When AI does production, the agency business model shifts from selling hours to selling judgement, owned category intelligence and accountable outcomes. Retainers priced on headcount lose their logic once a first draft takes minutes. Agencies that reprice around what clients still need from people, and run agents for the rest, protect their margin; agencies that simply discount hours to match AI savings shrink.
Key takeaways
- FTE-based retainers invite a simple procurement question: if AI saves a fifth of the time, why not a fifth of the fee?
- Holding companies are answering with AI platforms and outcome-linked contracts; independents are answering with output pricing.
- Clients still pay for judgement, accountability and distinctive work, and are wary of speed that produces generic output.
- The agencies that gain from AI reinvest the time it saves into more pitches, billable work and owned intelligence rather than giving it away.
Why is the retainer model under pressure?
Because most retainers are priced as people multiplied by time, and AI shrinks the time. When the input falls, the client reasonably asks why the price does not.
That question is already in the room. Digiday reported in 2025 that CMOs were asking agencies whether a 20% cut in the time needed to manage media should mean a 20% cut in fees, as S4 Capital described a move from billable hours towards pricing tied to outputs (Digiday).
Agencies often make it worse themselves. In a July 2026 piece on what procurement leaders want, one procurement leader described agency decks promising work "78% faster and 40% cheaper", which leads procurement straight to asking for lower fees (Campaign). The same piece notes that agencies are open to alternatives, including a move from fixed fees towards outcome-based models.
What are holding companies doing about it?
They are building AI platforms and starting to tie fees to results. Both moves try to make the value of the agency something other than hours.
The Media Leader's April 2026 overview lists the main platforms: WPP Open, Publicis Groupe's Marcel, Omnicom's Omni, Dentsu.Connect, Havas Converged.AI, Stagwell's The Machine and Brandtech's Pencil (The Media Leader). Several sit on large identity and data assets, such as Epsilon at Publicis and Acxiom within Omni.
WPP has gone furthest on pricing in public. In February 2026 it set out plans for performance contracts with fees tied to business results such as sales and brand performance, moving away from time and materials, and said it would pay its global client leaders on client growth (Digiday). Among independents, Digiday has reported shops such as Ace of Hearts pricing by output with no timesheets at all (Digiday).
What do clients still pay agencies for?
Judgement, accountability and work that does not look like everyone else's. Speed matters, but speed on its own is now cheap.
SOMIN's data shows where the friction is. In our ads-optimisation workspace, the insight engine holds an angle on goal misalignment, grounded in goal-setting theory: advertisers find that their objectives do not match the metrics their agencies prioritise. The insight built on it draws on 36 engagements across 10 posts. A related tension summary describes paid media agencies chasing vanity metrics rather than business outcomes.
A model-generated moment of truth in the same workspace says it plainly: "I need to evaluate my agency based on real results, not just ROAS or CPA." It is a model's reading, not a quote, but it is the brief every agency should expect at its next review. If AI makes production cheap, clients will judge you on what the production achieved.
Which pricing models fit agentic production?
Models that price outputs, outcomes or intelligence rather than time. Most relationships will end up as a hybrid of the options below.
| Model | What the client pays for | Fits when | Watch out for |
|---|---|---|---|
| FTE retainer | People's time, estimated as headcount or hours | Scope is stable and mostly human judgement | AI savings turn into requests for fee cuts |
| Output pricing | Deliverables: a pitch, a brief, a set of assets, a report | Production is repeatable and agents do much of it | Agreeing the quality bar for each output |
| Outcome fee | Agreed business results, such as sales or brand measures | Both sides trust the measurement | Attribution disputes and factors outside the agency's control |
| Intelligence subscription | Ongoing category data, monitoring and analysis | The agency owns data or tooling the client values | It must stay better than what the client can buy directly |
| Hybrid | A strategy retainer, priced outputs and an outcome bonus | Most mid-sized relationships | Complexity; keep it to three parts the client can explain internally |
The strategy retainer survives in most of these models. What changes is that it pays for thinking and accountability, while production moves to unit prices that reflect how the work is actually done.
What does the shift look like inside an independent agency?
In the agency case studies we have published, the time AI saved went into more pitches, more billable work and a sharper point of view.
- KPI Media cut pitch preparation from a one-week sprint to two days and prep cost from $4,000 to $1,000 per pitch, then closed a new $5,000-a-month retainer (KPI Media case study).
- NEO360 went from three people per proposal to one account manager, with cost per proposal down from $1,000 to $700, returning its media buyer and analyst to billable client work (NEO360 case study).
- Blak Labs reduced client-reporting effort from 4 FTEs to 1.75, with that workstream's monthly cost falling from $10,000 to $6,000 (Blak Labs case study).
- SAMY used a forensic category read to reframe a pitch for Nightcap Group around business impact and won a multi-million-pound retainer (SAMY case study).
The pattern matters more than any single number. Agents compressed the hours; the agencies chose where the hours went. Under a contract that simply billed fewer hours, most of that value would have gone straight back to the client.
How should an agency reprice without losing margin?
Separate the work by who should do it, then price each part on its own logic. A practical sequence looks like this.
- Audit the hours. Split current work into judgement, repeatable production and research. Be strict about which is which.
- Move research and production to agents. Measure the time saved on real briefs, not in a demo.
- Price outputs for repeatable work. A category audit, a brief or a set of tested concepts each gets a unit price.
- Keep a strategy retainer for judgement. Tie a share of it to outcomes you and the client both measure.
- Own something the client cannot buy alone. Category data, method and a record of what works are what make your outputs worth more than a tool's.
- Disclose how you use AI. Procurement will ask; a clear answer is better than a defensive one.
The fifth step is where many agencies stall, so we cover it separately in why category data beats a better prompt. For how agencies run the research and pitch loop on SOMIN, see our agency solutions.
FAQ
Will AI kill the agency retainer?
Not entirely. Retainers for strategy and accountability will remain, but retainers priced purely on hours or headcount will keep shrinking as AI takes over research and production.
What is outcome-based pricing for agencies?
It ties part of the agency's fee to agreed business results, such as sales, leads or brand measures, instead of the time spent. It works best when both sides trust the measurement.
Should agencies lower fees because they use AI?
Not automatically. If AI lets you deliver more or better work, price the outputs and outcomes; if it only makes the same work faster, expect clients to ask for a share of the saving.
What are holding company AI platforms?
They are group-wide systems such as WPP Open, Publicis Marcel, Omnicom Omni and Dentsu.Connect that combine data, AI models and agents for planning, creative production and optimisation.
How can independent agencies compete with holding company AI?
By specialising, pricing outputs instead of hours and building deep category intelligence in the markets they serve. Off-the-shelf AI agents and category data are available to independents too.
Pick one service line this quarter, usually pitch research or monthly reporting, and run it through agents on live briefs. Track the hours saved, decide where they go, and propose an output price for that service at the next renewal.
