Four platform changes and one regulation that quietly rewrote what your numbers mean.
Welcome to EcomifyAI. This is the operator's filter on ecommerce AI — five things a week that change what you should do, and nothing that only changes what you should think about. No back-issues to reference yet, so here is the standard: if we cannot link it to a primary source, we do not run it.
The Signal
From 17 August, Google changed how target-based bidding handles budget-constrained campaigns. If a tCPA or tROAS campaign was capped by budget, it used to skim the cheapest high-intent auctions and beat its stated target — a €10 tCPA quietly delivering €5. That is over. The system now optimises to the number you typed, and Google has confirmed it will not adjust targets on your behalf. The Bid Target Adjustment Tool is live in-account.
Most operators will read the coming dip as efficiency loss. It isn't. That outperformance was an artefact of throttling, and it has been sitting inside every benchmark you own. The planning damage is worse than the media damage: any Q4 forecast built on last year's realised ROAS from a budget-capped campaign is overstated, because that number can no longer be reproduced at scale.
Pull realised versus target CPA and ROAS for the last 90 days on every budget-limited campaign. Where realised beat target materially, reset the target to realised — not to the number you wish were true — and re-baseline Q4 before you commit peak spend.
Worth knowing
AI Max migration runs all September, not on the 1st.
Search campaigns using campaign-level broad match or standalone automatically created assets migrate to AI Max between 1 and 30 September; Google blocked new campaigns of that shape on 3 August. The rolling window is the problem — an unknown migration date will contaminate your before-and-after. More useful: AI Max experiments now run with brand and location guardrails enabled (20 August). Start one now, while you still control the baseline.
Merchant Center opened the PMax box at SKU level.
Since 24 August, product-level reporting spans all ad channels and formats, and YouTube affiliate results are split out from organic. The headline is transparency; the value is margin. ROAS-optimised Performance Max has been buying volume on your worst-margin SKUs, invisibly. Export product-level data, join it to unit margin, and build your first custom-label split from what looks efficient but isn't profitable.
Shopify gave every Liquid storefront an agent API — by default.
On 5 August, Shopify exposed WebMCP tools across Liquid and Hydrogen storefronts: agents can call search_catalog, update_cart and proceed_to_checkout directly. Live on every Liquid store, nothing to install, no opt-in. Discovery moved to your product data. If your variant is "Midnight" with no colour-family metafield, an agent asked for a navy shirt returns nothing — and you never see the loss, because there was no session.
The EU's AI transparency duties are live and they reach product creative.
Since 2 August, Article 50 places disclosure obligations on deployers, not only model providers, for synthetic content resembling real persons, objects, places or events. Everyone read "deepfakes" as political video; for ecommerce the live question is AI product renders and synthetic UGC. The "standard editing" carve-out is written for providers, not for you. Worth a specific question to your advisor — this is published regulation, not legal advice, and its ecommerce application is unsettled.
One thing to do this week
Audit your budget-limited campaigns for structural over-performance and reset the targets to what they actually delivered. Do it before you set Q4 budgets, not after — every downstream number depends on a baseline that just changed underneath you.
Thanks for reading issue one. Reply and tell me what you're seeing in your own accounts — it makes next week's issue sharper.
P.S. We're building an Agent-Readable Catalogue Audit off the back of that Shopify change — a scoring rubric for whether an AI agent can actually find and buy your products.