The pitch tells you who sells well. An audit of your own account tells you who can actually run it.
The way to evaluate a PPC agency is on work, not on a presentation. Give every finalist read-only access to your live ad accounts, ask each of them for a written audit of that same account inside the same week, and grade the audits against each other. The agency that finds the real problem in your account before it has a contract is the one most likely to fix it after.
This is written for B2B marketing leaders who are running real spend, have thin paid expertise on the internal team, and are about to choose between three or four agencies whose decks all look the same.
→ A pitch measures sales skill. An audit of your account measures diagnostic skill, which is the thing you are buying.
→ Read-only access is genuinely safe. Google's own access levels let a user view campaigns and run performance reports while being unable to edit anything.
→ Run all the finalists at once on the same account, in the same week, against the same questions.
→ Score the audits on what they found and what they refused to promise, not on how polished the document looks.
Why the pitch tells you almost nothing
Every agency deck has the same four slides. Logos, a team page, a process diagram, and two case studies with percentages and no starting numbers. None of that predicts what happens in your account in month three.
Credentials are real, and they are also not the thing. An agency full of ex-platform people can still hand your account to someone managing eleven others. A three-person shop can be the sharpest paid team you ever work with. The deck cannot tell those two apart, and neither can a reference call, because you are handed the references who were always going to say yes.
The claim that trips up the most evaluations right now is AI. Almost every agency currently describes itself as AI-forward, and for most of them that means they use the same automated bidding and asset generation everybody else uses, because it is built into the platforms. It is a positioning claim, not a capability. The only way to test it is to make the agency show you a decision its system made and a decision it stopped a human from making.
So stop asking for pitches earlier and start asking for work earlier.
Ready for paid ads that pay off?
Book your free auditGive every finalist read-only access, at the same time
The instrument you want is a pre-engagement audit of your live account, produced by each finalist independently, with no hand-holding from you.
Google Ads. Invite each agency as a user at the Read-only access level from Admin, then Access and security. Per Google's own access-levels documentation, a Read-only user can view campaigns, use planning tools, and edit and run campaign performance reports, and the same table shows that editing campaigns requires Standard or Admin. They can see everything and change nothing.
LinkedIn Ads. Campaign Manager assigns every user a role that determines what actions they can take, and LinkedIn's user roles and permissions page covers adding users with access to view, edit, or create campaigns and ads. Add the finalists at the viewing level only.
Analytics and your CRM. Give read access to the analytics property. Do not give CRM access during an evaluation. Instead, write down three numbers on one page and hand the same page to everyone: average deal size, lead-to-close rate, and the maximum you can pay for a qualified lead. If an agency does not ask for those numbers, that is your first data point.
Running the finalists concurrently is the part most buyers resist. It feels discourteous, and there is a worry about three strangers poking around at once. Do it anyway. Sequential evaluation over six weeks means each agency sees a different account in a different state, which destroys the comparison. Concurrent evaluation gives you three reads on one account in one week, and the differences between those reads are the entire signal.
This also filters for something useful. An agency that will not look at your account before a signed contract has already told you how it works. The audit is the least expensive thing we produce and the most useful thing a prospect gets, which is why our own paid-ads engagements start with one.
Grade the audit, not the document: five things a real one contains
Polish is free now. Anyone can produce forty clean slides. Here is what actually separates them.
A specific number with the mechanism attached. Not "there is significant wasted spend." Something closer to this. A stated share of last quarter's spend went to queries that cannot convert, here are eight of those queries, here is the match-type setting that let them in. A number without a mechanism is a guess dressed up as a finding.
A challenge to your conversion action. This is the single strongest tell in the whole process. Most reported conversion counts in B2B accounts are wrong in a way nobody has checked: a tag firing on the wrong element, a form that got replaced, four conversion actions all feeding bidding at once. An agency that accepts your conversion number at face value will spend the next year optimizing toward it. One that opens the audit by asking whether the number is real has done conversion-tracking work before.
A first-90-days sequence with an order and a reason for the order. Any list of twenty fixes is a list. A sequence tells you they understand dependency: tracking before bidding, structure before budget, page before scale. Ask why item one is item one. The answer is where competence lives.
At least one thing they would not do. An audit where every finding leads to more spend, more channels, and more scope is a sales document. Real audits contain subtraction. Turn this off, pause that, stop paying for this placement.
A clear statement of what they could not see. Read-only access has limits, and an honest audit names them. The agency that writes "we could not verify offline conversion imports because we had no CRM access, so treat the pipeline numbers in section three as unconfirmed" is the one whose reporting you will be able to trust later.
Five questions to ask each finalist after the audit
Send all five in writing, to everyone, and give them a day. Written answers are comparable. Live answers reward whoever presents best, which is the thing you are trying to stop measuring.
1.) Which finding do you fix in week one, and when will we know it worked? The second half matters more than the first. A credible answer accounts for how long a change takes to read. Google documents that after a bid strategy change it can take "up to around 50 conversion events or 3 conversion cycles" for the strategy to calibrate (source). An agency promising a clean verdict in two weeks on an account producing twelve conversions a month either has not done the arithmetic or is telling you what you want to hear.
2.) Who touches this account, how often, and how many other accounts do they carry? You want a name, a cadence, and a number. Vagueness here is the most reliable predictor of a quiet account six months in, which is the same failure mode as an incumbent agency going passive.
3.) Which decisions does your automation make on its own, and which stop with a person? Any answer that draws a real line is fine. An answer that treats AI as a general capability rather than a set of specific decisions tells you the claim is marketing. We wrote a whole diagnostic for the software-versus-agency version of this question.
4.) What would make you tell us to spend less? Almost nobody has an answer ready. The ones who do are describing a real threshold, usually something about lead quality, sales capacity, or a channel that stopped being incremental. An agency that cannot imagine recommending a budget cut is not going to recommend one.
5.) What does this cost, and what changes the price? Get the fee, the scope, what counts as extra, and what triggers a change. Our own pricing is published with what each tier includes for exactly this reason, and how an agency handles fixed fee versus flexible scope tells you what its incentives will be when your priorities shift mid-quarter.
Score it on one page, and do not average
Four columns. Diagnosis, honesty, delivery, and cost. Fill one row per finalist with a sentence in each column, not a score out of ten.
Averaging hides the thing you need to see, because the risks are not interchangeable. An agency with a brilliant diagnosis and a thin bench is a capacity risk you can manage by keeping the scope narrow. An agency with a deep bench and a generic audit is a judgment risk, and no amount of scope management fixes that. Rank the risks you would rather carry, then choose.
Two disqualifiers regardless of everything else. An audit that got a factual detail about your account wrong, because whoever wrote it did not look carefully and will not look carefully later. And any agency that would not tell you something you did not want to hear across four written answers and one live conversation.
What it means when every audit is mediocre
Sometimes three agencies produce three thin documents. When that happens, the fault is usually not in the shortlist.
The common cause is that the account genuinely cannot be read. No trustworthy conversion signal, spend spread across five channels nobody owns, and an unverified tag setup mean there is nothing to diagnose, so everyone defaults to generic advice. Fix the signal first, then run the evaluation. You will get three sharper audits.
The other cause is scope. Asking four agencies to audit search, paid social, programmatic, and review-site placements at once produces four surveys. Pick the channel carrying the most spend and make the audit go deep on that one. Depth on one channel is a better capability test than breadth across four.
The three-week version of this process
Week one is the shortlist. Grant read-only access to every finalist on the same day and hand each one the same page of internal numbers. Week two, collect the written audits and send the five questions. Week three is one working call per finalist with no slides allowed, spent walking their audit of your account, after which you fill in the one-page scorecard and decide.
Three weeks is enough. The reason to keep it tight is that a long evaluation selects for the agency with the most patient sales process, and that has never been the same thing as the agency that will run your account well.
If you want to see what a real audit of your account looks like before you commit to anything, that is how our paid ads work starts, and our Clutch profile is where the reviews live if you want to check the work first.
