What is the real difference between first-touch and last-touch attribution?

Closing metric versus discovery metric

Last touch is a closing metric. It tells you which channel stood in front of the customer at the moment they acted. That is genuinely useful for deciding where the next dollar of harvest spend goes, and it is stable enough to compare week over week.

First touch is a discovery metric. It tells you which channel introduced you to someone who eventually bought. It is the only common model that gives any credit to awareness spend, which is why the people running awareness campaigns tend to prefer it. In paid media planning the two models routinely support opposite budget recommendations from the same underlying data.

The bias each one carries

• Last touch over-credits harvest channels. Branded search, remarketing and direct visits show extraordinary conversion rates because they sit closest to a decision someone else caused.
• First touch over-credits whatever is observable first. If a customer heard about you from a neighbor and then clicked a paid ad, paid search is the first touch on record. The neighbor is invisible.
• First touch decays with time. Cookies expire and get cleared. A journey that started ninety days ago often looks like it started last Tuesday.
• Neither handles phone-first behavior. A call placed from a saved contact record has no session at all, so it attaches to nothing unless you resolve the caller to an earlier visit.

A diagnostic worth running once a quarter

Pull channel share of conversions under both models and put the two columns side by side. Channels whose share is much larger under first touch are demand creators. Channels whose share is much larger under last touch are demand harvesters. Channels with similar shares are largely self-contained.

That single comparison is more informative than either model alone, and it usually settles arguments about whether a channel is underfunded. It is also easy to automate once lead sources are joined to job outcomes through an integration layer instead of being read out of four separate dashboards.

What to do with the gap

Treat a large gap as a hypothesis, not a conclusion. A channel that looks like a demand creator under first touch might simply be a cheap channel that many people brush past. The way to confirm it is to reduce or pause that channel in a comparable market and watch what happens to total demand, including branded search volume.

Meanwhile, resist the urge to run the business on two models at once. Use one for the standing report and pull the other deliberately, as a diagnostic, the way you would run a specific test rather than change the gauge on the dashboard.