Why we killed the retainer and what replaced it

The month a client paid us $9,000 to do almost nothing

We had a retainer client on $9,000 a month. One quarter, their priorities shifted and there was genuinely little for us to do. So we invented work. We ran a brand audit nobody asked for. We rebuilt a landing page that was already converting fine. We sent a deck full of charts that changed no decisions.

The client paid the invoice. We felt sick about it. That month made the misalignment impossible to ignore: the retainer paid us to fill hours, and filling hours is the opposite of the job. So we killed the retainer across our whole book and replaced it with something that ties our revenue to what actually moves for the client.

Why the retainer misaligns everyone

A fixed monthly fee creates three quiet incentives that work against the client.

First, it rewards staying busy. When the fee is the same whether we ship one high-leverage thing or twenty low-leverage things, the safe move is volume. Volume looks like effort. Effort is what gets renewed. So agencies optimize for the appearance of motion.

Second, it punishes efficiency. We are an AI-native agency running 83 agents. When an n8n workflow does in four minutes what used to take a strategist two days, a retainer means we just absorbed that gain and the client saw none of it. The faster we got, the worse the deal became for them. That is backwards.

Third, it hides the unit of value. A client paying $9,000 a month cannot tell you what a single deliverable cost or what it returned. The retainer bundles everything into one opaque number, so nobody can prune the work that does not pay.

We ran the math on our own book before changing anything. Across 11 retainer clients, roughly 40 percent of logged hours went to work that produced no measurable change in pipeline, revenue, or retention. We were billing for motion.

What replaced it: outcome tiers plus a throughput meter

The model has two parts that sit side by side.

The first part is outcome pricing for the things we can attribute. A booked qualified meeting from a cold-outreach sequence has a fixed price. A net-new piece of pipeline above a threshold has a price. A retained client who would have churned, caught by a win-back flow, has a price. These are priced per result, and the result is logged in Atlas the moment it fires. When a positive reply gets classified, a sales-pipeline row is created automatically, and that row is the billing event. No invention required.

The second part is a throughput allotment for everything that does not attribute cleanly: design systems, content libraries, automation buildouts. We sell these as a fixed number of shipped units per month, where a unit is a defined artifact with an acceptance bar. A client buys, say, 12 shipped units a month. A blog article is one unit. A working n8n workflow with monitoring is two. We publish the unit menu so there is no ambiguity about what counts.

The key difference from a retainer: if we do not ship the units, we do not bill them, and they do not roll into a vague sense of work performed. Throughput is counted, visible, and capped. The client sees exactly what they bought arrive.

A concrete example

One client moved from a $7,500 retainer to the new model in March. Their plan: 8 throughput units a month plus outcome pricing on outreach meetings at a fixed rate per booked call.

In April we shipped 8 units (3 articles, a Klaviyo flow rebuild counted as 2, two landing pages, one reporting dashboard) and booked 14 qualified meetings through the cold-email engine. Their total invoice came in at $8,900. Higher than the old retainer, but every line was a thing that arrived or a meeting that happened. The client could see the 14 meetings in their own CRM.

In May their team got slammed and asked us to pause new buildouts. Under the retainer, we would have invented filler and billed $7,500. Under the new model, we shipped 2 units and booked 9 meetings, and the invoice was $4,200. We billed less because we did less, and the relationship got stronger for it. They trusted the number because the number matched reality.

What this forced us to fix internally

The model only works if attribution is honest, so it pushed us to clean up our own measurement. Outcome pricing means we cannot fudge what counts as a qualified meeting, so we wrote the qualification rule down and let the reply classifier enforce it. A meeting counts when the prospect is in the target segment, has booking authority, and shows up to a call that runs past five minutes. Those three conditions are checked automatically off the pipeline row, not argued about at invoice time. Throughput pricing means a unit has to have a real acceptance bar, so every unit type now has a definition of done that a human validates before it counts. An article is not a unit until it passes the editorial gate. A workflow is not a unit until it runs in production with monitoring attached.

It also changed our internal economics in a good way. Because efficiency now flows to the client through lower invoices when we do less, the gains from automation show up as margin on the outcome side, where we get faster and cheaper per result while the price per result holds. The 83 agents make us money on outcomes rather than by letting us bill more hours.

Handling the objections we expected

Two objections came up immediately, one from clients and one from our own team.

Clients worried about budget predictability. A retainer is a flat line they can plan around; an outcome-and-throughput model moves month to month. We solved this with a floor and a ceiling. The throughput allotment is a committed monthly minimum, so there is always a predictable base. The outcome side has a soft cap the client sets, and when a strong month would push outcome billing past that cap, we flag it before the work happens so they choose to greenlight the overflow or bank the leads for next month. Predictability comes from the floor; upside is opt-in.

Our team worried that outcome pricing would make income lumpy and stressful, tying paychecks to things partly outside their control. So internal compensation does not pass the volatility straight through. The team is paid on shipped throughput and on a rolling three-month average of outcomes, which smooths the spikes. The agency carries the month-to-month variance so the individual operator does not. That separation matters: the operator's job is to do excellent work, and excellent work over a quarter shows up reliably even when a single month is quiet.

How we transitioned the existing book

We did not flip all 11 clients overnight. We ran a parallel quarter where we kept billing the retainer but also tracked what each client would have paid under the new model, then showed them both numbers side by side. For most clients the new number was within ten percent of the old one, which made the conversation easy: same rough cost, far clearer accounting. Two clients would have paid noticeably more because we were doing high-outcome work that the flat retainer had been undercharging for. One would have paid less because the retainer had been padded. We moved that last one first and on purpose, because leading with the client who saves money builds trust for the rest of the rollout.

The parallel quarter also caught our own pricing mistakes. Our first unit menu mispriced workflow buildouts too low relative to the effort, and outcome rates for enterprise meetings too flat against the deal sizes they produced. We only found that by running real months through the model before committing to it. Anyone copying this should run the shadow quarter before announcing anything.

What we would tell another agency

If your fee is the same whether you ship the one thing that matters or twenty things that do not, you are being paid to look busy. Pick the handful of outcomes you can attribute cleanly and price them per result. Sell everything else as counted, capped throughput with a real acceptance bar. Then let your automation make you cheaper per unit instead of letting it pad your hours.

We have run this for a full quarter now. Revenue per client is up, churn is down, and we never again have to invent a brand audit nobody asked for. More on how we wire attribution and outcomes lives at arthea.ai.