How to choose a performance marketing agency.

An agency writing a guide to choosing agencies has an obvious conflict of interest, so let us defuse it the only honest way: by starting with the decision that can rule us out. Most selection processes begin by comparing agencies to each other, which quietly assumes the answer to a prior question nobody asked. There are four decisions here, they run in order, and the first one has arithmetic behind it.

• →Do you need outside help at all, or is your spend past the point where a hire is cheaper?
• →What shape of help — freelancer, individual specialist, agency, in-house, or a hybrid?
• →Which specific provider, judged on evidence rather than pitch quality?
• →How is the arrangement structured so that it can fail cleanly and early rather than drift for a year?

Decision one: the crossover arithmetic

The reason this comes first is that agency fees scale with spend while a salary does not, so there is a crossover point where hiring becomes straightforwardly cheaper. It is calculable. Take a fully-loaded senior paid media hire — salary plus payroll costs, tooling, and the management overhead of employing someone, which together typically run 1.25x to 1.4x the salary:

Senior in-house operator, salary        $110,000/yr
Fully loaded at ~1.3x                  $143,000/yr
                                       ~$11,900/mo

Agency at 12% of spend reaches $11,900 at
  monthly spend of                      ~$99,000

So the naive crossover sits near $100k/mo of spend.

Below roughly that level, an agency is usually cheaper than the equivalent internal capability — not marginally, but by a wide margin at smaller spends, where the fee buys a fraction of a senior operator you could not otherwise employ. Above it, the arithmetic starts favoring the hire, and past $250,000 a month it favors the hire decisively.

Except the naive crossover is too low, for three reasons that are structural rather than self-serving:

• →One hire is one person's capacity. A $100,000-a-month account across search, shopping and Meta with a real creative operation is more than one person's job, so the honest comparison at that spend is often two hires against one agency fee — which moves the crossover up substantially.
• →One hire is one person's pattern exposure. An operator sees your account; an agency operator sees dozens, which is where diagnostic speed comes from. That advantage is real but decays — after a year inside your business, the in-house operator's context advantage starts outweighing the agency's pattern advantage.
• →One hire is a single point of failure. Holidays, illness and resignation each stop the account, and the resignation case takes months to recover from. Agencies carry bench depth you are effectively renting, and the value of that insurance is proportional to how much revenue runs through the channel.

Decision two: what shape of help

If the crossover says outside help, the next question is what kind, and the options are structurally different rather than tiers of the same thing. A freelancer sells hours, an individual specialist sells judgment, an agency sells a system with bench depth, and a hybrid keeps strategy in-house while renting execution. Each has a spend band where it stops making sense and a specific failure mode; the comparison is worked in freelancer versus expert versus agency.

One shape-level filter is worth applying early, because it eliminates a lot of shortlists: on Meta the deliverable is a stream of creative rather than account management, and the volume required can be derived rather than guessed. If your derived requirement is seven net-new concepts a month and the arrangement you are considering cannot produce seven, the shape is wrong regardless of how good the people are. The derivation is in how much a Facebook Ads agency costs.

Decision three: which provider

Only now does comparing providers make sense, and the operative principle is that a question only helps if a weak provider cannot answer it well. "What is your process" fails that test — every agency has a process slide. "Show me the last report you sent a client whose performance was down" cannot be faked, because the artifact either exists or it does not.

The question that carries the most signal is the one to ask before any contract, in writing: how will we know if this is working? A serious answer is a measurement plan — which metric governs, how brand and non-brand are separated, how platform-reported results get reconciled against your revenue, and what the honest evaluation window is with a reason for its length. A weak answer is a ROAS figure from someone else's account. A disqualifying answer is any version of "we use the platform's reporting," because Google and Meta will each claim the same conversions and neither nets out demand that would have converted anyway — the problem in MER vs ROAS and brand vs non-brand.

The rest of the selection questions, each paired with the answer that should end the conversation, are in questions to ask before hiring an ads agency — including the four popular questions that reveal nothing at all.

Red flags visible before signing

• →Guaranteed results in a discovery call — nobody who has run real accounts guarantees outcomes before seeing your data.
• →Audit findings that are all crisis and no nuance, delivered with a countdown-timer discount. Manufactured urgency is a pricing strategy, not a diagnosis.
• →Reluctance to give you admin access to your own accounts "for technical reasons." There are no such technical reasons.
• →Case studies with percentages and no context — "+400%" of an undisclosed base, over an undisclosed timeframe, is decoration.
• →A pitch that never asks about your margins, sales cycle, or capacity. Anyone proposing spend levels without unit-economics questions is planning to optimize a number disconnected from your business.

Decision four: structure the fee and the trial

Two structural choices remain, and both determine what you experience monthly for years. The fee model sets the incentives — each of the four models creates a conflict with a calculable size, and each has one contract term that neutralizes it, laid out in PPC management pricing models compared. Judge the resulting number against your margin rather than against an industry average: the profit-share and break-even-improvement arithmetic is in how much a Google Ads agency costs.

Then design the trial so that it can fail. The pitch measured sales skill; the first ninety days measure the product. That means a defined initial scope with explicit deliverables, a foundation-first plan rather than immediate spend heroics, and — the part almost everyone skips — a decision rule agreed in advance:

• →What is expected to be visibly fixed by day 30 (tracking integrity, structural problems, brand separation) — these are verifiable regardless of performance.
• →What performance movement is expected by day 90, stated as a range rather than a number, on a metric you both named at the start.
• →How the transition dip is treated. A handover normally costs a few weeks of below-normal performance while structural fixes land; agree its expected size up front so month four is not spent arguing about whether the drag counts.
• →What happens if the day-90 read is inside the range, below it, or ambiguous. Deciding this in advance is what separates a trial from a rolling contract nobody has the information to cancel.

Agencies confident in delivery accept this readily. Agencies that resist a structured trial are telling you which part of their funnel is strongest, and it is not the part you are buying.

Where this framework has limits

• →The crossover arithmetic assumes you can actually hire the operator it describes. Senior paid media talent is scarce, and a crossover that says "hire" is worthless if the role sits unfilled for six months while the account drifts.
• →It uses a single salary point and a single fee percentage. Both vary widely by market and by scope, and the conclusion is sensitive to both — treat the ~$100,000 figure as a midpoint to recalculate, not a constant.
• →Decision three tests transparency and self-awareness, which correlate with competence without being identical to it. A candid provider can still be wrong about your account.
• →The whole sequence assumes the channel is worth running at all. If the unit economics do not support paid acquisition — a possibility the arithmetic in how much a Facebook Ads agency costs can surface — then no provider decision fixes it, and the correct move is to repair the margin or the offer first.

If you want to see how we answer decision three ourselves, that conversation is free — and if the crossover arithmetic says you should be hiring rather than outsourcing, we would rather tell you that early than bill you for a year first.