Geo-holdout studies across more than fifty brands found median retargeting incrementality of about 28%. Roughly seventy cents of every retargeting dollar funds purchases that were happening anyway. We still think you should run it — we just think you should know which seventy cents.
Retargeting, also called remarketing, is advertising shown specifically to people who have already interacted with your business — visited your website, viewed a product, abandoned a cart or opened an email. A tracking pixel records the visit, adds that person to an audience, and ads follow them across other websites, social platforms and connected TV until they convert or the audience window expires.
Retargeting companies sell it as the best-performing channel in almost every advertising account, and by every metric the platforms report, it is. Retargeting display converts at roughly double the rate of standard display. Cost per acquisition is usually the lowest in the account. Return on ad spend frequently looks spectacular.
And most of that is an illusion. Not fraud, not incompetence — an artefact of measurement. Retargeting audiences are, by definition, people who already showed interest. They visited your site. They looked at the product. Many of them were going to come back regardless. When one of them does come back and buy, the retargeting platform records a conversion and takes the credit, because the ad was the last thing it saw.
That is the single most important thing to understand about this channel, and it is why a remarketing agency page opens by telling you the numbers are inflated. We would rather have that conversation now than in month four.
A holdout test is simple enough to explain in a sentence. Take your retargeting audience, deliberately withhold ads from a random slice of it, and compare what the two groups do. The difference is the only thing your advertising actually caused. Everything above that line was going to happen anyway.
Press run and watch it play out on 1,000 returning visitors, with a 10% holdout.
A demonstration of the method using illustrative figures in the range published research reports for retargeting. Your own numbers will differ — the point is the shape of the result, and that you cannot know your shape without running the test.
Until you run a holdout you do not know your incrementality rate. But you can see what different rates would mean, and published research narrows the range considerably: analyses of retargeting campaigns commonly find incrementality between 20% and 40%, with one set of geo-holdout studies across more than fifty brands reporting a median near 28%.
Set your own numbers and an assumed incrementality rate below. The left column is what your dashboard says. The right is what is probably true.
Incrementality ranges drawn from published 2026 analyses including Recast geo-holdout studies, Northbeam research and Haus lift experiments. The 28% default is a reported median, not a prediction about your account. The only way to know yours is to measure it.
An incremental ROAS of two is a good business. It is simply a very different business from an incremental ROAS of eight, and every decision downstream depends on knowing which one you have. Budget allocation, bid caps, how hard you scale, whether to shift money toward prospecting — all of it changes.
The failure mode is not running retargeting. It is running retargeting on reported numbers, concluding it is your best channel because the dashboard says so, and moving budget out of the channels that actually create demand and into the one that harvests it. Research on this is consistent: prospecting campaigns typically show incrementality of 60 to 80%, roughly double retargeting's. The channel that looks worse on your dashboard is frequently doing more of the real work.
This is the question that should be asked before anyone builds a campaign, and it usually is not. Retargeting needs a pool large enough to deliver against, and the pool is much smaller than raw traffic suggests. Ad blocking removes some. Cookie restrictions and opt-outs remove more. And your audience window quietly expires people every single day.
A planning model. It applies your untrackable share to visitors accruing over the window you set, which approximates how audience membership actually behaves. Real pool sizes depend on returning-visitor overlap, platform-specific match rates and how aggressively you exclude converters.
| Window | What it means | Use when |
|---|---|---|
| 1–7 days | Smallest pool, highest intent. These people are still actively considering. | Short purchase cycles, cart abandonment, high traffic volume |
| 14–30 days | The usual default. Reasonable balance of size and intent. | Most businesses, most of the time |
| 60–90 days | Larger pool, considerably weaker intent. Frequency falls as the pool grows. | Considered purchases, long sales cycles, low traffic |
| 180 days | Mostly people who have forgotten you exist. Cheap impressions, thin returns. | Rarely. Usually a symptom of not having enough traffic |
If you find yourself stretching to 180 days to fill an audience, the honest answer is that you do not have a retargeting problem — you have a traffic problem, and the money belongs in paid search or social until you do.
Setting up a retargeting campaign takes about twenty minutes. Everything below is what separates a retargeting agency whose work compounds from retargeting companies that quietly bill you for demand you already had.
A permanent control group so incrementality is known rather than assumed. This is the whole argument and almost nobody runs it.
Removing people who already bought. Astonishingly common to find accounts advertising to existing customers at full price.
A cart abandoner and a blog reader are not the same person and should not see the same ad at the same bid.
Past a certain point additional impressions stop persuading and start irritating, while still costing full price.
Matching the audience window to your actual sales cycle rather than accepting a platform default.
Pixel, server-side events and match quality. Covered in depth under Meta Pixel setup.
Showing a different message at impression twelve than at impression one, because the same ad forty times persuades nobody.
Retargeting runs across display, social, search and connected TV from one audience strategy.
Sending someone back to a generic homepage wastes the click. See conversion optimization.
Send us your retargeting numbers. We will tell you what your reported ROAS is likely worth, and design a holdout test you can run yourself in about an hour.
Get my reviewEverybody has experienced badly run retargeting. You looked at a pair of boots once and they followed you for six weeks across every website you visited, long after you bought them somewhere else. That experience is what the phrase "creepy advertising" describes, and it is not caused by the technology. It is caused by three settings nobody adjusted.
| What went wrong | The fix |
|---|---|
| No frequency cap | Limit impressions per person per week. There is a point past which additional exposures persuade nobody and simply accumulate resentment. |
| No converter exclusion | Remove people the moment they buy. Following a customer with ads for the thing they just purchased is the most visible possible sign that nobody is minding the account. |
| Window far too long | Someone who visited five months ago and never returned has made their decision. Let them go. |
| One creative, forever | The same image forty times reads as automation rather than communication. Sequence it or cap it. |
This matters commercially, not just ethically. Reported conversions cannot see brand damage, so an account can be simultaneously producing an excellent ROAS and steadily annoying a portion of your market. It is one more reason the reported number should not be the only number you look at.
Incrementality varies enormously by how long the buying decision takes. The longer and more considered the purchase, the more genuine work retargeting does.
Long cycles and several stakeholders. Among the highest genuine incrementality anywhere.
💻SaaSTrial abandonment and feature-page sequencing. Real work, not just credit-claiming.
⚖Law FirmsHigh case value and slow decisions. Careful with sensitive categories and audience rules.
🏠Real EstateMonths of consideration. Listing-level sequencing outperforms generic brand ads.
🛒EcommerceHighest reported ROAS and lowest true incrementality. Test before you scale.
🏥HealthcareRestricted categories limit what you can build audiences on. Check rules first.
🦷DentalWorks for high-ticket treatment plans. Rarely worth it for routine appointments.
🍽RestaurantsShort windows and local radius only. Long-window retargeting makes no sense here.
Retargeting, also called remarketing, shows ads specifically to people who have already interacted with your business — visited the site, viewed a product, abandoned a cart or opened an email. A tracking pixel records the visit and adds that person to an audience, and ads then follow them across other websites, social platforms and connected TV until they convert or the audience window expires. It consistently reports the best performance in an advertising account, partly because it genuinely works and partly because it targets people who already intended to buy.
In everyday use, nothing. The two terms are interchangeable and both describe advertising to people who have already engaged with you. Historically remarketing was Google's term for the practice inside its own platform while retargeting was the wider industry word, and some people still use remarketing to mean email-based follow-up specifically. If a remarketing agency draws a hard distinction between them, ask what they mean by it, because there is no settled definition.
Because retargeting audiences are people who already showed intent, and a meaningful share of them would have converted with or without the ad. The platform records the conversion and takes credit because its ad was the last touchpoint. Published geo-holdout research across more than fifty brands reported median retargeting incrementality near 28%, and analyses commonly place it between 20% and 40%. Campaigns showing 4 to 8 times measured ROAS frequently show incremental ROAS closer to 1 to 1.5 once the audience's existing intent is accounted for.
A holdout test deliberately withholds advertising from a random slice of your retargeting audience and compares what that group does against the group that saw ads. The difference is the conversions your advertising actually caused. Meta's Conversion Lift and Google's ghost ads both support this natively and can be configured in well under an hour, and geo-based holdouts work across any channel. Published guidance suggests running at least four weeks and needing a control group generating a couple of hundred conversions before the result is meaningful. The animation on this page shows the method in operation.
Almost certainly not. An incremental ROAS of two is a good business — it is simply a very different business from an incremental ROAS of eight, and every decision downstream depends on knowing which one you have. The real failure is running retargeting on reported numbers, concluding it is your best channel, and shifting budget out of the channels that create demand into the one that harvests it. Prospecting campaigns typically show incrementality of 60 to 80%, roughly double retargeting's, which means the channel that looks worse on your dashboard is often doing more of the actual work.
Enough to build an audience the platforms will actually deliver against, which is smaller than raw traffic suggests once ad blocking, cookie restrictions and opt-outs are accounted for. Many platforms require a minimum audience size before a campaign will serve at all. The audience pool calculator on this page estimates yours. If you find yourself stretching the window to 180 days simply to fill an audience, you do not have a retargeting problem, you have a traffic problem, and the budget belongs in paid search or social until that changes.
Match it to your actual sales cycle rather than the platform default. One to seven days gives the smallest pool and the highest intent, suiting cart abandonment and short purchase cycles. Fourteen to thirty days is the sensible default for most businesses. Sixty to ninety suits considered purchases and long cycles at the cost of much weaker intent. A 180-day window mostly reaches people who have forgotten you exist and is usually a symptom of insufficient traffic rather than a deliberate strategy.
Because three settings were not adjusted. No frequency cap means the same person sees the same ad dozens of times. No converter exclusion means customers who already bought keep being advertised to. And an overly long window means people who decided months ago are still being followed. None of that is inherent to the technology and all of it is fixable in an afternoon. It matters commercially as well as ethically, since reported conversions cannot see brand damage — an account can produce an excellent ROAS while steadily irritating part of your market.
Effectively all of them, from one audience strategy. Google Ads covers search and the display network, Meta covers Facebook and Instagram, programmatic platforms reach the open web, and connected TV can serve an ad on the television in a household that visited your site. Running them from a single segmented audience strategy rather than as separate campaigns is what prevents the same person being hit from four directions at once.
Most retargeting companies charge a monthly management fee separate from ad spend. We price it to the work rather than as a percentage of budget — partly because a percentage fee gives an agency a reason to keep retargeting spend high, which is the opposite of what this page argues for. Retargeting is also rarely worth running as a standalone engagement, since it depends entirely on other channels supplying the audience. Call +1 720-712-8615 and we will tell you whether it makes sense on its own for you, or whether the budget belongs upstream first.
Send your retargeting numbers. We will tell you what they are likely worth and design a holdout you can run yourself.
Your information is never sold or shared. We respond within one business day.