A lot of agency owners are having the same confusing year. Client count is stable. Retention looks fine on the dashboard. Nobody has fired them. And revenue is flat or down, with no obvious event to point at.
The number that explains it
There's a name for what's happening, and once you have it the year makes sense.
Soft churn. The client doesn't leave — the retainer shrinks. A £25,000 monthly engagement becomes £15,000. The relationship continues, the logo stays on your website, and your retention metric records a win. Research through 2026 found roughly 60% of senior marketing leaders had reduced agency spend as a direct result of AI tooling, with the typical reduction landing in the 20–30% range.
This is the single most important dynamic in the agency business right now, and almost nothing in standard agency reporting can see it. Logo retention is healthy. Revenue is eroding underneath. You can hold every client you had a year ago and be materially smaller.
The measurement blind spot Agencies track whether clients leave. Very few track whether clients shrink. In 2026 the second number is doing more damage than the first.
The macro data is consistent with it. Promethean Research puts average digital agency revenue growth around 7.5% — positive, but well below historical norms. RSW/US reported that only 39% of agencies grew in 2025, down from 44% the year before. The global market is still expanding at low single digits, so this isn't a contraction. It's a redistribution: fewer agencies capturing the growth while more run flat.
The structural problem underneath
Here's the part that gets less attention than it deserves, and it's an arithmetic problem rather than a strategy one.
Industry research on agency headcount composition found that production employees make up roughly two-thirds of the typical agency workforce, rising to about three-quarters once project management is included. Revenue-generating roles — sales and marketing, excluding account managers — account for around 6.6% of total headcount.
That structure made complete sense in an era when production was the scarce, expensive, defensible thing. Clients came to agencies because making the work was hard and agencies had the people who could make it.
Production is no longer scarce. So a typical agency is now heavily staffed in the function that's deflating fastest, and barely staffed in the functions that aren't — strategy, judgement, and the ability to win new business.
| Metric | Where it sits | What it tells you |
|---|---|---|
| Average revenue growth | ~7.5% | Below historical norms; growth is concentrating |
| Agencies that grew in 2025 | 39%, down from 44% | Majority ran flat or shrank |
| Revenue per employee | ~$163k average | Below ~$120k signals structural trouble; specialists target $250k+ |
| Retainer churn | ~18% annually | ~56-month average client lifespan |
| Project churn | ~42% annually | ~24-month average lifespan — less than half |
| Production share of headcount | ~two-thirds | Heavy exposure to the function AI is deflating |
Revenue per employee is the most useful diagnostic on that list because it isolates which problem you have. Low RPE with high utilisation is a pricing problem — you're busy and underpaid. Low RPE with low utilisation is a headcount problem — you're carrying capacity the work doesn't need. A number of agencies checking this in 2026 have found both, which is uncomfortable but at least actionable, and it's the honest starting point for any conversation about pricing for profitable rates.
In-housing: relocation, not elimination
Around 32% of brands expect to handle nearly all creative work in-house within twelve months, and a large majority of big brands already run some form of internal agency.
The instinct is to read that as the work disappearing. It isn't. It's moving — and it's moving specifically away from agencies positioned as execution capacity, because execution is exactly what AI tooling has made viable for a small internal team. A brand that needed twelve people to produce a campaign can now do a version of it with three and a good stack.
What in-house teams still lack is what they've always lacked: outside perspective, specialist depth they can't justify hiring for, and the pattern recognition that comes from working across many accounts. The agencies retaining well are the ones that shifted from doing all the work to enabling the client's team — training, tooling, audits, strategic direction, and the specific pieces that genuinely need a specialist.
That's a smaller, higher-margin engagement than a full-service retainer. Whether it's better depends entirely on whether you've resized to fit it.
WebCatalog
Desktop tool that turns websites into standalone apps with multi-account profiles and workspaces.
Best for: Power users juggling many web apps & accounts
The proving problem
There's a second force compounding the pressure, and it's easy to miss because it looks like a measurement issue rather than a commercial one.
Agencies are being asked to justify spend at exactly the moment proving marketing's contribution has become hardest. Signal loss, multi-touch journeys, and a growing share of research happening inside AI answers where no click is generated have all degraded the evidence base — the picture set out in why attribution keeps getting harder.
The commercial consequence is direct. A client under budget pressure cuts what they cannot defend, and the agency with the weakest evidence gets cut first — regardless of whether it was doing the best work. Agencies that invested early in incrementality testing and honest measurement have found it doubles as a retention tool, because it gives the client something to show their own board.
Why clients actually leave
Delivery dissatisfaction was the leading stated reason for client departures in 2026, named by around 48%. Which is worth sitting with, because it's rarely about the work being bad.
In practice "delivery dissatisfaction" usually decomposes into three things: results that didn't match what was expected, communication that left the client uncertain what was happening, and reporting that didn't make the value legible. Only the first is really about the work.
Two under-discussed departure mechanisms worth planning for:
The client's marketing director leaves. New leadership arrives with their own relationships, their own view of what's underperforming, and a reasonable instinct to review inherited vendors. Agency relationships frequently don't survive that transition — and it has nothing to do with your work. The defence is having more than one relationship inside the account, and having your value documented somewhere other than in one person's memory.
Procurement gets involved. Once a marketing relationship becomes a procurement conversation, it's being evaluated on cost against comparable suppliers rather than on outcomes. That's a losable fight if you haven't established what makes you non-comparable.
A note on the data
Worth being straight about this, because it affects how much weight to put on the figures above.
A great deal of agency-industry research is conducted by companies selling to agencies — agency-management software, agency consultancies, agencies studying agencies. Sample sizes are often modest and self-selected, since the agencies who respond to benchmarking surveys skew toward the organised ones. Definitions vary: "churn" sometimes means logo loss and sometimes revenue loss, which produces wildly different numbers from the same underlying reality.
None of that makes the figures useless. The directions are consistent across independent sources, which is what matters. But treat specific percentages as indicative rather than precise, and be particularly sceptical of any statistic that arrives without a stated method — a caution that applies to most marketing benchmarking, not just this corner of it.
What's actually working
The agencies growing in 2026 look different from the ones that aren't, in fairly consistent ways.
Specialisation, narrower than feels comfortable
The clearest divide in the data. Specialist agencies command materially higher revenue per employee than generalists, because a specialist is not comparable to anyone and therefore not procurement-able in the same way. Generalist full-service positioning is precisely what an AI-equipped in-house team can now approximate.
Narrow can mean a vertical, a channel, a business model, or a specific problem. What it can't mean is "we do everything, but well" — that's the position under most pressure, and it's the one hardest to defend on price. Depth in a single discipline, like performance media or a particular vertical's content problem, is what makes a rate defensible.
Owning something the client can't replicate
Proprietary method, proprietary data, a genuine track record in a specific situation. Agencies retaining best are the ones whose value doesn't reduce to hours of production — because hours of production now have a visible market price and it's falling.
Selling outcomes and judgement, not deliverables
A retainer priced as "four blog posts and eight social posts" is priced as production, and production is deflating. The same work priced as ownership of a business outcome survives the conversation about whether AI could do it, because the client isn't buying the artefact — they're buying someone accountable for whether it works. This is the substance of how agencies are repricing AI-assisted work, and it's the most consequential unresolved question in the sector.
Visible expertise as the growth channel
With only around 6.6% of agency headcount in revenue-generating roles, most agencies have no real new-business function — growth comes from referral and reputation. That makes the visibility of the people doing the work a commercial asset rather than a nice-to-have, which is the argument in winning clients through your team's visibility.
It matters more now because of how buyers behave. B2B buying committees increasingly weight peer evidence and third-party signal over vendor claims — the pattern in why committees trust peers over ads — and a growing share of shortlists now form inside AI answers before anyone contacts you. An agency with no public footprint is invisible in both places.
Getting found before the pitch
One further shift worth planning around. Buyers increasingly build their shortlist before contacting anyone — and a growing share of that shortlisting now happens inside AI answers rather than search results or referral conversations. An agency that isn't described anywhere the models can read is absent from a step that occurs before you know a prospect exists.
Practically, that means published thinking with specific claims attached, presence on the review and comparison resources models draw from, and a site that states plainly what you do and for whom. It's the agency application of generative engine optimisation, and it's a channel most agencies have not yet treated as one — despite advising clients on it.
The talent side
Briefly, because it's the constraint most likely to bite next.
Agency turnover has historically run around 30% annually, with recent samples improving into the high teens. Roughly a third of agencies report meaningful burnout. And the composition of demand has shifted underneath: senior strategists and AI-fluent operators are in demand while traditional production and junior roles contract.
That last shift has a consequence nobody has solved. The junior production roles that are contracting were how the industry trained its seniors. If you remove the apprenticeship layer, you get a short-term margin improvement and a medium-term shortage of people capable of the judgement work that's now the whole value proposition. Agencies thinking three years out are treating junior development as an investment rather than a cost, and it's an unfashionable position that will probably look obvious by 2029.
What to do about it
- Measure shrinkage separately from churn. Contracted monthly revenue year over year, excluding new business. If you only track one new number this quarter, this one.
- Calculate revenue per employee honestly. Including founders. Then work out whether you have a pricing problem, a headcount problem, or both.
- Audit your positioning against an AI-equipped in-house team. For each service you sell, ask what a competent three-person internal team with good tooling could now do themselves. What survives that test is your actual business.
- Reprice away from deliverables. Anything sold as units of production is exposed. Anything sold as accountability for an outcome is considerably less so.
- Build relationships beyond your main contact. The marketing director's departure is a foreseeable risk and a fixable one.
- Fix onboarding, since delivery dissatisfaction is the leading exit reason. Most of it is expectation and communication rather than work quality — the ground covered in onboarding a new client properly.
The short version
The agency business isn't collapsing; it's being quietly resized. Clients aren't leaving, they're spending less, and standard retention reporting can't see it. Meanwhile the typical agency is two-thirds staffed in production — the function that's deflating — and barely staffed in strategy and new business, the functions that aren't. The agencies growing are narrow, own something that can't be replicated, and sell judgement rather than output. Measure shrinkage, work out your revenue per employee, and be honest about what a good in-house team with AI tooling could now do without you.
Weighing an agency against building in-house?
We work as the specialist depth alongside your team, not as replacement production capacity.
Explore Working With Us →Frequently asked questions
How is the agency business performing in 2026?
Growing slowly and unevenly. Promethean Research puts average digital agency revenue growth around 7.5%, well below historical norms, and RSW/US reported that only 39% of agencies grew in 2025, down from 44% the year before. The global market is still expanding at low single digits, so the sector is not contracting. What has changed is that growth has concentrated: fewer agencies are capturing it, and many are running flat while their client count stays stable.
What is soft churn and why does it matter?
Soft churn is when a client keeps the relationship but reduces the retainer, typically by 20 to 30 percent, rather than cancelling outright. Research in 2026 found roughly 60% of senior marketing leaders had reduced agency spend as a direct result of AI tools. It matters because standard retention reporting cannot see it: logo retention stays healthy while revenue erodes underneath. An agency can hold every client it had a year ago and still be materially smaller.
Is in-housing killing marketing agencies?
It is relocating work rather than eliminating it. Around 32% of brands expect to handle nearly all creative in-house within twelve months, and a large majority of big brands already operate an internal agency. The exposure is concentrated in execution-only agencies, because execution is precisely what AI tooling has made feasible for a small internal team. Agencies leading with strategy, proprietary method or genuine specialism are retaining at markedly higher rates than those positioned as production capacity.
What is a healthy revenue per employee for an agency?
Industry research puts the 2025 average around $163,000 per full-time employee, with figures below roughly $120,000 treated as a structural warning sign and specialist agencies typically targeting $250,000 or higher. The number is useful because it diagnoses which problem you have. Low revenue per employee with strong utilisation points to a pricing problem. Low revenue per employee with weak utilisation points to a headcount problem. Many agencies discover they have both.
Do retainers still retain better than project work?
Substantially, and the gap is wide. Analysis separating agencies by engagement model found retainer relationships churning at roughly 18% annually against about 42% for project-based work, with average client lifespans near 56 months versus 24. Retainers have also become the dominant primary model for digital agencies. The caveat for 2026 is that a retained client is no longer necessarily a stable one, because the retainer itself can shrink considerably while the relationship continues.
