Two years of AI investment, and the bill is coming due.
The spending is not slowing. The proof is not arriving. And a handful of the largest AI providers have just started doing something that should make every revenue leader sit up: they are offering to get paid only when the work actually gets done.
That is not a pricing experiment. It is a transfer of risk. And most go-to-market organizations are not instrumented to survive it.
The Proof Gap Is Now a Board-Level Problem
The numbers tell a consistent story across three separate studies published in the last month.
77% of finance organizations have deployed AI, but only 35% can confidently gauge the return. Worse, only 14% are deploying AI against a detailed strategy. Deployment has outrun definition.
The buyer side feels the squeeze. 92% of finance leaders report pressure to show that their AI investment yields a decent return. They are being asked a question they cannot answer with the instrumentation they have.
And the pilot graveyard keeps filling. 72% of executives say they have successfully scaled fewer than a quarter of their AI pilots, while two-thirds say their organization struggles to measure AI’s ROI at all. Only about half report having a balanced KPI framework for evaluating AI value. Nearly three-quarters say the scramble for short-term ROI is actively crowding out the transformative work.
Read that last finding again. The absence of a credible measurement discipline is not just a reporting problem. It is killing the ambitious initiatives that would have produced the returns.
Someone Is Getting Value. Very Few Can Prove It.
Here is where it gets interesting for anyone selling into this market.
McKinsey’s latest state-of-AI research found 80% of respondents reporting gains in individual productivity and 50% reporting improved decision making. Scaling is real at the top end: 40% of organizations above $1 billion in revenue are now scaling AI agents, up from 27% a year ago.
Yet just 6% report AI contributing at least 5% of EBIT. Meanwhile 20% have reduced AI usage because of operating costs, as token consumption becomes a line item someone has to own.
So value is being created at the desk level and evaporating before it reaches the P&L. McKinsey’s own framing captures the gap precisely: conviction in AI is growing faster than the financial returns organizations can attribute to it.
Attribute. That is the whole ballgame. The value is happening. Nobody built the chain of evidence to claim it.
Outcome Pricing Turns Your Value Gap Into a Revenue Gap
Which brings us to the shift.
OpenAI has reportedly begun letting select large customers pay only when its AI completes assigned tasks. Salesforce has been testing comparable models. The logic is straightforward: when customers cannot prove ROI, the vendor’s incentive to sell the result rather than the consumption becomes overwhelming.
If you sell software or services, absorb what this means. In a seat or token model, your revenue is protected by a contract. In an outcome model, your revenue is protected by evidence. The measurement system stops being a marketing asset and becomes the invoice.
Most vendors are nowhere near ready. They sell on a business case built in the deal cycle, hand the account to Customer Success with none of the promised metrics attached, and then rediscover the question ninety days before renewal.
The outcome economy does not tolerate that operating model. Neither, increasingly, do buyers.
Three Things to Do Now
- Retire the business case. Move to a Shared Value Plan.
A business case is a document you build to win approval. It is authored by the vendor, consumed by a committee, and abandoned at signature. A Shared Value Plan is a jointly owned agreement: which metrics matter, what the baseline is, who is accountable on each side, what the measurement cadence looks like, and what “success” numerically means at 90, 180 and 365 days.
The distinction matters more than the semantics. A business case is a promise. A Shared Value Plan is a contract of evidence, co-signed. When your champion moves on, and they will, the plan survives them.
Start with your top ten accounts. If you cannot state the agreed target metrics for each one without opening a slide deck, you do not have a value plan. You have a sales artifact.
- Instrument the solution to produce its own evidence.
You cannot quantify what you did not capture. Most value programs fail here, not in the modeling.
Go back to your product and services teams with a specific ask: for each outcome we claim in the market, what telemetry proves it, and is that telemetry available to the customer’s own team in a form Finance will accept? Cycle time reduced. Error rates. Win rate by cohort. Deflected volume. Hours redeployed, and where they went.
Note the second half of that requirement. The Protiviti research surfaces exactly why it matters: leaders can measure accuracy, time and money, but improvements in user experience or employee satisfaction stay stubbornly subjective. Build for the metrics that survive a CFO’s scrutiny, and treat the softer signals as supporting narrative rather than the core claim.
If the instrumentation does not exist, that is a roadmap item, not a value engineering problem. Raise it as such.
- Build the pre-to-post handoff and the Value Review discipline.
The handoff is where quantification goes to die. Sales commits to a value hypothesis. Nobody transmits it. Customer Success starts from adoption metrics because those are the ones they can see.
Fix it structurally. No account transitions without the Shared Value Plan, the baseline data and the named accountable owners on both sides. Make it a gate, not a courtesy.
Then replace the QBR ritual with genuine Value Reviews. A QBR reports on activity: tickets, usage, roadmap, health score. A Value Review reports on the agreed metrics against baseline, with variance explained and corrective action agreed. Quarterly at minimum, and never for the first time in the renewal quarter.
The organizations that will win outcome-based deals are the ones already running this cadence, because they have a track record of realized value to point at when a prospect asks whether the numbers hold up.
The Question Worth Sitting With
If a major customer came to you next quarter and proposed paying only for proven outcomes, could you say yes?
Not “would we want to.” Could you. Do you know what to measure, is it instrumented, is the baseline captured, and is anyone accountable for reviewing it between renewals?
The 65% who cannot confidently gauge AI ROI are not going to stay comfortable in that position. Their CFOs are already asking. The vendors who can answer on their behalf will take the market from the vendors who cannot.
Value realization stopped being a differentiator. It is becoming the price of entry.
Let’s discuss your value program and practices, so we can help get you through the AI ROI Reckoning with great success: Click here to schedule a consultation with us.
Sources:
- Protiviti Global Finance Trends Survey (902 finance leaders), reported by CFO Dive, August 27, 2026: https://www.cfodive.com/news/only-35-percent-finance-leaders-confidently-gauge-roi-ai-protiviti-cybersecurity-it/829015/*
- Infosys AI ROI research (1,000+ senior business executives), reported by CIO Dive, August 26, 2026: https://www.ciodive.com/news/CIOs-waiting-AI-savings-ROI/828883/
- McKinsey State of AI research, reported by TechRadar Pro, August 28, 2026: https://www.techradar.com/pro/well-its-about-time-mckinsey-report-says-ai-is-on-the-road-to-roi-at-last
- Avalara Finance Leaders survey, cited in CFO Dive, August 2026.
- OpenAI and Salesforce outcome-based pricing tests, originally reported by The Information, via Stocktwits, August 31, 2026: https://stocktwits.com/news-articles/markets/equity/open-ai-reportedly-joins-salesforce-others-in-testing-outcome-based-ai-pricing-astra-launch-in-focus/cZYyoe0RJs5