We Paid Meta to Reach Our Own Customers. Haus Says It Worked.

· Paid Social

We Paid Meta to Reach Our Own Customers. Haus Says It Worked.

We ran a Haus geo holdout on a retention campaign for a brand in September. Three weeks, half the country held out, about £80,000 of spend, all of it aimed at customers who hadn't bought in 60 days.

Most accounts we run put real effort into keeping existing customers out of the ads. Purchase exclusions, Klaviyo lists, subscriber lists, the account-level existing customer definition. This brand wanted to know what happens if you pay to reach the ones who've gone quiet.

Here's what came back.

One campaign, one ad set, one number to beat

The campaign targeted lapsed customers in the UK. People who had bought before but not in the last 60 days, and who weren't active subscribers. Recent purchasers were excluded via pixel and Klaviyo, and the active subscriber list was excluded on top.

The targeting held. Across the test window, 95% of the campaign's spend went to Meta's existing customer segment. Under 3% reached people Meta classes as new to the brand.

Before launch we agreed the bar. The brand needed an incremental ROAS of 1.5 on returning customer sales to call this worth doing. That's close to breakeven for them. Clear it and retention becomes a new, fully incremental line of spend in the account.

We also agreed what happens if it failed. Retest with smaller budgets and a longer exclusion window, 120 days instead of 60.

The test ran 4 to 23 September 2026, with a week of post-treatment measurement after.

1.93 against a 1.5 target

On the primary KPI, returning customer sales in the online store:

  • Incremental ROAS was 1.93.

  • Returning customer sales rose 4.17% in the treatment regions.

  • That's around £143,000 of returning customer revenue that wouldn't have happened without the ads.

  • It came to 4,929 incremental repeat orders at about £15 each.

It passed. Comfortably.

Total store revenue matters more than the returning line on its own. Add the new customer lift (more on that below) and total iROAS was 2.45, with about £182,000 of incremental revenue across both groups.

Meta said 6.2x. Haus said 2.45.

During the test window Meta reported a 6.2x ROAS on the campaign, 7-day click 1-day engaged view. It claimed around 15,800 purchases.

Haus measured about 6,300 incremental orders across new and returning customers. 40% of what Meta took credit for.

No surprise there. These are people who already buy from the brand. A chunk of them were coming back this month anyway, and Meta saw them on the way in. Platform attribution can't separate "reminded them" from "was in the room when they bought". A holdout can.

What I'd take from it: a retention campaign showing 6x in Ads Manager is not a 6x campaign. Here it was still good enough. If Meta had said 3x, the real number could easily have been under 1.5.

https://x.com/PhilKiel/status/2030000209800147079

Ads aimed at old customers brought in new ones

New customer sales rose 2.28% in the treatment regions. 1,344 incremental first orders, at about £55 each.

On new customer revenue alone, iROAS was 0.52. So nobody should run this as a prospecting campaign. It still moved the number.

Some of that is the 3% of spend that reached people Meta classes as new. Some of it is what you'd expect from a retention brand with a loyal base. A returning customer buys again, mentions it to someone in the house, a partner or friend orders their own. We can't see that path in the data. The holdout can only tell us it happened.

The Amazon halo, for anyone wondering, was nothing. A 0.15% lift. Retention spend on Meta pulled people back to the website, not to the marketplace.

What we're watching now that it's scaled

The test ran at roughly £3,700 a day. Since it finished, the brand has added retention ad sets for two more products and the campaign is spending several times that.

Fair decision off the back of a passed test. It also makes the 1.93 a historical number. Incremental ROAS was measured at one level of spend, and it won't hold at four times the budget. The pool of lapsed customers in the UK is finite.

You can see the pressure inside the test window. As weekly spend ramped from about £18,000 to £35,000, Meta's reported ROAS fell from 8.6x to 4.8x. Since scaling it has drifted down again, and the latest week is under 4x. Apply the same 40% discount and you're close to the 1.5 line.

So the plan is to retest at the new spend level rather than assume the old result carries over. CPMr and frequency on the retention ad sets are what we check weekly. A small pool shows saturation fast.

How to run this test in your own account

If you've been excluding all customers from everything, this is worth a proper look. The structure we'd use again:

  1. Define "lapsed" for your purchase cycle. We used 60 days since last purchase for a monthly-ish retention product. A longer cycle product needs a longer window.

  2. Exclude recent purchasers and active subscribers, using both pixel and CRM lists. Subscribers are already coming back, so paying to reach them buys revenue you'd get anyway.

  3. Agree the iROAS threshold before the test starts, and what you'll do if it fails. Picking the bar after you've seen the result isn't a test.

  4. Make returning customer sales the primary KPI, but read new customer and Amazon sales too. The side effects were a quarter of the total lift here.

  5. Ramp budget slowly and check the result holds at the spend you plan to run, not just the spend you tested.

Excluding customers is still the right default for prospecting. It just doesn't settle what a lapsed customer is worth to reach. Run the holdout before you decide the answer is zero.


Read more about CPMr: https://taikundigital.com/phil-kiel/cpmr-calculation


Stay up to date or book a call: https://taikundigital.com/linkbios/phil


Brand anonymised. Test: Haus geo holdout, 50% holdout, Meta only, 4 to 23 September 2026 with a post-treatment window to 30 September. Primary KPI: returning customer online store sales. Platform figures are Meta-reported, 7-day click 1-day engaged view, for the same dates. Post-test spend and ROAS from Meta data to 12 October 2026.


Back to all articles