We reached cold audiences for Aurora Fragrance at £0.51 per click.

The operating founder was buying the ads himself and knew it. “I’m running the meta ads at the minute, and I’m just throwing money to social media,” he said in June 2026, “it’s nothing, it’s not strategy there.” The business had launched in March 2026: 248 orders, about £5,000 a month, a 1.6 to 1.7% conversion rate. Over the 90 days to 16 July the account spent £4,339 and returned £3,329, a 0.77x. Cost per purchase was £39.81 against an average order of £30.54: he paid about £40 to win orders worth about £30.

The structure explained most of it: roughly £25 a day split across 9 campaigns and 20 ad sets, with 68 ads competing inside 30 days and more than half taking under £10 each, so nothing exited the learning phase. One campaign copy per country cut the data four ways, the UK taking 72% of spend at 0.74x. The four ad sets holding about 90% of spend had no audience exclusions, so new-customer budget was re-showing ads to past buyers. Attribution included engaged-video-view, which the audit says typically inflates reported ROAS by 10 to 20%. And no retargeting was running at all.

Step #1 Fix the number the account gets judged against

Three break-even figures were in circulation before launch: 1.12x from an unsubstantiated 89% margin, a 1.33x “blended” figure, and 1.64x computed on a recorded call from the founder’s own costings with shipping in it. We ruled 1.64x, the strictest and the only one built from real inputs. The same math put the US at 3.37x, which is why we went UK first.

STEP #2 REBUILD, THEN GATE THE LAUNCH

Nine fragmented campaigns became three lanes: a prospecting CBO, a retargeting campaign, a retention ad set. Purchaser exclusions went into every prospecting ad set, attribution reset to 7-day click, 1-day view. All 52 ads were built paused and verified by API read-back before anything could spend. The gate returned GO on 52 of 52. It caught two things: twelve ads with a claim baked into the image rather than the copy, three of them reading “money back if it isnt yours”, and a prospecting campaign configured at £117 a day against a planned £72, one click from spending.

Step #3 Read the account against the arithmetic, not against itself

Over the 30 days to 3 September the account took 324,455 impressions at a 2.05% click-through rate and £0.51 a click, and returned 0.573x on £3,369.25 of spend. Order value ranked the markets, not anything we had done to the ads: GB a £47 order at 0.54x, the US £65 at 1.57x, Australia £89 at 5.60x, same creative, same lanes. At a £48 basket the account needs a £29.41 cost per purchase. At the £87 basket it has actually produced, £53.05. At £110, £67.07.

Step #4 Publish our own waste before making the ask

Our memo to the founder on 4 September names what we wasted: £659.38 across 19 countries with no orders, 19.6% of spend; £288.79 in ads over £50 that returned nothing; £177.16 on a pack we flagged for pausing on 18 August and never paused. Fixing all of it buys a better number and the same answer. It then puts one condition on the relationship: cost per purchase under £67 by 10 October, or do not renew.