A CFO sat down for a low six-figure renewal with one of their software vendors. The vendor’s pitch leaned on the AI feature they had bundled into the contract last cycle. The CFO had a different question ready.
How much of this can we do ourselves with Claude?
The answer, after running it: about 95%. The Claude token cost to do it: roughly 15% of what the vendor was charging. The CFO churned the AI feature, kept the parts of the platform Claude could not replace, and walked out of the renewal with a 45% reduction in total spend on that vendor.
This is one renewal, written up by one CFO. It is not a market trend. But the conversation behind it is now happening at this CFO’s desk and many others, and it changes what every B2B software vendor needs to defend in 2026.
Claude Is the New Anchor Price
Procurement has always worked with anchor prices. The buyer asks: what does the comparable cost? The vendor’s job is to justify the gap.
For two decades, the comparable was another vendor. You priced against the alternative on the shortlist. The competitive set was knowable, the gaps were defendable, and the conversation was about features and roadmap.
The comparable is now Claude. Or GPT. Or the foundation model the buyer’s technical team can wire into a workflow over a weekend.
This is a different kind of anchor. A vendor competitor has overhead, sales cost, and a contract you can negotiate. A foundation model has none of those. It does not need to make margin on your account. It does not have a sales team to feed. The token cost is the floor, and the floor keeps moving down.
When the CFO does the math, they are not comparing your product to your nearest competitor. They are comparing your product to “what an internal engineer could prompt Claude to do in two weeks.”
That is a much harder benchmark to beat with a feature list.
The 5% Question
The CFO churned 95% of the AI feature. The remaining 5% was whatever Claude alone cannot deliver inside that vendor’s product. That 5% is the entire defensible surface area of the AI portion of the contract.
For most vendors, the 5% has not been named. It was bundled into the AI feature when bundling AI was a competitive necessity. Nobody asked which part of the bundle was actually defensible against a model the buyer could call directly.
The renewal cycle is now asking.
If you are a vendor CEO or product leader, the diagnostic is simple. List every AI capability in your product. Next to each one, write the answer to: what stops a competent engineer at the customer’s company from replicating this with a Claude API call and a weekend? If the answer is “nothing meaningful,” that capability is at the wrong end of the renewal conversation.
The defensible answers fall into a small set. Below is the working list.
The Defensible-Surface Table
| What Claude alone does well | What vendors can defend |
|---|---|
| Generic content and drafting | Compliance-stamped output |
| One-off analysis | Audit trail, change history |
| Single-user assistance | Multi-tenant, RBAC, SSO |
| API call from a script | Workflow integration, error recovery |
| Today’s accuracy | SLA-backed accuracy, eval pipelines |
Read the right column carefully. Every item is governance. Every item is the unglamorous infrastructure that makes AI safe to use at organizational scale. Every item is what the CFO cannot reproduce by handing a developer an API key and a long weekend.
This is the same observation we made in The AI Verification Tax, expressed from the other side of the contract. The cost of using AI well is not the cost of generating output. It is the cost of trusting it. Vendors who own the trust layer have something Claude cannot deliver alone. Vendors who own only the generation layer are selling against the price of tokens.
What the Counter-Anchor Says
A separate piece of analysis published the same week argued that the death of SaaS has been greatly exaggerated. The data point: the vast majority of software spend remains seat-based. Consumption pricing has barely grown.
Both stories are true at the same time, and the resolution matters.
Contract shape is sticky. The shift from per-seat to per-token is happening more slowly than the noise suggests. But contract size is not sticky. The CFO did not change the shape of the renewal. They cut the dollar value by 45% and kept the seat structure intact.
The pricing pressure shows up in the renewal number, not the contract model. Vendors who are watching for a dramatic shift to consumption pricing are watching the wrong indicator. The shift that is already happening is downward pressure on dollar value as Claude reframes what each seat is worth.
Re-Architect the Sell Motion Around the 5%
The strategic move for vendors is uncomfortable but clear.
Stop selling the 95%. The 95% is now a commodity, and pricing it as a premium feature is what makes the CFO churn it at renewal. If your AI capability is generic generation, drafting, or one-off analysis, it does not belong in the price ladder as a differentiator. It belongs as table stakes, priced near the cost of the underlying tokens, or unbundled entirely so the customer can bring their own model.
Sell the 5%. The 5% is governance, integration depth, certified compliance, multi-tenancy, audit trail, error recovery, evaluation pipelines, SLAs the CFO can show their board. This is what the customer cannot stand up alone, and this is what justifies the contract.
The renewal conversation a year from now will be about the 5%. The vendors who have already restructured around it will keep their accounts. The vendors who are still defending the 95% will discover that the CFO already did the math.
The CFO is not the threat. The CFO is the buyer telling you, in advance and in public, what the next renewal will sound like. The vendors who listen will redesign the product. The vendors who do not will lose 45% of the contract and call it pricing pressure.
It is not pricing pressure. It is the floor, and Claude set it.
This analysis builds on the OnlyCFO essay Pricing Pressure Will Crush You (April 2026) and the contrasting Death of SaaS Has Been Greatly Exaggerated (April 2026).
Victorino Group helps software vendors define and defend the part Claude alone cannot replicate. Let’s talk.
All articles on The Thinking Wire are written with the assistance of Anthropic's Opus LLM. Each piece goes through multi-agent research to verify facts and surface contradictions, followed by human review and approval before publication. If you find any inaccurate information or wish to contact our editorial team, please reach out at editorial@victorinollc.com . About The Thinking Wire →
If this resonates, let's talk
We help companies implement AI without losing control.
Schedule a Conversation