Question lists for agencies are easy to find. The ones we read were written by agencies, with the answers their writer would like to hear. We did not find one that shows how to recognise a dodge. So for each question below, we say what a straight answer contains and what an evasive one sounds like. Some are questions agencies would rather not hear in a pitch. Ask them anyway.
We are an agency too. The same list applies to us, and we would rather you used it. For the wider decision, see how to choose an ecommerce marketing agency.
One rule for reading everything below. A straight answer contains a method and a limitation. "It depends" is fine if it is followed by what it depends on.
Questions about measurement: how will you know it worked?
Ask: "How will we know the work added sales, rather than collected credit for sales that would have happened anyway?"
That is the incrementality question: the extra sales or conversions caused by an activity, over and above what would have happened without it.
A straight answer: Names how they would estimate it, such as a holdout, a geo test or a comparison against a baseline. It also says what the method cannot tell you, and what they would do if volume is too low for a clean test.
A dodge: "Our platform ROAS is strong." Platform ROAS is reported by the same platform the agency is buying from. A newsletter from a measurement company, which sells measurement reviews, argues that ROAS is a number the agency has a lot of influence over, and suggests asking for estimated incremental ROAS and how it is measured1. That is a seller's argument, not proof. The question holds up either way.
Follow-up: "What is the incremental ROAS (iROAS) on this account, and how was it measured?" iROAS is incremental revenue, meaning revenue caused by the ads, divided by ad spend.
Then ask about the big picture. "What is our MER (marketing efficiency ratio) today, and what would you expect it to do?" MER is total revenue divided by total marketing spend, without splitting credit between channels. It cannot say which channel worked. It does not depend on any platform's counting.
Also ask which attribution models the reports use. Attribution assigns credit for a purchase to the clicks and ads on the path before it. It is a rule for splitting credit, not proof of cause.
Good questions cover measurement, money, ownership and limits.
Questions about money: how are you paid, and what does that reward?
Ask: "How are you paid? What happens to your fee if we spend less?"
Every fee model rewards something. Your job is to find out what.
A straight answer: Describes the model plainly and says what it rewards. If the fee rises with spend, a good agency says so and explains how it handles the conflict. A revenue share fee means the agency receives a percentage of the brand's revenue, either total or incremental, often on top of a base fee. For that model, the answer must also say which revenue counts and what the baseline is.
A dodge: "Our fee scales with your success." Success is a fuzzy word. Spend is a number.
Why a fee tied to spend deserves a follow-up is covered in marketing agency red flags.
Ask next: "Tell me about a time you recommended cutting our spend."
If the fee rises with spend, this story cost the agency money, which is what makes it a useful test. A good answer has a specific channel, a reason and what happened to sales afterwards. "We have never needed to" is an answer too. You now know it.
Questions about access and ownership
Ask: "Who owns the ad accounts, pixels and data? Do I have admin access today?"
Refusing to give clients admin access to their own accounts is on agency-written red-flag lists, for example ATTN's2.
A straight answer: You own them. The agency is a user. The answer says what happens on exit, what you can export and who can see what.
A dodge: "It is simpler if we manage it all." Simpler for whom?
Ask next: "Will you work with our own attribution or test tools, or only yours?" A straight answer says yes and explains how it would handle a disagreement between the two. A dodge says the other tool is wrong.
Questions about reporting: what will I see each month?
Ask: "Can we see a report you send to a current client, with their details removed? Which decision did it lead to?"
A pitch is the best version of an agency. A real report is the everyday version.
A straight answer: A report that shows platform numbers next to the client's own sales, explains the gap and labels what is measured, what is assumed and what is unknown. It ends with a decision: increase, change, stop or keep checking.
A dodge: A dashboard demo and "you can see everything in real time". Seeing everything is not the same as knowing what to do.
Watch out for the word "scale". In agency reports it is usually shorthand for increase budget or scope. Ask for the result that justifies it.
Questions about limits: when would you tell me to spend less?
Ask: "What can't you tell us yet?"
Every honest agency has a list. If theirs is empty, they have not looked, or they are not saying.
A straight answer: Names open questions, such as what cannot be measured with your volume, which channels are hard to read and where the data is thin.
A dodge: "We have full visibility." With consent banners, several platforms and their own counting rules, nobody sees everything.
Ask next: "Which of the results on your website were measured by someone other than you?" Self-measured results are not worthless, but you should know which ones they are. Ask us the same.
And: "When would you tell me to spend less, or to stop a channel?" The answer shows whether the agency can say no to its own revenue.
What does a straight answer sound like compared with a dodge?
| Topic | Straight answer | Dodge |
|---|---|---|
| Measurement | A method, plus what it cannot show | "Our ROAS is strong" |
| Money | What the fee rewards, and how the conflict is handled | "We win when you win" |
| Access | You own the accounts | "We handle it all for you" |
| Reporting | Your shop numbers beside platform numbers, with the gap explained | Platform screenshots only |
| Limits | A list of unknowns | "We have it covered" |
| Method | Explained in plain words | "Proprietary" or "trust the algorithm" |
"Proprietary" and "trust the algorithm" are not answers. They are the end of the conversation.
Use the same list with every agency on your shortlist, in the same order. Write down the answers. A comparison only works if the questions did not change.
For ZFD: We want these questions asked of us. We would rather lose a pitch on a straight answer than win it on a dodge.