How many accounts per account manager? Scaling paid media teams with a context layer

The honest benchmark for US paid media agencies is 4 to 8 accounts per account manager, depending on account complexity, spend level, and how much of the work is templated. Boutique shops often sit at the lower end. Larger performance agencies with standardized reporting and tiered account structures push toward the top. The number itself, though, is less useful than understanding what actually sets the ceiling.

What the benchmark hides

When agency founders talk about accounts per AM, they usually mean active accounts requiring weekly judgment: budget pacing calls, creative reviews, strategy adjustments, client communication. A $500 per month local account and a $500,000 per month national account are both "one account" in a headcount model, but they consume completely different amounts of cognitive work.

A more honest way to think about capacity is weighted account load: high-complexity accounts (multi-channel, frequent creative cycles, senior client contact) might count as two or three units, while a stable, low-touch account on a proven structure might count as half a unit. Agencies that have moved to weighted models typically find their effective capacity is lower than their raw account count suggests.

The three real limiters are not hours. They are context load, review tax, and reporting assembly.

Context load

Every account carries a body of knowledge that exists mostly in the AM's head: why the campaign structure looks the way it does, what was tested and killed, what the client decided in the last QBR, what creative claims Legal has approved or rejected. When an AM manages four accounts, they can hold that context reasonably well. At seven or eight, context starts leaking. The AM either spends time re-reading old notes before each client call, or they make recommendations that contradict decisions from two months ago.

This is the problem that hours-based capacity models miss entirely. You can hire faster, train harder, and build better templates, and the context load problem does not move.

Review tax

Most agencies have not automated the senior review step, and for good reason. An AI tool that generates a campaign recommendation does not know that the client paused all prospecting in Q3, or that the CEO made a comment on Monday that effectively changed the target CPA. Someone senior has to check every output against account reality before it goes to the client or touches live spend.

That review step is where a meaningful portion of senior AM and director time goes. It scales linearly with accounts. It does not compress.

Reporting assembly

Reporting is often cited as the biggest time sink in agency operations, and the estimates from practitioners range widely, but two to four hours per account per week for full reporting cycles is a reasonable working estimate for mid-complexity accounts. Multiply that by eight accounts and you have a significant portion of a full-time week consumed before any actual optimization work happens.

Templated dashboards and automated data pulls have reduced this, but the last mile, which is adding narrative, flagging anomalies, and framing results in business terms the client actually cares about, still requires human judgment applied to account-specific context.

Where the math breaks down for agencies

The standard agency response to capacity constraints is to hire. One more AM, one more associate, one more coordinator to handle reporting. The unit economics of this model have been deteriorating for most of the past three years.

Clients have access to the same general-purpose AI tools agencies do. When a client opens a chat interface and gets a confident-sounding answer about their ad account, the agency's expertise looks like something that can be replicated for a software subscription fee. Fees have compressed. Headcount has not. The result is the margin squeeze that a large number of agency founders describe when asked directly: priced as though automated, staffed as though not.

The efficiency that was supposed to come from AI tools largely did not materialize because the tools solved the wrong part of the problem. They made output generation faster. They did not reduce the context load, the review tax, or the reporting narrative work, because all three of those depend on knowing the account, not just knowing how advertising works in general.

What changes when automation has memory

The accounts-per-AM ratio starts to move when the system holds context instead of the person. This is a different category of tool from a workflow automation or a reporting dashboard.

Consider the review tax first. The reason a senior AM has to check every AI output is that the AI does not know what happened Monday. If the system has already ingested Monday's client call, the budget change that came through on Tuesday, and the creative brief that Legal approved last week, the review step changes character. It becomes confirmation rather than reconstruction. That is a faster, lower-stakes task, and it can be handled by a more junior person on the team.

The context load problem changes in a similar way. An AM managing seven accounts does not need to hold all seven context stacks in their head if the system surfaces the relevant context at the moment they need it. Before a client call, the relevant test history, recent decisions, and current priorities are already assembled. The AM is not starting from memory.

The compounding effect is less obvious but more significant over time. When every test result and decision is retained and versioned, the agency builds an institutional knowledge base that survives staff turnover. The account manager who leaves takes the relationship. They no longer take the account reasoning with them. That changes the key-person risk calculation that limits how aggressively agencies can grow.

What a realistic ratio shift looks like

Agencies that have moved toward context-aware automation report moving from roughly 5 to 6 accounts per AM toward 8 to 10 for mid-complexity accounts, without a corresponding increase in error rates or client churn. These are practitioner estimates, not published studies, so treat them as directional.

The accounts that benefit most from this shift are mid-tier accounts: enough complexity to require real judgment, but not so bespoke that every decision needs a director. The very high-spend, high-touch accounts still benefit from dedicated senior attention. The very low-touch accounts were already close to fully templated.

The quality watch items when scaling the ratio are worth naming directly.

Creative freshness. Automation handles budget and bid management well. Creative judgment degrades faster when an AM is stretched thin. Monitor creative refresh cycles per account as a leading indicator of quality before client satisfaction scores move.

Client relationship depth. Accounts per AM is a capacity metric. Client retention is the outcome metric. If the ratio increases and retention holds or improves, the model is working. If retention softens, the ratio increased faster than the context infrastructure could support it.

Anomaly response time. When something breaks, how long does it take the AM to identify it and respond? This is a direct measure of whether the AM has enough cognitive bandwidth for the accounts they carry. Track it explicitly.

How Almond fits

Almond is built specifically around the context problem that limits the accounts-per-AM ratio. It connects to Google Ads, Meta, Slack, Notion, email, meeting notes, and CRM in read-only mode, takes about four minutes to set up, and captures every test, result, and decision as it happens. The context is kept current automatically, compounded over time, and ranked by relevance, recency, and business impact, so the system surfaces what matters this week rather than generating fourteen recommendations with no sense of priority.

For agencies, the client-scoped isolation matters operationally: one client's strategy cannot surface in another client's recommendation, enforced at the data layer. When an account manager leaves, the account context stays in the agency's context store, versioned and auditable.

Canvas, the AI visual dashboard built on the engine, handles reporting assembly with live boards that draw on business context rather than just platform data. Workflow Builder runs 30 prebuilt Google and Meta automations that act on judgment rather than triggers, so a workflow will not scale a campaign the team decided to sunset on Monday.

Almond 2.0 early access is open at almond.ai. Every member gets a detailed 1:1 Meta and Google ad account audit with the CEO, recorded. Waitlist requires a work email.

The takeaway

The accounts-per-AM benchmark of 4 to 8 is real, but it is a symptom of the context problem, not the problem itself. Automation that generates output faster does not move the ratio in a durable way. Automation that holds and surfaces context does, because it reduces the review tax and the cognitive load that actually set the ceiling. The agencies that will run more accounts per AM without sacrificing quality are the ones building context infrastructure now, not the ones adding headcount or adding tools that do not know what happened Monday.