Five questions for your agency instead of another report

An agency report can be complete and still not help. It covers spend, clicks, conversions, ROAS and the comparison with last month, yet at the end of the meeting nobody is sure what should change. The problem is rarely the quality of the table. The agency sees the ad account; the owner sees margin, stock and commercial priorities. A decision only happens where those two views meet, and five questions are enough to get there.

Why a report does not decide anything

The owner–agency relationship is a textbook principal–agent relationship. Agency theory describes how the agent knows more about its own work than the principal, and how reporting is one of the tools that narrows that gap. This is not about distrust. Each side simply holds a different half of the picture.

The agency knows what happened in the auction, with budgets and with ads. The owner knows product margins, what is running low in stock, how many orders the warehouse can handle and what the business wants to sell next month. A report that contains only the first half ends in a verdict (“it went well / less well”). The questions below push the meeting towards the second half and towards a decision.

Five questions and what a good answer looks like

1. What changed?

First separate the changes the agency made from changes outside advertising: season, prices, product availability, competitors or a change in tracking. Google Ads keeps a change history that shows who changed a budget, targeting or keywords and when, on the same timeline as performance.

• A good answer: two or three material changes, each with when it happened, who made it and which comparable period it is measured against.
• Warning sign: a list of metrics that went up or down with no distinction between the agency's decisions and outside influences.

2. Why does the agency believe it?

The type of evidence matters here. Platforms attribute conversions according to an attribution model – a rule for dividing credit among clicks and views. That is different from measuring how many orders would not have happened without the ads. Google and Meta both offer separate tests with a control group that did not see the ads.

The difference is not academic. A large experiment at eBay found that ads on brand keywords had no measurable short-term benefit there, because customers would have arrived anyway. That was a very well-known brand, and a smaller store whose name competitors bid on may be in a different position. It still shows that attributed revenue can overstate the effect. A comparison of experiments at Facebook also found that common observational methods often fail to reproduce the result of a randomised test.

• A good answer: the agency says whether this is platform attribution, timing that coincides, or a test with a control group, and speaks about certainty accordingly.
• Warning sign: “the algorithm is learning” and nothing more, or a before-and-after comparison presented as proof that a specific change caused the improvement.

3. What is the business effect?

ROAS, or its Czech inverse PNO (ad cost as a share of revenue), based on attributed revenue tells you how much revenue the platform credits to the ads relative to spend. The owner, however, cares about what is left after paying for goods and advertising. Translate the result into margin: gross margin from orders minus ad cost, ideally next to the store's total revenue rather than only the revenue credited to campaigns. Google Ads can report gross profit when it receives cart data and product cost in the feed, but the same calculation can be done outside the platform.

An illustration: a campaign with 100,000 CZK of attributed revenue and 15,000 CZK of spend has a PNO of 15%. At a 30% gross margin, 15,000 CZK remains after advertising; at a 20% margin, only 5,000 CZK. The same ratio means very different outcomes for different product ranges, so the campaign target should come from margin, not from a generic benchmark.

• A good answer: the effect in margin and total revenue, with a note on whether the share of new customers or the product mix has changed.
• Warning sign: only conversion counts or ROAS, or conversions that include micro-actions such as add-to-cart.

4. What exactly will you change?

A useful meeting ends with one or two decisions, not a promise to “keep optimising”. This is where the owner comes in: is the goal growth or return, what is the break-even margin, which item is running out, how many orders can the warehouse handle. Without that, the agency optimises towards a goal it had to guess.

• A good answer: a specific change (budget, bidding target, excluding part of the range, a new campaign), the expected direction of the effect and what the agency needs from you.
• Warning sign: a general plan with no change, or a change unconnected to the margin or constraint you named.

5. How will we know it worked?

Agree the success criterion in advance: which metric, over what period, against what comparison and what you will do if the result does not arrive. Where volume allows, Google Ads offers experiments that split traffic between the original and the modified campaign. Google recommends running them for at least four to six weeks and warns that low-volume campaigns produce inconclusive results.

• A good answer: a metric, a window, a comparison and a rule such as “if X, we roll back” – agreed today, reviewed at the next meeting.
• Warning sign: success is defined afterwards according to whatever happened.

Do not demand proof the data does not have

The opposite mistake is to demand that the agency prove every change caused the result. Research covering 25 large field experiments showed that even for large advertisers the confidence interval around advertising returns is often very wide, because individual purchases vary so much. A small or mid-sized store often lacks the volume for a conclusive test, and Conversion Lift in Google Ads is not available to every account.

The reasonable request is therefore not “prove it” but: state how strong the evidence is, make changes reversibly and review them against a rule agreed in advance. A change that coincides with a result is a useful signal. It just must not be presented as proof.

What to do now

1. Send the questions in advance. The agency has time to prepare answers instead of another table.
2. Add your half. Margin or break-even ad cost ratio, stock and fulfilment constraints, and one commercial priority for next month.
3. Ask for the type of evidence. Attribution, coinciding timing or a test.
4. Leave with one decision. Write down the change, metric, window and rollback rule.
5. Start the next meeting with the review. New numbers come after.

Korzaro puts ad spend and the revenue ads attribute to themselves next to the store's total revenue and margin after advertising. For changes in campaigns it compares equal periods before and after the change, without claiming that the change caused the movement. You bring a shared basis to the agency meeting; the decision stays with you.