Measure visits to this site with Google Analytics? Only with your OK. Privacy
Acteurly

Marketing attribution for small business: which channel pays

By the Acteurly team · · 8 min read

In short

To know which marketing channel really pays, don't trust platform conversions: each one counts every customer it touched, and together they almost always exceed your real customers. Each month, compare what they claim with the new customers your log confirms, then pause a doubtful channel for four weeks: if you lose no more than √(your usual customers), it isn't paying.

First of the month, 8 a.m.: 40 conversions, 31 new patients

First of the month, 8 a.m. The owner of a physiotherapy clinic opens her reports. Google Ads claims 26 conversions. Meta claims 14. That's 40. Yet her booking system shows 31 new patients, and she knows a good share were sent by two doctors down the street. Which number is right?

None of them, at least not on its own. Every ad platform grades its own homework, with its own rules, and each one counts any customer it touched. Add up what they claim and you will almost always land above the number of customers you actually signed.

Our view: for a small business without an analyst, marketing attribution isn't solved by picking the right model in Google Analytics. It comes down to two simple habits. A monthly reconciliation between what the platforms claim and what your customer log confirms. And, when real money is in doubt, a pause test: switch a channel off for four weeks and see whether customers disappear.

Why every platform takes the credit

It isn't bad faith; it's how they're built. Three things inflate the numbers.

  1. Each one only sees its own touches. In Google Ads, data-driven attribution is the default model for most conversion actions; it shares credit among Google ads, not with your neighbour's recommendation. Meta, according to Meta ads specialist Jon Loomer, counts conversions by default within 7 days of a click and within one day of an ad simply being shown, even with no click. The same patient can be claimed by both.
  2. A conversion isn't a customer. A 20-second call, a form submitted twice, an existing patient rebooking: all of these can count.
  3. Part of word of mouth is invisible. A link sent by text or Messenger often arrives with no referrer and gets filed as "direct" traffic. A writer at The Atlantic named this "dark social" in 2012. And the patient who calls straight from your Google Business Profile never visits your site; the profile only counts how many times someone tapped the call button.

What about Google Analytics 4? Only three attribution models are left: data-driven, paid and organic last click, and Google paid channels last click. First click, linear, time decay and position-based were retired in 2023. Data-driven attribution compares paths that convert with paths that don't, and a conversion can be reassigned for up to seven days. Useful, but GA4 only sees website visits. What matters most for a small business, the referral, the truck, the direct call, never reaches it. Switching models just reshuffles crumbs.

"How did you hear about us?", asked properly

The question everyone recommends is usually asked badly. Four rules make it reliable.

  1. Open first, no menu. "How did you hear about us?" Write down the answer in the customer's own words. A list offered too early invites them to tick the first plausible box.
  2. Two columns, not one. Then ask: "And the very first time, where was that?" The trigger (what made them call) and the discovery (how they first heard of you) are often different channels. A customer who "came from Google" was often searching for your name after hearing it at a family dinner.
  3. Be wary of the word "Google." To a customer, the ad, the map and the organic result are all the same thing. Write "Google" as said, then narrow it down with your other tracking.
  4. Ask every new customer, the same week. Three weeks later, nobody remembers. Add a "ChatGPT or another AI" option: ChatGPT adds utm_source=chatgpt.com to links to your site, but the person who reads the answer and then phones you leaves no trace.

That other tracking is cheap. Add UTM parameters to every link you control: newsletter, Facebook posts, the QR code on a flyer. Google requires utm_source, utm_medium and utm_campaign every time, and values are case-sensitive, so keep everything lowercase. For calls from your ads, the Google forwarding number, available in Canada, records the time, duration and campaign of each call.

On your Google Business Profile, go carefully with call tracking numbers. Google's guidelines don't mention them by name, but they require a number the business controls directly, local where possible. If you use one, keep your usual number among the two additional numbers the profile allows.

The first-of-the-month reconciliation: an example in Rimouski

Back to the physiotherapy clinic, in Rimouski, Quebec, in October. All figures are assumptions for the example. It spends $1,200 on Google Ads and $450 on Facebook and Instagram ads. Every new patient went into the log with their answer to the question.

The reconciliation table: what platforms claim, what your log confirms

ChannelPlatform claimsLog confirms
Google Ads26 conversions9 new patients
Facebook and Instagram (ads)14 conversions3 new patients
Google Business Profile41 taps on "Call"8 new patients
Referrals (doctors, word of mouth)No platform9 new patients
Don't knowNo platform2 new patients

Log total: 9 + 3 + 8 + 9 + 2 = 31. The two ad platforms claim 26 + 14 = 40 conversions; the log confirms 9 + 3 = 12 of them.

Next, work out each paid channel's reality ratio: confirmed customers ÷ claimed conversions. Google Ads: 9 ÷ 26 ≈ 0.35. Meta: 3 ÷ 14 ≈ 0.21. Along the way, the ranking flips. On the dashboard, Meta looks cheaper: $450 ÷ 14 ≈ $32 per conversion, against $1,200 ÷ 26 ≈ $46 for Google Ads. In the log, it's the reverse: $450 ÷ 3 = $150 per patient for Meta, $1,200 ÷ 9 ≈ $133 for Google Ads.

Read a ratio over three months, not one. If it holds steady, it becomes your converter: in a month when you can't fill in the log, multiply the platform's conversions by it. The cost per confirmed patient is the one that belongs in your customer acquisition cost, and the table itself fits in your one-page monthly report.

The pause test: proof that a channel pays

The table tells you which channel customers name. It doesn't tell you what would happen without that channel. The Google Ads patient might have called anyway. To find out, switch the channel off and watch the total.

Start with the campaign on your own business name, if you run one. In a large experiment at eBay, economists found that brand-keyword ads had no measurable short-term benefit: people who type your name are already looking for you. If a competitor bids on your name and shows up above you, the logic changes; turn the campaign back on.

The pause-test calculation, ready to copy

  • Customers lost = new customers in the 4 weeks before − new customers in the 4 weeks of the pause
  • Noise threshold ≈ √(new customers in the 4 weeks before)
  • If customers lost exceed the threshold, the channel pays. Cost per real customer = money saved ÷ customers lost
  • Example: the clinic pauses its brand campaign, $500 per 4 weeks. Before: 28 new patients. During: 27. Lost: 28 − 27 = 1. Threshold: √28 ≈ 5.3. Since 1 is below 5.3, there is no measurable loss: the clinic keeps $500, or $6,500 over 13 four-week periods
  • Had it lost 8: 8 is above 5.3, and $500 ÷ 8 ≈ $63 per real patient, an excellent price; switch it back on
  • Your numbers: ___ customers before − ___ customers during = ___ lost; threshold √___ = ___; $ ___ ÷ ___ customers lost = $ ___ per real customer

Three rules make the test mean something. One channel paused at a time. A quiet period, checked against the same weeks last year. And in Quebec, avoid the holidays, March break and, in the trades, the construction holiday at the end of July. If you serve several towns, there's a variant: switch ads off in one area, keep them in another, and compare the two.

What attribution will never tell you

Small numbers. The noise threshold is a rough rule. With fewer than ten or so new customers a month, a pause test only catches big effects. Read the table over a quarter, not a month.

Slow channels. A minor hockey sponsorship, a sign, regional radio: their effect takes months. Four weeks of pause says nothing about them. Judge them by the "discovery" column in your log.

Customers in two languages. Ask the question in the customer's language and sort the answers into the same channels. And because the log holds personal information, protect it and limit who can open it: Quebec's Law 25 applies to that spreadsheet too.

AI. More and more customers will read an AI answer before they call. No platform will tell you; only your question can. This month, check whether chatgpt.com shows up as a source in GA4, and compare it with the "AI" answers in your log.

If you spend tens of thousands of dollars a month across several platforms, an analyst and a proper incrementality test are worth paying for. Below that, the log and the pause are enough.

Your plan for the next 30 days

  1. Today: add two columns to your customer log, "trigger" and "discovery," and ask every new customer.
  2. This week: add UTM parameters to your newsletter and posts, and turn on the Google forwarding number for your call ads.
  3. On the first of the month: fill in the reconciliation table and work out the reality ratio for each paid channel.
  4. The month after: pick the channel with the lowest ratio, or your brand campaign, and start a four-week pause. Write the start date in the log.

Frequently asked questions

Which GA4 attribution model should a small business use?

Data-driven attribution suits most small businesses, since it is the model Google maintains and sets as the default. Last click is still handy for a simple read. Don't spend time comparing them: the gap between two models is small next to what GA4 can't see, such as direct calls and referrals.

Why do Google Ads and Facebook report more conversions than I have customers?

Each platform counts the conversions it touched, with no idea the other one saw the same customer. Their conversions also include short calls, duplicate forms and existing customers. Compare their figures to your confirmed new customers every month to learn the real gap.

How do I know which marketing channel brings the most customers?

Divide what you spend on each channel by the new customers your log confirms for it, not by the conversions on the dashboard. For a paid channel you doubt, run a four-week pause test and check whether total new customers drop by more than the noise threshold.

Do I need call tracking software?

Not to start. The Google Ads forwarding number already tracks calls from your ads, and your Google Business Profile counts taps on the call button. Software starts to pay off when the phone is your main source of customers and you spend on several channels at once.

In Acteurly

Read next