What is Retargeting?

Retargeting advertises to people who have already been somewhere: a site, an app, a product page, a checkout, a video on a platform.

In almost every account it reports the best numbers. The lowest cost per acquisition, the highest return on ad spend, the best conversion rate. On a dashboard sorted by efficiency it sits at the top, and the natural conclusion is to give it more money.

That conclusion is wrong often enough to be worth examining carefully, and the reason has nothing to do with the campaign being badly built.

The people in a retargeting audience have already demonstrated interest. Many of them intended to come back. Some had a cart open. A portion would have converted with no further advertising at all.

Every attribution system will credit the retargeting ad for those conversions, because the ad was served and the conversion followed. That is what attribution does: it assigns credit for observed sequences. It has no way to ask what would have happened otherwise.

So the reported cost per acquisition is genuinely low and substantially overstates the contribution. The spend is real, the conversions are real, and the causal link between them is weaker than the report implies.

The size of the gap varies enormously — by business, by product, by how long the purchase cycle is — which is why it cannot be estimated from a rule of thumb and has to be measured.

The measurement is a holdout, and it is one of the few genuinely clean tests available in advertising.

Split the retargeting audience at random. Advertise to one half and not to the other. Compare conversion rates.

The difference is the incremental contribution. Everything above it in the reported figure is credit for conversions that were happening anyway.

The test costs something — the held-out group may convert slightly less — and that cost is the price of knowing. Accounts that run it once usually change how they budget, and accounts that never run it keep scaling toward a number that cannot support the weight.

A geo split works similarly where audience-level holdouts are not available, and is worth the additional noise when it is the only option.

Even setting incrementality aside, retargeting cannot scale independently.

The pool is downstream. It is filled by prospecting, by organic traffic, by email, by whatever else brings people to the business in the first place. Its size is a consequence of those activities rather than of any decision made inside the retargeting campaign.

An account that moves budget from prospecting into retargeting improves its blended reported efficiency immediately, because it has shifted money toward the campaign that reports best. It also stops filling the pool. Several months later the retargeting audience is smaller, frequency against it is higher, and the reported efficiency is falling for reasons that look like fatigue and are actually starvation.

This is among the most common ways a well-managed account gets quietly worse while every number on the dashboard improves.

Frequency is impressions divided by reach, so a small pool does not absorb a budget — it converts it into repetition. The audience entry works that arithmetic through; the consequence here is that a retargeting pool reaches heavy repetition far sooner than its budget figure suggests, and when it does, response falls, negative feedback rises and the cost per result climbs.

Four things address it, in order of effect. A shorter membership window, so the pool refreshes and stops accumulating people whose interest has passed. Excluding converters, so the campaign is not advertising to customers. Sizing the budget to the pool rather than to ambition. Rotating creative faster than in prospecting, on a cadence set by how quickly frequency actually climbs.

Frequency capping is the fifth and is a trade rather than a fix: it limits repetition by declining delivery, which on a saturated pool means unspent budget.

Segment by depth. Cart abandoners, product viewers, category browsers, general visitors. Each is a different distance from buying and deserves a different budget, message and window.

Segment by recency within those, because two days and two months are different signals from the same behaviour.

Exclude across segments, or the segments are not real: a product-viewer audience that includes cart abandoners will behave like a cart-abandoner audience with extra noise.

Say something new. The most common retargeting creative is the product the person already looked at, which addresses none of the reasons they left. Shipping terms, returns, a comparison, an answer to the obvious objection, a review — these give the ad something to do besides reminding.

Keep it in proportion. A retargeting budget that grows faster than the traffic feeding it is buying frequency, whatever the reports say.

Retargeting depends on recognising somebody who was seen before, and the ways of doing that have narrowed considerably.

Third-party cookies have been restricted or removed across major browsers. Tracking prevention shortens the life of first-party storage in several of them. Consent requirements mean a share of visitors are never eligible to be added to a pool at all. App environments have their own permissions.

The practical effects are consistent: pools are smaller than they were, membership windows effectively shorten because identifiers expire before the stated window does, and the audiences skew toward the browsers and contexts where recognition still works.

Two responses help. Server-side event collection recovers some of the signal, because the event is sent from the advertiser's own systems rather than from a browser. And list-based audiences — built from customers and contacts the business already holds — depend on none of the browser machinery and have become the most durable form of this activity.

The strategic consequence is worth stating: an account whose retargeting depends entirely on anonymous browsing behaviour is building on ground that keeps shifting, and one that collects identity properly at signup and purchase is not.

There is no transferable share of budget, and the useful framing is a sequence rather than a ratio.

Fill the pool first. Retargeting cannot create demand, so the campaigns that bring people in have to be funded before the campaigns that follow them up.

Fund retargeting to the pool, not to its reported efficiency. The question is how much the audience can absorb before frequency climbs past where response falls — which the account can observe directly.

Then check the increment periodically, because the answer moves as the pool, the creative and the market change.

An account that follows that order will usually end up spending less on retargeting than its dashboard suggests it should, and making more money.

Retargeting and remarketing name the same practice, and the honest answer is that nothing separates them. Both describe advertising to people who have already interacted with a business, both are used of the same campaigns, and the platforms are inconsistent about which word they print — most social platforms say retargeting, Google Ads has long said remarketing, and one ad set is called either depending on who is describing it. Where somebody does draw a line, it is usually that retargeting is paid media to an anonymous cookie or device while remarketing reaches a known list on an owned channel such as email. That split is coherent, it is not universally observed, and an advertiser who assumes the other party is using it will misread them about half the time. The discipline that pays is not choosing a side. It is saying what is actually being described: which audience, built from which behaviour, over which window, and on which surface it is reached.