Retargeting ads take credit for buyers who were already coming

Retargeting shows ads to people who already found you, so most of the sales it claims were coming anyway. Judge it on two things: matched sales from leads it was first to record, and a holdout if the budget is large. Never on platform-attributed revenue, which counts view-throughs and last touches it did not cause.

A kitchen remodeler spends $6,000 a month on retargeting: display banners across news sites, a carousel on social, a programmatic line from an agency. The dashboard says 120 conversions and a return of more than ten dollars per dollar, which is three times better than paid search and five times better than paid social. Every quarter someone suggests moving budget into retargeting, and every quarter it looks like the obvious call.

It is not obvious. Nearly everybody that campaign reached had already been on the website, because that is how retargeting works: it can only follow somebody who has visited. Most of them arrived through search, social, a referral or a yard sign. Some had already called. The retargeting ad was the last thing many of them saw before buying, and last is where the credit lands. This post is about separating what retargeting adds from what it only collects.

What retargeting is, and why its numbers look so good

Retargeting, or remarketing, shows ads to people who have already interacted with you: visited the site, viewed a product, started a form, or appear on a customer list. Its audience is selected for intent, so it converts well whatever the ads do. A high conversion rate here proves the audience was good, not that the ads were.

The mechanics differ by platform, but the audience is always built from something you already had:

• Display retargeting. A tag on your site adds visitors to a list; banners follow them across publisher sites for a set number of days.
• Social retargeting. The same idea inside a social feed, built from a pixel, from people who engaged with a post or video, or from an uploaded customer list.
• Programmatic retargeting. A demand-side platform buys impressions for your visitor list across exchanges, often bundled into a larger buy so the retargeting share is hard to see. Our post on retail media and programmatic covers how those reports are assembled.
• Search remarketing lists. Higher bids or different ads for past visitors when they search again. These reach the people most likely to buy of all.

None of this is a scam. Reminding an interested person can tip a decision, and the experimental evidence below says it sometimes does. The problem is narrower and more expensive: the standard way of counting retargeting results cannot tell a sale it tipped from a sale it merely stood beside. Prospecting display, which reaches people who have never visited, has a different problem, covered in display advertising attribution.

Three ways retargeting claims a sale it did not make

Retargeting inflates its own results in three ways: it counts sales after impressions nobody clicked, it is the last touch for people already deep in a decision, and it claims the same buyers as the channel that found them. Each is a reporting rule working as designed, not a fault.

1. View-through credit

A view-through conversion is a sale or lead recorded after someone was served an ad and did not click it. For retargeting, view-through credit is especially generous, because the audience is people who are likely to convert anyway and the ads are served to them repeatedly. Show a banner to every recent visitor, and some of them will buy within a day for reasons that have nothing to do with the banner. Each of those purchases can become a view-through conversion.

Read the defaults as a hint from the platform itself: view-throughs are a weak signal, so they are kept out of the main column. The trouble starts when a report switches to All conversions, which is the bigger number, or when a programmatic vendor reports a combined figure with a long view window and no breakdown. Ask for click-through and view-through separately, every time.

That change removed the longest view windows from reporting tools that read Meta's API, and anyone comparing retargeting results across early 2026 should check whether the series changed underneath them. It did not remove view-through credit: one day of it is still there, and one day is plenty for an audience that was about to buy. How window length tilts every channel's numbers is covered in attribution windows; for retargeting the short version is that even the shortest view window is long enough to be flattering.

2. The last touch on people already in the funnel

Retargeting is built to be the last thing someone sees. It follows a visitor for days or weeks and, if bids are set to maximize conversions, the platform's algorithm spends most heavily on the visitors most likely to convert. That is exactly what it should do to hit its target, and it means the budget concentrates on people who were nearly certain to buy. In a last-touch report, which is what a platform's own conversion count effectively is, each of those buyers is a retargeting sale. The attribution models post explains why last touch rewards whatever sits closest to the purchase.

3. Overlapping claims with the channel that found them

A buyer who clicked a search ad in week one, browsed, left, and then saw retargeting banners for two weeks will appear in the search platform's report and in the retargeting report. If both run on the same ad platform, its own deduplication picks one; across platforms nothing does. That is the general problem of overlapping channel claims, and retargeting is its purest case, because every retargeting conversion by construction began with some other touch.

What experiments say retargeting really adds

Controlled experiments show retargeting can work, and also how small the real effect is next to what attribution claims. In one well-known field test, retargeting raised a retailer's sales by about 11% against a holdout: real money, and roughly a tenth of the sales the campaign would have been credited with.

Work through what 10.8% means. If the control group spent $100, the group that saw the ads spent $110.80. The extra $10.80 is what retargeting caused, which is $10.80 ÷ $110.80 = 9.7% of the sales made by the people it targeted. The other 90.3% would have happened without a single banner. That campaign was a genuine success, statistically strong, and still about nine in ten of the purchases a naive report would attach to it were coming anyway.

The second study is a reminder that retargeting's effect depends on where the person is in the decision, which is the same thing that decides whether they would have bought anyway. An ad that follows someone who is still browsing can be ignored; an ad that follows someone comparing two final quotes may tip them. None of that is visible in a conversion count.

The same month, counted three ways

The gap between what retargeting claims and what it adds is usually large enough to change a budget decision. The illustration below uses one month at one business and counts retargeting's revenue three ways: by platform claim, by matched sales from the leads it was first to record, and by an experiment-sized lift.

The figures are illustrative assumptions, not benchmarks. The remodeler spends $6,000 on retargeting. The platforms report 120 conversions: 35 after a click, 85 view-through. Its lead file, built from the call tracker and web forms with a source column on each row, contains 70 leads tagged as retargeting, because only clicks carry a tracking parameter. Matching those 70 on normalized phone and email against everything else in the file shows that 52 people were already there from search, social or a referral before the retargeting lead appeared. Eighteen were recorded for the first time through retargeting. The average job is $4,000.

The arithmetic. Of the 52 people who were already leads, 13 bought; of the 18 new ones, 3 did, so 16 buyers in all have a retargeting lead somewhere in their history. Assume another 7 buyers never clicked anything but were served a banner the day before they called, and appear only as view-through conversions. The platform-style figure is 16 + 7 = 23 sales × $4,000 = $92,000, or $92,000 ÷ $6,000 = 15.3 per dollar. Credit each sale to the earliest matched lead instead, and the 13 return to the channels that found them: 3 sales × $4,000 = $12,000, or 2.0 per dollar. If the lift were the same as the published experiment's, 16 × 9.7% = 1.55 sales, about $6,200, roughly what the campaign cost.

Two cautions keep this honest. The middle row is not a causal number: some of those 3 buyers might have come back anyway, and some of the 13 might have been nudged by an ad. The third row borrows another business's result and is only a sense of scale. What the table does show is that the platform claim sits far above both, and that the difference is decided by people who were already on the way.

A retargeting report that does not say how many of its buyers were already leads from another channel is not a report on retargeting. It is a report on your other channels' customers, routed through the last ad they saw.

Which leads did retargeting actually create?

The most useful question you can answer without an experiment: for each lead tagged as retargeting, was this person already in your lead file from another source? If yes, retargeting reached someone another channel found. If no, retargeting recorded them first, and that subset, matched to closed sales, is the fairest attribution figure retargeting can earn.

Retargeting cannot create a visitor; it needs a prior visit. What it can create is the first contact you can see. Plenty of visitors browse and leave without calling or filling in anything, and nobody knows who they were. If a banner brings one back and they call, that call is the first record of them. Something else brought them to the site the first time, and you will rarely know what. Crediting the first recorded lead gives retargeting the people it converted from anonymous to known, and nothing more.

1. Tag retargeting clicks separately. Use a source value on retargeting campaigns that is distinct from prospecting on the same platform, so the lead file can tell a retargeting click from a cold one. A social lead with no distinction between the two cannot be judged.
2. Build one lead file with dates. Calls, forms and chats from every source in one export, each row with the contact's phone or email, the source and the date. A file per channel hides the overlap you are looking for.
3. Normalize before comparing. Phone numbers in one consistent international format, emails lowercased and trimmed. Otherwise the same person looks like two, and retargeting appears to have found people it did not.
4. Find each person's earliest lead. Group by phone or email and sort by date. Mark every retargeting lead as first recorded or already known, and record which channel came first for the already known ones.
5. Match closed sales to the people, not to the clicks. A sale belongs to the earliest lead for that buyer. Count retargeting's sales only where its lead came first, and list separately the sales where it appeared later.

The already-known share is the single most informative number in the exercise. If three quarters of retargeting's leads were known before it touched them, three quarters of its platform-reported value belongs, at least in part, to other channels. The same grouping is what keeps a person who filled in two forms from being counted twice; see duplicate leads for how to group them without losing the first source.

CloseRev does this match from two CSV exports: your closed sales and your leads. A sale counts toward a channel when its normalized phone number or email appears in the lead file, credited to the earliest such lead, so a buyer who was a search lead before clicking a retargeting ad is credited to search. Sales with no matching lead stay in Direct / Unknown. It does not read ad platforms, impressions or view-through conversions, and it does not run holdout tests.

One consequence of matching on contacts is worth stating plainly. View-through conversions never appear in a lead file, because nobody clicked and no source was recorded. A buyer who saw a banner and then called your main number shows up under whatever source that number carries, or under Direct / Unknown. That is not an undercount of retargeting; it is a refusal to credit an impression that left no trace.

When a holdout is worth it for retargeting

Retargeting is one of the best candidates for a holdout test, because its attributed returns are high, its audience would buy at a high rate anyway, and the audience is a list you control. If retargeting is a meaningful share of spend, a holdout is the only way to learn what it causes.

The general design, including the geographic version run on your own closed sales, is in our post on incrementality tests. Three points are specific to retargeting:

• Hold out people, not just places, where the platform allows it. Because the audience is a list, some platforms can split it at random and withhold ads from one part. A random split of the same audience is a cleaner comparison than two regions.
• Exclude buyers from the audience first. A list that still contains last month's customers spends on people who have already paid, and makes any test noisier.
• Measure in closed sales, not platform conversions. A lift in site visits, which the experiment above also found, is not a lift in revenue. Compare the two groups' matched sales over the test and a few weeks after.

If volume is too low for a test to detect a 10% difference, and for many local businesses it is, the first-recorded-lead figure is the best guide you have. It will be imperfect in both directions, but it will not credit retargeting with your other channels' customers.

How to report retargeting honestly

Report retargeting on matched sales from leads it recorded first, show the share of its leads that were already known from other channels, and keep view-through counts out of revenue entirely. Where a holdout has been run, report its incremental estimate beside the attributed figure, never multiplied into it.

• Matched sales from first-recorded leads, with revenue and cost per sale, in the same table as every other channel.
• Already-known share: retargeting leads that existed in the file from another channel, and which channels those were. This credits search, social and referrals for the people they found.
• Platform claim, labeled as such, with click-through and view-through separated and the window stated.
• Unmatched revenue, stated, rather than spread across channels; see unattributed revenue.
• Lift, if tested, with the test dates and the confidence interval, as a separate line.

If the honest report says retargeting is mostly a reminder service for people your other channels already found, that is useful, not a failure. It suggests a smaller budget, a tighter audience (recent visitors who did not leave contact details, not everyone who ever loaded a page) and a test before any increase. And it moves credit back to the channels that are actually starting conversations, which is where the next dollar of growth usually comes from.

Retargeting is a reminder, and reminders are worth something. Pay for it on the leads it recorded first and the sales those leads closed, test it if it is large, and never let it inherit the customers your other channels found.