Marketing attribution is broken: measure with first-party data
Short answer Click tracking misses many customers: people who decline cookies or app tracking, and everyone who calls, walks in, or hears of you from a friend or an AI answer. Measure with data you own: ask every lead how they found you, count paid jobs by source, divide spend by new customers, and pause a channel to test it.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callWhy marketing attribution is broken
Attribution means deciding which ad, search or post gets the credit for a customer. Ad platforms and Google Analytics do it by following clicks, and they see fewer of them each year. They never saw the customers who phone, walk in or come by referral.
| What happens | What the tracking sees |
|---|---|
| A visitor declines your cookie banner | With Google's basic consent mode, "no data is transferred to Google at all, not even the consent status" (Google Ads Help: About consent mode) |
| Someone uses Facebook or Instagram on an iPhone | Apps need "the user's permission through the App Tracking Transparency (ATT) framework in order to track them", and tracking includes linking app data with other companies' websites "for targeted advertising or advertising measurement purposes" (Apple Developer: User privacy and data use) |
| Someone browses in Safari | "Safari limits third-party cookies and data by default" (iPhone User Guide); it has blocked them outright since March 2020 (WebKit) |
| Someone calls after seeing your truck, a referral or an AI answer | No click, so nothing to follow |
| One person uses a phone, then a laptop | Two strangers, unless they sign in or give you their email |
Google estimates part of what it can't see. Consent mode uses signals from visitors who said no to "model your metrics", but "your tags need to meet a certain data collection threshold" (Google Ads Help: About consent mode). A small site may not reach it.
An accounting firm we run marketing for sent each ad booking back to the ad platform only with the booker's permission. On October 6, 2026, only 2 of 6 Google-ad bookers had accepted ad measurement, so only 2 were sent.
What didn't happen: Chrome kept third-party cookies
Many articles still say Google is phasing them out. It dropped that plan:
"Chrome will maintain our current approach to offering users third-party cookie choice in Chrome." (Privacy Sandbox blog, October 17, 2025)
The same post retired the Attribution Reporting API and other tools built to replace cookies, citing "low levels of adoption". For a local business the gaps are consent, iPhones and leads that never click.
Measure with first-party data: five numbers that need no click
First-party data is what your business records itself: who got in touch, how they say they found you, and what they paid. It doesn't depend on cookies or app permissions, so a tracking gap can't erase it.
| Number | Where it comes from | What it answers |
|---|---|---|
| Source each lead names | "How did you hear about us?" on every form, booking and call | Where people found you, word of mouth and AI answers included |
| Paid jobs and revenue by source | A lead log or CRM, one row per lead | Which sources bring paying customers, not just inquiries |
| Cost per new customer | All marketing spend ÷ new paying customers, each month | Whether marketing as a whole pays, tracked or not |
| Searches for your business name | Search Console's branded queries filter | Whether more people know you |
| What drops when you pause | A pause or an area holdout | Whether a channel causes leads or only collects credit |
Ask every lead how they found you
Make it a required pick list on your contact form and booking page, and ask it on every call. Specific choices give usable answers: Google search, Google Maps, an ad on Google, Facebook or Instagram, ChatGPT or another AI assistant, a friend or neighbor, our truck or sign, other (type it).
People may name the last step they remember, not the first: a neighbor's tip followed by a search for your name can come back as "Google". So log two sources per lead: the one they name and the one your tracking shows, such as the UTM tag (a label on an ad link that tells analytics where the visit came from).
Count paid jobs by source, not platform conversions
Each ad platform counts its own "conversions", usually form fills or calls, and takes credit for them. Only your records show which inquiries became paying customers. Keep one row per lead: date, source named, source tracked, booked (yes or no), amount paid. Total it by source each month; the report to ask for shows a layout.
The same accounting firm: across Google, Meta and ChatGPT ads it spent about $6,100 from October 1 to November 4, 2026. Those ads brought about 13 booked calls and $0 in new-client revenue. Every client who paid in that period came from ChatGPT answers or a Reddit comment, at no ad cost.
Once you know which leads paid, tell the ad platforms, so their bidding chases customers instead of form fills. For Google, see tracking a lead from Google Ads to a booked job. Meta's Conversions API lets you "send down-funnel event data from your CRM back to Meta" (Meta for Developers: Conversions API for CRM integration).
Divide total spend by new customers
Add up everything marketing cost this month: ads, agency fees, tools, mailers. Divide by new paying customers. This blended cost per customer ignores which channel got credit, so no tracking gap can distort it. Compare it with what a new customer is worth to you. It can't say which channel works; the tests below can.
Watch how many people search your name
Someone who hears of you anywhere and later searches your name shows up in Search Console, even when the first contact left no click. Since November 20, 2025, its Performance report has a branded queries filter that separates searches for your brand from the rest. Since March 11, 2026, it is "available to all eligible sites" (Google Search Central blog). Two limits: only top-level properties with enough searches qualify, and Google warns "some queries may occasionally be misidentified." Watch the monthly trend; Google Search Console for a small business covers setup.
Test a channel: pause it or hold out an area
To learn whether a channel causes customers, compare a time or place with it against one without it. Google calls this an incrementality experiment:
"It's the way to measure the causal impact of ads." (Google Ads Help: About Conversion Lift)
| Test | How | Fits | Watch for |
|---|---|---|---|
| Pause | Turn one channel off for a few weeks; compare new leads and paid jobs with the weeks before and the same weeks last year | Any budget | Seasons, weather, a price change or a second change in the same weeks |
| Area holdout | Keep ads in some of your towns, stop them in similar ones; compare leads per town from your lead log | Businesses serving several towns | Towns must be alike and large enough to bring leads |
| Platform lift study | The platform splits people or regions into a group that sees your ads and one that doesn't | Larger budgets | "Conversion Lift isn't available for all Google Ads accounts" (Google Ads Help: About Conversion Lift); Meta runs its own Conversion Lift tests |
Decide before the test what result means keep, cut or change, and change one thing at a time. If a pause costs no paid jobs, the channel was taking credit for customers who would have come anyway.
Collecting first-party data: US and Canada rules
Asking how someone found you is ordinary business. Using their details for marketing has rules, and they differ by country.
| Rule | US | Canada |
|---|---|---|
| Marketing email | CAN-SPAM: every message must explain how to opt out, and "You must honor a recipient's opt-out request within 10 business days" (FTC) | CASL: consent comes first; see CASL consent for texting and email |
| Personal information | California's CCPA, for businesses it covers (below) | PIPEDA, the federal privacy law: "organizations are generally required to obtain meaningful consent" (Office of the Privacy Commissioner) |
| Texts to leads | See missed-call text back consent rules | Same guide, plus CASL |
California's CCPA covers for-profit businesses with gross revenue over $26,625,000 (from January 1, 2025, per the California Privacy Protection Agency), that buy, sell or share data on 100,000 or more California residents or households, or that earn half their revenue selling it (California Attorney General). Its main tool is an opt-out: "You may request that businesses stop selling or sharing your personal information."
Example
Illustrative cases. Numbers are invented for the example.
A plumber who paused Google Ads in two towns (US dollars)
She serves six towns, spends $3,000 a month on Google Ads plus $500 on other marketing, and wins 30 new paying customers: about $117 each. Google Ads reports 45 conversions; her lead log shows 22 callers who named Google, 9 of whom paid. She stops ads in two towns for a month. Leads there fall from 20 to 12 while the other four hold steady, so she turns the ads back on.
A Toronto med spa on Instagram (Canadian dollars)
Meta reports 12 bookings from ads in a month; 21 new clients pick "Facebook or Instagram" on the intake form. Searches for the spa's name rise from 300 to 390 a month after the ads start. A three-week pause cuts new clients from 30 to 22 against the three weeks before. The ads did more than Meta's count showed. Before emailing those clients offers, she checks she has CASL consent.
A tax preparer with clients from AI answers
Of 40 new clients this season, 9 pick "ChatGPT or another AI assistant". Analytics shows almost no visits from ChatGPT; most of those people searched the business's name or phoned. Without the question, Google would have looked like the source.
FAQ
Did Google get rid of third-party cookies?
No. In October 2025 Google said Chrome keeps letting people choose. Safari already limits them by default.
Can I trust "How did you hear about us?" answers?
As one input, not the whole answer. Compare them with tracked sources and with what changes when you pause a channel.
Do I need a customer data platform?
Not for these five numbers. A CRM or spreadsheet with one row per lead, its source and what it paid is enough.
What is marketing mix modeling, and do I need it?
A model estimating each channel's effect from spend and sales. If data is thin, Google's Meridian suggests "three years of weekly data instead of two."
Different for you?
- Your leads aren't tracked in Google Analytics yet: GA4 key events for a small business.
- You want Google Ads to learn which leads paid: tracking a lead from Google Ads to a booked job.
- You want a monthly report built on paid jobs: how many leads marketing should generate.
- Many leads phone and some calls go unanswered: what missed calls cost.
- You email or text Canadian customers: CASL consent.
- You'd rather have lead tracking and follow-up run for you: Conversion Agent.