Retargeting with a shrinking pool: how to keep it working

Retargeting used to be simple: a pixel followed almost everybody, and you bid to show them the thing they looked at. Two changes broke it. Browsers restricted third-party storage, and consent rules meant a large share of visitors never get a pixel fired at all.

What remains is real but smaller, and the shops still getting value from it have moved the centre of gravity from the pixel to their own customer data.

How much has actually gone

The honest answer is that it depends on your consent rate, and that most shops have never measured theirs.

1. Measure the consent rate first. The share of sessions where analytics and advertising storage are granted. This is the single number that determines how much retargeting you can do at all.
2. Compare pixel-observed sessions to server-side session counts. The gap is the invisible portion.
3. Check the decay. Even consented identifiers have shorter lifetimes than they used to, so a 30-day retargeting window may in practice be a 7-day one.

Shops doing this for the first time often find their retargetable audience is a third of what the platform's audience size suggests, because the platform is counting modelled and inferred members.

Move the base to data you own

• Customer match on consented segments is now the sturdiest audience you can build. Match rates are lower than the raw list, but the members are real.
• Suppression is the underused half. Uploading recent buyers to exclude them stops you paying to advertise a thing somebody already owns — usually the highest-return change available.
• Lookalikes from a consented seed extend reach without relying on following individuals.
• Server-side event sending, with consent, restores measurement the browser no longer provides — but it is not a way around a refusal, and treating it as one is both a compliance problem and a data quality one.

With a smaller pool, creative matters more

When the audience was huge and cheap, frequency did the work. A small audience seen many times becomes irritating quickly, and the same dynamic ad shown twelve times is how a brand gets muted.

1. Cap frequency deliberately — a handful of impressions per person per week, not per campaign.
2. Cap duration. Somebody who looked at a product three weeks ago and has not returned is not converting because you showed them a fourth ad.
3. Separate the audiences by intent. A cart abandoner and a category browser deserve different creative; merging them wastes the one thing you know.
4. Exclude converters everywhere, on every campaign, refreshed daily rather than weekly.

When to stop spending on it

Retargeting is the format most flattered by attribution, because it targets people already far down the funnel. A holdout test on retargeting is uncomfortable and usually clarifying.

• Run a geographic or audience holdout for a month and compare conversion rates in the held-out group.
• Expect a smaller true effect than the platform reports, especially where view-through is being counted.
• If the pool is below the platform's minimum and you are relying on modelled expansion to reach it, you are no longer retargeting — you are running prospecting with a retargeting label and a retargeting bid.

The channel that quietly replaced it

For the specific job retargeting used to do — reaching somebody who looked and did not buy — an email or a message to a consented subscriber does the same thing, costs a fraction, reaches the person reliably, and is measurable.

It requires that you identified them, which means an email address and consent, which means the browse abandonment flow needs a subscriber rather than an anonymous session. That is a good argument for putting effort into getting the address early, and it is where the budget that used to go to retargeting is most usefully redirected.