Executives: proving your advertising caused the sale

Your advertising report attributes sales to your campaigns. It does not demonstrate that these sales would not have happened without them. Google's own documentation acknowledges this explicitly. Measuring that difference changes how a budget gets allocated far more than any campaign optimization.

On this page

Attribution is not causality, and Google says so

This article is written for executives and marketing managers whose annual advertising budget exceeds the threshold where a mistake gets expensive, say one hundred thousand dollars. Below that, the methods described here remain valid, but the exercise pays off less.

An attribution model splits credit across the touchpoints it was able to observe. It can be consistent, stable and refined while staying silent on the question that matters: would these sales have happened without the advertising?

The clearest case is brand search. A customer who already knows you, types your name, clicks your ad and buys will be counted as an advertising conversion. They probably would have found you without you paying for that click. The report did not lie, it simply answered a different question.

Google acknowledges this in its own documentation. It describes its incrementality measurement tool as measuring the true causal impact of advertising, and states that it intentionally ignores conversion tracking parameters and standard attribution rules. In other words, the platform itself distinguishes the number on your dashboard from the causal effect.

What a test actually measures

The test follows the logic of a controlled trial: one group sees your ads, a comparable group does not, and you compare sales between the two. The measured gap is the causal effect of the advertising.

What you get at the end is not a conversion count. Two numbers decide a budget. The incremental cost per action, that is, what a sale you would not otherwise have had costs. And the incremental return on ad spend, what each additional dollar returns once you exclude the sales that would have come without advertising. Google publishes both metrics in its Conversion Lift tool.

The practical difference is stark. A campaign showing a cost per conversion of $90 can have an incremental cost per conversion of $240, because two-thirds of those conversions would have happened anyway. You are not making the same decision depending on which number you look at.

Three methods, from simplest to most rigorous

The first method is the one most companies can run as early as this quarter. It proves nothing on its own, but it produces an order of magnitude that is worth infinitely more than no measurement at all. The condition is to cut cleanly, over a period that covers the buying cycle, and to change nothing else during that period.

The geo test, practical in Quebec

It is the method that suits the largest number of companies here, for three reasons.

It works with sales closed offline, which is the case for any company selling by quote, by rep or in-store. It requires no individual-level data, which settles a good share of consent questions. And its geo-split methodology is public, so a third party can verify it.

The main trap is contamination. If a buyer sees your ads in an exposed region but buys in a control region, the measured gap shrinks and the real effect gets underestimated. Google names this risk explicitly in its documentation. In Quebec, contamination mostly lurks between neighbouring regions and around major hubs.

Concretely, a manufacturer that ships across the province can expose some regions and not others for eight weeks. It then compares the inquiries received and the sales closed by region, accounting for population and history. It is a serious, doable exercise, and almost nobody offers it in this market.

The long sales cycle case

If your sales close in six or nine months, a four-week test measures nothing useful. The window has to cover the cycle, which raises two difficulties worth naming in advance.

The first is duration. A test that runs nine months crosses seasons, price changes and sometimes staff turnover. The longer the window, the more other factors muddy the measured gap.

The second is organizational patience. A leadership team that agrees to deprive a region of advertising for three quarters is rare, and it has good reasons to resist.

The approach that holds up is to measure an intermediate step, the quote request or the qualified opportunity, which arrives within a reasonable window. You then need to check that the close rate stays stable between the exposed group and the control group. It is not perfect, and it needs to be stated that way rather than presented as a sales measurement. The logic for connecting the request to the sale is covered in our article on cost per signed quote.

The limits worth stating

An incremental result is never a certainty. It comes with a confidence interval, and the point of the exercise is to reduce the risk of a decision, not eliminate it. A vendor who presents a conversion lift, the sales gap measured between the exposed group and the control group, of 23% with no margin of error is not giving you a measurement, they are giving you a number.

Second limit: a test measures a situation, not a law. The incrementality of a brand campaign in January does not predict that of a product campaign in May. Tests get repeated, they do not get filed away as permanent truths.

Third limit: volume. Below a certain number of conversions per region and per period, the measured gap drowns in noise. A low-volume company gains more from properly measuring its cost per sale than from launching a test that will conclude nothing.

Opening a territory raises the same causality question, developed in entering a new region.

The overall view of paid channels is in choosing and concentrating advertising budgets.

Frequently asked questions about incrementality