are now hitting renewal cycles for the first time. Pricing that was easy to approve when everything was experimental becomes hard to defend when the Chief Financial Officer wants to know exactly what was delivered. Vendors who defined outcomes vaguely will feel this first.

What Buyers Should Do Before Signing

The Bessemer playbook is written for founders. Here is the translation for the enterprise buyer's side of the table.

1 Define the outcome in writing before agreeing to outcome-based pricing — not the spirit of the outcome, but the exact, measurable, auditable definition. What system records it. Who can dispute it. What happens when the AI completes a task incorrectly and the human still has to step in.

2 Ask the vendor to model three scenarios — low, expected, and high usage — and show you the invoice in each. Then ask what controls you have if actuals track toward the high scenario. Budget caps, usage alerts, and approval gates are reasonable asks. Vendors who push back on these are telling you something important.

3 Close the internal gap between the person approving the purchase order and the team whose behaviour drives the usage. In hybrid models, make sure everyone knows which portion is fixed and which is variable — before the overage invoice arrives.

4 At renewal, ask for data — not promises. What was the actual outcome volume? What was the resolution rate? What would this have cost under an alternative model? Sustainable pricing should feel like a no-brainer in retrospect, not a mystery.

---

The Bessemer piece ends with confidence: AI will monetize outcomes. That is probably right as a direction of travel. But the journey from here to there runs through a lot of poorly-defined contracts, surprised Chief Financial Officers, and renewal conversations that nobody planned for carefully enough.

Outcome-based pricing works. The condition is that both sides first do the hard, unglamorous work of agreeing on what the outcome is. That conversation is less exciting than the technology. It is also the one that determines whether the deal survives its second year.

Works Cited  ·  MLA 9th Edition Bessemer Venture Partners. “The AI Pricing and Monetization Playbook.” BVP Atlas, Bessemer Venture Partners, 9 Feb. 2026, www.bvp.com/atlas/the-ai-pricing-and-monetization-playbook. Accessed 28 Feb. 2026.