Gain-share and outcome-based pricing for AI transformation: how it works and what to check in the contract
Gain-share pricing moves AI delivery risk to the vendor, but it is only honest if your team owns the baseline, the attribution rule and the sign-off.
Gain-share, outcome-based and value-based pricing all mean the same thing: the vendor is paid from a measured improvement rather than for hours or deliverables. It moves delivery risk from the buyer to the vendor. Whether the deal is honest depends on five things: the baseline, the measurement window, what counts as caused by the work, who signs off the number, and when payment follows. Get those wrong and you have bought time and materials with a better name.
Three pricing families, three risk positions
Time and materials, or the retainer, is the oldest model. You pay for time and carry the whole risk, and the vendor's incentive is for the work to continue. HBR's guide for consultants calls hourly billing useful when nobody knows how long a project will take (HBR, 2019).
A fixed fee per deliverable moves one risk: the vendor carries the cost of building the thing. You still carry the risk that it does nothing. KPMG notes that outcome-based contracting is commonly confused with output-based models, which is what most fixed-fee deals are (KPMG, 2022).
Gain-share is the third family. HBR describes it as a model in which the consultant makes nothing unless the client improves, and can usually charge a premium for it. Business Insider reported in November 2025 that about a quarter of McKinsey's global fees come from outcomes-based pricing (Business Insider, 2025).
The table is illustrative.
| Dimension | Time and materials or retainer | Fixed fee per deliverable | Gain-share or outcome-based |
|---|---|---|---|
| Upfront cost to the buyer | Full, from day one | Full or staged, before value shows | None, or a small mobilisation fee |
| Who carries delivery risk | Buyer | Shared: vendor builds, buyer bets it matters | Vendor |
| Incentive to pick the highest-value work | Weak; any work bills | Weak; scope is fixed in advance | Strong, if the vendor is held to the ranked list |
| Incentive to finish and make it stick | Weak; finishing ends the billing | Ends at handover | Strong; nothing is paid until the number is real |
| How value is verified | Usually not | Acceptance of the deliverable | Client measures against a baseline and signs off |
What the contract has to define
The baseline is the number before the work, measured the same way afterwards and frozen at signing. Twelve months washes out seasonality. The measurement window is how long after go-live the improvement is counted, and whether as actual savings or an annualised run rate.
Attribution is the rule for what counts as caused by the work. Volume fell, a supplier cut its price, two people resigned and were not replaced: none of that is the vendor's. KPMG calls demonstrating impact the biggest challenge, because the link from the work to the KPI has to be undeniable and demonstrable.
Verification is who signs off the number. The honest answer is your finance team, with the vendor's calculation as a proposal. Deloitte's accounting guidance on outcome-based AI contracts asks the same of vendors: the contract has to define a successful outcome specifically enough that both parties can tell when it has occurred, or revenue cannot be recognised as the outcomes arrive (Deloitte, 2026).
Payment follows verification, in arrears. Payment on go-live is a fixed fee, whatever it is called.
Where these deals go wrong
Baseline gaming comes first: the vendor who helps set the baseline picks the expensive quarter and values hours at a rate your controller would not recognise. Then come savings that would have happened anyway. The AP clerk was retiring in March, the automation went live in April, the vendor claims the headcount.
Attribution disputes follow when two initiatives touch the same process, which in a real company is always.
Cherry-picking is subtler. A vendor paid on results gravitates to the fix everyone knew about, the invoice matching in the shared inbox, and away from the planning process run out of a private spreadsheet, worth three times as much and twice as slow. Lock-in: if the automations run on the vendor's platform and the savings are read off the vendor's dashboard, you cannot leave and you cannot check.
The larger risk is the pilot that never reaches the P&L. Fortune's account of the MIT NANDA study puts the share of generative AI pilots that achieved rapid revenue acceleration at about 5 percent (Fortune, 2025). A gain-share deal on one of the others pays nothing, but only if the contract says so.
What a CFO should insist on
Put the baseline in the contract as a schedule drawn from your own ledger. Describe the measurement method so a new controller could reproduce it without calling the vendor. Write the attribution rule before the work starts, with named exclusions for volume, price and headcount changes the work did not cause. Make verification a step your finance team owns, with disputes defaulting to no payment until the number is agreed, and pay only after sign-off. Oblige the vendor to work the ranked list of opportunities rather than choose its own. Keep the code, the data and the measurement in systems you control. Refuse any minimum fee or true-up that turns the deal back into a retainer.
Our view
Gain-share is only honest when the client's own team verifies the number and payment follows that verification. A gain-share deal where the vendor measures the gain is a retainer with extra steps. The measurement schedule tells you more about a vendor than the pitch does.
Whatever the commercial model, what makes an honest deal workable is discovery before building. Transcript, built by Lightbloom AI, values every problem in the company's own hours and cost per hour before anything is built, so the baseline is written down before the build starts.
Questions people ask next
Is outcome-based pricing the same as value-based pricing?
Mostly. Value-based pricing usually means a fixed fee set with reference to expected value, paid whatever happens. Gain-share means the fee itself moves with the measured result.
What if the value is real but hard to measure?
Convert it into something the ledger already tracks: rework hours, error rates that cost money, days of working capital. If it cannot be expressed in a number finance already reports, leave it out.
Should the vendor share in revenue uplift or only in cost savings?
Cost savings are easier to attribute and verify, so start there. Revenue uplift has too many parents: pricing, sales hiring, the market.
What happens when the numbers come in low?
In a genuine gain-share the vendor is not paid, and the contract should say so plainly. Watch for a minimum fee, a mobilisation charge that grows, or a right to re-baseline.
How Lightbloom AI works with companies is set out at /service-offering.
References
- HBR, 4 Ways for Coaches and Consultants to Price Their Services (2019), https://hbr.org/tip/2019/09/4-ways-for-coaches-and-consultants-to-price-their-services
- KPMG, Outcome-Based Contracting in IT: finding the Holy Grail? (2022), https://assets.kpmg.com/content/dam/kpmgsites/ch/pdf/outcome-based-contracting-it.pdf.coredownload.inline.pdf
- Business Insider, AI is reshaping how McKinsey makes money, as syndicated by Yahoo Finance (2025), https://finance.yahoo.com/news/ai-reshaping-mckinsey-makes-money-195132745.html
- Deloitte, Technology Spotlight: Accounting for Outcome-Based Pricing in an Agentic AI Software Product (2026), https://dart.deloitte.com/USDART/home/publications/deloitte/industry/technology/accounting-outcome-based-pricing-agentic-ai
- Fortune, MIT report: 95% of generative AI pilots at companies are failing, reporting on MIT NANDA, The GenAI Divide: State of AI in Business 2025 (2025), https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/
Keep reading.
All notes- How do you capture tribal knowledge before someone leaves? Tribal knowledge is how the company actually runs, held by the people who run it. Why handovers fail, what works, and why it is the first layer of a world model. Read →
- Why AI agents need a world model of your company AI agents stall inside companies because nobody has described how the work runs. What a company world model gives an agent, and why to build it before you deploy. Read →
- What is a company world model? A company world model is a living picture of how a company really works: every process as it runs, what hurts most, and who everything depends on. Why AI agents need it. Read →