Key takeaways
- AI cut agency fees by 15 to 30 percent at many US agencies while headcount per account stayed roughly flat, compressing margins from both sides.
- The efficiency stalled because context re-entry and senior verification were never automated, only the output generation step was.
- Automating context capture eliminates the briefing ritual and lets workflows check actions against current account reality, reducing the senior review tax.
- Pricing models that separate AI-assisted execution from accumulated strategic judgment give agencies a defensible fee structure clients cannot replicate for twenty dollars a month.
- Context must live at the agency level, not inside individual account managers, or key-person risk deepens rather than shrinks with every AI adoption.
The squeeze is real, and the tools are not the problem
If AI cut your fees but not your costs, you are not alone. Across US performance marketing agencies, the pattern is consistent: clients pushed retainers down 15 to 30 percent (a rough estimate based on conversations, not a published benchmark) once AI became visible in the workflow, while headcount per account stayed roughly flat. The tools are genuinely good. That is exactly what makes this confusing.
The problem is not the AI. The problem is where efficiency was supposed to come from, and why it stalled before it reached your P&L.
Why the efficiency never arrived
The standard promise was straightforward. AI handles the repetitive work, senior people focus on judgment, the same team runs more accounts, margin expands. Two years in, that math has not closed for most agencies. Three things explain why.
Context still gets typed in by hand
Every AI tool on the market is good at the 30 percent: platform mechanics, bid logic, creative structure, keyword patterns. What none of them hold is the other 70 percent: why the account is structured the way it is, what you tested and killed last quarter, what the client decided on Monday, what Legal rejected last month.
That context lives in meeting recordings, Slack threads, account manager heads, and three-year-old Notion docs nobody has updated. Before a senior person can use AI productively on an account, they spend time, often 15 to 25 minutes per session by common report, typing that background in. The AI is fast. The briefing ritual is not. So the time saving is real but smaller than expected, and it falls entirely on the person who already knows the account.
The output still needs a senior check against account reality
AI is confidently wrong in ways that are hard to spot unless you know the account. It will propose a test you already ran and killed. It will optimize for a KPI you dropped at the last quarterly business review. It will write copy around a claim Legal rejected. None of those errors are visible to a junior reviewer. They require the same senior person who would have done the work manually.
So the review cycle did not compress. It just moved later in the process. The deliverable gets produced faster and checked just as carefully. Net time saved per account: modest. Net cost saved: close to zero.
Clients see the AI and reprice accordingly
This is the asymmetric part. Clients do not see the briefing time or the senior review. They see that you are using the same tools they can access for twenty dollars a month. The conversation shifts from "what does your expertise cost" to "what is the AI markup."
A typical US performance agency retainer for a mid-market account ran somewhere between eight and fifteen thousand dollars a month before 2023. Anecdotally, many agencies report pressure to reduce that by 20 percent or more, with clients explicitly citing AI. The fee dropped. The cost structure did not.
What the P&L actually looks like now
Run the rough numbers. An agency with ten accounts at an average of ten thousand dollars per month is doing one million dollars in annual retainer revenue. A 20 percent fee reduction brings that to eight hundred thousand. If headcount costs stayed flat at, say, sixty percent of revenue, the old cost base was six hundred thousand against one million. The new cost base is still six hundred thousand against eight hundred thousand. Margin compressed from 40 percent to 25 percent, and that is before the new line item: AI tool subscriptions, API spend, and in some cases internal engineering time to build and maintain proprietary tooling.
Those are illustrative figures, not published data. But the direction is consistent with what agency founders describe.
Where the fix actually lives
The efficiency promise was never going to come from faster output alone. It had to come from eliminating the two tax lines that absorbed the time savings: context re-entry and senior verification.
Eliminate those two, and the math changes.
Automate context capture, not just task execution
If the context is already in the system before anyone asks a question, the briefing ritual disappears. That requires something that reads meetings, Slack, email, ad platform data, and client decisions as they happen, and keeps a current, structured record per account without anyone writing a brief.
When a workflow runs on that context layer, it does not just execute a trigger. It checks the action against what the team actually decided. It will not scale a campaign the team agreed to sunset. It will not recommend a test the account already ran. The senior review tax collapses because the system already holds the account reality the senior person was checking against.
Let context compound across the portfolio
The second efficiency is the one most agencies have not yet priced. Nine years of pattern recognition across forty accounts is the actual product. It currently lives in people. When it lives in a system, versioned and retained per client, two things happen.
First, onboarding a new account manager goes from three weeks of tribal knowledge transfer to something much shorter. Second, when a client asks why they are paying agency fees when they have access to the same AI, you can show them something concrete: the accumulated test history, the reasoning behind account structure, the decisions already ruled out and why. Their AI knows the internet. Yours knows their category across forty accounts and can demonstrate it.
Pricing model options for US agencies in 2026
Fixing the cost structure creates room to rethink how you charge. A few models that fit the new reality.
Retainer plus performance tier
A lower base retainer, closer to what clients now expect given AI, plus a performance fee tied to agreed metrics. This works when you have enough confidence in your system to accept outcome risk. It also reframes the conversation: you are not charging for AI access, you are charging for judgment that produces results. The base covers costs; the performance tier is where margin lives.
Typical structure: a base retainer covering account management and reporting, plus a fee equal to a percentage of media spend above a performance threshold, or a flat bonus tied to ROAS or CPA targets. Specific percentages vary widely by account size and category.
Outcome-based project fees
For clients resistant to retainers, scoped projects with defined deliverables and clear success criteria. An account restructure, a creative testing sprint, a feed optimization. These price the output, not the hours. If context capture reduces the hours required, margin expands without the client seeing it.
Portfolio pricing
For agencies managing multiple brands for a single holding company or franchise group, a portfolio rate that covers all accounts under one agreement. This works when your context layer holds cross-account learning, because the value you deliver scales faster than your cost.
Tiered service levels
A structured menu: a lower tier for accounts that need platform management and reporting with AI-assisted execution, a higher tier for accounts that get senior strategy, cross-account learning, and direct access to accumulated test history. The lower tier is priced to compete with in-house AI. The upper tier is priced on expertise that cannot be replicated with a twenty-dollar subscription.
The condition that makes any of this work
None of these pricing models hold unless you can actually deliver on the efficiency side. The reason agencies are stuck is not that they chose the wrong pricing model. It is that the context problem was never solved, so the senior review tax never went away, so the cost structure never moved.
The sequence matters: fix context capture first, demonstrate that the same team can run more accounts without the quality degrading, then reprice from a position of genuine margin rather than optimistic projection.
Where Almond fits
Almond is built specifically for this problem. It is a Business Context Engine for performance marketing: it reads ad accounts, Slack, Notion, Sheets, email, meeting notes, and CRM in read-only mode, captures every test, result, and decision as it happens, keeps it current, and ranks it by relevance, recency, and business impact.
For agencies, every client context store is isolated and scoped at the agency level, not the account manager level. When someone leaves, they take the relationship. They do not take the account history. Workflow Builder runs 30 prebuilt Google and Meta automations that act on that context, so they will not execute against a decision the team already reversed. Canvas surfaces the accumulated learning in client-ready boards.
The early access program includes a detailed 1:1 Meta and Google ad account audit with the CEO, recorded. Waitlist at almond.ai requires a work email.
Frequently asked questions
- Why did AI reduce agency fees without reducing agency costs?
- Clients saw AI tools in agency workflows and repriced retainers downward, assuming the work was largely automated. But the core cost driver, senior review of every deliverable against account-specific reality, did not change. Context still gets typed in manually before each session, and outputs still need expert verification, so headcount per account stayed flat while fees fell.
- What is a realistic agency retainer pricing model in 2026 given AI pressure?
- Most US performance agencies are moving toward a lower base retainer combined with a performance tier tied to agreed metrics, or tiered service levels that separate AI-assisted execution from strategic judgment. The key is pricing the outcome and the accumulated expertise, not the hours or the tool access, since clients can access the same tools independently.
- How does context capture reduce the senior review burden in agency workflows?
- When a system continuously captures meeting decisions, test results, and account changes, it holds the account reality that senior reviewers were checking AI output against. Automations that run on that context layer will not recommend killed tests or deprecated KPIs, which eliminates the most time-consuming verification steps and lets senior people focus on genuine judgment calls.
- How can a performance agency demonstrate that its AI is different from what a client can access directly?
- The differentiator is accumulated, client-specific context: the test history, the reasoning behind account structure, the decisions already ruled out and why. A generic AI knows the internet. An agency with a context engine knows that client's category across multiple accounts and can show the evidence. That is a concrete, demonstrable asset rather than an assertion about expertise.
- What is the key-person risk problem with current agency AI tools?
- Most AI tools are only useful on an account when operated by the person who already holds its context. That person is also the only one who can catch confident errors. So the tool deepens dependency on the account manager rather than reducing it. When that person leaves, the AI becomes unreliable on those accounts until someone else rebuilds the context, which takes weeks.
Almond 2.0 early access
Get a 1:1 Meta and Google audit with the CEO.
Every waitlist member gets a 1:1 audit of their ad accounts. Work email only.