James keeps three host-performance sheets by hand. We rebuilt them in the DataHub so they update themselves. This note answers one question: do the numbers agree — and where they don't, why not?
Yes, they match. We checked around 100 numbers, one by one. Only 9 came out different — and every single one of those traces back to one phone call. Nobody's maths is wrong.
The one part that doesn't match yet is the advertising cost. James has told us his rule for that (thanks James), and we've been able to check part of it. Three puzzles are left — they're at the bottom of this page.
We only compared full months (April, May, June). July is different everywhere for a boring reason: the sheet was printed partway through July, and more calls have happened since.
| Where | Sheet | Screen | Why |
|---|---|---|---|
| Alex — April (2 numbers) | 53% | 54% | The sheet counted 94 calls for Alex that month. We count 93. One booking. |
| Reanna — May (2 numbers) | 20% | 21% | The sheet counted 18 customers for Reanna. We count 19. One person — probably became a customer after the sheet was printed. |
| Reanna — June (2 numbers) | 21% / 4% | 20% / 5% | A timing fix. See below. |
| Rachel Apr, Millie May, Muskaan May | 1% / 3% / 2% | 2% / 4% / 3% | Three cancellations that were recorded after the sheet was printed. |
The Reanna June one is worth knowing about. The old report decided when a call was cancelled by looking at the last time the booking record was edited. Any later admin edit could push that time forward — making a cancellation look like it happened days after the call. The new screen uses the booking system's own cancellation log instead. For one of Reanna's June calls, that moves the cancellation from "after the call started" to "before" — which is fairer on her. We double-checked: using the old method gives James's numbers exactly.
James's sheet shares the month's Google Ads spend evenly across every scheduled call. His note says calls and cost from A/B tests are left out. He's now told us which tests count: only tests where the form sends people to the placement team's booking calendar.
The sheet's April rate (£20 per call) only works if about 60 calls were left out. But the website's test system didn't exist until 5 June — so whatever was excluded in April, it wasn't tracked by that system. James: what was the April test, and how did you spot its calls?
May's cells work out to exactly £13 per call, using the true call counts. But that needs ad spend of about £8,660 — and the Google Ads account spent £9,146 in May. No tests ran in May, so nothing should have been left out. James: did May's spend come from somewhere other than the account total — certain campaigns only, maybe?
This is the odd one. The sheet's June cost figures work out to 136 calls for Mkaya — but only 118 happened. Every host is higher, about 760 calls in total against 654 real ones. Leaving test calls out makes numbers smaller, not bigger — so these columns must count something extra. We tested everything we could think of (counting by booking date, week buckets, the hosts' personal calendars, deleted bookings) and none of it fits. James: where do the June and July "scheduled" counts in the cost tab come from?
Checked against the live warehouse on 28 July 2026 · the new screen lives at datahub.uktgdata.com/luft-host-analysis