Attribution in 2026: what actually matters

Multi-touch is dead and last-click lies. Here's the attribution model that actually works for service businesses spending $5k-$150k/month.

Every founder we talk to has the same attribution wound. Meta says 40 leads, GA4 says 22, the CRM says 31, and the sales team swears half the deals came from referrals. Nobody is wrong. Nobody is right. Everyone has a different definition of a 'lead.'

After running attribution stacks across hundreds of service-business accounts, we've landed on a model that works for companies spending between $5k and $150k a month. It's not elegant. It is honest.

Why the old models broke

Multi-touch attribution required a clean device graph. iOS 14.5 ended that. Last-click was always wrong, but at least it was wrong consistently — now even the click data is noisy because half of it never reaches the platform. Marketing mix modeling works, but it needs $5M+ in annual ad spend and a year of clean history. Most service businesses have neither.

What's left is a triangulation model: combine three imperfect signals, weight them by what each is good at, and stop pretending any single dashboard is telling you the truth.

The three signals worth trusting

We run all three side by side. When two of them agree, we trust the read. When all three disagree, we investigate before changing anything.

Self-reported is the cheat code

Adding one required field to your booking form — 'How did you hear about us?' with five preset options plus 'Other' — recovers 60–80% of the attribution platforms lost. It's free. It takes 20 minutes. Almost nobody does it.

Pre-set options matter. Open-text fields produce garbage. Use: Google search / Instagram or Facebook ad / Referral from a friend / Saw your truck or signage / Other. Track the rolling 90-day mix. When 'Instagram or Facebook ad' rises and Meta-reported leads stay flat, you've just found a chunk of attribution Meta is no longer crediting itself.

If you don't have a 'how did you hear about us?' question on your booking form, you don't have an attribution problem — you have a measurement problem you chose to have.

The pixel + CAPI + CRM stack

On the technical side, the minimum viable stack is: Meta pixel and Conversions API both firing, GA4 with enhanced conversions, and CRM events flowing back to Meta via offline conversions or a CAPI gateway. This is table stakes — and it's part of every funnel build we ship.

The thing nobody talks about: pixel and CAPI need to be deduplicated correctly. We see accounts that look like they're double-counting because event_id isn't passed through. CPL looks artificially low. Then someone fixes it and 'CPL goes up' — except it didn't, the count was just always wrong.

Blended ROAS is the north star

If you can only watch one number, watch blended ROAS at the channel-spend level. Total monthly closed revenue divided by total monthly ad spend across all channels. It's the only metric that survives platform changes, attribution drift, and the universal tendency of dashboards to flatter themselves.

Set a target — for most service businesses, a blended ROAS of 4–6x is healthy depending on margin. Watch the trend, not the daily number. When blended ROAS slides for three weeks in a row, something is broken. When it lifts for three weeks in a row, double the budget and find out where the ceiling is.

How to install this in a week

This stack costs nothing beyond the time to set it up and is the foundation under every account in our case studies. Without it, you're optimizing on noise.

The bottom line

Attribution in 2026 is not a tool problem. It's a discipline problem. Stop chasing a perfect dashboard, start triangulating three imperfect ones, and put the 'how did you hear about us?' question on your booking form by tomorrow. If you also want to fix what those numbers are measuring, the CPL article is the next stop.