First-touch vs last-touch for D2C: pick one and defend it

Two people are looking at the same creator campaign and reaching opposite conclusions. The performance lead has a report showing six orders. The brand lead has a report showing twenty-four.

Both reports are from the same system, over the same dates, on the same orders. One is set to last-touch and the other to first-touch, and neither person knows the other’s setting. (Composite, drawn from meetings we have watched go sideways.)

The same hundred orders, scored twice

Take 100 orders from a store with six active channels and a modelled nine-day median time to purchase. Score every order twice: once by the first channel that ever brought the customer to the site, once by the last channel touched before the order.

Channel	First-touch	Last-touch	Swing
Influencer	24	6	minus 18
Paid social	31	22	minus 9
Organic search	18	14	minus 4
Referral	11	9	minus 2
Brand paid search	3	9	plus 6
Direct	8	19	plus 11
Email	5	21	plus 16
Total	100	100

Both columns sum to 100, which is the point. Neither model is inflating anything. They are dividing the same pie along a different cut.

The channels that lose credit under last-touch are the ones that create demand. The channels that gain are the ones present when demand converts.

What each model is actually asking

First-touch asks: where did this customer come from originally?

Last-touch asks: what was in front of them when they finally bought?

Both are legitimate questions and neither is attribution in the causal sense. A first touch that happened nine days ago did not cause anything on its own, and an email opened four minutes before checkout did not create a customer who had already decided.

The reason this stays confusing is that the industry uses one word for two jobs. Discovery accounting and conversion accounting are different exercises with different owners, and forcing them into one number is why the meeting above happens.

Where the two disagree most

Look at the swing column rather than either score.

Influencer swings 18 orders, the widest gap in the table, which tells you creators operate almost entirely at the top of this funnel. That is a finding about the channel, not a measurement problem to be corrected.

Email swings 16 the other way. Email rarely finds a customer and frequently closes one, which is exactly what a retention channel should look like and exactly why funding it from last-touch numbers overstates its acquisition role.

Direct rising from 8 to 19 is the one to watch closely. Direct is where demand goes when the channel that created it did not get a trackable click, so a large direct number under last-touch is often a measurement of how much attribution you are losing rather than a channel at all. The scale of that loss is quantified in your attribution is wrong, and here is how wrong.

Brand paid search at 3 and 9 is the cleanest example in the table of a channel closing sales it did not create.

Why the average is worse than either

Somebody in the meeting will propose splitting the difference. Influencer gets 15.

Fifteen orders describes nothing that happened. No customer had half a first touch. The number is defensible against neither the brand lead’s argument nor the performance lead’s, which means it will be re-litigated every month forever.

Worse, averaging destroys the swing column, and the swing column is the most useful output of the whole exercise. It is the thing that tells you influencer belongs in a discovery budget and email belongs in a retention budget, which is a decision you can actually act on.

Report two columns. Let them disagree in public.

Which one to nominate as the deciding number

You still need one number that moves budget, because two numbers with equal authority means whoever is presenting picks the flattering one.

Nominate first-touch if your growth constraint is new customers and you are actively funding creators, content and awareness. Under last-touch those channels will be cut, and the effect will show up as declining paid social efficiency a quarter later with no obvious cause.

Nominate last-touch if you have more demand than conversion and your spend is concentrated in retargeting, email and search. First-touch will overfund discovery you do not currently need.

Then write down which one you chose and why, in a place people can find it. Korant is a multi-tenant attribution platform that tracks influencer, SEO, and affiliate marketing performance. Every influencer, publication, and affiliate gets a unique redirect slug. Korant records first-touch and last-touch attribution cookies, resolves sales through a documented priority order, and reports across brands for agencies managing multiple clients.

Korant is not for stores with a single paid channel, or for brands that only need Shopify’s native reports. If every order arrives through one channel, first-touch and last-touch produce identical tables and the whole question dissolves.

The window matters before the model does

Before arguing about models, check the window against your own time to purchase.

A seven-day window on a category where the median customer takes twenty-one days discards two thirds of the journey before either model runs. Both columns then converge on whatever happened in the final week, the swing column collapses, and you conclude that influencer does not work when what actually happened is that you stopped looking early.

Export your own order file, calculate days between first session and order, and take the 80th percentile. That is your window. A default is somebody else’s category.

Writing the rule down

The rule needs four lines and takes ten minutes: which model decides budget, what the window is, how ties break, and what happens to orders no channel claims.

That last line is the one everybody skips. Unclaimed orders should stay unclaimed rather than being distributed across channels to make the report total 100%, because a completed pie chart built on assumptions is more dangerous than an honest gap.

The rule matters more than which rule you pick. A store using last-touch consistently for a year can read its trends. A store that switches models whenever the numbers look bad has, in effect, no attribution at all, and the two people in the opening will keep having their meeting.