Partly, and only for results you can measure cleanly. A fixed fee for the work plus a capped bonus tied to one agreed measure usually serves a small business best. Pure payment on results rarely does: savings are hard to attribute, and a consultant carrying that risk will price it in, so the total can exceed a fixed quote.
The results that suit a success fee are narrow: something the consultant controls, that you can count from your own records, against a baseline agreed before work starts, over a period long enough to smooth out the seasons. Most small-business AI projects meet two of those four conditions at best, which is why the bonus should stay small and the definition tight.
Six ways to pay, and what each costs you in the end
Results-based pay is one of several structures, and it helps to see them side by side before judging it. The internal-time column is the one owners forget: every structure that depends on measurement costs you hours as well as money.
| Structure | How it works | Your internal time | Who carries the risk |
|---|---|---|---|
| Fixed fee | An agreed price for defined deliverables | Agreeing scope up front | Consultant, on effort; you, on results |
| Hourly or day rate | You pay for time spent | Reviewing timesheets | You, on both effort and results |
| Fixed fee with holdback | Part of the fee is held until acceptance tests pass | Running the tests | Shared, on delivery |
| Fixed fee plus capped bonus | A bonus if one agreed measure hits its target | Measuring one figure | Shared, on results, with a ceiling |
| Gain-share | A percentage of measured savings or revenue for a period | Monthly reconciliation | Shared, with no ceiling unless you add one |
| Pure contingency | Paid only if results arrive | Monthly reconciliation and likely disputes | Consultant up front; you, at the top end |
If you're still choosing between the first two, fixed-scope vs hourly AI consulting covers that decision. The rest of this tutorial is about the bottom four, where payment depends on something beyond the work itself.
What the consultant needs to accept a success fee
A results clause is only fair when the consultant can actually move the result. Before proposing or accepting one, check these five conditions:
- Control: the measure depends mainly on the consultant's work, not on your pricing, stock, marketing or staffing.
- Access: the consultant can see the data needed to track it, without you handing over more than is necessary.
- A stable baseline: last year's or last quarter's figure is a fair comparison, with no big change in between.
- A long enough period: long enough to smooth out busy and quiet weeks, short enough that other changes don't swamp the effect.
- A cap: a maximum bonus, so a result you'd have got anyway can't produce a bill you didn't budget for.
When any one of these is missing, a success fee turns into a bet on factors neither side controls. The toy shop below shows the most common one to go missing.
A toy shop's Christmas problem with results-based pay
The shop and its figures are invented; the trap is worth planning for anyway. An independent toy shop with a Shopify store asks a consultant to set up AI-drafted product pages, an AI agent for routine customer questions and abandoned-basket emails. The consultant offers three ways to pay:
- Option 1: a fixed $3,000.
- Option 2: $1,500, plus 15% of the increase in online revenue from September to February compared with the same months last year.
- Option 3: $2,000, plus a $750 bonus if, over eight weeks, at least half of routine customer questions are answered without staff involvement.
Option 2 looks attractive: half the fixed price, and the consultant only earns more if the shop does. Here's how it plays out. Last year's online revenue for those six months was $60,000. This year, a toy line becomes the craze of the season and online revenue reaches $78,000. The increase is $18,000, the consultant's share $2,700, and the total $4,200: $1,200 more than the fixed fee.
How much of the $18,000 did the AI work cause? Nobody can say. The craze would have sold anyway; the product pages may have helped a little; the abandoned-basket emails recovered some sales, which the shop can count, but not most of them. Option 2 paid the consultant for Christmas. It fails the control test and the stable-baseline test at once, because a toy shop's revenue is dominated by the season and by whatever happens to be popular.
Option 3 measures something the consultant does control: whether the agent handles routine questions. The shop can count it from its own inbox, the cap is clear, and the most it can cost is $2,750, less than the fixed fee. The owner took Option 3 with one change: the eight weeks run from mid-January, after the peak, so the measure isn't distorted by the busiest weeks of the year. The shop also kept its own tally of recovered baskets. It wasn't part of the fee, but it told the owner whether that part was worth keeping.
Cheap, middle and generous structures, priced out
The clearest way to compare structures is to price each one at three outcomes. Take a $3,000 project and a measure of gain (savings or extra profit, however you've agreed to count it) over six months: $2,000 if the result is poor, $10,000 if it's as expected, $25,000 if it's strong. All figures are illustrative, not market rates:
| Structure | Poor result | Expected result | Strong result |
|---|---|---|---|
| Fixed fee of $3,000 | $3,000 | $3,000 | $3,000 |
| Hourly, about 30 hours at $100 | $4,000 (overrun to 40 hours) | $3,000 | $2,800 |
| $3,000 with 15% held back until acceptance | $2,550 if tests never pass | $3,000 | $3,000 |
| $2,000 plus $1,000 bonus if gain reaches $8,000 | $2,000 | $3,000 | $3,000 |
| $1,500 plus 15% of the gain | $1,800 | $3,000 | $5,250 |
| 30% of the gain, nothing otherwise | $600 | $3,000 | $7,500 |
Every structure costs $3,000 at the expected result, which is how a consultant pricing honestly would set them. The differences are all at the edges. The capped bonus protects you in both directions. Gain-share and pure contingency are cheap when things go badly and expensive when they go well, and "well" is precisely when you're least able to prove how much the consultant caused. If you can't budget for the right-hand column, you can't afford the structure.
There's a second reading of the table. Hourly billing is the only structure where a poor result costs more than an expected one, because poor results often come with overruns. That's a point in favour of fixed fees generally, whatever you decide about bonuses.
Where a bonus fits, and where a holdback is fairer
The five conditions sort projects quickly. Three short cases show how the answer changes with the kind of result:
A members' club's renewal emails: a bonus fits. The club asks a consultant to set up personalised, AI-drafted renewal reminders. Renewals are countable, the season is fixed, and attribution can be solved properly: the club keeps its old single reminder for a random fifth of members and pays a bonus only on the difference in renewal rate between the two groups. With roughly 80 members in the comparison group, small gaps are noise, so the clause sets a clear threshold (say, at least three percentage points) and a cap. The comparison group does the attribution work that argument would otherwise have to do.
A church office's admin automation: a holdback fits. The office wants hall-hire replies drafted and bookings entered automatically. There's no revenue to share, and the time saved depends as much on how the part-time administrator uses the system as on the build. A bonus on "hours saved" would rest on estimates. A holdback is cleaner: 15% of the fee is paid once the system drafts correct replies to at least 18 of 20 test enquiries written from real past emails. The consultant is paid for delivering what was promised, and the office judges the time saved for itself.
A charity shop's listing tool: a bonus, measured per hour. A consultant sets up AI-drafted online listings from volunteers' notes and photos. The total number of items listed depends on how many volunteers turn up, which the consultant doesn't control. Items listed per volunteer hour, taken from the shop's rota and listing records, does depend on the tool. That's a measure a small bonus can safely hang on, provided the clause names both records as the source.
The pattern: pay a bonus when you can isolate the consultant's effect with a comparison group or a per-unit measure; use a holdback when the honest measure is "did they deliver what was specified"; use neither when the result is swamped by things outside the project.
Borrow the lesson from per-resolution software billing
AI software vendors have already run into the problem of defining a "result", and their billing rules show how slippery it is:
- Intercom charges $0.99 per outcome for its Fin agent, and counts an "assumed resolution" when a customer goes quiet for 24 hours after Fin's last answer.
- Zendesk closes a messaging conversation after 2 hours of inactivity by default (72 hours on email and web forms), then bills only resolutions its AI check verifies; since 18 May 2026 hand-offs to a person and unverified closes are free.
- HubSpot's Customer Agent uses 50 credits, about $0.50, per resolved conversation.
Silence isn't satisfaction. A customer who gives up looks exactly like one who got an answer, and a bonus tied to "questions resolved" has the same weakness. So when you write a results clause, borrow the vendors' method but tighten it: say what counts, what doesn't, and how you'll check. In the toy shop's case, a question counts as handled only if the customer didn't contact the shop again about the same thing within seven days, confirmed from a weekly sample of twenty conversations. Outcome-based AI pricing goes further into how each vendor counts, and how to measure whether your AI chatbot is working shows how to set up the sample.
Costs that hide inside a results clause
A success fee changes more than the invoice. Budget for these as well:
- Measurement time. Someone has to pull the figures, check them and agree them with the consultant each month. At two hours a month for six months, valued at $25 an hour, that's $300 of owner time the fixed fee wouldn't need.
- Data access. Revenue-based fees mean sharing sales data. Limit access to the figures the clause needs, and make sure the engagement letter covers confidentiality.
- Disputes. A vague measure invites an argument at exactly the moment the relationship should be ending well. One dispute can cost more in time and goodwill than the bonus it's about.
- Behaviour that chases the measure. A consultant paid on revenue may push for discount codes that lift sales and cut margins. One paid on "resolved" chats may tune the agent to close conversations quickly.
- Cash-flow timing. A bonus that falls due just after your peak season can land when cash is tight, or when you'd rather reinvest.
To cut these costs: use one measure, not three; take it from a system you already run; sample instead of auditing everything; and set the payment date in the clause. Keep a record of what the project cost and delivered overall too; how to judge whether your AI consultant delivered value covers that review.
Wording for a bonus clause that survives a dispute
The clause has to answer every question a dispute would raise, before anyone is in a dispute. Illustrative wording, based on the toy shop's Option 3; it isn't legal advice, and a solicitor should review anything significant:
Performance bonus. The Client will pay a one-off bonus of $750 if, during the eight weeks from 15 January, at least 50% of routine customer questions received through Shopify Inbox are handled by the AI agent without staff involvement. "Routine" means questions about delivery, returns, product age suitability and order status. A question counts as handled only if the same customer does not contact the Client about the same matter within seven days. The Client will measure this from a random weekly sample of 20 conversations, shared with the Consultant each Monday. Questions arising from stock errors, website outages or changes to the Client's policies during the period are excluded. The bonus is capped at $750 and payable within 30 days of the end of the period. If the Client materially changes the agent's settings without the Consultant's agreement, the bonus is assessed on the weeks before the change.
Every sentence is there for a reason: the measure, what counts, how it's checked, what's excluded, the cap, when it's paid, and what happens if you change the setup. The last one protects the consultant, and a fair clause needs that too; a consultant who sees their protection written in is more likely to agree the rest without haggling. For the clauses that sit around this one, see what to check in an AI consulting contract.
Metrics that invite gaming, and safer replacements
Some measures are easy to agree and easy to game. Swap them for versions that are harder to move without doing the real work:
| Tempting measure | How it gets gamed | Safer version |
|---|---|---|
| Chats closed by the AI | Closing conversations early | No repeat contact about the same matter within 7 days, from a weekly sample |
| Revenue growth | Discounts; riding a busy season | Conversion rate on pages the project changed, against pages it didn't touch |
| Hours saved, self-reported | Generous estimates | Time logged for the task over two weeks before and after |
| Emails sent | Volume without quality | Orders or replies from tracked links in those emails |
| Satisfaction score | Surveying only happy customers | The same survey, sent on the same trigger, before and after |
The cost of getting this wrong is easy to underestimate. Think of a small online shop that agrees a bonus on "customer chats resolved by AI", counted by its helpdesk. The number hits target in week three. A month later, repeat emails from customers asking the same question twice are up by a quarter. The agent has been tuned to give a short answer and end the chat, which the helpdesk counts as resolved. The bonus gets paid, the customers are worse off, and the shop spends the next month undoing the tuning. The safer version in the first row would have caught it in the first weekly sample.
If none of the safer measures fits your project, that's useful information too. It usually means the fair answer is a fixed fee with a holdback, and a separate conversation, after the work is done, about what it achieved.
Success fees: questions before you agree one
Is a success fee the same as outcome-based software pricing?
They're cousins. Software vendors charge per unit of outcome, such as Intercom's $0.99 per resolved outcome for its Fin agent, and keep charging for as long as you use the product. A consultant's success fee is usually a one-off bonus tied to a project result over a set period. Both depend entirely on how the outcome is defined, so read that definition first.
How much of the fee should depend on results?
My guidance for a small business is a quarter to a third of the total at most, with a cap. That's enough to show the consultant believes in the result, and small enough that a dispute about measurement doesn't sour the whole relationship. If a consultant wants most of the fee at risk, run the strong-result sum before agreeing; generous upside can cost more than a fixed quote.
Is a holdback simpler than a bonus?
Usually. A holdback keeps back part of a fixed fee, often 10 to 20%, until agreed acceptance tests pass. It ties payment to delivering what was promised rather than to business results the consultant only partly controls. For most first projects it's the fairest risk-sharing tool, and it avoids arguments about attribution entirely.
What if a consultant offers no results, no fee?
Ask three things: exactly what counts as a result, how it will be measured and from which system, and what you'll pay if the result is strong. Then do the sum at that strong result. For narrow, countable work the offer can be fair; for broad goals such as growing sales, you may end up paying for gains the consultant didn't cause.
Further reads
- How Much Does an AI Consultant Cost for a Small Business? — What AI consultants charge before any results clause.
- How to Negotiate an AI Consulting Quote Without Cutting Corners — Negotiate the fee structure without cutting what matters.
- How Soon Should AI Pay for Itself? Payback Periods by Project — Payback targets that help set a sensible bonus threshold.
- Did Your AI Pilot Work? How to Set Success Criteria That Hold Up — Write success criteria that hold up before any bonus depends on them.
- Is an AI Consulting Retainer Worth It After Your First Project? — Whether ongoing support should be paid by result or retainer.
- How to Calculate AI ROI for Your Business (Worked Example) — Work out the return that any success fee is a share of.
- How to Calculate the ROI of an AI Automation Before You Build It — A nine-step pre-build ROI method, a copyable worksheet, a roofing contractor's quote follow-ups costed, and a removals firm's idea that failed the test.
- AI Consultant Red Flags: 12 Warning Signs to Walk Away From — Twelve warning signs when hiring an AI consultant, each with an example, the question to ask, its innocent version and a scored two-proposal comparison.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Intercom pricing page and billing definitions for Fin; Zendesk and HubSpot documentation on how automated resolutions are counted and billed.