Does retargeting actually sell anything? What a Journal of Interactive Marketing study found
A mattress company let two researchers move $5.7 million of its ad budget around 210 American media markets, then counted the orders. Chasing people who had already visited the site produced no measurable sales at all.
Facebook reported a return of 7.47 on retargeting. A field experiment across 210 markets found the retargeting budget could be cut to zero without losing a single order.
The audience that actually moved the numbers: people who had never been to the site. Photo: Ryoji Iwata on Unsplash.
The short answer
In the only experiment of its kind, no. Two researchers took over the display budget of a large mattress company and randomly varied spending across 210 American media markets for two months, $5.7 million in all. Ads shown to people who had never visited the site moved orders reliably on both Google and Facebook. Ads shown to previous visitors moved nothing, at any budget level, not even going from nothing to half the usual spend. The platforms, meanwhile, were reporting retargeting as the best money the company spent.
Key takeaways
Markets where Google retargeting was switched off entirely grew orders by 12.6%. Markets kept at the full budget grew 13.8%.
Prospecting behaved as advertising should: Facebook markets at zero grew 6.3%, markets at full spend grew 16.9%.
Someone who clicked a prospecting ad stayed in the retargeting pool for 45 days and took around two weeks to buy. The last ad they clicked got the credit.
At a 50% contribution margin the model puts the right Facebook prospecting budget at $2.46 million. The company was spending $3.53 million.
What the researchers did
Almost nobody gets to run this experiment, because almost nobody will let a researcher switch off a working ad
channel. Jeffrey Larson and Jeffrey Dotson, both at Brigham Young University, persuaded a large
direct-to-consumer mattress company to let them do exactly that.
The setup is the kind of thing an agency talks about and never builds. They split the United States by Nielsen
media market, 210 of them, and assigned each market to one of 19 budget recipes. A recipe set the spend
separately for four things: Google ads to strangers, Google ads to previous site visitors, Facebook ads to
strangers, Facebook ads to previous visitors. Some markets got nothing in a channel, some got half the usual
budget, some the usual amount, some half again as much. Then everything ran untouched for nearly two months
while the orders came in.
In the industry's language, ads to strangers are prospecting and ads to previous visitors are retargeting. The
company, like the three agency heads the authors interviewed, was certain retargeting was the better buy. It
was certain enough that it refused to let the researchers raise the prospecting budget above normal, on the
grounds that spending more there would dilute the targeting.
Design
19 budget recipes assigned across 210 US media markets, about 10 markets per recipe
Budget moved
$5.7 million of display spend over nearly two months, against roughly $58 million of sales
Four channels
Google and Facebook, each split in two: ads to strangers, ads to people who had already been to the site
Levels tested
Nothing, half, normal, and half again, so saturation could be seen as well as lift
Category
Direct-to-consumer mattresses, average online order around $1,000
What the platforms were reporting
Start with the numbers the company was actually managing to. During the experiment itself, the two platforms
reported their own returns on every dollar: Facebook claimed its retargeting ads were returning more than
three times what its prospecting ads returned, and Google reported retargeting ahead too.
Those figures had shaped years of decisions. They are also the reason the industry treats the combination of
prospecting and retargeting as a one-two punch, with retargeting doing the finishing. The belief is not
unreasonable on its face. Someone who has been to your website has told you something real about their
interest, and nobody in the prospecting pool has told you anything at all.
There was a quieter warning sign in the data, and the company had not read it. In the markets told to spend
half again as much on retargeting, the platforms could not actually spend the money. The pool of previous
visitors simply did not generate enough impressions to absorb it. A budget that cannot be spent is a budget
that was already too big.
What each platform reported it returned, per dollar spent, during the experiment
What each platform reported it returned, per dollar spent, during the experiment
Group
Value (x)
Google, ads to strangers
0.99
Google, ads to past visitors
1.27
Facebook, ads to strangers
2.21
Facebook, ads to past visitors
7.47
The platforms' own attribution during the experimental period. Facebook retargeting spent $1.5m, Google retargeting $1.2m.
What happened to the orders
Now the part the platforms cannot see. Because budgets were set by geography and the company had never run
regional campaigns before, the markets work as controls for each other, and the order counts can simply be
compared.
On prospecting, more money meant more orders, cleanly and in both channels. On retargeting, nothing. Not a
smaller effect, nothing measurable at all, in either channel, at any level.
The detail that kills the obvious objection is the step from nothing to half. If retargeting were merely
oversold, if the company were buying the tenth impression when the second would have done, then cutting the
budget in half should still have hurt less than cutting it entirely, and the markets at half spend should sit
somewhere in between. They do not. Going from no retargeting at all to half the usual retargeting produced no
improvement that came anywhere near significance.
Two further results are worth having. Advertising in no channel changed how much people spent once they
decided to buy, only whether they bought. And the experiment was built to detect synergy between prospecting
and retargeting, the idea that prospecting fills the pool that retargeting then converts. No such interaction
showed up, though the authors are careful to say the test was underpowered even at this budget.
Measured effect of each channel on order volume (advertising elasticity)
Measured effect of each channel on order volume (advertising elasticity)
Group
Value ()
Google, ads to strangers
0.153
Facebook, ads to strangers
0.091
Facebook, past visitors (not detectable)
0.033
Google, past visitors (not detectable)
0.008
Model 1, aggregated to the 19 experimental conditions. The two prospecting figures were significant at p = .001; neither retargeting figure came close.
Why the dashboard was wrong
The authors' explanation is mechanical, and once you see it you cannot unsee it. Someone sees a prospecting
ad, clicks, lands on the website, and does not buy, because nobody buys a mattress in an afternoon. For the
next fortnight they are shown retargeting ads constantly, and the ads work the way a shopfront works: they
recognise the product they are already thinking about, so sooner or later they click one. Then they buy. The
platform looks at that final click and awards itself the sale.
Retargeting, on this reading, is not causing purchases. It is standing next to them. And the longer the
purchase cycle, the more reliably it gets to stand there, which means the bias is worst in precisely the
categories where the stakes are highest.
There is a second reason, less about measurement and more about what an ad is for. If advertising works mostly
by telling people things they did not know, then the retargeting pool is the worst audience you have, because
they already went to your website and read it. The prospecting pool is the only group left who can still learn
something. That logic also puts a question mark over any channel that targets people the platform has already
judged likely to buy. The authors name Google's Performance Max as the obvious candidate for the same
flattering arithmetic.
What it costs, and what to do
Turn the elasticities into money and the picture is uncomfortable for the budget as it stood. One extra order
cost far more through Facebook prospecting than through Google prospecting, and during the experiment Google
prospecting was the smallest of the four budgets at $0.76 million.
Run the same arithmetic on the other side and it says Google prospecting, which had been running at $1.79
million before the experiment, should have been somewhere near $4.14 million. More than double, into the
channel that was getting the least.
The authors are realistic about most firms not being able to run a test like this, so their fallback
recommendation is a single setting: frequency caps. Cap the retargeting and the spend falls without the
returns falling, since the returns were not there. Cap the prospecting and the same budget reaches more people
instead of the same people more often, which is what you want if information rather than persuasion is doing
the work.
One finding that the authors flag and I would underline: Google prospecting ads also lifted the company's
Amazon sales, where it had no ads at all and sold only its cheap accessories. People saw an ad, went looking,
and bought where they already had an account. Nothing in any dashboard would ever have shown that.
Extra advertising spend needed to produce one more order, in dollars
Extra advertising spend needed to produce one more order, in dollars
Group
Value ($)
Google, ads to strangers
216
Facebook, ads to strangers
717
Applied to the pre-experiment spending levels. Average online order value was about $1,000.
Marketing in 3 steps
Ready. Set. Market.
1
ReadyDecide and prepare
Write down the two numbers. Open your ad accounts and find what you spent last month on people who have already visited your site, and what you spent on people who have not. Write the two numbers side by side. Most teams have never seen them that way.
Check how long people take to buy. Find the gap between a first visit and a purchase. If it is longer than a few days, the ads shown during that gap are collecting credit they did not earn.
2
SetBuild it
Cap how often you chase people. Set a frequency cap on your retargeting: three or four times a week per person, no more. Make the change this week. Your spend will drop straight away.
Move that money to strangers. Put everything the cap saves into reaching people who have never been to your site. Move it in one go. Dripping it across both audiences tells you nothing.
3
MarketLaunch and measure
Switch it off in a few cities. Pick five cities. Turn retargeting off completely there for a month and change nothing else. Leave five similar cities running as they are.
Count orders, not return on spend. Add up total orders in the five test cities and compare them with the five you left alone. Ignore the return-on-spend figure in your dashboard, because that is the number you are testing.
What to expect
A smaller bill before you see anything else. Capping retargeting cuts spend immediately. The orders are the slow part, so give it a month before judging.
Your reports will look worse while your business does not. Credit moves away from the channel that was claiming it, so reported returns fall even if total orders hold or rise.
Prospecting pays off on a delay. Here the typical gap between first visit and purchase was a fortnight, so a week of data will show you the cost and none of the benefit.
It changes who buys, not how much they spend. No channel in this experiment shifted order value, so do not expect a bigger basket out of a budget move.
Where it does not work
One advertiser, one category. A high-priced mattress bought after weeks of thinking. Cheap, impulsive, repeat purchases may behave differently, and the authors say so.
Collected in 2019. Before Apple's tracking prompt and the wider privacy changes, which have since made retargeting pools smaller and blunter rather than better.
The budget only went down. The client refused to let prospecting rise above normal, so the case for spending much more there is an extrapolation from the measured elasticity, not a tested result.
No synergy found is not no synergy. The test for prospecting feeding retargeting had standard errors wide enough that only an enormous effect would have registered.
Questions marketers ask
Does retargeting work?
In the one large field experiment that has measured it against sales rather than clicks, no. Turning retargeting off entirely in a randomly assigned set of American media markets left order volumes indistinguishable from markets running it at full budget (Larson & Dotson, 2026, Journal of Interactive Marketing). That is one advertiser in one category, so it is a warning rather than a verdict.
Why does my dashboard show such a high return on retargeting?
Because a retargeting ad is usually the last ad a buyer clicks, and the platform credits the last click. Someone who visited your site weeks ago and is already going to buy will see and click your retargeting ads in between, which makes the channel look decisive when it is merely present.
Should I stop retargeting completely?
Nothing here shows retargeting does harm, only that it was not adding sales in this case, and it was absorbing a budget that could not even be fully spent. The proportionate move is an aggressive frequency cap rather than a shutdown, with the saved money redirected and the result measured.
Is prospecting on Google or Facebook better value?
In this experiment Google prospecting had both the stronger effect on orders and by far the smallest budget, which is the combination that makes a channel underfunded. Facebook prospecting worked too, but each extra order cost roughly three times as much.
How can I test this on my own account without a research team?
Use geography. Switch the channel off in a handful of similar cities, leave comparable cities untouched, hold everything else constant for a month, and compare total orders rather than platform-reported returns. That is the whole method, and it needs no modelling.
A modelling paper on what happens when a marketplace bids for sponsored slots automatically, adjusting each bid to how likely a given shopper looks to buy. Sellers respond by raising their prices, competition for placements softens, and conversion rates can fall even as the targeting gets sharper. The platform tends to lose ad revenue and gain commission, so whether it comes out ahead depends on its commission rate. Sharper targeting is not the same as a healthier marketplace.
Ads you can buy from directly surprise people, and the surprise cuts both ways: interviews turned up praise for convenience and information alongside complaints about intrusion. Across streaming television and social media the format read as more unexpected and more negative overall, yet it landed better on streaming than in a social feed. Where the viewer came looking for information, the reaction turned positive.
Source
Larson, J. S., & Dotson, J. P. (2026). Prospecting Versus Retargeting: Insights from a Geography-Based Field Experiment. Journal of Interactive Marketing, 61(3), 257-274. https://doi.org/10.1177/10949968261422740
This brief is written for practitioners from the published article. Numbers are as reported by the authors; read the article for full methods.
Shooters are the favourite of 36% of console players and 17% of phone players; strategy games 35% on laptops and 4% on consoles. How to read the gaming channel, from 116,047 players.
A meta-analysis of 1,536 results from 290 studies finds personalization lifts responses with almost no backlash. Moderate levels do worst.
Bruno Schivinski is Associate Professor of Marketing at Gdańsk University of Technology. He studies how consumers engage with brands in digital and social media.
