What Attribution Cannot Tell You
Attribution reports only count what can be tracked. What they miss, why the missing part is not random, and a simple way to fill the gap by asking customers.
5 min read
An attribution report is a record of the visits your tools were able to see. It is useful, and it is not a complete account of why anyone bought. Knowing what is missing keeps you from cutting the things that work.
What Tracking Cannot See
Conversations. A recommendation from a colleague, a mention in a private chat group, a comment at a conference. These produce a visit with no source, or a search for your name.
Things people hear or watch. Podcasts, videos, talks. The listener types your address in later.
Other devices. Someone reads about you on a phone and signs up on a laptop. Those look like two people.
Old visits. Browsers delete tracking data. Safari caps cookies set by scripts at seven days, so a first visit from a month ago may be gone.
Visitors who decline tracking. How many accept depends heavily on how the question is asked. One study of more than 80,000 visitors to a German website found that the design of the consent notice had a large effect on what people chose.
Other people at the same company. One person does the research, another signs up, a third pays.
The Missing Part Is Not Random
If tracking lost visits evenly across channels, the report would be thin but fair. It does not. It loses the early, indirect, word-of-mouth contact and keeps the final click. So the channels that get credit are the ones closest to the purchase: direct visits, searches for your brand name, email, and retargeting ads.
That creates a specific trap. A team sees that content and community show little revenue, cuts them, and watches branded search and direct traffic fall a few months later, because those were the result and not the cause.
Ask Your Customers
The simplest way to fill the gap is to ask. Add one question to your signup or demo form: "How did you hear about us?"
A few details make it work:
- Use a free-text field. A dropdown gets the first option. People will type "a friend at another company" or the name of a podcast, which no tracking tool would have told you.
- Make it optional and short. One line.
- Read the answers monthly and group them by hand. There will not be many, and the reading is the point.
- Compare with the tracked source. Where the two disagree is where you learn something.
Self-reported answers have their own bias. People remember the most recent or the most memorable contact, and "Google" often means they searched for you after hearing about you somewhere else. Treat it as a second view, not a replacement.
Is Multi-Touch Attribution Worth It
Multi-touch models split the credit for a sale across several visits. They are built on the same tracked visits, so they inherit every gap above and add a set of weights that someone had to choose.
They can be useful when you have a large number of sales, long journeys with many tracked visits, and paid spend across enough channels that the split changes a decision. For a small team with a few dozen customers a quarter, the model produces precise-looking numbers from too little data.
A reasonable setup for a small team:
- First touch and last touch on every customer. See First Touch vs Last Touch Attribution.
- A "how did you hear about us" question.
- When you want to know whether a channel works, change one thing and watch. Pause it, or double it, for long enough to see the effect on new customers.
The third one is the only real test of cause. The first two tell you where to look.
How Leapio Fits
Leapio reports first touch and last touch on revenue, by channel and by account, and it does not offer multi-touch models. Any property you send with an event, such as a self-reported source on signup, is stored on the event and can be used in breakdowns. See Revenue Attribution.