What AI-Assisted Delivery Does to Agency Pricing
When AI-assisted development halves the hours to ship, hourly and time-and-materials pricing make the agency cannibalise its own revenue — the value held, the hours fell. The case for repricing on value, what happens to estimates and utilisation, and a worked example of a 200-hour build that now takes 90.
AI-assisted delivery has done something specific and awkward to agency economics: it has cut the hours required to ship a given piece of work without cutting the value of that work to the client. A feature that took 200 hours now takes 90. The client still wants the feature as much as they did, and it is worth the same to their business. Only the hours fell.
That is fine if you price on value or outcome. It is a slow act of self-harm if you price on hours. Under hourly and time-and-materials billing, AI agency pricing has the agency cannibalising its own revenue: you invest in tooling and senior judgement to deliver faster, and the reward for delivering faster is a smaller invoice. The more efficient you get, the less you earn for the same result.
The honest answer is not to pretend AI changes nothing, and not to overclaim that it changes everything. It is to recognise that the link between hours and value — which was always loose — has now visibly broken, and to move pricing towards value and outcomes so the agency keeps the upside of its own efficiency. This piece covers why hourly billing punishes you here, what happens under each pricing model when hours halve, the knock-on effects on estimates, utilisation and the cost base, and a worked example of a build that dropped from ~200 hours to ~90.
A developer on my old team shipped a piece of work last year that the estimate had put at three weeks. He did it in a bit under one. He had used an AI assistant to scaffold the boilerplate, draft the tests, and clear the kind of repetitive plumbing that used to eat the first week of any build. The work was good. It passed review. The client was happy.
And the invoice, because the engagement was time-and-materials, was a third of what the estimate implied. We had got dramatically better at the work, and the reward for getting better at the work was getting paid less for it. That is the moment the AI agency pricing problem stopped being abstract for me.
This piece is for the agency owner or CFO watching AI-assisted delivery compress the hours on real projects and wondering what it does to the model. The uncomfortable centre of it: when AI halves the hours to ship and you bill by the hour, you are cannibalising your own revenue. The value you delivered did not fall. The hours did. If the price is tied to the hours, the price falls with them — and you have paid for the privilege.
The cannibalisation problem: same value, fewer hours
Pricing on hours always rested on a quiet assumption: that hours were a reasonable proxy for value. Spend more hours, deliver more value, charge more. It was never exactly true — a senior who solves a problem in two hours that a junior would take twenty hours to fumble delivers more value in fewer hours — but it was close enough to run a business on, because effort and output moved roughly together.
AI-assisted delivery breaks that proxy in the open. The hours and the value now move in different directions on the same project. The feature is worth what it was worth to the client's business; the effort to produce it has dropped sharply. If your price tracks the effort, your price drops while the value you handed over stayed flat.
Run that forward and the incentive is upside-down. Every improvement in your delivery speed — better tooling, sharper seniors, reusable patterns — translates directly into a smaller invoice for the same result. You are paying to reduce your own revenue. This is the same structural flaw I wrote about in why your hourly rate is the wrong number to optimise: hourly billing punishes efficiency. AI did not create the flaw. It made it impossible to keep ignoring, by making the efficiency gains too large to absorb quietly.
Worth saying plainly, because the topic attracts overclaiming: AI does not deliver projects on its own, it does not remove the need for senior judgement, and it does not halve the hours on every kind of work. It is very good at the repetitive, well-specified middle of a build and much weaker at the ambiguous edges where the real risk lives. The compression is real and uneven. The pricing response has to handle the part that genuinely compresses without pretending the whole job did.
What happens to each pricing model when hours halve
Take a single engagement and watch what the same efficiency gain does to your revenue under each pricing model.
Hourly and time-and-materials
The hours are the price. Halve the hours and you halve the invoice. The client gets the same outcome for less money, which they will quite reasonably enjoy, and the agency has converted a productivity gain into a price cut it did not negotiate and did not intend. T&M is the purest form of the trap: there is no fixed envelope to protect you, so every hour saved is a pound not billed. Of all the models, this is the one where AI-assisted delivery does the most damage to the agency and the least to the client.
Fixed price
A fixed-price quote protects you in the short run. You quoted the outcome at a price; you deliver it in fewer hours; the margin on that specific project improves, sometimes dramatically. That is the good news, and it is real.
The catch is the next quote. Fixed prices are usually anchored to an estimate of hours, and estimates compress as everyone — clients included — learns that the work is faster now. Procurement starts asking why a build that was £80k last year is still £80k when "AI does half of it". Fixed price buys you a window of improved margin, but the window closes as the market re-anchors its expectation of what the work should cost. You keep the gain until the next negotiation resets the baseline.
Value and outcome pricing
When the price is tied to what the outcome is worth to the client rather than to the hours it took, the efficiency gain stays with the agency. The feature that saves the client's team 200 hours a quarter, or unlocks a revenue line, is worth what it is worth regardless of whether you built it in 200 hours or 90. Deliver it faster and your margin improves without your price falling, because the price was never about your time. (The trade-offs between fixed and time-and-materials structures, before you even add value pricing on top, are worth reading in fixed-price vs time and materials margin.)
Value pricing has real conditions — you have to credibly own the outcome, and it does not fit every engagement — but AI-assisted delivery shifts the cost-benefit firmly towards it, because it is the only model where getting faster does not mean getting paid less.
| Pricing model | Hours fall from 200 to 90 | Who keeps the efficiency gain |
|---|---|---|
| Time-and-materials | Invoice falls with the hours | Client keeps all of it |
| Hourly | Invoice falls with the hours | Client keeps all of it |
| Fixed price | Margin improves this project; baseline resets next time | Agency keeps it until the next negotiation |
| Value / outcome | Price unchanged; margin improves | Agency keeps it |
A 200-hour build that now takes 90
Here is the worked example, anonymised and scaled to round numbers. A build that historically took ~200 engineering hours now takes ~90 with AI-assisted delivery, for the same delivered outcome. Assume a blended real cost of ~£55/hour and look at what each pricing model does.
Before AI — 200 hours:
- Time-and-materials at £140/hour: revenue £28,000, cost ~£11,000, margin ~£17,000 (≈61%).
- Fixed price anchored to that estimate: revenue £28,000, similar margin.
After AI — 90 hours, same outcome delivered:
- Time-and-materials at £140/hour: revenue falls to £12,600. Cost ~£4,950. Margin ~£7,650. You delivered the same thing and your revenue dropped by ~£15k. The client paid less for an identical result.
- Fixed price still at £28,000: revenue holds at £28,000, cost drops to ~£4,950, margin jumps to ~£23,000 (≈82%). Excellent — until the renewal, when the client has learned the work is faster and pushes the fixed price down towards £18k–£20k, settling the margin somewhere in between.
- Value priced at the outcome (say the feature is worth ~£45,000 to the client's business): revenue £45,000, cost ~£4,950, margin ~£40,000. The price never referenced your hours, so the 110 hours you saved did not leak to the client.
The spread is the whole argument. The same efficiency gain produces a £15k revenue loss under T&M and a £17k revenue gain under value pricing, on identical delivered work. The hours fell by 55%. What happened to the agency depended entirely on whether the price was attached to the hours or to the outcome.
What shifts underneath: estimates, utilisation, cost base
Repricing is the headline, but AI-assisted delivery moves three other numbers that a CFO needs to expect.
Estimates compress, and so does estimate accuracy for a while. The hours-to-ship for familiar work drop, and your historical estimating baseline goes stale faster than usual. For a transition period your variance widens — not because estimating got worse, but because the ground moved. Re-baseline your role-rate estimates against recent AI-assisted projects rather than the two-year history, and expect the compression to be uneven: the well-specified middle of a build compresses hard, the ambiguous front end barely at all.
Billable utilisation looks worse even as margin can improve. This is the counter-intuitive one and it catches people. If a developer ships the same outcome in 90 hours instead of 200, their billable hours on that project fall. Read utilisation as a headline number and it looks like your team has gone soft. But margin per project can be up, because the cost of delivery fell faster than the price did (under fixed or value pricing). Utilisation as a target actively fights you here: it rewards filling hours, exactly when the point is to deliver outcomes in fewer of them. Watch margin per project, not chair-hours.
The cost base shifts towards senior judgement. AI is strongest at the work juniors used to do — boilerplate, first-draft tests, repetitive plumbing — and weakest at the architectural and ambiguous work that needs experience. So the human hours that remain skew senior. Your blended cost per remaining hour goes up even as total hours go down. That is fine, and it is another reason to leave hourly billing behind: the remaining hours are the expensive, high-judgement ones, and they are worth more than an hourly rate set against a junior-heavy mix will capture. Tracking this drift — who is actually doing the remaining work, at what real cost — is exactly the role-rate-versus-real-cost variance that decides whether a faster project is actually a more profitable one.
The agencies that come out ahead of this are not the ones that adopted AI fastest. They are the ones that adopted it and then repriced, so the efficiency gain landed in their margin instead of leaking to the client through an hourly invoice. The ones that get hurt are the genuinely good shops that got faster, kept billing by the hour, and quietly shrank their own revenue while congratulating themselves on the productivity.
If you want to see whether your AI-assisted projects are actually more profitable — total hours down, but margin up once the senior-skewed cost base and the pricing model are accounted for — that is the variance a real-time margin layer is built to show. The Saldo pricing page lays out what it costs, and the demo runs on your real Jira data so you can read the before-and-after on your own projects rather than on a slide. The hours fell. Make sure your pricing kept the value, because nothing else will.
Continue inside Saldo