Agency pricing models in 2026 come down to five: hourly, retainer, fixed fee, productized subscription, and outcome-based pricing. Hourly still works where scope is uncertain or procurement demands it, but it pays the agency less every time production gets faster. A twenty-hour task done in four hours loses 80 percent of its revenue on an hourly contract. Most agencies will run a portfolio: retainers for ongoing capability, fixed fees for defined builds, subscriptions for recurring production, and outcome pricing where the result can be measured.
The case for hourly billing
Hourly billing lasted as long as it did because it solved real problems. Clients know what they’re buying, procurement has a number to compare across vendors, and your team has a meter that makes scope creep visible. It answered a hard question, how to price knowledge work with no clear unit of output, by making time the proxy everyone could agree on.
It also protected the agency. When a project ran long or the work was harder than expected, the meter kept running, so the downside was capped at the hours billed. McKinsey’s study of more than 5,400 large IT projects found average cost overruns of 45 percent, with one in six carrying overruns that averaged 200 percent, which is a good explanation for why time-and-materials survived for decades.
Where hourly breaks, and what the client doesn’t see
The billable hour assumes time is a reasonable proxy for effort and effort is a reasonable proxy for value. A task that took twenty hours can now take four while producing the same platform or outcome, so the meter runs for four and the engagement loses 80 percent of the revenue attached to that work. On one twelve-month time-and-materials engagement at our agency, worth roughly a million dollars in potential revenue, the first milestone came in about 75 percent under budget. Good for the client, bad for the contract.
Clients see it from their side. Productive’s 2025 survey of 180 agencies found that 27 percent had been asked to lower prices because of AI and nearly half expected the request. Its June 2026 follow-up of 174 agencies found that 61 percent of agencies still implementing AI hadn’t settled on a pricing model, up from 47 percent.
The client also can’t see much of what now creates the value. They see the deliverable and the turnaround, not the specification library, review systems, or accumulated context behind it. In a traditional agency, weeks of visible labor helped explain the invoice. As execution compresses, that explanation becomes less convincing.
Earlier this year we wired a workflow from Slack to ticketing to code generation to review to deployment and watched it run in minutes, work that used to occupy a project manager, two developers, and a QA lead for two or three days. Someone asked how we bill for that. We didn’t have an answer yet.
Retainer pricing
The retainer was the easiest place for us to start because clients already understand it and procurement already knows how to approve it. The change that matters is what the fee represents: access to a defined level of ongoing capability, responsiveness, and continuity, not a predetermined number of hours.
If the agency gets more efficient underneath that arrangement, the economics improve without passing every gain through as fewer billable hours. The risk is a retainer that’s secretly hourly, a fee with an hours cap that the client audits, which is just time-and-materials with a lower ceiling. Define the retainer by scope of capability and response commitments instead.
Fixed fee pricing
Fixed-fee work handles the efficiency problem directly. The client agrees to a price for a defined outcome, and the agency decides how to produce it. If a workflow that took three weeks can be delivered in one, the agency keeps that gain as long as scope and quality hold.
The risk moves the other way. If the project is poorly specified or expands, the agency absorbs the cost, which makes specification work far more important than it was on hourly engagements.
AI adds a specific hazard. OpenAI rolled back a GPT-4o update in April 2025 after it turned sycophantic, and a Stanford study found developers using AI assistants wrote less secure code while feeling more confident about it, so a plausible plan can hide risk you’ve already agreed to absorb.
SPI Research’s 2026 benchmark found fixed-price project margins at 37.2 percent compared with 36.4 percent for time-and-materials, so the two models perform similarly on average. The difference appears at the edges. A well-specified fixed-fee project built on a mature workflow is where the efficiency gain lands. A vague one is where it disappears.
Productized subscription pricing
Subscription models have emerged for recurring creative and production work. Designjoy, founded by Brett Williams in 2017, charges roughly $5,000 a month for unlimited design requests with one active task at a time and delivers most within 48 hours, with Williams running it himself at well over $1 million a year. Its published rate in September 2026 is $4,995 a month discounted and $5,995 at list.
Superside sells a monthly budget spread across creative services, rolling over unused amounts, at a $15,000 monthly minimum on an annual term plus a $1,000 monthly software fee, with a dedicated tier from $30,000. Across that range, productized subscriptions for SEO, paid search, content, and development run from the low thousands a month to $30,000 and above.
The structure is the same at every price: a predictable monthly fee, a managed queue, and controlled delivery capacity instead of a running clock. It works for work that repeats, and badly for work that doesn’t, because an empty queue is a subscription the client cancels.
Outcome based pricing
Outcome based pricing charges for a measured result rather than for time, capacity, or a deliverable. Intercom’s Fin AI agent is the cleanest example outside the agency world, priced from $0.99 per resolved conversation alongside its per-seat plans. The customer pays for the result, not the labor or compute behind it.
FIG Agency moved off time-based billing five years ago after building a creative data system that can evaluate an entire category within 48 hours of a brief. Its CFO, Richard Tan, has said plainly that a time-based price wasn’t appropriate remuneration for that kind of result. FIG also restructured around senior people rather than junior leverage, because if you aren’t selling hours there’s no reason to staff for them. Tan calls outcome-based pricing the next frontier rather than a finished transition, which matches what we’ve found.
Outcome pricing needs three things an hourly agency may not have: a result the client agrees is measurable, a production system reliable enough to deliver it repeatedly, and enough history to price the risk. It fits paid media, conversion work, and anything with a number attached. It fits brand work and exploratory strategy badly, at least for now.
Compute as a cost of goods, and the ninety-day transition
Whatever model you choose, compute now sits inside delivery cost. Token pricing, API metering, and inference are the cost of goods for an orchestrated agency, and Anthropic’s own figures put agents at roughly four times the tokens of ordinary chat use and multi-agent systems at roughly fifteen times. At September 2026 list prices, Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens. Google’s third-generation Pro model costs $2 per million input tokens and $12 per million output tokens for prompts up to 200,000 tokens.
We treat it the way we treat freelancer costs or software licenses, because the agencies that discover compute after they’ve set prices end up back in the efficiency trap.
Some agencies bill agent work as labor, so a twenty-hour estimate at $150 stays a $3,000 invoice whether a developer took twenty hours or an agent took two. It preserves margin, and it raises awkward questions about what an agent hour is and whether the client is paying for compute or a result, plus disclosure exposure under Utah’s AI Policy Act and California’s AI Transparency Act, in effect since August 2, 2026. A retainer or fixed fee avoids most of that ambiguity.
We treat the pricing change as a ninety-day process. The first stretch is internal: which clients are already buying continuity or outcomes even though the contract says hours, and which types of work have become far more efficient in our delivery history. Then the conversations, framed around how delivery has changed and what the client is buying, not around AI making the work cheaper. Some clients move immediately, and some stay hourly because procurement requires it, which isn’t a failure.
Cash flow is the immediate risk. A retainer or fixed fee can improve margins over time and still create a timing problem if several large accounts change at once, especially when hourly billing was predictable and the replacement changes when cash arrives. Model it first, and decide in advance how you’ll answer a client who asks you to “just use AI” to make the work cheaper. That’s a much harder call to make while you’re trying to keep a large account.
| Model | Who carries the efficiency risk | Margin behavior as production speeds up | Best client fit |
|---|---|---|---|
| Hourly | Client gets the gain, agency loses revenue | Falls: 20 hours to 4 cuts revenue 80% | Uncertain scope, procurement-mandated, legacy support |
| Retainer | Shared, tilted toward agency | Improves if fee is capability-based, not hours-capped | Ongoing production, continuity, responsiveness |
| Fixed fee | Agency absorbs overruns | Improves with mature workflows, 37.2% avg project margin | Defined builds with strong specifications |
| Productized subscription | Agency manages queue capacity | Improves with throughput, $5,000 to $30,000+ per month | Recurring creative, SEO, content, development |
| Outcome-based | Agency carries delivery risk, client pays only for results | Highest upside, needs reliable system and measurable result | Paid media, conversion, anything with a number attached |
| Agent-as-labor markup | Agency, plus disclosure exposure | Preserves margin short term, ambiguous unit | Transitional at best |
What this means for an agency founder
I don’t think every agency should stop billing hourly tomorrow. We still have relationships where it makes sense. What changed for us is that we no longer assume hours are the natural unit for any engagement, because once the production system compresses days into minutes, the pricing model has to describe what the client is buying.
The full chapter on this, including the milestone that came in 75 percent under budget and what we did about it, is in The Cognitive Agency. If you want help modeling a transition for your own client list, our AI consulting and development team can help you work through the economics before you change the contracts.
