How to optimize Meta lead ads for qualified leads, not lead volume

Optimizing Meta lead ads for qualified leads means giving a campaign on Facebook and Instagram a different job: not “bring a lead” but “bring a lead that sales will confirm”. Meta’s name for it is conversion leads optimization. Below: what the algorithm learns from, which CRM stage to send, how many events are enough, the set-up and the numbers to read. The examples come from a real report: a driving school, 2–8 October 2026.

What does Meta’s algorithm learn from?

Only from the outcome that reaches it. In a lead-form campaign there is one such outcome: the form was submitted. The algorithm multiplies it and makes it cheaper — it finds people who fill in forms easily. Whether they pick up the phone or buy becomes known in the CRM, and never reaches Meta by itself.

In one week at the driving school Meta counted 863 leads. That week’s 36 sales were closed in the CRM, and Meta learns of a sale only if the CRM tells it. Without that signal all 863 leads are the same “result” to the algorithm, and it looks for people who fill in forms; with it, for people like those who turned out qualified and bought. The Conversions API carries that signal from the CRM to Meta.

Which CRM stage should count as a “qualified lead”?

The stage has to meet two conditions: it happens often, and it happens early. The guideline in Meta’s developer documentation is a stage that 1–40% of leads reach, within 28 days of the lead. The logic is simple: a stage nearly everyone reaches tells the algorithm nothing, and a rare one gives it too few examples.

The driving school’s qualified share for the week is 39% — at the top edge of Meta’s guideline; over 30 days (9 September – 8 October 2026) it was 44%, and 48% at a learning centre. If the share stays above 40%, consider moving the signal one stage later: in CAPI Platform any stage can carry an event under a name you choose.

Product screen · capip.uz

The sale signal goes out either way: Purchase carries the amount and currency when the deal has an amount. The only question is which event the campaign optimizes for.

How many events are enough?

Two Meta pages answer differently. The developer documentation asks for at least 200 leads a month and signals sent at least daily. The newer marketing page says there is no minimum volume, 200+ is only a recommendation, and a single conversion event can be optimized for. In practice: treat 200 as the target, but start sending signals below it.

The same arithmetic helps to choose the stage. At the driving school the qualified-lead stage can give 335 signals a week — about 48 a day; the sale stage 36, about 5 a day. With a small flow of leads, a late stage gives too few examples.

On timing: Meta’s documentation speaks of a learning phase of 2–4 weeks. Do not judge by the first days’ numbers.

How do you switch a campaign to qualified leads?

1. 1

   Tie the CRM stage to a signal

   On CAPI Platform’s Funnel page the qualified-lead stage is given QualifiedLead, the sale stage Purchase. When a lead enters the stage, the event goes to Meta by itself. It works with Bitrix24, amoCRM (Kommo) or capiCRM.

2. 2

   Check that the signal goes out

   The “Send a test event” button checks the token, the dataset and the whole path. Then every signal is in the CAPI events log with its time, name and status. Let signals accumulate first.

3. 3

   Create a custom conversion

   QualifiedLead is not a standard Meta event; the platform sends it under that name. So in Ads Manager (Events Manager → Custom conversions) you build one from it: source — your dataset, event — QualifiedLead.

4. 4

   Change the goal in a new ad set

   As the performance goal, choose qualified leads rather than the number of leads — Meta’s own page words it “maximize number of qualified leads” — and select the custom conversion you created. The dataset must be connected to that ad account.

5. 5

   Test it beside the old ad set

   Same creatives, form and audience; only the goal differs. Leave the running ad set unedited, or there is nothing to compare with.

6. 6

   Read the result by CPQL

   Once the learning phase has passed, compare the two ad sets in the report: qualified share, CPQL and sales.

A lead-form campaign can also be created on CAPI Platform’s Launch campaigns page, with the goal “Quality leads (CAPI)”. The condition is the same: CAPI switched on and the dataset connected to the ad account.

What changes in the numbers, and how do you read them?

Leads may become fewer and the cost per lead may rise. That is expected: the algorithm now looks for a person who will turn out qualified, not for the cheapest form-filler. So after the goal changes, decisions are not made on cost per lead.

Product screen · capip.uz

• Qualified share moves first. In the driving school’s week it is 39%: 335 of 863 leads.
• CPQL is the main measure. If the lead got dearer but the qualified lead got cheaper, the optimization is working.
• CAC has the last word, but it lags: do not conclude before a sales cycle has passed. At the driving school sales are 11% of the qualified leads.
• Cohort mode. Switch the report to “Cohort (by lead creation date)”: it shows how many of that period’s own leads became qualified and bought.

What have published tests shown?

Three public sources. None is a CAPI Platform result or a promise: the figures are given as published.

Read TECOBI’s two figures together: the conversion rate was 68% higher, while the cost per qualified lead was only 8% lower. The gain is in who the leads are, not in a cheaper lead. Univo’s figures rest on metrics the client shared with Meta.

Which mistakes spoil the result?

• The stage is too broad. If everyone who answers the phone becomes “qualified”, the signal is no different from a plain lead.
• The stage is too late. A handful of sales a week does not give the algorithm enough examples.
• Cards are not moved. The signal goes out when the stage changes: while a lead sits still for a week, Meta knows nothing for a week.
• Every manager has their own “qualified”. Agree on one written criterion.
• Expecting old leads to count. The 30 days of history loaded on connection go into the report only, not to Meta. The algorithm learns from the day signals start.
• Edits during learning. Change the stage, the goal and the budget every day and there is no period left to compare.