Why Agency Pricing Is Shifting From Hours to Outcomes in the AI Era

Because the billable hour has always contained a quiet conflict, and AI just made it loud. When you pay by the hour — time and materials, the model where you're billed for actual hours worked at a set rate — the vendor earns more the longer the work takes. Efficiency costs them money. For decades that tension was tolerable because hours were a rough proxy for effort and value.

AI breaks the proxy. When a tool can compress part of the work dramatically, ‘hours spent' stops describing the value delivered — and billing for them starts to look like billing for slowness. Venture analysts have named the underlying shift bluntly: ‘software is becoming labour,' and the per-seat or per-hour unit is ‘no longer the atomic unit' of value (a16z, December 2024). So agencies that want their incentives aligned with the client's are moving toward pricing the result, not the clock.

Value-based pricing sets the fee by the economic value the work creates for the client — productivity gained, revenue added, cost removed — relative to the client's next-best alternative, rather than by what it cost the vendor to produce. McKinsey frames it as pricing ‘according to the measurable economic value your offer creates for specific customer segments.' The price answers ‘what is this worth to you?' not ‘how many hours did this take?'

Outcome-based pricing is the stricter sub-type, where some or all of the fee is contingent on hitting a measurable result — a price per resolved support ticket, per qualified lead, or a share of the savings created. It's no longer theoretical: AI customer-service products now bill per successful resolution and charge nothing for failed attempts, shifting performance risk to the vendor. The healthiest versions usually keep a fixed floor plus variable upside, so neither side bears all the downside.

By making hours an unreliable measure of either effort or value. The evidence cuts both ways, which is precisely the point. In a controlled task, developers using GitHub Copilot finished 55% faster (GitHub, 2022). Yet the most rigorous independent trial found experienced developers were about 19% slower with AI on large, mature codebases — while believing they were faster (METR, July 2025). Google's DORA research, meanwhile, has linked heavier AI adoption to slightly lower delivery stability.

Put those together and the lesson isn't ‘AI makes everyone faster' or ‘slower' — it's that AI makes hours swing wildly depending on the work. When the same feature might take a third of the time or half again as long, the hour stops being a fair unit to price against. Billing by it means either the client overpays for AI-accelerated work or the vendor is penalised for using AI well. Pricing the outcome sidesteps the whole problem.

Further in talk than in invoices, so far — this is an early shift, not a finished one. As of the early 2020s, value-based pricing was only about 10% of agency revenue, well behind fixed-price (just under half) and retainers (around 44%), with hourly still at roughly 30% (SoDA/Productive agency data). Most agencies still bill mostly by time or fixed scope.

But the leading edge is moving fast. McKinsey reported it now earns roughly a quarter of its global fees from outcomes-based pricing — a notable departure from billable hours (reported by Hunt Scanlon, December 2025). In software, pure per-seat pricing is shrinking while usage- and hybrid-based models grow (Bessemer, 2026) [VERIFY]. The direction is consistent across professional services and software: away from time, toward value. The honest summary is that outcome pricing is crossing from novelty to mainstream — not that hourly is gone.

Each model parks the risk somewhere different, and the table above maps it. With hourly/time-and-materials, the client carries the risk — you pay for every hour, however long it runs. With fixed-price, the vendor carries it — they absorb any overrun, which is why they pad the estimate to cover the unknowns. Value- or outcome-based pricing shares the risk, tying the fee to a result, which is why it both rewards vendor efficiency and demands a clean way to measure success.

A dedicated team or retainer is closer to hourly in disguise: you rent capacity at a fixed monthly cost, direct the work, and carry the utilisation risk yourself. None of these is cheaper in the abstract — the cheapest is the one whose risk profile matches your certainty. Where do AI tools and freelancers sit? AI/vibe-coding tools are flat subscriptions; freelancers are still mostly hourly or small fixed-price, because a solo provider can rarely absorb the downside that real outcome pricing requires.

When the outcome can't be cleanly measured or attributed to the vendor's work. ‘Increase revenue' is a terrible outcome metric if marketing, sales, product and the macro economy all move it too — attribution disputes are the single most common way these deals sour. Watch, too, for a vendor who picks a vanity or proxy metric (raw clicks, ‘leads' of unknown quality) and dresses up ordinary work as ‘value pricing.'

Be cautious when there's no agreed baseline, when the buyer can't or won't share the data needed to measure the result, or when the time horizon is so long and multi-causal that nobody can later say what caused what. And be wary of any structure where one side bears all the downside: a vendor demanding a pure success fee is often over-promising, while a buyer pushing all risk onto the vendor simply pays a large risk premium baked into the price. Good outcome deals keep a fixed floor plus shared upside.

Hourly still wins when the scope is genuinely unknown — discovery, R&D, prototyping, incident response, untangling legacy code. Pinning a fixed price or an outcome on work nobody can yet define forces the vendor to pad heavily or cut corners. Hourly (often as staff augmentation) also suits buyers who want close control and visibility and are willing to manage the work, and inherently variable work like ad-hoc maintenance.

And sometimes the honest answer is to hire no agency at all. If you're validating an idea, building a demo, or shipping an internal tool a few trusted people will use, AI tools or a single freelancer are usually enough — no pricing-model debate required. Engage an agency, on any model, only when something real depends on the build. When you do, the pricing question worth asking isn't ‘what's your hourly rate?' but ‘how do your incentives line up with my outcome?'