Last-touch attribution gives credit to whatever the customer clicked most recently, which on a car deal is almost always a branded search or a click on your own website. That systematically flatters the bottom of the funnel and starves everything that created the demand in the first place — mail, events, radio, word of mouth. The fix is not a better attribution platform. It is three cheap mechanisms a dealership can run itself: unique tracked numbers and offer codes, a match-back of sold units against the mailed list, and a holdout control group that tells you what would have happened anyway.
The Problem, Stated Precisely
A customer gets a mailer on a Tuesday. On Thursday they mention it to their spouse. On Saturday they search the dealership by name, click the result, browse inventory, and fill in a form. The CRM records the lead source as organic search, or paid search if the brand term was bid on.
Nothing in that chain is a recording error. The last touch really was a search. But the mailer caused the sale, and under last-touch the mailer gets nothing while the brand term gets everything — a term that, by definition, only gets typed by someone who already knows you exist.
Run a marketing budget on that data for two years and you will have shifted spend into the channels that harvest demand and away from the channels that create it. The numbers will look better every quarter while the store gets quieter.
Why Dealerships Have It Worse Than Most
Three structural reasons.
The purchase is slow and multi-touch. Vehicle consideration runs weeks to months. The more touches in a journey, the more last-touch distorts.
A great deal of the journey is offline. The mailer, the drive past the lot, the conversation at work, the phone call. None of it generates a click, and anything without a click is invisible to the analytics that most attribution rests on.
The lead source field is filled in by a person. Ask a salesperson to select a source and they will select the one that pays, or the first in the dropdown, or whatever they selected last time. This is not cynicism, it is what the data looks like in almost every store we have reviewed.
The Three Mechanisms
1. Instrument the offline channel
Give every offline campaign its own tracked phone number and its own offer code, and point any URL at a coded landing page rather than the homepage. This does not solve attribution, but it converts an invisible channel into a countable one. Cost is trivial — a set of tracked numbers is tens of dollars a month.
The discipline that makes it work is refusing to reuse a number across two campaigns, and recording the code at the desk rather than hoping it appears in the CRM note.
2. Match back sold units against the mailed list
After the campaign, take your sold log for the window and match it — by name, address and household — against the list you mailed. Every match is a customer who received the piece and bought, whether or not they called the tracked number, mentioned the offer, or remembered the mailer at all.
A match-back typically finds two to four times the buyers that tracked response alone reports, because most people who receive a mailer and act on it do not use the mechanism you gave them. This is the single most under-used measurement available to a dealership, and your mail vendor can run it for you.
Its limitation is important: a match-back proves exposure and purchase. It does not prove causation. Some of those matched buyers were coming anyway, which is what the third mechanism is for.
3. Hold out a control group
Take a random, matched sample of the list — a few thousand records — and deliberately mail them nothing. At the end of the window, measure the purchase rate in the holdout and in the mailed group.
The difference is the campaign’s incremental effect. If the mailed group bought at 0.33% and the holdout at 0.06%, the campaign caused the difference and you can state that with a straight face.
It costs nothing except the units you might have sold to the holdout, which is precisely the point — if the holdout sells at the same rate as the mailed group, the campaign was not working and the money was already wasted.
What To Do About Branded Search
The specific test worth running, because it settles an argument most stores have annually. Branded paid search almost always shows a spectacular return under last-touch, because the people clicking it were already looking for you.
Pause it for a defined period and watch total sessions and total leads, not paid sessions. In many stores, organic traffic absorbs most of the loss and the total barely moves — which means the paid spend was buying clicks you were getting free. In others, aggressive competitor bidding on your name means the paid line is genuinely defending traffic. Both results happen. The point is that last-touch data cannot tell you which one you are, and a two-week pause can.
A Reporting Format That Survives Contact With an Owner
Report three columns per channel rather than one:
- Attributed — what the platform or CRM claims. Useful for trend, not for truth.
- Matched — units where the customer demonstrably received or engaged with the campaign.
- Incremental — units above the control, where a control exists.
Label each honestly and never present the first as the third. An owner who is shown the same number three ways, with the assumptions stated, will trust the one you recommend acting on. An owner shown a single flattering figure learns, eventually, to discount everything.
Frequently Asked Questions
Do we need an attribution platform?
Most single-rooftop stores do not. Tracked numbers, a match-back and a holdout will get you further than a platform that is still, underneath, reading clicks. Multi-rooftop groups with large digital budgets may get value from a proper multi-touch model, but only if offline is instrumented first.
Is first-touch better than last-touch?
It has the opposite bias — it over-credits whatever introduced the customer and ignores what closed them. Neither single-touch model is right. The reason to prefer incrementality is that it asks a different and better question: what changed because we spent this money?
How long should the measurement window be?
Long enough to cover the consideration period, which for most campaigns means 60 to 90 days rather than the calendar month the invoice covers. Measuring a mail drop on 30 days will understate it.
Our CRM source data is a mess. Where do we start?
Do not start by cleaning it. Start by instrumenting one campaign properly and holding out a control. One well-measured campaign is worth more than a year of tidied-up guesses, and it gives you a reference point to judge the CRM data against.
Summary
Last-touch attribution is not wrong about what it measures; it is wrong about what it implies. On a slow, largely offline purchase it credits the harvest and ignores the planting. Instrument offline campaigns with tracked numbers and codes, match sold units back against the list to find the buyers who never used them, and hold out a control group to separate the sales you caused from the sales you would have had regardless. Report attributed, matched and incremental as three labelled columns. The tooling is cheap; the discipline is the expensive part.
