The annual report, rebuilt from an export that spanned two systems.

01 · What was in the export

The reason each enquiry closed had not come across.

The export was taken mid-migration, so it spanned the old enquiry system and the one replacing it. Most of the trouble came from that, and none of it was anybody's fault. Things do not survive a migration intact.

The important loss was the closed reasons. Whether an enquiry ended in a booking, a cancellation, or the client going elsewhere is the question the report exists to answer, and for a large share of the year that field arrived empty. That single gap decided what could and could not be published, and most of the judgement calls further down this page follow from it.

The rest was the usual migration debris. Dates stored as text rather than dates, so sorting them put April before August. Delegate numbers held as a range instead of a figure. Rows sitting outside the reporting year because the export had not been trimmed. A country field that read GB for an address in Beijing, so the overseas figures had to be rebuilt from postcode formats and the addresses themselves.

None of it is visible when you open the file. It looks like a spreadsheet full of enquiries. You find it by going through the thing column by column and asking what each one is actually counting.

02 · What I built

The raw export is never touched. Every rule and every fix is on the record.

The export sits in the workbook exactly as it arrived. The rules live in one place, the corrections live in another and are keyed to the enquiry reference, and everything downstream reads from those. So any figure can be traced back to the row it came from and the rule that shaped it, and a corrected export can be dropped in without unpicking a year of edits.

From that, one clean row per enquiry, and the report tables and charts built off it. The previous year's published figures were typed in with their page references, so the comparisons can be checked against the PDF rather than taken on trust. The workbook ties itself back to the exact number of rows that were sent, and says so on its own health check.

Every enquiry that needed something from the client is listed one to a row, with the cell for the answer sitting next to it. Their replies are kept verbatim, so a decision made in August is still readable, and still attributable, in a year's time.

03 · What I would not publish

A figure that flatters you is worse than no figure.

With that many outcomes missing, there was an obvious way to make the report look healthier: quote every outcome as a percentage of the ones that were known. The percentages would have been excellent. They would also have been meaningless, so that column does not appear anywhere in the report, and the reason is printed under the tables that would have carried it.

The same went for the year-on-year value comparison, which sits on two different price bases and cannot be made like-for-like retrospectively. It is published with both faults stated underneath rather than quietly left to speak for itself. Twelve caveats were assessed one at a time and each was either held back or published with the caveat attached.

And where the client said a set of figures was wrong but could not say what the right ones were, they were left alone and flagged. Correcting them would have moved two other tables, and a confident wrong number is the thing this whole exercise exists to prevent.