AI adoption in agencies stalls after the pilot because the obstacles are human and operational, not technical. People fear for their roles, perfectionists rewrite AI drafts by hand, juniors feel overlooked and nobody has protected time to learn. Agencies restart adoption by saying plainly what AI will and will not change, picking one measurable workflow, funding practice time and treating review as a recognised craft.
Key takeaways
- Most agencies already use generative AI, but most use it on less than half of their work, so the stall is about depth, not access.
- Agency staff name skills gaps and lack of time as their biggest barriers, ahead of the technology itself.
- Fear of replacement, perfectionism and undervalued juniors recur in SOMIN's social-listening data, and each needs a different fix.
- Frame AI around one workflow with a before-and-after number, and move the hours saved into billable or strategic work rather than cutting the people who saved them.
- Make checking AI output a named, credited skill, backed by a shared library and a written definition of good enough.
Where does AI adoption in agencies actually stall?
It stalls between experimentation and routine use. Almost every agency has licences and a few enthusiasts; far fewer have changed how a typical week of client work gets done.
The 4As and Forrester put numbers on this in their second State of GenAI Adoption study. Three quarters of respondent agencies used generative AI in 2025, up from 61% a year earlier, yet nearly 80% used it on less than half of their work. The share citing a lack of understanding of the tools more than doubled, to 42%.
Sentiment is cooling at the same time. The 2026 Advertising Agency Report from Basis, a survey of more than 200 agency professionals, found that the share of AI pessimists rose from 19.9% to 32.0% in a year. Broad, shallow use combined with growing scepticism is what a stalled rollout looks like from the inside.
Why do agency teams resist AI tools?
Because AI touches identity: what people are paid for, what they take pride in and how they expect to be promoted. A licence does nothing to answer those questions, so people quietly protect themselves.
SOMIN's own data shows the pattern. In our social-listening workspace, a 30-day whitespace study of 455 posts (August 2026) grouped 75 of them under one tension theme, "Navigating Professional Challenges". It describes anxiety from overthinking, the pressure of perfectionism, a fear of relying on AI, and interns and junior staff who feel undervalued. Three of those threads explain most stalled rollouts.
Fear of being replaced
SOMIN's moments-of-truth model, which generates the micro-moments a persona is likely to act on, surfaces the thought "I want to understand how AI can improve my workflow without replacing my role." It is a model-generated moment, not a quote from a real person, but it captures the trade-off people are weighing. If leadership never answers it, staff answer it themselves by using AI as little as possible where managers can see.
Perfectionism and the revision trap
In SOMIN's audience-research workspace, a September 2026 whitespace study of 220 posts put 91 of them under the theme "Navigating the AI Marketing Maze". A central complaint is AI-generated content that needs substantial human revision. A perfectionist who rewrites every draft concludes, reasonably, that it is faster to start from a blank page.
A related SOMIN insight, "Authenticity thrives where insecurity ebbs", is built on the imposter-phenomenon angle across 64 posts: self-doubt makes marketers less confident in their decisions. AI output that nobody has agreed how to judge feeds exactly that doubt.
Juniors who feel undervalued
The tasks AI absorbs first, such as desk research, first drafts and reporting, are usually the ones juniors owned and learned from. Take them away without redesigning the role and juniors disengage, which removes the people most likely to experiment. We cover the fix in detail in how to train junior marketers on AI.
Is the real barrier skills or time?
Both, and time is the one leaders underrate. In an agency, learning hours compete directly with billable hours, and client work wins unless someone ring-fences the learning.
EMARKETER reported in January 2026 on AI Digital's Ad Agency Generative AI Adoption Benchmark. Skill gaps and training needs were the top barrier (60.7%), followed by being too busy with daily work (51.8%) and technology integration (44.6%). Only 16.1% of agencies had comprehensive AI policies and training in place, and 58.9% had no shared prompt or best-practice library.
Read together, the figures describe agencies asking individuals to figure AI out alone, in their spare time, with nothing written down. That is a management problem, which is good news: management problems can be fixed by management decisions.
How do you restart a stalled AI rollout?
Match each symptom to its cause, then fix the cause rather than buying another tool. The table below maps five common patterns to a specific response.
| Symptom | Likely cause | Fix |
|---|---|---|
| People avoid the tools when managers are watching | Fear for their role | Write down what AI will and will not change, and where saved hours go |
| AI drafts get rewritten from scratch | Perfectionism, no agreed standard | Define good enough with an evidence checklist |
| Juniors stop volunteering ideas | Their learning tasks were automated away | Give them review and insight work, and credit what they catch |
| "We'll try it after this pitch" | No protected time | Ring-fence learning hours as a non-billable investment |
| Everyone prompts differently | No shared library | Build one team library of prompts, workflows and examples |
- Frame the change in writing. A SOMIN insight built on the framing-effect angle across 105 posts makes the point that how AI tools are presented shapes how teams judge them. Present AI as the end of a role and you get resistance; present it as the end of a specific chore, with the freed hours going somewhere named, and you get trials.
- Start with one repeating workflow. Choose research-heavy, internal work such as competitor audits, pitch research or monthly reporting. Time it before and after.
- Redeploy, do not cut. When NEO360 moved proposal prep from three people to one account manager, the media buyer and analyst went back to billable client work rather than out of the door, as the NEO360 case study describes. Staff watch closely what happens to the first hours saved.
- Protect practice time. Put a few hours a fortnight in the resourcing plan and treat them like a client commitment.
- Make review a craft. Agree what a checked output looks like: sources traced, claims tested, audience fit confirmed. Then credit the people who catch errors, not only the people who ship fast.
Which workflow should an agency automate first?
The one that repeats every week, eats senior hours and never reaches the client in raw form. Competitive research and reporting fit that description in most agencies.
That is where the agencies on our SOMIN for agencies page started: category audits, whitespace maps and pitch briefs built from live competitor data, with a person signing off at each step. The same loop runs through SOMIN's AI agents, and teams that want a low-risk way to practise can run it in Claude with the free Marketing Mondays plugin before changing any client process.
What should agency owners measure in the first 90 days?
Measure adoption depth and time returned, not licences issued. Five numbers are enough:
- Share of eligible work in the chosen workflow that now runs through the new process.
- Hours per cycle on that workflow, before and after.
- Where the saved hours went: billable work, pitches or strategy.
- Entries added to the shared library, and how often they are reused.
- A short monthly pulse on how confident people feel using AI on client work.
Once those move, you can translate them into money. Our companion piece on how to measure AI ROI in agencies shows how, with figures from five agency case studies.
FAQ
Why do agency employees resist AI tools?
Mostly because AI changes what they are valued for. Fear of replacement, perfectionism about AI drafts and juniors losing the tasks they learned from are the most common reasons, and none of them is solved by buying more software.
What is the biggest barrier to AI adoption in agencies?
Skills and time. In AI Digital's benchmark, reported by EMARKETER in January 2026, 60.7% of agency professionals named skill gaps and training needs, and 51.8% said they were too busy with daily work.
Should agencies cut headcount after adopting AI?
Cutting the people who produced the first savings is the fastest way to stop everyone else from producing more. Redeploying freed hours into billable work, pitches and strategy keeps adoption moving and usually shows up in revenue.
Who should lead AI adoption in an agency?
A senior owner with authority over resourcing, supported by a practitioner who runs the workflow every week. Adoption stalls when it sits with an enthusiast who cannot protect anyone's time.
This week, pick one repeating workflow, time it, write a one-page note on what AI will change and where the saved hours will go, and block two hours of practice time for the people who run it. Review the five numbers above at day 30, 60 and 90, and expand only when the first workflow has stuck.
