Shared-savings AI pricing: how outcome-based fees work

What is shared-savings AI pricing?

Shared-savings pricing is a fee arrangement where the AI provider is paid an agreed share of a measured financial result, such as labor hours saved, money recovered or costs avoided, instead of hourly rates or a fixed project fee. If the result doesn't materialize, the fee is reduced or disappears.

The idea isn't new. Energy service companies have long been paid from measured utility savings, and recovery auditors in accounts payable and telecom have long worked on contingency, taking a percentage of what they recover. What's newer is applying the model to AI implementation, where the buyer's biggest fear is paying for a project that never works.

You will see several names for variations of the same idea: outcome-based pricing, performance-based pricing, gain-sharing, contingency fees and value-based pricing. The common thread is that the fee follows a measured outcome, not effort.

How does outcome-based pricing work, step by step?

Every shared-savings arrangement rests on four agreements: what the baseline is, how the result will be measured, what share the provider earns, and for how long. Get those four right and the rest is administration.

1. Agree on the baseline. Measure the current state before anything changes: hours spent per week on the task, error rates, dollars lost to duplicates or missed discounts, or the current cost of a contract. Both sides should sign off on the numbers and the source data.
2. Define the measurement. Decide exactly which number will be tracked, where it comes from (time logs, accounting reports, carrier statements, refund checks) and how often it will be reported.
3. Set the share and the basis. Agree what percentage of the measured result the provider earns, and whether it is calculated on gross savings or net of running costs.
4. Set the term. Decide how long the provider shares in the result: a fixed number of months, a single payment on a one-time recovery, or an ongoing arrangement while the system operates.
5. Add caps, floors and exits. Many agreements cap total fees, define a minimum threshold before any fee is due, and spell out how either side can end the arrangement.
6. Reconcile on a schedule. At each reporting period, compare actual results to the baseline, calculate the fee and share the underlying data so you can check it.

What outcomes can be measured?

Shared savings only works for results you can count with data both sides trust. In small business AI, that usually means one of three types of outcome.

Outcome type	Examples	How it is measured	Typical fee structure
Recovered money	Duplicate vendor payments, health premiums billed for former employees, unapplied credits, carrier refunds	Refunds, credits or checks actually received	One-time percentage of the amount recovered
Cost avoided	Cancelled auto-renewals, reversed price increases, captured early-pay discounts, a lower SUTA rate	Difference between the old and new cost over a defined period	Percentage of the avoided cost over an agreed term
Labor saved	Hours no longer spent on draw requests, quote follow-up, invoice matching or reporting	Baseline hours minus current hours, times a loaded hourly rate	Ongoing share of measured savings while the system runs

Recovered money is the easiest to verify because a check or credit memo either arrives or it doesn't. Labor savings are real but softer, because they depend on time tracking and on what the freed-up hours are used for. Agree up front how you will treat that.

Illustrative example: how the math works

The percentages below are hypothetical, chosen only to make the arithmetic easy to follow. They are not a quote and are not Tarvora's rates. Real shares vary by provider and by how hard the result is to produce.

Illustrative example 1, a one-time recovery. A review of 12 months of accounts payable finds $9,000 in duplicate payments and unapplied vendor credits. Vendors refund or credit $7,500 of it; $1,500 can't be recovered. With a hypothetical 30% share, the fee is calculated on the $7,500 actually recovered: $2,250. The business keeps $5,250 it would otherwise never have seen. If nothing had been recovered, the fee would be zero.

Illustrative example 2, ongoing labor savings. An office team spends 10 hours a week on invoice matching at a loaded cost of $45 an hour, about $23,400 a year. After an agent goes live, the team spends 3 hours a week reviewing exceptions, about $7,020 a year. Measured savings are about $16,380 a year. With a hypothetical 40% share over the agreed term, the provider earns about $6,550 a year and the business keeps about $9,830, with no upfront build fee.

Notice that in both cases the contract terms matter more than the percentage: whether the fee is based on amounts found or amounts actually recovered, and whether labor savings are measured on reviewed time logs or estimates.

Shared savings vs. hourly, fixed-fee and subscription pricing

The main difference between pricing models is who carries the risk that the project doesn't deliver. With hourly and fixed-fee work you carry it; with shared savings, the provider does.

Model	What you pay for	Who carries the risk	Main advantage	Main drawback
Hourly	Time spent	You	Flexible scope	Open-ended cost, no link to results
Fixed project fee	A defined deliverable	Mostly you	Known upfront cost	Delivered does not mean working or saving money
Subscription or retainer	Access and maintenance	You	Predictable monthly cost	Fees continue whether or not value does
Shared savings	A share of measured results	The provider	Little or no upfront cost; fees follow value	Requires a clear baseline; total fees can exceed a fixed fee if results are large

That last drawback is worth taking seriously. If a project produces large savings, you may pay more under shared savings than you would have under a fixed fee. That is the price of transferring the risk, and a fee cap or a defined term limits it. For a wider view of AI pricing, see how much AI implementation costs for a small business.

What to put in a shared-savings contract

A good shared-savings agreement makes the fee calculation boring and checkable. Before signing, make sure the contract covers each of these points in writing.

• Baseline: the exact starting numbers, the period they cover and the source documents.
• Measurement method: which reports or records are used, who produces them and how often.
• Fee basis: percentage share, and whether it applies to amounts identified or amounts actually recovered or saved. Prefer the latter.
• Term: how long the provider shares in ongoing savings, and when the share ends.
• Cap and threshold: any maximum total fee and any minimum before fees apply.
• Attribution: how to handle savings that would have happened anyway, such as a vendor that was already refunding a known overpayment.
• Approval rights: you approve every contact with your vendors, carriers or customers before it happens.
• Ownership and data: you own what gets built and the findings, and your data isn't pooled, resold or used to train models for others.
• Exit: what happens to fees, systems and documentation if either side ends the arrangement.

This article is general information, not legal, tax or accounting advice. Have your attorney or accountant review any fee agreement, especially how recovered amounts and fees are recorded on your books.

Red flags in outcome-based AI deals

Most problems with shared-savings deals come from vague measurement. Be cautious if you see any of the following:

• Fees calculated on "identified" savings rather than money actually recovered or costs actually avoided.
• A baseline the provider sets alone, or one based on estimates rather than your records.
• An open-ended term with no cap, so you pay a share of the same savings indefinitely.
• Large "setup" or "onboarding" fees that turn a shared-savings deal back into a fixed fee.
• No right to approve outreach to your vendors, insurers or customers.
• Reports you cannot trace back to source documents.

For a broader checklist that applies to any AI provider, see our 12 questions to ask before hiring an AI implementation partner.

When shared-savings pricing is, and isn't, the right fit

Shared savings fits best when the result is measurable, the provider can actually influence it, and you would rather pay from results than from your operating budget. It fits poorly when the goal is hard to quantify.

Good fits: recovering overpayments in accounts payable or health insurance billing, cutting unwanted renewals, and automating a well-measured, repetitive task like invoice matching or pay app assembly.

Poor fits: general productivity tools, brand or marketing work where results have many causes, one-off strategy projects, and processes nobody has ever measured and nobody is willing to measure now.

Tarvora is paid a portion of the cost savings its work creates and measures: you see a working prototype or audit findings before paying, and you know the exact share before committing. Our Back Office Audit works on a contingency basis, so if nothing is found, nothing is owed.

Frequently asked questions