Written by: Nicole Ogonowska, IT Growth Manager, Digital Colliers
In July 2026 the European Commission issued a preliminary finding that Facebook and Instagram breach the Digital Services Act on addictive design. Autoplay and infinite scroll are named. If the finding holds, Meta either changes the surface where your ads live or pays DSA fines that scale with global turnover. Either outcome moves the price of attention. Your paid social forecast for the back half of 2026 probably assumes neither happens.
What actually changes if the feed changes
Start with the mechanic. Autoplay and infinite scroll are the reason a user sees ten ads in a session instead of two. Kill either one and impressions per session drop. Meta will not accept lower revenue per user quietly, so CPMs rise to defend inventory. CPMs for DTC advertisers were already climbing through 2024 and 2025, and customer acquisition cost across DTC brands is up roughly 40% since 2023. A forced UX change on top of that is not a small edit to the model. It is a reprice.
The knock-on for eCommerce operators is uncomfortable but predictable:
- Prospecting campaigns get more expensive first, because cold audiences depend on volume of impressions.
- Retargeting holds up longer, because it depends on intent not scroll depth.
- Creative fatigue accelerates, because each impression has to work harder.
- Attribution windows look worse, because the funnel gets shorter and choppier.
None of that is speculative. It is just what happens when you shrink the number of ads a user sees per session and keep advertiser demand constant.
The forecast most teams are running is already wrong
If your 2026 plan pencils in flat or improving blended CAC on Meta, ask what it assumes about the feed. Most media plans I see assume the surface is a constant. It is not. The DSA case is one input. The EU AI Act's Article 50 transparency obligations kick in on 2 August 2026 and touch generative ad creative. GDPR fines still reach up to €20M or 4% of global turnover, which shapes what data Meta will and won't let you use for lookalikes. Every one of these is a small squeeze on the machine you rent from Meta to find customers.
Meanwhile the demand side is soft. UK eCommerce grew around 3% in 2024 versus 2023, and single-digit growth is the new baseline. You cannot outgrow a CAC problem in a 3% market. You have to fix the unit economics, and roughly 30% of SKUs at a typical multi-channel brand already lose money per order once returns and ad spend are counted.
Own the model, don't rent it
The operators who look calm in this cycle are the ones who stopped treating Meta as the customer database. They treat it as a distribution channel with a shrinking half-life. Practically, that means a few things:
- First-party data as the source of truth. Every order, every session, every return, joined to a customer ID you own. Not a Meta pixel event, not a GA session, a row in your warehouse.
- Predicted LTV that runs on your data, not the platform's. If Meta is optimising bids on a 7-day purchase signal and you have a 12-month LTV curve, you bid smarter than the auction.
- Retention economics that don't need paid social to work. Email, SMS, post-purchase, product bundling, returns reduction. Boring, and the reason some brands survive a CPM spike.
- Channel diversification that is real, not a slide. Search, retail media, affiliate, organic. Each with its own creative pipeline.
The throughline is that your model of the customer lives on infrastructure you control. When Meta changes the feed, or the DSA forces the change, or the AI Act narrows what generative creative can claim, your acquisition math still works because the inputs are yours.
What to do this quarter
Run the forecast twice. Once with today's Meta CPMs and once with a 20 to 30% CPM increase and a shorter attribution window. If the second version breaks the P&L, you have a data problem, not a media problem. The teams that ship through 2026 are the ones treating the DSA finding as a scheduled event, not a surprise. The rest will find out in the quarterly review.