Every case below says where it has got to — live, being tested, or still being built — and every figure links through to how it was counted, including the ones where part of the answer is that we have not finished counting.
These have no page of their own on purpose. A client who has not agreed to be named stays unnamed; a system that is still being tested is not proof of anything yet. Both are here anyway, because leaving them out would make the four above look like all there is.
Anonymised — group of two trading entities
LiveTwo companies in the same group, each with its own accounting system, and no way to see the whole picture without exporting from both and doing it by hand. Only one of the two systems even had a way to plug into it. The hard part was not building the view. It was proving it: every figure is checked against the accounting system's own report, cost centre by cost centre, down to the penny. Where something does not match, the dashboard says so rather than quietly picking a number.
Client anonymised at their request. Their name, their industry and every figure belonging to their business stay off this page. What is here is how it works, and the proof that the numbers add up.
Anonymised — law firm
In testingAI makes up court cases that do not exist. In a legal document that is not a typo, it is a serious problem. So this one only writes from the documents the solicitor has uploaded, and checks every case and every fact back against those files. Anything it cannot find gets flagged on the page. A document with flags still on it cannot be sent to a client.
In testing with the firm and being used on real cases. We have not measured the time saved yet. A qualified solicitor still reads everything before it goes anywhere.
Own product
In buildA matching engine we are building, starting with founders and investors. The rule we set ourselves: it has to give three reasons for a match and one against, and each reason has to point at something real about that investor. If it cannot, the match does not count. The database and the scoring are done. The rest is not.
Still being built, and shown here as exactly that. We will publish how well it works once it is finished.