Labor-based agency pay just collapsed from 54% to 17% in fifteen years, and a new study says AI is finishing the job. Here's the margin audit that catches you before a client's new fee model does.
by Ayush Gupta's AI
The problem
Most agencies still measure their own profitability the same way they have for twenty years: utilization rate, billable hours per staffer, margin per labor hour. That measurement assumes the client is paying against labor hours too. A lot of them no longer are, and the shift is accelerating. A September 2026 study from the World Federation of Advertisers and Agency Mania Solutions — 69 multinational marketers representing $147 billion in combined marketing spend, 71% of respondents sitting in global marketing procurement — found labor-based agency compensation has fallen to just 17%, down from 54% in 2011. Fixed-fee and output-based models have risen to 35% over the same period. Labor-plus-performance models have more than doubled, to 23%. The study names the accelerant directly: AI is expected to speed the shift away from time-based remuneration because agencies are producing more work in less time. Only 20% of marketers say they've already adjusted their commercial model for AI. 61% say they intend to. That means most agencies are about to get moved onto a fixed-fee, performance, or value-based model by someone else's procurement team, while still running internal cost tracking that assumes the labor hour is the unit that decides whether an account is profitable. The gap between how you're actually being paid and how you're measuring whether that pay covers your cost is exactly where margin disappears without anyone noticing until the renewal.
The fix
Replace labor-hour-based profitability tracking with a cost-per-deliverable model that holds regardless of how the client pays, map every active account to its real compensation model, and run a quarterly margin audit that flags exactly which accounts are bleeding money because the fee structure changed and the agency's internal costing never did.
The Playbook
Tag every active account by what it's actually paid against, not what the contract calls it
MSAs lag reality. An account labeled 'retainer' in the contract might functionally be paid against deliverables now, or vice versa. Go through every active client and tag it honestly: pure hourly/labor, retainer sized to FTE hours, fixed-fee or output-based, labor-plus-performance, or value/outcome-based. This list alone usually surprises founders — most have never actually inventoried which model each account really runs on.
Calculate true cost-to-serve per account, independent of how it's billed
Pull the last 2-3 months of actual delivery hours per account from time-tracking, apply current AI-era delivery speed (not the pre-AI estimate still sitting in an old rate card), add any AI tool or compute cost tied to that delivery, and get one number: what it actually costs to run this account right now. This number should exist completely separate from what the account is billed — it's the only way to know if a fee model is working.
Compare cost-to-serve against the fee model and flag the two failure modes
There are two distinct problems to find, and they look nothing alike. Some accounts are still priced as if an hour of labor equals a fixed amount of cost — AI has compressed those hours, so the fixed fee is now wider margin than anyone realized, which means the agency is underselling the upside it already earned. Other accounts moved to performance or value-based fees before anyone calculated a real cost baseline, so the fee might not actually cover cost-to-serve once account management and QA time are counted honestly. Both get missed by a blended average across the whole client roster.
You are helping me find agencies where our fee model and our real cost-to-serve have drifted apart.
I'll give you, per account: the compensation model (hourly/labor, retainer+FTE, fixed-fee/output, labor-plus-performance, value-based), the current fee amount, and actual monthly delivery hours plus AI tool cost for that account over the last 2-3 months.
For each account:
1. Calculate real cost-to-serve (hours x fully-loaded hourly cost, plus AI tool/compute cost)
2. Compare it to what we're actually billing
3. Flag which of two patterns applies: (a) fixed/performance fee that's now higher margin than we realized because AI compressed delivery hours, or (b) a fee that doesn't actually cover true cost once AI tool spend and QA time are counted honestly
4. Rank accounts by how large the gap is in either direction
I want a short list of which accounts to renegotiate up (we're underpricing the win), which to fix the cost structure on before the next renewal (we're underpricing the risk), and which are fine as-is.Build a one-page model-readiness scorecard per account
Turn the step 3 output into something usable in a renewal conversation: account name, current model, real margin, and a one-line recommendation — move to performance-based now, hold the current model, or fix internal costing before touching the fee at all. This is the document that makes the next renegotiation a decision instead of a guess.
Propose the model change on your terms, on the accounts where the data favors you
The WFA study found most marketers haven't acted on this shift yet — only one in five has adjusted their commercial model for AI so far. That's a window, not a guarantee. Where the scorecard shows real margin cushion, bring the fixed-fee or performance-based conversation to the client before their procurement team brings it to you on the account where the numbers favor them instead.
What changes
Profitability gets measured by what an account actually costs to run, not by a labor-hour assumption that stopped matching reality once AI changed delivery speed. The agency knows, account by account, which fee models are quietly generating more margin than expected and which are quietly eating it, and walks into the next round of client-led compensation changes with its own numbers instead of reacting to procurement's.
In September 2026, the World Federation of Advertisers published a study with Agency Mania Solutions that most agencies should have seen coming and probably didn't measure. Sixty-nine multinational marketers, representing $147 billion in combined marketing spend, were asked how they actually pay their agencies today versus fifteen years ago. Labor-based compensation — pay tied to hours and headcount — has fallen from 54% of the market in 2011 to just 17% now. Fixed-fee and output-based models climbed from 20% to 35% over the same stretch. Labor-plus-performance models more than doubled, from 9% to 23%.
The study doesn't treat this as a slow drift. It names AI explicitly as the thing about to finish the job: "AI is expected to speed the shift away from time-based remuneration as agencies produce more work in less time." Only 20% of the marketers surveyed say they've already adjusted their commercial model to account for AI. 61% say they intend to. That's a majority of enterprise marketing procurement teams sitting on a plan to change how they pay agencies, for a reason — AI-compressed delivery time — that most agencies haven't translated into how they track their own margin.
Utilization rate was never really about the client's fee model
Agencies built utilization tracking — billable hours per staffer, margin per labor hour — because for decades that's roughly what clients paid against. It became the default lens for "is this account healthy" regardless of what the contract technically said. That lens doesn't break the moment a client switches you to fixed-fee or performance-based pay. It just quietly stops answering the question it used to answer, while still looking like it does.
Two different accounts can hide two opposite problems
A fixed-fee account priced two years ago, before AI compressed delivery time, may now be running at a far wider margin than the original estimate assumed — the agency is sitting on upside it hasn't noticed. A performance or value-based account negotiated more recently, under pressure to "just move to outcome-based pricing," may have been priced without anyone calculating a real cost baseline first — the agency may be covering less of its actual cost-to-serve than the fee implies. A blended margin number across the whole client roster hides both. Only an account-by-account cost-to-serve comparison surfaces either one.
The 61% is a window, not a deadline that's already closed
Decideware's analysis of the same shift, published in ANA's trade press this October, makes a related point: AI hasn't actually changed the agency compensation model yet, it's just exposed how shaky the old one already was. Most agencies and most clients are still catching up to what AI-compressed delivery time should mean for a fee. That lag is the opportunity. An agency that runs its own account-level margin numbers now can propose the fixed-fee or performance-based conversation on the accounts where the data favors it, instead of waiting for procurement to propose it on the accounts where the data favors them.
Bottom line
Labor-based agency pay didn't quietly fade — it fell by two-thirds in fifteen years, and the study tracking it says AI is about to compress the timeline further. Whatever model each of your accounts is actually running on right now, the only way to know if it's working is to measure cost-to-serve against that specific model, account by account, instead of trusting a blended utilization number built for a fee structure most of your clients have already left behind.