As below, consolidating my MCP submission into 1 instead of 2, each solving a seperate problem 👇 1️⃣ "Liam", my AI media buyer that runs our Meta account What it does: Claude Code + the HYROS MCP, working like a full-time media buyer who never forgets: it reads HYROS itself, makes the kill / keep / scale calls, and logs why. The problem: I was the media buyer. Every ad read meant pulling reports, judging on one attribution model, and forgetting why we'd made a call three weeks earlier. So sometimes I relearned the same lessons and held losing ads too long! How I solved it: connected Claude Code to the HYROS MCP and gave it a memory. Every day it: • Pulls HYROS itself and reads every ad on first- AND last-click • Makes kill / keep / scale calls against rules we set before the results came in • Remembers every decision and why, so it never relearns a lesson!!! • Turns its own mistakes into checklists so it doesn't repeat them Best catch: one ad showed 0% upsell on first-click and was the best in the account on last-click (16.7%). On one attribution model, it would have been killed. The prompt that started it: "You're my media buyer. Pull HYROS yourself, judge every ad on first- and last-click, follow our rules, and log every call and why." Feedback for HYROS • The CAC curve can't tell a budget change from a new ad launching. Our "cheap week" was a new ad, not the extra budget. Add a "new ads launched" marker per spend level? 2️⃣ Give Meta cAPI what a buyer is WORTH over their lifetime What it does: sends Meta's cAPI what each new buyer is likely worth to us (from our lifetime HYROS data), so Meta hunts for buyers who become our highest-LTV members, not just more cheap front-end buyers. Like Meta's value optimisation, but fed with my own HYROS data. The problem: our front end is a cheap cookbook. At checkout, a buyer who never comes back and a buyer who becomes a long-term 1-1 coaching member look identical to Meta. Meta's own predicted-value bidding needs ~100 conversions a week, which our 1-1 program may never hit!

Hi all, Looking for advice from anyone who's dealt with Meta's Health & Wellness data-sharing restrictions. Context : - We sell dietary supplements, based in France (EU market). Legit business, compliant products, no grey-hat stuff. - ~€150k/month on Meta, 20 full-time employees, Meta is our main acquisition channel. - Meta classified our domain under Health & Wellness / pharmaceutical. Ads still run and get approved, but lower-funnel events (Purchase, AddToCart) are blocked. Ad delivery is fine, only the conversion data is cut. - We never saw the warnings. Notifications went to the BM email and the Events Manager banner, and we only noticed when attributed conversions dropped. - Our Meta account managers say there's nothing they can do: the classification is automated and considered accurate because we sell supplements. - We only need to optimize for Purchase. Our read on what triggered it : Most likely a mix of health claims in our copy / landing pages and what Meta's crawlers picked up when scanning the site. Nothing structural, but enough for the classifier. Our plan 1. Clean the whole ecosystem: site, landing pages, product pages, ad copy, claims. Then re-submit the category review in ~90 days with everything in order. 2. Rebuild tracking properly: move to CAPI server-side, with strict data minimization. We'll stop sending Meta everything about our customers and only pass what's needed for matching and optimization, with product/health signals stripped from the payload. 3. Domain architecture: a clean domain that Meta "sees" in the ads, a buffer domain in between, then our actual store. Click IDs/UTMs captured server-side, persistent ID across the hops, events sent via CAPI from the clean data source with a neutral custom event name. What we understand so far - The restriction is at the root-domain / data-source level, so CAPI on the same domain doesn't help (events get dropped on receipt) and subdomains inherit the classification. - Appeals rarely succeed for brands that genuinely sell supplements, so our 90-day re-appeal is a bet, not a plan A.