How to track lapsed customer reactivation to repeat purchase for Meta Ads

Architecture

The exact event sequence

Five stages. Nothing breaks technically — the tracking works perfectly. The problem is that a campaign shown only to people who already like you will always look effective. The identifier has to survive all of them.

1. Step 1

   Lapse defined

   The first real decision. Ninety days for a coffee subscription and ninety days for a mattress company mean entirely different things.

2. Step 2

   Audience built

   From your own customer list, uploaded to the platform. Everyone in it has bought before.

3. Step 3

   Ad served

   To a known past customer, who may already have been intending to reorder.

4. Step 4

   Return visit

   Frequently direct, because they know your brand and type the URL.

5. Step 5

   Repeat purchase

   Credited to the campaign, whether or not the campaign caused it.

Diagnostic sequence

What's specific to Meta Ads

The funnel logic above is the same wherever you advertise. These are the parts that change because it's Meta Ads.

1. 1

   Click identifier: fbclid, which becomes the _fbc cookie

   fbclid arrives on the landing URL and is normally written into the _fbc cookie. If it isn't captured on landing, _fbc never exists — and _fbc is the strongest click signal Meta has, so its absence caps your match quality regardless of what else you send. Forward _fbp too, from the browser rather than regenerating it server-side.

2. 2

   Import mechanism: the Conversions API

   Server events go through CAPI with an action_source describing where the conversion happened. Offline outcomes flow through the same API rather than a separate product. Verify current field names against Meta's documentation before building — this surface changes.

3. 3

   Deduplication

   This is the mechanism that decides whether adding CAPI helps or doubles your conversions. Both the Pixel and the server event must carry the same event_id — the order or booking reference works best, because both sides can produce it without coordinating. A case difference in the event name alone produces two events even with a matching ID. Read the Deduplicated indicator in Events Manager over a full week, not one session.

4. 4

   Window and timing

   Meta's default attribution is a 7-day click and 1-day view. View-through is included by default, which is a large part of why Meta claims more conversions than your other platforms — that overlap is expected rather than a fault.

5. 5

   The Meta gotcha

   Match quality is where most Meta accounts lose attribution, and it's almost always normalization rather than coverage. An email hashed with a capital letter, or a phone hashed with brackets, produces a perfectly valid hash that matches nobody and errors nowhere. Normalize first, then hash, and read the score per event rather than overall.

Source-of-truth validation

How to prove the chain works

One real journey, followed all the way, then a reconciliation. Neither on its own is enough.

• A live Meta ad clicked, and the identifier confirmed present on the landing URL.
• The same identifier confirmed on the repeat purchase from a lapsed customer after a full real journey through every step.
• Repeat purchase confirmed arriving at Meta Ads, verified by the Deduplicated indicator and the per-event match quality score, both read over seven days.
• Your order system, with customer lapse status at the time of the ad reconciled against Meta Ads per record over a fully closed window.
• The same reconciliation repeated 30 days later.

Straight talk

When this doesn't apply to you

Worth checking before you build anything, because two of these are common and both mean the work isn't worth doing.

• If you have no repeat purchase behaviour at all — a one-time high-value product — there is no reactivation funnel to build.
• If you're advertising a business type Meta restricts data collection for — health being the obvious one — the parameters you can legitimately send are narrower, and that constraint should shape the build rather than be worked around.
• If your cycle from click to repeat purchase regularly exceeds the import window, the final event can't drive bidding whatever you build. Optimize toward repeat purchases, reported against a holdout rather than in isolation instead and use the later data for budget decisions rather than for bidding.
• If almost none of your reactivated purchases carry a click identifier today, that's the whole project. Fix capture first — the import is the easy half and no tooling rescues a record with nothing to reference.