First-Touch vs Last-Touch Attribution: Which to Use

Every attribution article explains the models. Almost none of them tell you what happens when you point those models at a business that closes 30 jobs a month instead of 30,000 checkouts a day.

So here is the version with the arithmetic in it: one $9,000 roofing job, four readings, four different channels taking the credit — and the honest call on which reading to run your budget off.

What do first-touch and last-touch attribution actually do differently?

First-touch assigns 100% of a conversion's credit to the first channel in the path; last-touch assigns 100% to the last one. Both are single-touch rules — they hand the whole job to one channel and hand zero to everything else in between.

The bigger fact most posts skip: you cannot select first-touch in Google's tools anymore. Google's documentation states plainly that the first click, linear, time decay, and position-based models are no longer available as of November 2023. GA4's attribution reports now offer exactly three models — data-driven, paid and organic last click, and Google paid channels last click. Google Ads offers two: last click and data-driven.

Two more mechanics that decide who gets paid, straight from Google's docs: every GA4 attribution model excludes direct visits from receiving credit unless the entire path is direct. And paid and organic last click and last non-direct click are two names for the same model.

So the real choice in front of you is not first vs last. It is: last non-direct click, Google-paid-only last click, or data-driven — plus whatever your CRM says, which is usually a fourth answer nobody reconciles.

How does one $9,000 roofing job get credited four different ways?

Take a real-shaped path. A homeowner reads your storm-damage blog post from organic search on day 1. On day 8 they search 'roof replacement cost' and click your Google Ads ad. On day 19 they click a Meta retargeting ad. On day 23 they type your domain in directly and call. The job closes at $9,000.

Nothing about that path is unusual. What is unusual is that nobody shows you the payout table:

Four models. Four channels. One job. And in each of those four worlds, a different line item survives next quarter's budget cut. That is not a rounding difference — it is the whole decision.

Meanwhile Google Ads is claiming the full $9,000 conversion value inside its own interface, and Meta is claiming it inside its own. Add the two dashboards up and you booked $18,000 on a $9,000 roof. We cover why that happens in more depth in why most service businesses can't tell which ad dollar made them money.

Which channels does last-touch systematically over-credit?

Last-touch over-credits two things almost every time: branded search and retargeting. Both sit at the end of the path by construction, and last-touch pays 100% to whatever sits at the end of the path.

Think about who clicks a branded search ad. Someone who already knows your name. Something else taught them the name — a truck, a neighbor, a blog post, a Meta ad, a map pack listing. The branded ad was the door handle, not the door. Under last-touch it takes the entire job.

Retargeting has the same shape. It can only reach people who already visited you, which means it is structurally the last paid click before a decision that was already forming. A retargeting campaign showing a 9x return on ad spend under last-touch is often re-billing you for demand you already paid to create.

This is why we tell owners to read branded-search and retargeting ROAS with a heavy discount, and to judge non-brand search and cold prospecting on cost per booked job instead. If you want the ugly version of that test: pause branded search for two weeks and watch whether total booked jobs move or just move channels.

Does GA4's data-driven model work at low conversion volume?

It runs, but Google itself tells you the data you need to make it meaningful — and it is more than most service businesses have. Google's data-driven attribution documentation states that all conversion actions are eligible for DDA regardless of volume, but that Google recommends at least 200 conversions and 2,000 ad interactions within a 30-day period for the model to analyze your data accurately.

Read that against your business. Two hundred conversions in 30 days, from a contractor closing 30 jobs a month, means your 'conversion' has to be a form fill or a call — not a job. Even then, most $2,500-to-$25,000-ticket service businesses do not generate 2,000 ad clicks a month on a sane budget.

Google is blunt about what happens next: the same doc says that depending on data availability, the last click and data-driven attribution models can have the same results in certain situations. In plain English — at low volume, your fancy machine-learning model frequently outputs last click with extra steps. It is not lying to you. It just has nothing to learn from.

There is a second wrinkle worth knowing before you trust a Monday number: GA4 states that conversions can be reattributed for up to 7 days after the conversion. Your channel mix genuinely changes underneath you for a week.

Which model should you run if you close 30 jobs a month?

Run paid and organic last click (last non-direct click) as your reporting model, and treat it as a directional ledger, not truth. At 30 jobs a month you do not have the sample size to distinguish a data-driven model's fractional credit from noise, and you will burn hours arguing about decimals that describe 4 conversions.

The reason is not that last-touch is accurate. It is that last-touch is stable and legible. You can explain it to a partner in one sentence, it does not silently re-weight itself, and its biases are known and correctable — you already know it inflates brand and retargeting, so you discount those and move on.

Keep data-driven ON for bidding inside Google Ads if you are running Smart Bidding, because that is what the bid strategy consumes, and Google Ads defaults most conversion actions to it anyway. Use last non-direct click for the report you actually read. Those are two different jobs and they do not have to match.

Set the lookback window deliberately, too. GA4's default lookback for non-acquisition key events is 90 days, with 30- and 60-day options. For a roof or an HVAC replacement, 90 is right — people shop for weeks. For an emergency drain clear, a 30-day window stops giving credit to a click from two months ago that had nothing to do with the burst pipe. If your sales cycle runs long, our guide to tracking Google Ads conversions with long sales cycles goes deeper.

What does a self-reported field fix that no model can?

A 'How did you hear about us?' field on your form and on your call script captures the touches no attribution model can see: the yard sign, the referral, the Facebook group, the ChatGPT answer that never passed a UTM. No model can attribute a channel that never fired a cookie.

It also catches the AI-search problem that is growing every quarter. If a buyer asks an assistant for the best roofer in Fort Worth, reads your name, and later types your domain in directly, GA4 records that as direct traffic and every model in the stack shrugs.

Make the field a required dropdown with a short list plus an 'Other' text box, pipe it into the CRM alongside the UTM data, and compare the two columns monthly. Where they disagree consistently, the self-reported column is usually closer to the truth about the first touch and the platform is closer to the truth about the last. Our breakdown of what to put in a 'how did you hear about us' dropdown covers the option list that actually gets answered.

How do you stop your ad platforms and GA4 from double-counting?

You cannot make the platforms agree — you can only stop adding them together. Google's own documentation names three separate reasons the same conversion appears twice with different totals.

• Different creditable channels: Google's docs state Google Ads reports default to Google Paid attribution, while Google Analytics reports default to Paid and Organic. Same conversion, different eligible channels.
• Different time zones: Google says GA4 reporting uses the property time zone and Google Ads uses the ad account time zone — mismatch them and your daily numbers will never tie out.
• Different models: Google Ads defaults most conversion actions to data-driven; your GA4 report may be on last non-direct click.
• Different platforms entirely: Meta counts view-through conversions that Google never sees, so both can claim the same booked job.

The fix is procedural, not technical. Pick ONE system as the scoreboard — we recommend your CRM, because it is the only one that knows a job actually got paid for — and demote every ad platform to an optimization tool. Google Ads' conversion count exists to feed Smart Bidding. It is not your revenue report.

Then reconcile monthly on one number: cost per booked job, per channel, from the CRM. If Meta's view-through claims bother you, read our take on whether view-through conversions are worth counting and on the iOS ATT hit to Meta attribution.

What's the honest verdict for a service business?

Verdict: last non-direct click wins — not because it is right, but because it is the least-wrong model you can actually staff. First-touch is gone from Google's tools. Data-driven needs volume you do not have and, by Google's own admission, can return the same answer as last click when data is thin. Last non-direct click is stable, explainable, and its biases are known.

So the stack we recommend for a business closing a few dozen jobs a month: last non-direct click for reporting, data-driven for Google Ads bidding, a self-reported field on every form and call, and one monthly number — cost per booked job from the CRM — that outranks every dashboard.

And discount brand and retargeting ROAS on sight. Any agency showing you a 12x return that is 80% branded search is not showing you performance. It is showing you a rounding artifact of the model they picked.

If your ad reporting and your bank account disagree, the reporting is wrong. We build paid ads programs that report on booked jobs rather than platform-claimed conversions — month-to-month, no lock-in, and you own the ad accounts. Get my free audit and we'll tell you which channels your current attribution setup is quietly over-paying.