Blog Inside Sneeze It · Inside Studio · Part 4
We judge ads by cost per lead, not by what Meta chose to spend
Meta grades your ads by where it puts the money. We grade them a second time, by what a lead actually cost, and that second grade decides what our team makes next.
Meta has an opinion about every ad you run. It shows that opinion with money: the ads it likes get the budget, and the rest quietly stop. Most agencies take that opinion as the verdict. If an ad spent a lot, it must be good. We stopped working that way, and the reason is one number from our own data.
We graded 18,920 of our own Meta ads. 1,175 of them became winners, meaning Meta put at least ten times the account's typical ad spend behind them, and at least $500. Then we asked a second question of those winners: did a lead from this ad cost less than the account's average? For 47 of every 100, yes. For 53, no.
Source: Creative Benchmarks
We wrote up what that data says about ad formats in What 18,920 Meta ads say about cost per lead. This post is about what we did with it: how cost per lead became the second grade on every ad we run, and how that grade shapes the work in Studio, our creative tool, every week.
Two grades for every ad
The first grade is Meta's. Studio's Creative Direction reads every ad a client's Meta accounts ran in the last 120 days, with spend, leads and run dates, and sorts each one with plain arithmetic. A winner spent at least ten times the account's median ad and at least $500. A middle ad ran 28 days or more without becoming a winner. A loser stopped before 28 days. Anything younger is too new to judge.
That is the same rule our industry benchmark uses (578,750 Meta ads from 2026), so a client's mix can be compared fairly with it. But it only tells you what Meta backed.
The second grade is ours. For every ad, Studio compares what a lead from it cost with what a lead cost across the whole account:
- Strong: at least three leads, at or under the account's cost per lead.
- Weak: spent the price of two leads and brought none, or brought leads at more than one and a half times the account's cost.
- Thin: not enough evidence either way, and it says so.
Meta's spend tells you what the auction liked. Your cost per lead tells you what your business liked.
The two grades together give four kinds of ad, and they lead to four different decisions. A winner with strong leads is the ad to copy. A winner with weak leads is the one to watch closely, because Meta is spending heavily on something that is not paying back. An ad Meta never backed but that brought cheap leads is worth a second try with a better start. And an ad that failed both tests is finished.
Graded by spend alone
The ad Meta spent the most on is called the best ad. The team makes more like it. Roughly half the time, that means making more of an ad whose leads cost more than average.
Graded by spend and cost per lead
Every winner is checked against the account's own cost per lead. The team copies winners that also brought cheap leads, and treats an expensive winner as a warning, not a template.
What happens to the grade
A grade on its own is a report. What makes it useful is that it feeds three things the team uses every day.
Labels. Claude labels every ad from a fixed list: what it shows, how it opens and how it was made (for example offer-led, member story, class in progress, logo only). For video, Studio listens to the sound as well as looking at the picture, so a voiceover ad is not filed as "text only". The labels come from a closed list on purpose. A free-form label cannot be counted, and a label that cannot be counted cannot carry a cost per lead.
Findings. Studio then adds up the cost per lead for each label in that client's account and compares it with the account as a whole. The findings read like "make more of this", "stop this" and "test this", and each one carries its evidence beside it. The judgement is arithmetic that a test can pin down. Claude only writes the human-readable brief on top, and its explanation has the numbers stripped out so it cannot restate them wrongly.
The next brief. Every Direction read ends with a creative brief for the next round, and "Start this ad in Studio" opens a draft with the copy already in it. So the path from "member stories bring cheaper leads in this account" to a new member story ad on the designer's screen is one click.
- 1Meta resultsEvery ad from the last 120 days, read only
- 2Two gradesWinner, middle, loser or new, then strong, weak or thin on cost per lead
- 3LabelsWhat it shows, how it opens, how it was made
- 4FindingsMake more, stop, test, each with its evidence
- 5Next adA brief and a draft, tested before the client sees it
Creative Direction has been live since September 23, 2026. As of October 10, Studio's public homepage counts 97 creative direction reports written from real results.
The grade follows the next ad, too
Cost per lead does not stop at last month's ads. Since October 9, 2026, every new ad can be tested against the client's latest Direction read before it goes to the client. We cover that feature in full in Test the ad before you see it. The part that matters here is how much weight cost per lead carries in it.
Studio labels the new ad the same way, then scores each label. If this client's own ads with that label brought leads at 85% of the account's cost or less, the label scores in its favor. At 125% or more, it scores against. Direction's own "make more" or "make less" call counts too, but never twice for the same evidence. Only when a client has fewer than three ads of that kind does Studio fall back to the industry benchmark, and the result says plainly that it is benchmark data, not the client's.
Two rules keep the test honest. If the newest Direction read is more than seven days old, Studio refuses to test and asks for a fresh read, because a test against stale costs is a guess. And if the new ad breaks a lesson the team has taught Studio about that client, an "on direction" verdict drops to "mixed".
Why we chose the harder grade
Judging by spend is easier. Meta does the work, the numbers are already in the dashboard, and the answer always looks confident. Judging by cost per lead is slower. It needs enough leads per ad to mean anything, which is why Studio says "thin" so often instead of guessing. It also exposes our own work: an ad our team was proud of can come back weak.
We accept both costs because a franchisee does not pay for impressions or for Meta's confidence. They pay for enquiries at a price that works for their location. When the platform's favorite and the business's favorite disagree, which our data says happens about half the time with winners, we side with the business.
What a client sees reflects that care. In One, our client portal, a customer sees a Direction read that a member of our team has reviewed and published, not a raw grade. That has been live since September 26, 2026.
