Before you pick a local service marketing attribution method

What to take away

A Leeds plumbing firm that counts every job from a Google search ad as ad-driven will overstate that channel and underprice the work its van signage and repeat customers bring in.
Last-click attribution is cheap, auditable and defensible, but it gives all credit to the final touch before the enquiry.
Modelled or multi-touch attribution spreads credit across the journey, yet it depends on consent, identifiers and assumptions you must be able to explain.
Neither method fixes a weak baseline, so agree what counts as a lead and what counts as a sale before you compare channels.
Whichever you pick, write down the rules and keep them stable for at least a quarter.

Last-click or modelled: the two families

How last-click attribution works in a local service business

Last-click gives the whole enquiry to the final recorded touch. A caller who saw a leaflet, searched your name, then clicked a paid ad is credited to the ad. It is simple, and the numbers reconcile to your invoicing system.

Last-click vs modelled attribution

Last-click

Credit: Final touch
Complexity: Simple
Reconciles to invoicing: Yes
Hides demand creators: Yes
Lawful basis needed: No

Modelled

Credit: Split across touches
Complexity: Complex
Reconciles to invoicing: No
Hides demand creators: Less so
Lawful basis needed: Yes

The weakness is that it rewards channels that close and hides those that create demand. For a short, high-intent journey such as emergency callouts, that distortion is modest. For considered work like a kitchen refit, it is severe.

How modelled attribution spreads the credit

Modelled attribution uses rules or statistical models to split credit across several touches. It can show a directory listing started a journey your ads finished. The trade-off is complexity: consistent tracking, enough volume for patterns to mean anything, and a written explanation of the model.

Under the Data Protection Act 2018, you also need a lawful basis for the personal data that feeds it; consent and retention sit inside the method, not beside it.

Matching the method to the evidence you need

What each method can and cannot prove

Last-click proves a recorded interaction came before an enquiry. It cannot prove the interaction caused it. Modelled attribution estimates influence across a journey. It cannot prove causation either, and its output changes when you change the model. Treat both as evidence about recorded behaviour, not proof of what persuaded a customer.

What each method can prove

Last-click

Proves: Recorded touch before enquiry
Cannot prove: Causation
Output stability: Stable
Budget defence: Simple, reproducible

Modelled

Proves: Estimates influence
Cannot prove: Causation
Output stability: Changes with model
Budget defence: Hard to reproduce

If you must defend a budget line to a partner or lender, a simple, stable method beats one nobody can reproduce.

Checking your landing pages before you trust the numbers

Attribution data is only as good as the pages it points at. If a mobile visitor lands on a slow, cluttered page, the click is recorded but the enquiry never happens, and your model reads that as a weak channel.

Google's guidance on optimising your website for mobile covers the landing page basics worth fixing first. The measurement and reporting guide sets out how to define leads, sales and time windows so both methods use the same definitions.

Running a fair comparison

Set the rules before you compare

Pick a window, such as 30 days from first touch, and apply it to both methods. Define a lead as a completed enquiry form or a tracked call over 60 seconds, and a sale as a booked job with a deposit.

Then run both side by side for a quarter. Expect differences in channel rankings; that gap is the useful output, because it shows where last-click flatters or punishes a channel.

Keep the numbers honest and the records lawful

Record which method produced each figure, and who signed it off. Keep marketing consent records separate from job records, and delete identifiers once the window closes. A reporting dashboard that labels each metric by method stops a modelled estimate reading as a measured sale.

To separate a genuine channel shift from a seasonal or wider market change, the ONS publishes economic output and productivity data.

Glossary

Last-click: full credit to the final recorded touch before an enquiry.
First-click: full credit to the first recorded touch in the journey.
Linear: equal credit split across every recorded touch.
Modelled: rule-based or statistical credit allocation across touches.
Lookback window: how far back a touch can sit and still count.

Common questions

Is modelled attribution more accurate than last-click?

Not automatically. It is more detailed, but accuracy depends on tracking coverage, volume and the assumptions in the model. Last-click is narrower but easier to audit and explain.

Can a small firm run both methods at once?

Yes, and it is the cheapest way to see the gap. Run last-click from your existing call and form tracking, then add a simple rule-based model: 40 per cent to first touch, 60 per cent to last, for one quarter.

What data protection rules apply to attribution tracking?

You need a lawful basis for any personal data you process, usually consent for marketing cookies, plus clear retention periods. The Data Protection Act 2018 and ICO guidance on UK GDPR set the framework.

How long should I keep one method before switching?

At least a full quarter, so seasonal demand does not distort the comparison. Changing method mid-quarter makes every channel trend unreadable.