€7 Per Lead on Meta Ads. Here's What Actually Changed.
January 2026. A furniture client in Italy. 103 leads. €7.30 average CPL. Best performing adset at €4.38.
That’s not a typo.
Two months earlier, the account was bleeding. Not catastrophically — no single disaster you could point to. Just the slow drain of an account stuck in a loop: audiences that used to work, campaigns that never got turned off, budgets spread thin across too many things.
I took it over in November. Here’s what happened.
The Account Was Stuck in Its Own History
When I got access, the first thing I did was pull two years of data. Not to analyze it — just to understand the shape of it.
What I found was an account that had been run by gut feel and inertia. Lookalike audiences built off a pixel that had seen better days. The same campaigns running month after month because nobody had a reason to stop them. CPL bouncing around — some months decent, some months you’re paying way too much for leads that go nowhere.
The account wasn’t broken. It was just… frozen. Optimizing inside a logic that had stopped working.
The instinct when that happens is to add things. New creatives, new copy angles, new audiences. I did the opposite.
The Switch: Broad + Interest Over Lookalike
This is the part that sounds too simple to be real, but it’s what moved the needle.
Lookalike audiences are built on who already converted. If your pixel data is old or thin, your lookalikes are built on a distorted picture. You’re cloning the wrong person, or cloning someone from 2021 who doesn’t reflect today’s buyer.
We killed the lookalikes. Moved to Broad + Interest targeting.
Broad means you let Meta’s algorithm figure out who to show the ad to, with minimal constraints. Interest means you layer in relevant signals — in this case, design and home improvement communities on platforms like Houzz.
The “Houzz + Design” adset ended up at €4.38 CPL. That’s the best result in the account’s history.
Why does this work? A few reasons. Meta’s algorithm has gotten very good at finding buyers when you give it room. Lookalikes add a constraint that can actually narrow the algorithm’s search in unhelpful ways, especially on smaller budgets. And interest targeting, done right, gets you in front of people actively thinking about the category.
What AI Actually Does Here (And What It Doesn’t)
I want to be specific about this because there’s a lot of noise around “AI for ads.”
I use Claude Code connected to the Meta Ads API and GA4 directly. Not screenshots. Not CSV exports I analyze manually. The actual raw data — adsets, spend, CPL, frequency, audience overlap, everything.
Think of it like this. Before, managing an account meant doing your own homework and then making decisions. Now the homework is done before I sit down.
Every week, the system pulls the data, runs analysis, and gives me a report: what’s working, what’s not, where budget is being wasted, what to test next. I read it. I decide what to act on.
Every decision is still mine. The AI doesn’t touch the account. It reads, it reports, it suggests. I make the calls.
What the AI Actually Caught
Three things that I might have missed, or caught much later, on my own.
Stale audiences from 2022. There were custom audiences in the account that hadn’t been refreshed in years. Still active, still eating impressions, still affecting lookalike quality. Nobody had flagged them because nobody was looking at that level of detail systematically. Gone.
Two underperforming adsets with no path to improvement. Not just “low ROAS right now” underperformers — adsets that had been given enough time and budget to prove themselves and hadn’t. The data made it obvious. Killed them. Budget redistributed.
A retargeting campaign eating 40% of the retargeting budget at 3x average cost per lead. One ad. Forty percent. Three times the cost. This is the kind of thing that hides in plain sight when you’re managing an account by looking at top-line numbers. The granular pull surfaced it immediately.
That last one alone probably funded the difference.
The Numbers
January 2026:
- 103 leads — target was 70
- €7.30 average CPL — best month on record
- €4.38 CPL on the “Houzz + Design” adset
- +176% year over year on lead volume, same budget
The previous average was hovering around €17. Some months were better, some months you’d rather not talk about.
That’s roughly a 94% reduction in cost per lead. Same budget. Zero.
The Honest Take
I want to be clear about what I don’t know.
Q1 is naturally strong for furniture. People are coming off the holidays, they’ve been sitting in their living rooms thinking about what they want to change. January tends to convert well in this category. Some of this result is the season doing its job.
Two months is also a small sample. I’ve seen accounts perform brilliantly for one quarter and then normalize. We don’t know yet if this holds.
And isolating the AI effect from the targeting change? Genuinely hard to do. The two things happened at the same time. The targeting switch was probably the bigger driver. The AI layer made me faster and caught things I’d have missed — but I can’t give you a clean percentage of attribution.
What I can say is that the combination of systematic data review and the targeting change produced results I wasn’t seeing before. Whether it’s reproducible over six months, I’ll tell you then.
The Principle
Most accounts don’t fail because of bad creative or wrong audiences. They fail because the people managing them are too busy to do the homework. Auditing every audience. Checking every adset for hidden budget drains. Refreshing what’s stale before it becomes a problem.
That work is boring. It doesn’t feel strategic. So it doesn’t get done.
The AI layer doesn’t replace judgment — and I’d be suspicious of anyone who tells you it does. What it does is make the boring homework automatic, so when you sit down to make decisions, you’re working with complete information instead of guessing.
I still make every call. But I walk in prepared.
The question I keep coming back to: how many accounts right now are sitting on €4 CPL opportunities, hidden under two years of stale data nobody’s looked at?
This client is an Italian furniture brand — not named by agreement. All numbers are real. Q1 seasonality caveat stands.