First-touch vs last-touch attribution, explained with one customer
Two ways to credit the marketing that produced a customer, shown with one journey, with the arithmetic of what each would have you cut and why the honest answer is to keep both numbers side by side.
Sohail Safdar · · 5 min read
Maria sees your Instagram ad in March, visits the site, and leaves without doing anything. In April she searches for a dentist near her, taps your Google Business Profile, and books an appointment.
One customer. Two defensible stories about what produced her. First touch says Instagram found her. Last touch says the profile closed her. Both are true, and if you pick one and act on it you will make a specific, predictable mistake.
What each one would have you do
Put the arithmetic on it, because the stakes are not symmetrical.
Say you spend $600 a month on Instagram and it shows zero bookings on last touch, which is exactly what happens when the ad creates awareness and the booking happens weeks later through search. Cutting it saves $600 a month, and the saving is immediate and visible in the account.
What it costs is invisible for about a quarter. If a third of the people who find you through search in any month first heard of you through that ad, and you take 30 new customers a month, you have quietly removed the top of a funnel that was producing ten of them. At a lifetime margin of $800 each, that is $8,000 a month of eventual revenue traded for $600 of visible saving, and the two land in different quarters, so the decision looks good for a while.
The reverse mistake is cheaper but real: crediting everything to first touch and neglecting the booking page, the profile and the reply speed that turn interest into an appointment.
What each is actually for
First touch answers: where do people first hear of us? Use it to decide where to spend to reach strangers.
Last touch answers: what makes people act? Use it to improve the moment of booking, which is usually the cheapest thing on the list to fix.
Where each one lies to you
First touch ignores everything after the first visit. The review that reassured them, the reply that came in four minutes, the reminder that stopped them forgetting: none of it appears.
Last touch ignores everything before the last click, which means it systematically over-credits whatever people use to find a business they have already decided to contact. Search usually wins last touch. That does not mean search created the demand, only that search is how people arrive when they already know your name.
What you need before any of this works
Campaign parameters on your links, carried through to the form, chat or booking that creates the lead, and stored as two values rather than one: the first one seen and the most recent.
Two values rather than one is the whole trick. Overwrite on each visit and you have last touch and no way back. Store only the first and you can never see what closed. Keeping both costs one extra column and is the difference between having the argument above and guessing at it. See lead source tracking without spyware.
Adding cost, which is where it becomes useful
Attribution alone tells you where customers came from, which is interesting. Add what you spent per channel and it tells you cost per enquiry and cost per customer, which is what decisions actually need.
The comparison to make is cost per customer against lifetime margin, not against first visit value. A plumber paying $120 for a customer worth $140 on the first job is not breaking even, they are buying a relationship worth several times that at a 15 percent discount. The same $120 is reckless for a business whose customers come once.
What neither can tell you
Word of mouth. The van. A sign. Somebody who typed your name because their neighbour mentioned you.
These arrive as direct or unknown, and the instinct is to treat the unknown slice as a measurement problem. It mostly is not. It is the part of your marketing that happens in conversations you are not present for, and in most service businesses it is the largest single source. Allocating the entire budget from the attributed slice means optimising a minority of your demand and assuming the majority takes care of itself.
The monthly habit that uses both
Fifteen minutes, once a month, and the point is to read the two side by side rather than in turn.
Put first touch and last touch next to each other for the same period and look for channels that appear in one and not the other. A channel with first touches and no last touches is doing awareness work, and its value is real and delayed. A channel with last touches and no first touches is harvesting demand that something else created, which is worth having and is not a growth engine. A channel in both is the straightforward case and needs no debate.
Then do the only calculation that matters: cost per customer against lifetime margin, by channel, using the first touch view for anything you are thinking of increasing. Not cost per click, not cost per enquiry, and not cost per customer on last touch alone, which is exactly how the awareness channel gets cut.
What not to do in this meeting is rank the channels into a league table. Attribution is too crude at small volumes to support an ordering, and a table invites a decision about the bottom row. The questions worth asking are narrower and answerable: is anything here obviously losing money across both views, and is anything obviously working and starved.
In Frontiva
The analytics report first touch and last touch, with the reader choosing which, rather than a single blended number that hides the choice. Marketing spend entered per channel becomes cost per enquiry and cost per customer, and every figure has the list behind it, so a number that looks wrong can be opened rather than argued about.
One deliberate limit worth knowing before you ask for it: nothing is recorded about visitors who never get in touch. The parameters travel with the enquiry at the moment somebody becomes a lead. That gives up some funnel analysis, and it means the product never accumulates a browsing history of people who only read a page. See speed to lead statistics.