Online advertisinghow to measure the effectiveness of an advertising campaign

How to measure the effect of an advertising campaign on your sales

Separate attributed sales from incremental sales. Choose a reliable test and use its results to adjust your advertising budget.

Mehdi B.

Co-founder · SEO & digital strategy

How to measure the effect of an advertising campaign on your sales
In brief

Sales attributed to a campaign do not prove that it causes them. Compare similar groups with and without the advertising being tested, then weigh the estimated additional sales against their cost and their uncertainty.

Your ad report shows sales. How many of them would you have made without the advertising? That is the whole difficulty when looking at how to measure the effectiveness of an advertising campaign. Attributed conversions are useful for running ads day to day. To defend a budget, you need something else: the number of sales the campaign really adds, with its margin of error.

Take an online shop that shows ads to its former visitors. Its report counts purchases after each exposure. The owner, however, wants to know how many of these orders would disappear if the campaign stopped. We follow this shop throughout the article.

Why does an attributed sale not prove effectiveness?

Attribution links a sale to an advertising interaction, a click or an ad viewed, according to rules and a window set by the tool. It shows that the ad was on the path to purchase. It does not show that it triggered it.

A shop customer discovers a product on a social network, receives an email, then clicks on a Google ad. Each of these tools can count the order as its own. If you add up their reports, you get more sales than the shop actually made.

Even without double counting, the figure misleads. A returning customer who already planned to come back may click an ad just before buying. The sale is attributed to the campaign, although it would have happened anyway. The incrementality test measures what remains once this existing demand is set aside.

How do you measure the effectiveness of an advertising campaign with a control group?

How to measure the effect of an advertising campaign on your sales

We compare two similar groups over the same period. The first sees the ads being tested. The second, the control group, does not. The gap between their purchases gives an estimate of what the campaign contributes.

When the tool allows it, people are split at random before the ads are shown. Comparing those who click with those who do not does not work: those who click often already had more desire to buy.

For the shop, the test covers the ads aimed at past visitors. The control group keeps access to the site, the e-mails and every other way to buy. So we measure what this campaign adds to everything else.

The analysis compares all purchases in each group, including those made without a single click on an ad, taking the size of the groups into account. It gives an overall estimate. It does not say, order by order, which one comes from advertising.

Which campaign and which method should you choose?

Start from a concrete decision: keep a campaign, reduce its budget or test an increase. Then choose an item big enough for the answer to change how you split your budget. Testing a small line of the account most often gives a figure you cannot do anything with.

The simplest option, when available, is the platform’s built-in test, often called “Conversion Lift”. It splits the groups itself and measures conversions. Check that it is available in your account before planning anything: Google states that it is not open to all advertisers.

The geographic test compares areas where delivery differs. It works if you track your sales by area and if your history allows a credible comparison. Before the experiment, the areas must have evolved in a similar way.

The shop can also exclude part of its audience from ads, provided it controls how visitors are split and how purchases are tracked. The control group must then be properly isolated. Switching off the whole campaign and comparing the following month with the previous one does not allow this.

The choice depends on your data, your sales volume and the cost of setting it up. Also count the sales you risk losing in the private ad group. This is the kind of framing we do in our Google Ads campaign management.

Do you have enough sales to get an answer?

Before launching the test, have it estimated what it is able to detect. This calculation relies on sales history, its variations, the planned duration and the size of the control group. It tells you whether the test can answer your question with the means available.

The question to ask fits on one line: “What minimum gain can this test detect?” If your decision rests on a small difference, a test that only picks up large increases is of no use to you. Extending the duration or reviewing the groups can improve the measurement.

The duration also depends on the time between the ad and the purchase. For its user-level tests, Google recommends more than 14 days, to be adapted to the conversion delay. That benchmark is not enough for a business whose customers take several months to buy.

For a geographic test, the GeoLift documentation recommends a stable history of at least four to five times the length of the test. The data must therefore be prepared before ads are switched off. Without a usable history, no table, however detailed, makes up for the gap.

If the volume stays too low, note it as “inconclusive”. Failing to detect an effect does not mean the campaign is useless. Extra visits or form submissions tell you about the journey, but they do not prove additional sales.

Which mistakes make the result misleading?

Comparing before and after mixes the effect of advertising with that of demand. A promotion, the holidays or a product launch are enough to move sales from one period to the next. A control group observed at the same time lets you separate the two.

During the test, avoid any change that affects one group more than the other. For the shop, a discount limited to certain areas or a local stock shortage is enough to skew the comparison. Note these events and check their effect before reading the gap.

Also check that the control group is truly excluded. If another campaign shows it the same ads, the gap between the groups narrows. Your other channels can keep running, as long as they do not blur the question being tested.

Finally, set the reading date and the decision rules before launch. Do not stop the test as soon as a figure pleases you. The range of uncertainty matters as much as the central estimate, especially if it includes a zero effect.

How do you turn the result into a budget decision?

In the dashboard, add the estimated additional sales, their range, the corresponding spend and the date of the test. Keep attributed conversions in a separate column. The manager can then see at a glance the gap between the usual tracking and the measured effect.

The cost per additional sale is the amount spent on the test divided by the estimated additional sales. If the test was about a budget increase, only take the added spend. Then compare this cost with your margin, once the variable costs of each order have been deducted.

More sales does not mean more profitability. For the shop, returns, discounts and processing fees reduce what is left to pay for acquisition. Work from validated orders and keep the same definition of margin from one test to the next.

If the result supports your business objective, keep the budget. If the estimated effect is too small to cover the cost, and the measurement is precise enough, consider reducing it. If the range allows a conclusion either way, do not move a large budget on the strength of this test alone.

For an increase, stay cautious: the measured return does not guarantee the return on the added budget. Increase in steps and review after each significant change in spend or targeting. This approach complements the allocation of budget across marketing levers.

Which checks should you run with your team?

Before setting aside a test budget, bring together the people in charge of campaigns, sales and data, and answer these questions together in a shared note.

  • What decision is waiting on the result? Name the campaign, the amount at stake and what you will do depending on the answer.
  • Which sale are we measuring? Specify where the orders come from, how you handle cancellations and how long it takes to see the purchases.
  • Which comparison holds up? Designate the control group, the exclusions to configure and the commercial changes planned during the test.
  • Who applies the decision? Name the person who updates the dashboard and presents the result at the next budget review.

Sources

AdvertisingSales measurementMarketing budget
Mehdi B.
About the authorMehdi B.

Co-founder of Smart Impact. Hooked on the web from the start, he launched his first project in 2006: an online music magazine still running today. With close to 20 years of SEO experience, a Swiss federal diploma in marketing and a solid geek streak, he turns clients' ideas, sometimes hazy ones, into concrete digital projects with his team.

LinkedIn
Further reading