Skip to content
← All posts
7 min read

Attributing revenue to email, and saying what the number cannot mean

A founder sends a campaign to four hundred people. Over the next two weeks, eleven new subscriptions start. How much of that revenue did the campaign earn?

The honest answers range from "all of it" to "none of it", and no method available to a company this size can narrow that to a point. What a method can do is state a rule, apply it the same way every time, and be explicit about which way it is wrong. This is the rule we use, and its faults.

What is last-touch attribution over fourteen days?

Last touch credits a conversion to the most recent campaign message the person received or clicked before converting, provided that message was within the window. The window is fourteen days. Nothing else is credited, and a conversion with no message inside its window is credited to nobody at all.

Three details make the rule precise enough to argue with.

What counts as a touch. A send and a click. A campaign email going out through your own sending account is recorded against the person who received it; a click on a tracked link is recorded the same way. Opens are not among them, because they are not collected. Open tracking requires a webhook from the sending provider that is not received, and an open rate is a poor signal in any case now that mail clients prefetch images.

What counts as a conversion. A trial start and a first paying subscription, read from your billing system. Not a form submission, not a demo request: money, or the beginning of the path to money.

What time is used. The conversion is dated by the billing system, not by the day a sync noticed it. This sounds pedantic and is not. A timestamp meaning "when we first looked" would be wrong for every customer who existed before the connector was plugged in, and it would answer a different question than the one asked.

There is one tie-break worth knowing about. A send and a click can be recorded in the same second, because two different systems both stamp to the second. When that happens the click wins. Somebody who clicked did more than somebody who was merely written to.

Why is the number meaningless without a baseline?

Because "the campaign is credited with €1,200 of new monthly revenue" and "the campaign caused €1,200 of new monthly revenue" are different sentences, and only the first is a measurement.

Some of those eleven subscriptions were going to happen. A person who had been evaluating you for three weeks, who received your email on Tuesday and subscribed on Thursday, gets full credit under last touch. The email may have been the thing that tipped them. It may have been a coincidence of timing. The rule cannot tell the difference and does not pretend to.

The correction is a comparison period. Alongside the attributed figure, the same measurement is taken over the window immediately before, of the same length: if the report covers the last thirty days, the baseline is the thirty days before those. Not the same month last year, not a seasonal average, which would require a history a two-year-old SaaS does not have. Just the previous window, defined once so the interface and the calculation cannot disagree about what it means.

The difference between the two is labelled an estimate, because that is what it is. And it is clamped at zero: a fortnight worse than the one before it means the campaign did not visibly lift anything, not that it destroyed revenue.

This comparison is crude. It has no seasonality correction, no trend adjustment and no control group. That is deliberate. A company with two hundred customers has neither the volume nor the history for anything more sophisticated, and a more elaborate model would produce a number that looks more precise without being more true. Trading an understandable bias for an unfalsifiable one is a bad trade when the reader has to act on the result.

What does last touch over-credit?

Two things, systematically, and both push the number up.

The last message takes everything. Somebody who read three emails over a month and subscribed after the third gives all the credit to the third. The first two did the work of getting them to open the third, and none of that shows up anywhere. If you run a nurture sequence, its early messages will look worthless and will not be.

Coincidence looks like causation. The person who was always going to subscribe, and who happened to receive something in the fourteen days before they did, is counted as converted because of it. There is no version of last touch that avoids this. It is the main reason the baseline exists, and the reason the difference is shown next to the raw figure rather than instead of it.

What does it miss entirely?

Three things, and these push the number down.

Anyone who is not already a contact. The join is by contact identity: this person received this message, this person started paying. A stranger who arrives from a link, browses, and signs up without ever having been in your list is invisible to it. There is no UTM matching here, so a click from a shared link resolves to nobody.

Cross-device and cross-person conversions. Somebody reads the email on their phone, subscribes on their laptop: fine, because the join is on billing identity rather than on a browser. But somebody who forwards the email to the colleague who actually holds the company card is not counted, and forwarding is common in B2B.

Everything outside fourteen days. A long sales cycle breaks this entirely. If your customers typically take two months from first contact to payment, the message that started it will be outside every window and you will attribute almost nothing. The window is a parameter, not a law of nature, but a longer one buys more coincidence along with more coverage.

Put the two lists together and you get the honest summary: the attributed figure is not an estimate of the campaign's true contribution. It is a floor on the revenue that is connected to it, biased upwards by coincidence and downwards by everything it cannot see. Which is why "at least" is the right word, and why the screen that shows the number says so.

Should a small team bother with a flawed method?

Yes, and the argument is comparative rather than absolute.

The alternative in practice is not a better measurement. It is no measurement. It is a founder in month four trying to recall whether the March campaign felt like it did anything, which is a method too, with worse properties: it is biased by recency, by whichever campaign they enjoyed writing, and by whatever else was happening that month. At least last touch is wrong the same way every time, and a consistent bias is something you can reason around. Recollection is not.

The second argument is about what the number is used for. Nobody should make an irreversible decision on a single campaign's attributed revenue. What the figure is good for is ranking: this campaign to this segment did better than that one to that segment, measured identically. Comparisons between campaigns are far more robust than any individual campaign's absolute number, because the biases are shared and largely cancel.

The third is about volume, and it cuts the other way. Below about thirty recipients, a campaign's result is noise, whatever the method says. Our own playbook refuses to draw a lesson from a campaign that small, and marks it inconclusive regardless of how good the numbers look. A confident lesson drawn from eight people is worse than no lesson, because it gets applied.

The rule for reading any attribution number

Ask three questions of it, including of ours.

What is the rule, exactly? What is it compared against? Which direction is it wrong in? A tool that cannot answer all three is not measuring, it is reassuring, and the two look similar right up to the point where you spend money on the difference.

If you want to see what this looks like in practice, with the window, the baseline and the method printed on the same screen as the figure, that is our attribution page.

See your own numbers first

Connect Stripe and your real revenue is on screen in five minutes, before an agent has said a word. Create your account, answer the questions, and your strategy gets written.