Guide
Guide · Agency Evaluation
Choosing a marketing agency well begins with defining what you are actually buying, because most mismatches between clients and agencies are category errors, not quality problems. This guide covers how to read an agency's numbers honestly, the questions that separate operators from resellers, the warning signs worth walking away from, and the circumstances in which the right answer is not to hire anyone at all.
First principle
The category of service matters more than the quality of the agency within it. An exceptional agency in the wrong category will underperform a mediocre one in the right one.
Marketing is not one thing. The term covers at least four distinct activities that are often quoted under the same brief:
These are complementary but not interchangeable. An agency that is strong at channel management will grow your Instagram audience; it will not, on its own, drive a release week. An agency built around paid media will move your cost-per-click; it will not build organic credibility. Before you evaluate a single agency, write down the one number you need to move and then identify which of these four activities is most directly responsible for moving it.
If the answer is unclear, that is a signal to slow down. A brief that does not name the outcome tends to produce a proposal that names everything, a retainer that charges for everything, and results that are attributed to everything. None of those are in your interest.
The most common failure mode is not fraud or incompetence. It is a client buying channel management when they needed distribution, or buying creative production when they needed media buying. Both parties can perform their side of the contract competently and the business outcome can still fail because the service purchased was not the lever that needed to move. Naming the category first eliminates this before the contract is signed.
Evaluation framework
These are not the only criteria, but they are the ones most commonly skipped and most reliably predictive of whether an engagement will produce a result you can act on.
A named client, a stated outcome, and a window. If a case study names the client but not the result, or the result but not the method, it is marketing copy, not evidence. Ask which of their results you could independently verify and what you would need to do so.
The agency's incentive structure should point at the same number yours does. A retainer paid regardless of performance is not automatically a problem, but you should understand what the agency optimizes for within that retainer, and whether it is the same thing you are optimising for.
How do they count a conversion? Which attribution model? Over what window? Any agency presenting results without stating their measurement methodology either does not know or does not want you to know. Both are problems.
Ask who actually does the work. Agencies that white-label third-party services are not inherently dishonest, but you should know you are paying a margin on top of a service you could access directly, and you should know what happens to your account if that relationship ends.
An agency that will take any brief is an agency that will take yours even when it is wrong. Ask what they have declined and why. The answer tells you more about their operating standards than any case study.
What did the last campaign that underperformed teach them, and what changed as a result? Operators learn and adapt. Resellers do not, because the work is not theirs. An inability to answer this question in specific terms is itself an answer.
Due diligence
A figure without its window and its method is a claim rather than evidence. The distinctions below are not pedantic, they change whether a number tells you anything useful about what your campaign will do.
Every headline metric an agency puts in a pitch deck is a choice. The same underlying data can produce a figure that looks dramatically different depending on four variables: the window it covers, the counting method, whether it is an average or a typical result, and how attribution is handled. Understanding each lets you ask the right follow-up.
"560 billion views since 2019" and "560 billion views per campaign" are different claims by several orders of magnitude. Cumulative figures since founding look large but tell you nothing about what a campaign at your budget, in your category, over the next ninety days, will actually produce. Ask for per-campaign figures and ask what the distribution looks like, a mean can be pulled far from the typical result by one outlier.
Platform-reported figures come from the dashboard the platform controls. Attributed figures run through the agency's own model, which may assign credit to the agency's activity that the platform alone would not have given. Neither is wrong in principle, but they are different numbers, and conflating them overstates the agency's contribution. Ask which figure you are looking at and what the attribution model is.
An average is a mean. If an agency ran ten campaigns and nine returned 1× and one returned 100×, the average is 10.9×, a figure that accurately describes none of the ten. Ask for the median and ask how often campaigns in your category have landed at or above the average they are quoting.
Every figure needs both stated explicitly. A "18.6× average ROAS" figure, for example, needs to say: ROAS over what window? Measured by what method, last-click, multi-touch, view-through? Including or excluding organic lift? These are not gotcha questions. Any agency presenting performance data should be able to answer them without pausing.
The four cells above are not metrics from a campaign. They illustrate the four components a properly-qualified figure must carry. Use them as a checklist when reviewing an agency's pitch deck.
Due diligence process
These five questions are not a formality. Each one separates an agency that operates its own capability from one that manages relationships and marks up third-party services. The quality of the answer matters as much as the answer itself.
This question has no wrong answer except no answer. Every legitimate operator has a list, purchased engagement, misleading attribution claims, campaigns they consider off-brief, categories they do not serve. A fluent, specific answer tells you the agency has made these decisions deliberately. A hesitant or vague answer tells you they will take any work that comes in the door. Both are useful to know before you commit.
An operator who does the work has a direct answer to this. They can name the campaign (or describe it without identifying the client), explain what they expected, what happened instead, and what they adjusted. A reseller cannot answer it specifically because the learning belongs to the party who actually ran the work. Vague answers about "iteration" and "optimization" that do not describe a concrete change are a signal.
Ask for the name and role of the person who will manage your account, and ask whether they are an employee or a contractor. Ask which parts of the service are delivered in-house and which are delivered through a third party. The answers to these questions change what happens when something goes wrong, who owns the relationship, who has the data, who can make a decision without escalating.
This tests whether the agency sees its job as executing instructions or producing results. An agency that only executes instructions will execute yours without objection even when the brief is pointing at the wrong problem. An agency that has told a client their brief was wrong, and can describe the conversation, is an agency that will tell you the same if warranted. That is the relationship worth paying for.
Ask what reporting access you will have and whether you will be able to cross-reference their figures against a source you control. Platform dashboards, GA4, a dedicated analytics tool. An agency that reports only through its own dashboard, with no path to independent verification, is asking you to trust numbers that have no external check on them. That is a structural problem regardless of whether the numbers are accurate.
Red flags
These are not stylistic preferences. Each one indicates either an inability to deliver what is being promised, or a deliberate choice to obscure what is being delivered.
No legitimate agency guarantees a specific outcome in advance of seeing the data, the creative, and the competitive landscape. A guarantee stated in a pitch, "we will get you to X", is either a sign that the outcome is so modest it is trivially achievable, or that the number will be produced by whatever means necessary, including artificial ones. Ask what happens to the guarantee if the method is questioned.
Bought followers, views, and likes produce numbers in a dashboard and nothing in the market. They also actively damage the account they are applied to: platforms identify inauthentic engagement and reduce the distribution of content from accounts that carry it. The damage persists after the purchased engagement is removed. Walk away from any agency that offers this service, explicitly or implicitly.
"Client results on request" and anonymised case studies are not case studies. They are claims that cannot be checked. An agency that has produced real results for real clients can name at least some of them, even under NDA constraints, and can state the outcome in measurable terms. Generic testimonials and stock photography are not a substitute.
An agency that agrees with everything in your brief before any real conversation has happened is an agency that has not evaluated your brief. The whole value of an external operator is that they have seen enough briefs and enough results to know which ones are pointing in the wrong direction. An agency that will not tell you that has not earned the relationship.
Honest assessment
Distribution exposes the underlying asset. If the asset is not ready, distribution accelerates the problem rather than solving it. An agency that takes this work anyway is not doing you a favour.
Marketing moves people toward a product. If the product produces a poor experience on arrival, the marketing has done its job and the product has failed at its job. More reach does not fix that, it scales the negative feedback faster than organic word of mouth would have done. The correct sequence is product first, distribution second.
The first campaign an agency runs for a new client is rarely the highest-performing one. The first flight establishes what works in your specific context, what hook the audience responds to, which channel converts, what the real cost per outcome is. That learning is what makes the second and third flights perform. A budget that can only support one flight produces data but no ability to act on it, and the client has paid for research rather than results.
There is no universal threshold for what "enough" looks like, it depends entirely on the category, the platform, and what a single customer is worth. But the question to ask is: if this campaign underperforms its projection, can we afford to run a revised version? If the answer is no, the engagement is a single bet, not a business decision.
Pricing is wrong. Distribution is broken. The sales process loses prospects that marketing delivers. In each of these cases, more marketing activity makes the underlying problem more expensive, not less. An honest operator will say so. If you have described your situation to an agency and they have not asked whether marketing is the right lever for the problem you have described, that is itself a warning sign.
Even when conditions are not ideal, a first engagement can be structured to produce usable intelligence. Define one testable hypothesis, not "let's try paid social" but "let's find out whether our target audience responds to problem-led or outcome-led copy, at this CPM range, on this platform." Define in advance what a successful test looks like, and what a failed one will tell you. Then whatever the outcome, you have information you can act on, and the engagement has served a purpose even if it did not hit its revenue target.
FAQ
These mirror the phrasings people and engines use, and each answer matches the FAQPage schema on this page word for word.
Define the one number the engagement is meant to move, and how you will measure it independently of what the agency reports.
Most agency relationships that end in disappointment do so because the client was buying one thing and the agency was delivering another, and neither party named the mismatch at the start.
Ask for a window, a method, and a baseline. A figure without all three is a claim rather than evidence. Cumulative totals since founding look large but tell you nothing about what a campaign in your category, at your budget, in the next ninety days will produce.
Per-campaign averages are more useful, but only if the methodology, what counts as a conversion, which attribution model, which window, is stated alongside them.
Five questions that separate operators from resellers: (1) What would you refuse to do, even if a client asked? (2) What did the last campaign that underperformed teach you, and what did you change? (3) Who actually does the work, employees, contractors, or a white-labelled third party? (4) Name a client whose brief was wrong and describe what you told them. (5) How will I be able to verify your reported numbers against a source I control?
Four patterns that should stop a conversation: guaranteed results stated in advance of seeing your data; engagement or follower metrics with no named methodology; case studies that name only the client and not the outcome; and an unwillingness to say a brief is wrong.
Legitimate operators know the limits of their service and will say so before taking money.
There is no honest benchmark that applies across categories, and any figure cited without knowing your sector, deliverable type, campaign length, and market will mislead you.
The more useful question is whether the fee is structured to align with outcomes. A retainer paid regardless of performance gives the agency no incentive to push hard. A performance component, even a small one, changes the conversation. Ask how the agency is paid and what happens to their margin if results are below the projection they gave you.
Most are not. The majority of ranked agency lists are paid placements, award submissions reviewed by the agency itself, or popularity votes with no verified outcome data behind them. A list that does not publish its methodology for ranking should be treated as advertising.
Verified third-party recognition, an Inc. 5000 rank by revenue growth, an audit by an independent measurement firm, is substantively different from a self-nominated award. The distinction is whether someone outside the agency checked the underlying numbers.
Keep exploring
Start a campaign
Share the campaign, release, event or launch you are planning.