Marketing attribution
What is marketing attribution: marketing attribution explained
Marketing attribution is the practice of assigning credit for a sale to the marketing touches that preceded it, so spend can be judged by what it produced. A car purchase involves many touches over several weeks, so every attribution model is a rule for splitting credit, not a measurement of cause.
It is also written as marketing attribution modelling, media attribution, or just attribution. The output is always the same shape: a table of channels with units and revenue next to them.
What does marketing attribution mean in a dealership?
Take one real-shaped buyer path. Day 1, an organic search lands on your used inventory page. Day 5, the same shopper clicks your listing on a third-party marketplace. Day 12, they open a follow-up email from the BDC. Day 19, they search your dealership by name and click the paid ad. Day 21, they show and buy.
Four paid or owned channels touched that deal. The store has one car to hand out credit for. Attribution is the rule you pick for splitting it, and there is no rule that recovers what actually changed the customer's mind.
How does marketing attribution work? The main models
- First touch. All credit to the earliest recorded touch. Organic search takes the sale above. See first touch attribution.
- Last touch. All credit to the final touch before the sale. The branded paid ad takes it. See last touch attribution.
- Linear. Credit split evenly. Each of the four channels books 0.25 of a unit. Simple, and it assumes the email and the first search mattered equally, which nobody believes.
- Time decay. Credit weighted toward the touches closest to the sale, usually on a half-life (a touch 7 days out counts half as much as one on the day). Reasonable for a 21-day purchase, and it still leans last-touch by design.
- Position based. Often 40 percent to the first touch, 40 percent to the last, 20 percent shared among the middle. A compromise between the two single-touch camps rather than an argument for either.
- Data driven. Weights derived statistically from paths that converted against paths that did not. Better in principle, and it needs conversion volume most single rooftops do not have.
Marketing attribution explained: a worked example
A store spends $40,000 in March and delivers 40 units. Spend breaks down as paid search $12,000, third-party marketplace $10,000, SEO $10,000, social $6,000, and CRM email $2,000.
Under first touch, the 40 units credit out as SEO 16, marketplace 14, social 8, paid search 2, email 0. Under last touch, the same 40 units credit out as paid search 12, marketplace 10, email 10, SEO 6, social 2.
Now run cost per sale on paid search. First touch says $12,000 divided by 2, or $6,000 a car. Last touch says $12,000 divided by 12, or $1,000 a car. Social goes the other way: $750 under first touch, $3,000 under last. Email looks worthless under one model and like the best channel in the store at $200 under the other.
Nothing changed except the rule. Same 40 cars, same $40,000, six times the difference on one channel.
Why attribution got less reliable, not more
Click-level attribution used to work reasonably well because browsers and phones let one company follow a person across other companies' properties. That assumption has been steadily withdrawn.
- App Tracking Transparency. Since iOS 14.5, Apple has required apps to ask permission before linking user data with data collected by other companies, and Apple's own definition of tracking explicitly covers advertising measurement, not just targeting.
- The Privacy Sandbox reversal. In October 2025 Google confirmed it would keep third-party cookie choice in Chrome rather than remove cookies, and retired ten Privacy Sandbox technologies for low adoption, including the Attribution Reporting API that was meant to replace cookie-based measurement.
- Walled gardens. The large ad platforms measure and report conversions inside their own systems, using their own attribution windows. Add up what each platform claims and the total routinely exceeds the units the store actually delivered.
None of that makes attribution useless. It does mean a channel report is an estimate produced by a chosen rule under partial data, and it should be spoken about that way in a management meeting.
What to do with it anyway
- Pick one model and keep it. The trend inside a consistent model is worth far more than the level. Switching models mid-year invalidates every comparison you have.
- Run first and last touch side by side. When they disagree wildly on a channel, that is the useful signal. It usually means the channel does one job well and the other badly.
- Test by holding out. Turn a channel off in one rooftop or one market for a month and watch total units. It is blunt, and it measures something no model can infer.
- Ask the customer. A single question at write-up is unreliable, and it catches the touches your tracking never saw, like a neighbour's recommendation.
One consequence worth naming: mid-funnel touches are the ones single-touch models erase. A personalized video sent to a lead on day 12 sits between the first click and the last one, so first touch and last touch both score it at zero, even in a month where the store can see response rates move. If a touch never appears in the report, it never gets budget, and that is a modelling artefact rather than a finding.
