Arman paid for a course at an IELTS prep centre two weeks ago and comes to class on Tuesdays and Thursdays. Instagram still shows him "Free trial lesson and level test". He sends the manager a screenshot: "Am I supposed to come to the trial again?" The same day the centre's owner writes to the media buyer: "Let's retarget everyone who's been on the site. A competitor has been following me around with ads for a week."
Arman's complaint and the owner's request are solved by the same setup. Retargeting means showing ads to people who have already been on your site or profile, and the same audiences double as exclusions: they tell Meta who should no longer see the sign-up ad. Below we build the audiences for the centre from the Meta lead test, set up the campaign and work out how much of its result the ads actually earned. The centre's numbers are illustrative.
What Meta builds audiences from
In Ads Manager these are called custom audiences. The centre can use three sources:
| Source | Who gets in | How long people stay |
|---|---|---|
| Website | Visitors to all pages or selected ones, people who fired a pixel event such as Lead | Up to 180 days, up to two years for purchases |
| Engagement | People who watched a video, engaged with an Instagram account or Facebook Page, or opened an instant form | Up to 365 days for an Instagram account and a Facebook Page |
| Customer list | Phone numbers and emails from the CRM that Meta matched to its accounts | No window: whoever is on the list is in the audience |
Three details are worth knowing before you set anything up. A website audience starts filling from the moment you create it, and Meta won't backfill it (create a website custom audience). So create your audiences on the first day the pixel is live, even if retargeting starts a month later.
The second detail is size. Only visitors Meta recognises make it in, minus those who opted out of ads based on their activity. That is why the ad account shows fewer people than GA4 shows visits. Until a website audience has a few hundred people in it, Meta advises against using it (website custom audience size).
The third is the customer list. Meta recommends at least 1,000 people, and if the list goes more than 30 days without an update, the audience gets the "Outdated" status (how to use custom audiences). And a number someone left so a manager could call them is not permission to upload it to an ad platform: personal data law sets its own conditions for that.
Whom to follow and whom to exclude
Describe the audiences in words first, then build them in the ad account. The centre ended up with four:
| Audience | Rule | Window | Purpose |
|---|---|---|---|
| Viewed the course, didn't sign up | Website: URL contains /ielts, minus anyone with Lead and students | 90 days | Retargeting |
| Submitted the form | Website: Lead event | 180 days | Exclusion |
| Students | Website: the /students page the manager sends after payment | 180 days | Exclusion |
| Watched the centre's videos | Engagement: Instagram video | 30 days | Add to retargeting if the website audience is too small |
The window is how many days a person stays in the audience after their last visit. The longer it is, the bigger the audience and the further people are from the day they were interested. The centre chose 90 days: some people prepare for IELTS around an exam date they have already booked and may look at the curriculum months before they decide.
The "Students" row matters most, and it is exactly what was missing for Arman. He signed up by phone and never filled in the form, so he has no Lead: the centre had no exclusion, and one based on leads wouldn't have caught him anyway. The centre made a page for students with the group schedule: the manager sends the link after payment, and everyone who opens it lands in the "Students" audience without exporting phone numbers. The other route is a CRM list, but it has to be re-uploaded every month and needs the students' consent.
"Submitted the form" and "Students" go into every campaign aimed at new people, including the campaign that grew out of the week 2 test. Excluded custom audiences are part of the audience controls, which Meta treats as hard limits even with Advantage+ audience on. But Meta can only exclude people it recognises, so a few students will still see the ads. The manager should know in advance what to tell them.
Setup: the audience as a control, not a suggestion
The retargeting campaign is built like the test: the Leads objective, Website as the conversion location, the Lead event. What changes is the audience and the ads.
With the Leads objective, you add a custom audience in the Audience section through "Add suggestions (optional)". But it is only a suggestion: Meta will also show the ad to people outside the audience if it expects more results that way. You end up with an ordinary campaign that has a hint, not retargeting. To show the ad only to "Viewed the course, didn't sign up", click "Further limit the reach of your ads", select the audience again and untick "Use as a suggestion". Exclusions go in through "Show more controls", in the "Custom audience exclusion" field (how to use custom audiences). Put "Students" here as well if the audience rule doesn't already subtract them: otherwise Arman and everyone who signed up without the form end up in retargeting.
A retargeting ad doesn't repeat what the person has already seen. They opened the course page, so they know the offer, but didn't sign up. A visit doesn't tell you what stopped them, so the centre made two ads around the two questions the manager hears on WhatsApp every day: how much the course costs and when the groups start. Under the tagging rules they are named rt_price and rt_schedule, and the URL parameters string is the same as in the test, with the ielts_trial campaign.
Frequency: a small audience wears out fast
In the week 2 test a thousand impressions cost 822 ₸: 9,860 ₸ for 12,000 impressions. Suppose retargeting impressions cost the same, though on a narrow audience the price may differ: what follows is a rough estimate, not a forecast. At 1,000 ₸ a day that is about 1,200 impressions a day, or 8,500 a week.
The course page gets about 300 visitors a week. Over 30 days that is roughly 1,300 visitors, and say Meta recognises about 1,000 of those who didn't submit the form. 8,500 impressions over 1,000 people is more than eight impressions each per week, and that is the average. Over 90 days there are about 3,900 visitors, roughly 3,000 people in the audience, and frequency comes out at about three a week. That is why the centre went with 90 days.
Average frequency hides the spread: one person sees the ad once, another fifteen times. The centre set itself a ceiling of four impressions a week: people have already seen the offer on the site. Meta offers a frequency cap in campaigns with the Awareness and Engagement objectives (objectives and frequency controls, frequency controls). In a leads campaign you hold frequency down with the budget and the audience size: watch the Frequency column for the last seven days, and once it crosses your ceiling, cut the budget or widen the window.
What the report showed and how much of it is real
The centre ran retargeting for two weeks: a 90-day window, 1,000 ₸ a day.
| Metric | Value |
|---|---|
| Spend | 14,000 ₸ |
| Reach | 2,300 |
| Impressions | 17,500 |
| Frequency | 7.6 over two weeks |
| Leads in the ad account | 12 |
| of which 7-day click | 7 |
| of which 1-day view | 5 |
| Came to the trial lesson | 3 |
For the last seven days the Frequency column shows 3.9, under the ceiling. The CPL in the ad account is 1,167 ₸, cheaper than the scaling steps (1,400 and 1,615 ₸). Each person at the trial lesson cost 4,667 ₸ against a target of 8,000 ₸. The owner wants to move money into retargeting from the campaign aimed at new people.
First split the leads: "Columns: Performance" → "Compare attribution settings" (compare attribution settings). The ad set uses the default window: 7 days after a link click or 1 day after a view (default attribution window). The centre's list has seven leads tagged rt_, the same as the click count in the ad account. Meta credited the other five on view: these people arrived some other way, and in the list they sit among the leads from search or without tags, and there's no telling which ones. Had the third window, for engagement without a link click, been switched on in the ad set, some of the five could have come from it as well. The data doesn't show whether the impression nudged them, so they stay out of the report.
Seven people tapped the ad. But all of them had already been on the course page and knew the centre, and some would have come back without the ad. How many, the report can't say: by design, retargeting shows ads to people who are already closer to applying, so its CPL can't sit next to the CPL of a campaign aimed at new people. And three people at the trial lesson are too few to price a trial visit: with two it would be 7,000 ₸, with four 3,500 ₸.
What the centre did: it kept retargeting at 1,000 ₸ a day as an add-on, reports only click leads tagged rt_, and left the budget for new people alone.
How many leads the ads added
Only a comparison with a group that wasn't shown the ad answers that: in Ads Manager this is a conversion lift test under Experiments, where Meta randomly splits the audience into test and holdout groups (about lift tests and holdouts). Meta recommends a budget of at least $5,000 and a run of at least 28 days for it (how to set up a conversion lift test), and at the official exchange rate for 13 October 2026, 451.31 ₸ to the dollar, that is over 2.2 million ₸, 25 times a month of the centre's Meta ads at 3,000 ₸ a day.
You can't split a pixel audience in half yourself. But you can test a different one: CRM leads who never made it to the lesson. Shuffle the list, upload one half to Meta, leave the other half alone, and a month later compare in the CRM what share of each half booked a trial lesson. A push notification holdout works the same way. The catch is size. Suppose three in a hundred book a trial within a month without ads, and you hope the ads lift that to 4.5. The A/B test planner tab on the calculators page with a 3% baseline and a 50% minimum detectable effect asks for 2,171 people per group, 4,342 in total. The centre gets about 10 such people a week, about 500 a year.
The conclusion for the centre is modest: it can't measure the lift from retargeting now or a year from now. Once the list holds thousands of people rather than hundreds, this kind of test becomes feasible.
Exercise: audiences and exclusions for your own project
Frequency works out like this: divide the daily budget by the cost per thousand impressions, multiply by 1,000 and by 7, then divide by the audience size. For the centre's 30-day window that is 1,000 / 822 × 1,000 × 7 / 1,000, about 8.5 impressions per person per week. If your number is also above your ceiling, widen the window or add an engagement audience. And if even the lowest budget the ad account accepts gives too high a frequency, you don't need a separate retargeting campaign yet. Create the audiences anyway: you need them for exclusions.
If the buyers audience doesn't fill up, the purchase is happening outside the site, as it did for Arman: by phone, at the office, through Kaspi. Then look for a page only customers open, or arrange a CRM list together with consent for that kind of transfer. And if your tagged leads match the clicks in the ad account and Meta credited the rest on view, leave those people out of the client report: the data doesn't show whether the impression nudged them.
Video: events and audiences
An overview of remarketing in Google and Facebook, in Russian. Check Ads Manager section names against the lesson text.