Building an attribution model your team actually trusts

Most attribution problems are not math problems. They are trust problems. Sales says the leads are junk, the platforms each claim the same conversion, and the founder looks at the bank account and sees a number none of the dashboards agree with. A model your team actually trusts is rarely the most sophisticated one. It is the one whose logic everyone understands and whose totals reconcile against real revenue.

Why platform numbers never add up

Meta and Google both report on their own conversions using their own attribution windows, and both are incentivized to claim credit. Add a click-attributed lead in Google, a view-through in Meta, and a duplicate from a returning customer, and your platforms can collectively report 140 percent of the leads you actually received. This is not fraud. It is overlapping windows and modeled estimates. The fix is to stop treating platform-reported conversions as truth and start treating them as inputs.

Anchor everything to one source of truth

For a local service business, the only number that cannot lie is the one in your CRM or booking system. Every real lead, call, and booked job lives there. Make that system the spine of your model and force the ad platforms to map back to it, not the other way around.

• Assign every inbound lead a single record with its first-touch source captured at creation.
• Use call tracking numbers per channel so phone leads are not invisible.
• Reconcile platform-reported conversions against CRM records weekly and investigate any gap over a small threshold.

If your attribution model cannot reconcile to the lead count in your CRM, it is a story, not a measurement.

Pick a model that matches your decisions

Multi-touch attribution is fashionable, but most local accounts make budget decisions at the channel level, not the touchpoint level. Start with a model proportional to the size of the choices you make. First-touch tells you what generates demand. Last-touch tells you what closes it. For most home-service advertisers, capturing both first and last source on every lead answers ninety percent of the real questions without an expensive data-science build.

Choose the model that changes what you do. If a sophisticated model produces the same budget allocation as a simple one, the simple one wins because your team will actually use it.

Feed the truth back to the platforms

Once your CRM knows which leads became real jobs, send that signal back. The Meta Conversions API and Google's offline conversion import let you tell the platforms which leads were qualified and which closed. This does two things at once. It makes their optimization smarter, because they bid toward booked revenue instead of form fills, and it tightens the gap between what they report and what you record.

• Pass a qualified-lead event, not just the form submission.
• Where possible, pass deal value so the platforms optimize toward higher-ticket jobs.
• Re-check reconciliation after the feedback loop is live, because the numbers will move.

Make the model legible to non-analysts

A model earns trust when the founder and the sales lead can explain it in a sentence. Build one shared report that shows leads, qualified leads, booked jobs, and cost per outcome by channel, sourced from the CRM and annotated where platform numbers diverge. When everyone reads the same page, the weekly argument disappears and decisions get faster.

Trust is also built by admitting uncertainty. Some leads will be unattributable, and word-of-mouth will quietly inflate your best channel. Label that bucket honestly rather than forcing it into a paid channel. A model that says we do not know about ten percent of leads is more credible than one that pretends to know everything.

Revisit it on a schedule

Attribution is not a project you finish. Tracking breaks, platforms change windows, and new channels enter the mix. Put a recurring check on the calendar to validate that events still fire, totals still reconcile, and the model still drives the budget. The teams that trust their numbers are the ones who keep proving the numbers are still true.