Every figure below traces to a monthly report I wrote at the time. Where something is unfinished, it says so. On a call I will screen share the account and you can look at it yourself.
E-commerce retailer · Shopping and PMax · taken over from an agency
May 2026 → August 2026 · four monthly readings on one pinned basis
The account was profitable on the report and thin in the bank
What it looked like
Every month produced a different answer to the same question, and nobody could say which answer was real. The interface said one thing, the revenue report said another, and the decisions kept getting made against whichever number was open at the time.
What was actually wrong
One custom conversion goal counted gross profit. The account default counted order value. Three campaigns carrying 53% of the spend ran on the profit goal and everything else ran on the default, so the same month read 1.83 in the interface and 3.01 on purchase revenue.
Neither number was wrong. They were answers to different questions, blended into one figure, and Smart Bidding was optimising toward the blend. The bidding targets confirmed it: one campaign targeted revenue, one targeted profit, and two maximised profit with no target at all.
What changed
Four things changed across those months. The conversion basis was pinned: one definition, written into the configuration and checked on every run. The products taking budget and never converting were cut. The feed titles and the labels the campaigns route on were rebuilt. And budgets stopped being argued against the interface and started being argued against the pinned basis.
| On the pinned basis | May 2026 | August 2026 |
|---|---|---|
| POAS, profit on ad spend | 0.79 | 1.40 |
| Monthly spend, indexed | 100 | 54 |
| Monthly revenue, indexed | 100 | 98 |
POAS 1.00 is break-even by definition, so this account went from losing money on every euro to making forty cents on it. The path was 0.79, 1.14, 1.26, 1.40. No month reversed.
The part that surprises people
Spend fell by nearly half and revenue moved less than two percent. The budget that came out was not buying sales. It was buying the appearance of sales, on a basis that counted some of them twice.
Still open
The account still reads drift. The pinned basis reads 64% higher than the account’s own primary total, so the two definitions still disagree, and three campaigns remain on the profit goal. That is the next fix and it is not done.
On your account
The first week is the same first week: read the account, find out what the reported number is actually counting, and agree in one line what the forty-five days will be judged on.
E-commerce retailer · Shopping and PMax
August 2026 · break-even re-measured 26 August 2026
Every winning decision had been scored against a number that was wrong by half
What it looked like
The account was performing well against its target and the target was fiction. Nobody had checked where it came from, because it had been in the brief since the beginning and briefs do not usually get audited.
What was actually wrong
The brief said break-even ROAS was 1.54. That figure had been derived from an average order value which had itself been derived, from a CPA multiplied by a ROAS, rather than measured. Measured against actual orders, the average order value gave a real break-even of 3.00.
So the account had spent months being judged against a bar set at roughly half its true height. Campaigns that looked like winners at 2.2 were losing money. The most useful thing I did here was not a campaign change. It was recomputing the break-even.
What changed
The break-even was rebuilt from measured orders and agreed in writing. Everything got re-scored against it, which moved some campaigns out of the winners column and others in. The account also scaled hard that month, and to be exact about it: the scaling was executed by the agency running the budgets, not by me. What changed on my side was the bar they were being scaled against.
| August 2026 | Value |
|---|---|
| ROAS on the purchase basis | 6.99 |
| Break-even ROAS, measured | 3.00 |
| Multiple of break-even | 2.33× |
| Month-over-month spend | +86% |
The part that surprises people
The account scaled 86% in a month and got more efficient, not less. That is the opposite of what scaling normally does, and it only happens when the thing being scaled was underfunded rather than overfunded. The wrong break-even had been holding it back.
Still open
There is no profit target agreed above break-even, so "good" has no ceiling defined yet. That conversation has not happened.
On your account
If your target came from a brief rather than from measured orders, it is worth an hour before it is worth a campaign change.
What is not here
No client names, no logos, no screenshots. The accounts are not mine to expose, and a redacted screenshot is not anonymous: campaign names alone identify an advertiser.
No testimonials either, because I do not have one worth showing yet. When a client agrees to be named and quoted, it will sit at the top of this page and it will outrank everything below it.
Other work, without the detail
Four more engagements, named only by their shape, because the accounts are not mine to describe in full. They are here so the two cases above are not the only thing you have to go on.
- Black Friday, carried. A week that decides the quarter, planned in advance and watched hour by hour, with everyone in the business under pressure at the same time.
- An annual revenue target made in a bad year. Found the weeks where competitors had eased off and bought them, instead of paying the crowded weeks at any price.
- Accounts taken over that were not profitable, and made profitable. Nearly always the same order of work: what the conversions really count, then where the budget goes.
- Launches in categories nobody wanted. Hard verticals, from nothing, producing leads the sales side actually kept.
No client is named anywhere on this site, and no figure is attached to these four. If you want the detail, I will open a real account on a screen share.
See what this would look like on your account Or write to me instead
Thirty minutes, one screen share. I will open a real account and you can ask what you like about it.