What Meta’s ad platform actually does — written from campaigns we ran ourselves, with the numbers attached. No growth-hacking listicles.
Andromeda changed which ads get pulled into the auction, not the 50 events an ad set needs to leave learning. On a small budget that second number still decides your week. Here is what to change and what to leave.
2026-09-11 · 5 min read
Most junk leads come from a form that submits in one tap, on placements where taps are accidental, with no question a real prospect has to answer. Higher intent, placement breakdown, one qualifying question, then judge on cost per accepted lead.
2026-09-11 · 4 min read
Short answer: Audience Network, Marketplace and right column can still be switched off for the whole account. Every other placement can only be weighted: a value rule cuts its bid by up to 90 percent, which changes its share, not its existence. Here is the arithmetic and the two columns to read.
2026-09-10 · 6 min read
Meta keeps recommending Maximize value of conversions. It only pays off when order values differ and volume is there. One product at one price has nothing for it to prefer.
2026-09-10 · 5 min read
Creative limited and Creative fatigue compare cost per result against your own past ads for the same event. Two columns tell you whether it is fatigue, or a comparison against a cheaper week.
2026-09-09 · 4 min read
The ad I had last seen marked "in review" was the one that delivered. The one marked Active spent nothing all day. Status is not proof. Spend is.
2026-09-03 · 7 min read
'My Meta ads suddenly stopped converting' is usually small-sample variance, not a broken account. The arithmetic behind it, and the one rule that survives a bad week.
2026-09-02 · 3 min read
CPM is a thermometer, not a dial. When a high reading means the auction is finding buyers, when it means your setup boxed you into an expensive corner — and why forcing it down usually buys cheaper people who don't purchase.
2026-09-02 · 3 min read
All purchases missing and some purchases missing are different diseases. The two-minute count that tells them apart — and why the Ads Manager column was never supposed to match your store's order total.
2026-09-02 · 3 min read
Most 'clicks but no sales' panics are sample-size problems wearing a costume. The expected-sales math that settles it in a minute, and the funnel walk for when it doesn't.
2026-09-01 · 4 min read
When events arrive fine but the conversion dropdown is empty, the problem is usually assignment, not tracking: dataset-to-ad-account links, permission levels, custom conversion sharing.
2026-09-01 · 4 min read
Claude, Meta's own rules, or a paid tool — a run that found nothing to do and a run that never happened look identical. The one log in Ads Manager that tells them apart.
2026-08-31 · 7 min read
Connecting proves who you are; it does not grant access to an ad account. The four messages people actually get, which of them you can fix, and the one-question test that tells you whether the problem is Meta's side or yours.
2026-08-31 · 5 min read
A campaign that crushed it for 35 days and then drifted is usually not broken — it ran out of new people. Frequency, first-time impression ratio, why duplicating the winner makes it worse, and what to change instead.
2026-08-29 · 4 min read
An ad set that reads Active with $0.00 spent is not a bug report. The checks in the order that finds it fastest, and the one that caught my own account on its first campaign.
2026-08-29 · 9 min read
The toggles vanish, the enhancements come back on every new ad, and a photographer's product shots get their contrast changed. Where the setting actually lives in the Marketing API, the one feature that defaults to on, and the read-back that tells you it is really off.
2026-08-28 · 4 min read
Agencies ask how to manage several Meta ad accounts with Claude. The project-per-client answer is fine; the part nobody answers is which user Claude is logged in as. System Users, assigned tasks and per-account rate limits, from Meta's own API.
2026-08-28 · 5 min read
Meta's MCP lets Claude read and write your ad account — for as long as the chat is open. What the API says about tokens, enhancements that switch themselves on, rate limits, and the hours nobody is watching.
2026-08-27 · 6 min read
Meta ships free automation that turns ads off, adjusts bids and increases budgets on a 30-minute cycle. Where that is the right tool, where a threshold stops being enough, and the one thing we built differently.
2026-08-26 · 4 min read
Everyone blames site speed. Meta's own documentation lists two reasons the numbers diverge, and the other one takes two minutes to rule out. Here is how to tell them apart.
2026-08-26 · 3 min read
Short answer: not by itself. Learning limited is Meta telling you the ad set cannot reach about 50 optimization events a week. What the label means, how to check it in thirty seconds, and the three moves that get an ad set out of it on the budget you actually have.
2026-08-25 · 5 min read
Meta optimizes on a window of at most 7 days. If you get paid back over 60 days, flat CPA after a budget increase is not evidence the increase worked. Here is what to gate on instead.
2026-08-24 · 5 min read
An honest comparison from the people who built one. Three situations where running Meta ads by hand is the right call, three where a tool earns its keep, and what ours still can't do.
2026-08-22 · 3 min read
A fresh dataset has no purchase history, so Meta has no one to aim at. What actually happens, why the campaign looks healthy while delivering nothing, and the event ladder that fixes it.
2026-08-22 · 3 min read