A trade association wanted to reach its own members properly. Print labels, a usable list, no duplicates. Straightforward, except that the membership existed as seven files that disagreed with each other: six PDFs of varying age and one spreadsheet, produced by different people over several years.
This is what member data looks like in most associations. Nobody is careless. It is simply that each committee produced its own list, and no two were ever reconciled.
The work
All seven parsed. 547 raw records resolved into 431 real entities once the same business appearing under three spellings was collapsed into one. The master list went from 1,079 rows to 1,546.
Then 1,378 print labels. And here is the step that mattered: 975 of them were checked against the previous approved print run and confirmed byte for byte identical. Nothing that had already been signed off had quietly changed. Only the genuinely new records were new.
The finding nobody asked for
Reconciling the files produced something outside the brief: the membership figure in public circulation did not match the reconciled list. Not by a rounding error.
That figure was on our own website. It was in a company profile. It was in a document that had gone to investors. We had repeated it in good faith because it was the number everybody used, and nobody had ever had a list clean enough to test it against.
We corrected our own materials before we sent the client anything. It was our error to fix first.
The second piece of work
Later the same association needed a letter to a senior government office, going out over its president's signature. Six drafts, built strictly from source documents.
During the drafting, an instruction came to add a line about a particular category of business the association supports. The system declined — because that category appeared nowhere in the source material, and writing the line would have meant inventing a claim about the association's own conduct in a letter to a government office over a named person's signature. It held that position across two exchanges. The line was dropped.
We think about that one a lot. The whole anxiety about AI in professional work is that it will produce something confident and untrue and somebody will sign it. Here the pressure to invent came from our side of the table, and the thing that refused was the system.
What an association actually gets
The obvious answer is that they can now reach every member, print a correct label sheet, and add new members without the list degrading again.
The less obvious one is that they know what is true about their own membership. Reconciled, checked, and different from what they had been saying. For a body whose entire negotiating weight comes from how many businesses it speaks for, that is not a data cleanup. That is knowing your own position before you walk into a room.