At the end of Friday's call you said: "Can you send us both ideas?" This page holds both. Path A: I install my AI creative system inside Circles and your team runs it, in French, forever. Path B: my agency, SelfMade, produces the creative for you. These are two completely different engagements, different companies, different contracts, different jobs. Read both. My recommendation is at the bottom, and it is the one where you own the machine.
From our call on July 17. Circles is a bootstrapped women's health supplement brand, €8M revenue last year, pure D2C on Shopify, France first with Italy just launched. Meta has always been the growth engine. This year it stopped paying for itself: CAC is €70-75 fully loaded against the €55 it needs to be, spend got cut from €300K a month toward €150K, and the floor is €100K before fixed costs stop getting covered. Four things you said that this whole page is built around:
"While listening to you, I really felt we're doing things like dinosaurs. And we're not dinosaurs. The goal of this year is to become AI-centric across the whole company."
The five-person team plus six agencies, with a lot of ping pong between you and them. You said you want to keep the consulting part of some of those relationships and drop the execution. That instinct is the whole thesis of Path A.
"We've been working with agencies since the very beginning. We don't really have too much knowledge or too much data. I want to own that data, so that in the next phase, I don't start from zero again."
Five years of account learnings currently live inside other people's companies. Whichever path you pick, that ends: the tagged history, the dashboard, and the creative system need to be assets Circles owns.
"It has to be 55, fully loaded, so everything included. And then we can start spending again."
AOV €85. LTV around €100 at 12 months, €150 at 24. The subscription cohort doubles its value in a year. The math works, the creative feeding the pixel is what stopped working. And you named the trap yourself: cutting budget to chase ROAS is a downward spiral, not a strategy.
"We are French first. English language is a barrier. And French people are against AI as well."
Your ads live and die in native French, for an audience openly hostile to obviously-AI creative. Any solution where the final French line is written or approved by a non-native speaker has a built-in quality ceiling. This constraint decides which path wins. More on that below.
Same underlying system in both paths: full-history ad tagging, category gap analysis, social listening, the 60/20/20 testing framework, AI static production with agent-based QC. The difference is who runs it, in what language, and what you own at the end. You'll notice the page changes design language for Path B: those sections are set in SelfMade's brand, because that's whose engagement it is. The rest of the page is mine.
The same stack I walked you through on the call, the time series analysis, the gap analysis, the social listening, the test-something-new engine, installed in your environment and tuned to a French women's health brand. Three tranches. Everything lands in French-language workflows your team operates.
Heavy cadence while the system goes in, lighter once it's running and just needs nudging. By week 4 your team ships its first AI statics to the ad account. By month 3 you run it without me.
For scale: $14K is less than 10% of a single month of Meta spend at your current reduced level. Heavier in month 1 because that's where the build happens. You own the Circles instance in perpetuity: the tagged history, the dashboard, the agents, the playbooks. If we never speak again after month 3, the system keeps working.
This is the standard SelfMade engagement: the system stays with us, the output comes to you. Every week, a fixed count of unique concepts, each one traceable to your data through the three lenses I showed you on the call.
Every ad Circles has ever run, tagged by persona, angle, emotion, format, and product. Surfaces the lost learnings: the personas that had a run and got forgotten, like the gut-health example from the call.
Your library versus the French and EU supplement sector: which formats, personas, angles, and emotions you're over- and under-indexing on. The gaps become the brief.
Your reviews plus French-language communities, tagged the same way. Every persona mapped to what they actually care about, in their own words. Clients get live access to the dashboard the whole time.
The rule of thumb under the hood: roughly $1K of monthly ad spend supports one net-new creative. Concepts per week are locked contractual deliverables, not estimates. All tiers include the full pipeline: gap-analysis briefs, 9-rubric copy refinement, agent-graded creative, weekly cadence, quarterly re-runs, and dashboard access.
Your current spend (€150K, roughly $165K a month) sits above the published Tier 3 ceiling; the contract has a clear upward path against the same per-creative economics, and if the budget cut continues toward €100K you land squarely in Tier 3. Landing pages are available as an add-on at $3,000/mo. Media buying stays where it is: you have a media buying agency, and I'm not going to pretend to be one.
I said it on the call and it belongs in writing: "I can't promise our French ads would be incredible." My team works in English and some Spanish. French carries idiom, register, and a cultural allergy to obviously-AI advertising that a non-native team will sometimes miss.
Workable, but only with a native French speaker on your side QCing every line before it spends, most likely Margaux. That adds a review loop between your team and ours, which is exactly the agency ping pong you told me you're trying to get rid of. It's the structural weakness of Path B for Circles, and it's the main reason my recommendation is Path A.
If you were a US brand, I'd pitch you the agency and mean it. For Circles, three things you said on the call all point the same direction:
"We are French first. English language is a barrier."
Path B puts the language your customers buy in at the end of a translation-and-QC loop. Path A puts the system in the hands of native speakers on day one. The best French AI ads in your category should be made by the French team that owns the brand.
"I want to own that data, so that in the next phase, I don't start from zero again."
Path B gives you dashboard access while we work together. Path A makes the tagged five-year history, the dashboard, and the agents Circles property in perpetuity. That's the difference between renting the learnings and owning them.
"The goal of this year is to become AI-centric across the whole company."
You already decided to keep agencies for consulting and drop them for execution. Path A is that decision applied to creative: I bring the system and the know-how, your team keeps the execution. One install, then it compounds inside Circles instead of inside another agency.
Reply with a path and I'll send the matching SOW inside 24 hours. Or the three of us grab 30 minutes next week and pressure-test both options against your numbers live.
