The basket shrank first. Then Meta's return followed.

A European bedding and sleepwear brand switched Releva off for a test, in stages. Its own orders and ad account show what the agent had been doing.

Book 30 minutes All case studies

Bedding and sleepwear · WooCommerce · 67,000 products
Switched off in stages, June to July 2026
Store orders and the brand's Meta account

After the switch-offtest

Items per order

−10%

within two weeks

Meta ROAS

−34%

six weeks later

CPM and CTRunchanged

Average item priceunchanged

−10%items per order within two weeks, at unchanged item prices

+51%larger orders with a recommended product than from an ad alone

−34%Meta ROAS six weeks later, with CPM and CTR unchanged

16 dayshalf-life of the effect: it faded, it did not stop

The client

A catalog of 67,000 products, sold mostly through Meta

The brand sells bedding, pajamas, blankets and sheets on WooCommerce, with Meta as its main acquisition channel. Releva ran the recommendations on the site, the abandoned-cart and abandoned-browse emails, and the audience and catalog sync into Meta.

In mid-2026 the brand switched Releva off for a test, in stages. That made its effect measurable in the store's own orders.

What happened, and when

Switched off in three steps

26 June: bulk email stops

Bulk campaigns stop. The automations with personal recommendations keep running.

30 June: recommendations cut back

Recommendation impressions fall 37% over two weeks. Banners on the same pages run normally, up 8.7%, so the site itself works.

15 July: the script is removed

Every personal function stops, on the site and in the inbox.

First effect

Shoppers put fewer products in the basket

The last two weeks of full service against two weeks of fewer recommendations.

Store data	15 to 29 June	30 June to 15 July	Change
Recommendation impressions per day	38,020	23,920	−37.1%
Orders	2,973	3,154	+6.1%
Items per order	1.93	1.74	−10.2%
Average item price, indexed	100	99	−0.9%
Average order value, indexed	100	89	−11.0%

Fewer items per order explains 93% of the fall in average order value. Repeat customers, who got recommendations on the site and in email, lost 9.9% of their basket. New customers lost 2.4%.

What a recommended product is worth

Orders with a recommended product were 51% larger than orders from an ad alone

In June 2026, orders with a product from a recommendation were 31% larger than orders with no touch, and 51% larger than orders that came from an ad alone.

As recommendation coverage rose from 54% of sessions in April to 90% in June, the average order rose 16%.

June orders · average order valueorders

With a recommended productindex 151

No touchindex 115

From an ad onlyindex 100

Second effect

Meta spent 7% more for a third less return

CPM and CTR describe what Meta does before the click. Both held, so the loss sits after the click.

Meta account	Change, 1 to 15 June against 1 to 2 August
Purchase ROAS	−33.7%
Cost per purchase	+30.0%
Purchases per day	−17.9%
Ad spend per day	+6.7%
CPM	−1.4%
CTR	+2.9%

By early September, Meta-attributed sales per day were down 41% against June on the same budget, and a purchase cost 70% more.

Why it was not the season

The brand's summer dip comes every year. In 2025 it began on 16 June. In 2026 the account held its June level until 15 July, the day the script was removed.

In 2025, return on ad spend rose 13% into September. In 2026 it fell 9%, at 2.5 times the daily spend of the year before.

Why it faded rather than stopped

Shoppers who had seen recommendations kept coming back, less and less. Measured on the real visits of 60,000 users, the effect halved every 15.9 days, like a TV campaign after it stops. The cost was largest a month after the switch-off.

How we counted

Every figure is a completed order joined to what the shopper did, or a relative change inside one ad account. The account counted 2.11 purchase events per real order, so absolute ROAS is not shown; the relative changes are unaffected. This is a before-and-after test, not a holdout.

Get the method and setup notes

How each number on this page was counted, the definitions, and what was switched on in the store, in order. Enter your work email and the notes open; we also send you the link.

Thanks. Open the method and setup notes.

We use your email to know who reads our case studies, and to send the link. Privacy policy