The CFO Veto: 49% of Software Buyers Lost an Approved Deal to Finance Last Year

49% of B2B software buyers say their CFO reversed a purchase the buying team had already approved in the last 12 months. That is the central finding of G2's 2026 Buyer Behavior Report, a June 2026 survey of 1,038 decision-makers. A CFO veto on a software purchase is no longer an edge case. It is a coin flip, and the report tells you roughly where in the cycle the coin gets tossed.

What the 49% CFO veto figure measures, and what it does not

Read the wording carefully, because most of the coverage has not. G2 asked buyers whether their CFO had reversed a deal the buying team had already approved in the last 12 months. 49% said yes. That is a share of buyers who experienced at least one reversal. It is not a share of purchases, and it is not a count of deals.

Half of the people you sell to have watched finance kill a deal their own team had signed off in the past year. One reversal is enough to change how a buyer behaves on the next 5 purchases, which is the part that matters for anyone selling to the CFO.

The survey itself is an online poll of 1,038 B2B decision-makers, from individual contributors to the C-suite, across North America, EMEA and APAC, fielded in June 2026. The report is titled The Evaluation Maze, and the title is the thesis: discovery got faster, approval got slower.

Direct answer

A CFO veto on a software purchase is finance reversing a deal after the buying team approved it. In G2's June 2026 survey of 1,038 buyers, 49% had experienced one in the last year. The reversal lands at budget approval, after vendor selection, and buyers who have been through it demand six-month payback and sub-12-month terms.

Where in the B2B software buying process the veto lands

G2 splits the journey into three stages, research, evaluation and decision, and asks buyers which one takes longest. Then it asks a separate question about what delays a purchase after a vendor has been selected. The veto shows up in the second question, and that placement is the most useful thing in the report.

Stage 1: research is now the short part

Only 36% of buyers name research as their longest stage, down from the year before. Review sites (38%) and AI chatbots (37%) are now level as the top shortlist influences. Finding candidates is fast. That is the part AI fixed.

Stage 2: evaluation is the longest stage for 40%

Evaluation overtook research as the longest stage for the first time, rising to 40% from 36% a year earlier. Decision trails at 22%. This is where the team forms the view that finance will later be asked to fund. How that shortlist gets built, and why 94% of buyers now start it before talking to a vendor, is covered in the analysis of the AI-built shortlist.

Stage 3: after selection, two gates and a planning step

Here is where the deal dies. Asked what delays a purchase after a vendor is chosen, buyers name IT security review first at 39%, rising to 50% among enterprise buyers. Budget approval is second at 32%. Implementation planning is third at 25%. The CFO reversal is by definition a budget-approval event: the team has approved, the vendor is chosen, and finance says no.

Why finance moved upstream: the token budget link

The report's sharpest finding is not the veto rate. It is what predicts it. Among organisations with a dedicated token or LLM budget, 54% had a CFO reverse an approved purchase. Among organisations without one, the figure is 29%. Nearly double.

80% of buyers say their organisation gives developers or technical teams a dedicated token or LLM usage budget. Once a company has a line item whose cost moves with consumption, finance has a reason to look at every adjacent software purchase. G2 puts it plainly: stronger AI spending creates more necessity for financial scrutiny, not less willingness to buy.

That is the mechanism behind the 31% to 46% jump in finance involvement, and the 32% to 25% fall in security involvement over the same year. In G2's phrasing, the gate moved from "is it safe" to "is it worth the potential cost overruns". The security question still tops the post-selection delay list. It has stopped being the question that reverses deals.

What the Zylo data adds

G2 does not measure whether the overrun fear is justified. Zylo's 2026 SaaS Management Index does. 78% of IT leaders reported unexpected charges tied to consumption-based or AI pricing models in the last 12 months, and 61% cut projects because of unplanned SaaS cost increases. The sample is 218 IT leaders, which is small, and Zylo sells spend-management software, so read it as directional. It still points the same way as G2: finance is reversing deals because recent deals cost more than the approval said they would.

What a vetoed buyer asks for next

A reversal changes the buyer. G2 isolates respondents who have experienced a late-stage veto and compares their demands with everyone else's. The differences are large and they all point at contract structure.

Six-month payback becomes the test

3 in 4 buyers who have been through a veto now expect positive ROI within six months of signing a contract. Not 12, not 18. If your ROI model shows breakeven at month 14, it is not a model the champion can carry into the budget meeting. My position is that you should publish the six-month case honestly, including the scenario where it does not clear, rather than stretch assumptions to hit it. A finance reviewer who finds one stretched input discards the whole document.

Terms get shorter across the board

70% of all buyers say the pace of AI innovation is pushing them toward shorter contracts, and 29% want terms under 12 months. In the vetoed group, 79% lean shorter and 40% strongly want under 12 months, against 18% of other buyers. The three-year prepaid deal with a discount is exactly the structure a vetoed buyer has learned to refuse. Which concessions buyers are actually winning on usage-based terms is set out in the review of usage-based pricing concessions.

Pricing structure is already in play. 49% of buyers say a current vendor has offered them a variable-cost option based on outcomes, consumption or tokens, and another 42% have been told it is coming. What finance negotiates once the structure is variable, and why discount ranks last, is in the breakdown of the six AI contract clauses CFOs actually fight over.

The finance artifacts that survive a CFO software purchase review

G2 did not ask which documents saved a deal, and nothing below should be read as if it did. The report gives you three things. The delay data: security 39%, budget 32%, implementation 25%. The post-veto demands: six-month ROI, sub-12-month terms, variable pricing. And one qualitative finding, that vetoed buyers over-index on customer proof and third-party validation. The checklist maps an artifact to each. The mapping is mine, and the table says which is which.

"Basis" is the honesty column. Nobody has measured which of these artifacts actually changes a veto outcome, including G2.

The consumption ceiling is the one most proposals omit

Of the seven, the TCO ceiling is the artifact I would build first, because it addresses the thing that produced the veto. A CFO who has been surprised by a token bill is not asking whether your product works. They are asking what the worst month looks like and whether they will find out before or after it happens. A TCO table that stops at the expected case answers neither question.

Customer proof has to carry a sample. Third-party validation is the only artifact G2 explicitly links to vetoed buyers. A case study with a 40% improvement and no baseline, period or exclusion list is marketing. A reference with a stated sample survives a finance review because it can be checked. Why the format still works when it carries verifiable numbers is examined in the piece on case studies with verified outcomes.

The security pack is a closing document now. G2's phrasing is that security documentation is a closing tool, not a compliance checkbox. I agree, with one addition: the AI disclosure matters more than the certificate. 87% of buyers say they are more likely to buy from a vendor offering transparent AI than from a cheaper black-box competitor.

Finance-facing content: what to publish before the deal exists

Everything in the checklist can also be public. Publishing it moves the finance conversation earlier, into the evaluation stage you can still influence, and it puts the material where the agents and chatbots doing TCO comparisons can read it.

A grid where you win everything is read by a finance reviewer as a grid they cannot trust, and by an AI agent as a page with no information in it. The structure that gets cited is set out in the template for B2B comparison pages. The argument for publishing metering rules is in the proposed transparency standard for rate limits and pricing.

The 9% agent ceiling, and why it matters for finance content

The second headline is about AI agents in the buying process. It is more conservative than the agent discourse suggests. 61% of buyers currently use or plan to use agents somewhere in software buying. The tasks they trust them with are the analytical ones: comparing total cost of ownership (51%), building shortlists (51%), researching solutions (49%) and evaluating shortlisted vendors (46%).

47% would let an agent research and recommend while humans make every final decision. 21% would limit it to research only. 9% are comfortable letting it execute purchases within approved guardrails, and 2% would allow purchases without pre-approval. Less than one-third will let an agent run unsupervised on anything.

The practical consequence is that the agent is a reader, not a buyer. Suppose your pricing is behind a form, your TCO assumptions sit in a PDF image and your security disclosure is a logo wall. The agent's comparison will then be built from your competitor's numbers and a guess about yours. The finance artifacts above need to be machine-legible for that reason alone, before any argument about search.

I would also resist the conclusion that the 9% is a floor that will rise quickly. The same survey shows internal resistance to AI adoption growing from 16% to 29% in a year, the largest single-year shift in the whole study. Buyers are using agents more and trusting them with authority less.

Where this argument is weakest

The post rests on one survey, and that is the first problem.

One vendor, one poll, self-reported

Every headline number here comes from a single G2 survey of 1,038 respondents, fielded online in June 2026. G2 is a review site that sells to software vendors, and the same report finds review sites to be the top shortlist influence, one point ahead of AI chatbots. That is not evidence of bias, but it is a reason to want a second source for the veto rate, and none exists yet.

"Reversed" is doing a lot of work. A buyer could reasonably count a deal that was delayed a quarter, cut in scope or sent back for a shorter term. The report does not separate a reversal that killed the purchase from one that reshaped it. If most vetoes are reshapings, the right response is a better term-length menu, not a crisis. The data cannot tell you which.

The counter-case: the veto is working as intended

There is a reading in which none of this is a problem to solve. Forrester's 2026 State of Business Buying research puts procurement professionals as decision-makers in 53% of business buying cycles, engaged from the start. It also finds the typical decision now involves 13 internal stakeholders and 9 external influencers. In a year when 78% of IT leaders were surprised by consumption charges, a late reversal may be the right outcome. A finance function that reverses one in two team approvals may simply be the control the other 12 stakeholders failed to be. On that reading the vendor's job is to be the one whose numbers the CFO did not have to reverse. I hold that view, and it is the reason this post is a checklist and not a sales play.

Frequently asked questions

How to build the finance pack this week

Start from the veto, not from the deck. Pull the last 3 deals that stalled or shrank after your champion had said yes, and write down the question finance asked in each case. In my experience it is one of three: what does the worst month cost, when does this pay back, and who else like us has the number.

Then build the consumption ceiling before anything else. One page, three usage scenarios across three years, a hard cap and a notification rule. If a prospect's finance team can find the worst case themselves, before the budget meeting, you have moved the veto conversation to a stage where you are still in the room.

Last, add a 12-month option to your order form and price it honestly. 40% of the buyers who have been vetoed strongly want it. Refusing to offer it does not make them sign for three years. It makes them pick a vendor who did.

Related reading