Ad measurement: knowing what is actually working

You run ads for a month. The ad screen shows 200 purchases.

You check your bank account. Revenue went up, but not by 200 orders worth. Closer to 80 orders.

Ad platforms use broad attribution models — rules that determine which ad gets credit for a sale.

These models tend to count more sales than the ads actually drove, because they are built to connect ads to purchases whenever a connection is possible, even when it is uncertain. Here is why.

There are several reasons why platform numbers and store numbers can differ.

Cross-device tracking gaps

Someone sees your ad on their phone. Later, they buy on their laptop. The platform may not connect these two actions to the same person.

When tracking gaps exist, some platforms use statistical modeling — estimating conversions (purchases that resulted from an ad) based on patterns rather than direct tracking.

These estimates can overlap with sales already recorded through other channels, leading to a higher total than what actually happened.

People see multiple ads

Someone sees your Facebook ad, then your Google ad, then your email.

Who gets credit for the sale?

Often, all three platforms claim it, because each one sees that the buyer interacted with its ad before purchasing and credits itself with the sale.

Tracking gets blocked

Many people block ad trackers. When they buy, the platform does not know. To fill this gap, some platforms estimate purchases they cannot directly see, using patterns from the data they do have. These estimates are not always accurate.

Time limits on credit

If someone sees your ad today and buys in 30 days, did the ad cause the sale? The platform says yes if the purchase falls within the attribution window — the time period during which a sale is credited to an ad. But the person may have bought regardless.

Instead of relying on platform numbers, compare your business metrics before and after running ads.

The question: When you started running ads, did total revenue go up by more than you spent?

If you spend $1,000 on ads and your total revenue increases by $3,000, the ads are likely contributing — because if ads are the main thing that changed, they are the most likely cause of the increase.

If you spend $1,000 and revenue stays flat, the ads may not be driving new sales — regardless of what the ad screen reports.

(This comparison has limits — other factors like seasonality can also affect revenue. Section 6 covers when this method does not work.)

How to do this:

1. Note your average weekly revenue before ads
2. Run ads for at least 4 weeks
3. Calculate your new average weekly revenue
4. Compare the increase to what you spent

This is a rough comparison, but it uses data that comes from your own systems rather than from the ad platform. Store and payment data records actual completed transactions, not estimates.

For a more detailed view, these three metrics (numbers you track to measure performance) help:

Cost per sale

How much you spend to get one purchase.

Formula: Total ad spend ÷ Number of actual purchases (from your store, not from the ad screen)

Example: You spend $500 and get 25 real orders. Cost per sale = $20.

Profit per sale

After all costs (product, shipping, payment fees), how much remains from each sale.

If you keep $50 per sale and spend $20 to get it (as in the example above), you make $30 profit — profitable.

If you keep $25 per sale and spend $20 to get it, you make $5 profit — barely profitable.

If you keep $15 per sale and spend $20 to get it, you lose $5 — unprofitable.

Break-even cost

The maximum you can spend per sale and still make money. In the example above, you keep $50 per sale, so your break-even is $50. Anything above that is unprofitable.

This number helps determine how much room you have before ads become unprofitable. (Break-even is also defined in Smart Words below.)

To use these metrics accurately, you need to know whether the platform numbers are correct. Here is a way to check if the platform is over-reporting:

Step 1: Get the platform number

Look at your ad screen. How many purchases does it claim?

Step 2: Get your real number

Look at your store or payment system. How many actual orders came in during the same period?

Step 3: Compare

If the platform says 100 and you see 100, the tracking is accurate.

If the platform says 100 and you see 60, it is over-reporting by 40%.

If the platform says 100 and you see 120, it is under-reporting (missing some).

What to do with this:

Once you know the gap, you can adjust. If the platform over-reports by 40%, reducing platform-reported numbers by 40% gives a closer estimate of actual performance. This is not exact — the over-reporting rate can vary across campaigns — but it provides a more realistic baseline than accepting the platform number at face value.

Comparing numbers is one way to check. Another approach is to turn ads off for a week and observe what happens.

What to look for:

If sales drop significantly when ads stop, the ads were likely the source of those additional sales — because if ads are the only change, they are the most likely cause of the drop.

If sales drop slightly, the ads may be contributing some revenue, but less than the platform reported.

If sales stay about the same, the ads may not be driving as many sales as the platform reports.

This test requires enough sales volume to produce a reliable result. With very few sales, random day-to-day variation (normal fluctuations that happen regardless of ads) can look like a real change. Other factors — a competitor running a sale, a holiday weekend, or a viral social media post — can also affect the results, so this test works best when repeated over multiple periods.

These methods have limits.

Very low sales volume

If you get 5 sales per week, comparing ads-on vs ads-off does not produce reliable data. At low volumes, random day-to-day variation — one customer telling a friend, a social media post getting shared, a slow week for no clear reason — can easily account for a difference of a few sales. As a rough guideline, at least 30-50 sales per week are typically needed to detect a meaningful difference, because at that volume, random variation becomes small enough relative to the data that a real pattern can emerge.

Long sales cycles

If people take months to decide (expensive products, services sold to other businesses, big purchases), this is called a long sales cycle — the time from when a potential buyer first learns about you to when they actually purchase. A one-week on/off test does not produce useful data for long sales cycles, because that week's sales came from ads shown months ago. Weekly comparisons only work when the typical purchase decision happens within a few days.

Sales that happen offline

If customers find you online but buy by calling, visiting, or emailing, you cannot track the connection. A dentist gets leads from ads, but the appointment happens by phone. A consultant gets interest on Instagram, but closes deals in meetings. Online tracking misses these.

Seasonal businesses

If your business has huge swings (holidays, tourist seasons, back-to-school), comparing this week vs last week does not isolate ad impact, because the seasonal change may be larger than any effect the ads have. You need to compare this December vs last December — but you probably changed your ads too.

For all these cases, longer time frames, different comparison methods, or accepting that the numbers will remain approximate may be necessary.