First touch or last touch: which channel actually brought them in

Attribution arguments usually get framed as a technical debate between models. It is not really technical. It is a question about what you are trying to learn, and the two common answers lead to opposite budget decisions.

What each model actually answers

Last touch answers: what was the final thing this person interacted with before they converted?

First touch answers: what brought this person into contact with us in the first place?

Both are true statements about the same journey. They are not interchangeable, and the difference matters most for exactly the businesses that most need attribution: the ones with a considered purchase and a multi-week decision.

The branded search trap

Here is the mechanism, and once you have seen it you will see it everywhere.

Somebody encounters you through a non-brand search, a LinkedIn ad, a referral, or a piece of content. They do not convert. They are not ready. They go away.

Three weeks later they have a live requirement. They remember your name, or half of it. They type your company name into Google. Your brand campaign serves an ad, or your homepage ranks first, and they click. They convert.

Last-touch attribution credits the acquisition to branded search.

Branded search now looks like your single most efficient channel, because it is converting people who had already decided. It has a spectacular cost per lead, because bidding on your own name is cheap and the click-through rate is enormous.

At budget time, the channel with the best cost per lead gets more money. The channels that actually created the demand look expensive by comparison, because they generate awareness that converts weeks later and gets credited elsewhere.

So you shift budget from demand creation into demand harvesting.

For a quarter or two nothing bad happens, because there is a backlog of demand still working its way through. Then the pipeline thins, and nobody can explain why, because every channel is still hitting its cost per lead target. The report looks fine right up until it does not.

Why this survives despite being well known

Three reasons, and none of them are stupidity.

It is the default. Most platforms report last-touch or last non-direct click unless somebody deliberately changes it. Defaults win.

It is observable. The last touch is the one you can see in the same session as the conversion. First touch requires stitching a journey together across sessions, devices and channels, which is more work and less certain.

It flatters. Last touch makes performance channels look efficient, and performance channels are usually what an agency is being paid to run. That is not a conspiracy, it is an incentive gradient, and incentive gradients do not need anybody to act in bad faith to produce a consistent bias.

The case for first touch, stated fairly

First touch is not a perfect model. It has real weaknesses and it is worth being honest about them.

It over-credits the top of the funnel. A person who encountered you once through a low-intent channel and then took six months and four more interactions to convert gets attributed entirely to that first encounter. The four interactions that actually moved them get nothing.

It is also harder to observe reliably. Cookie lifetimes, cross-device journeys and privacy controls all degrade your ability to see the genuine first touch, which means some proportion of your first-touch data is really best-observable-touch.

So why prefer it?

Because the errors point in a more useful direction. First touch tells you where demand originates. That is the question a budget allocation decision actually needs answered, and it is the question last touch structurally cannot answer.

If you get first touch slightly wrong, you over-invest in demand creation. If you get last touch wrong, you defund it entirely. Those are not symmetric mistakes.

What about multi-touch models

Multi-touch, time-decay, position-based and data-driven models all exist, and they are all more sophisticated than either endpoint.

They are also, in most mid-market deployments, a way of producing a number with more decimal places and no more truth. A data-driven model needs enough conversions to learn from. A position-based model needs somebody to choose the weights, and those weights are usually chosen because they produce a result that looks reasonable, which is circular.

The honest position: if you have the conversion volume to fit a model properly and the discipline to hold it to out-of-sample validation, use one. If you do not, a clearly stated first-touch rule that everybody understands will serve you better than a weighted model nobody can explain.

Clarity beats sophistication when the sophistication is unvalidated.

The rule worth adopting

Credit for a lead goes to that person’s first touch.

One sentence. Publishable. Applies in a good month and a bad one. Anybody in your business can understand it and anybody can check whether it was applied.

Combine it with a proper definition of a lead as a distinct person rather than a touch, and your channel mix starts describing something real.

What you should expect to see when you switch

Be ready for this, because it will look alarming for a week.

Branded search will collapse as a source. It will still be worth running, because if you do not bid on your own name somebody else will, but it will stop looking like your best acquisition channel because it never was one.

Organic and referral will grow, sometimes a lot. People who found you through a search or a recommendation and converted later were previously being credited elsewhere.

Your paid non-brand channels will look more expensive per lead and more important. Both of those are true, and the second one is what you needed to know.

The trend lines will mostly survive. The levels will move. That is what happens when you change the question.

The point

Attribution is not really about accuracy. Perfect accuracy is not available and pretending otherwise is how people end up with elaborate models they do not trust.

It is about which systematic error you would rather live with. Last touch systematically defunds demand creation. First touch systematically over-credits it. One of those failure modes is survivable and one of them quietly ends your pipeline.

Pick your bias deliberately, write it down, and apply it consistently. That is most of what good measurement is.