Selected projects, described in numbers.

The problem

Cases in one programme, invoices in another, deadlines in a spreadsheet, and the agreements themselves in the inboxes of whoever was handling them. None of those places held the whole picture and every one of them held a piece of it, so answering “what is happening with this case” meant tracking down one particular person. People running 100 or 200 cases could not correct so much as a typo in a record, and in 8 out of 10 records the basic details were missing, which is exactly what you cannot issue a document without.

What we built

We built a single register of cases, clients, documents and tasks, with permissions split along two axes: who can see and who can edit. On top of that, automatic notifications about events in a case, and a place where the team submits improvements and can see what happened to them. The migration covered the entire existing database in one pass, including tidying up accounts and merging duplicates.

The problem

Client details were collected in a meeting, then topped up by email and by phone, and finally retyped into an official form. The form had 477 cells to fill in, including 75 basic-detail fields and 12 tables of specifics. The justification was written from scratch on every case, even though in half the instances it differed only in details. The client gave the same details several times, and the firm retyped them several times.

What we built

We built an online questionnaire the client fills in once, at home, at their own pace. The answers land straight in the system, the person handling the case gets a verification screen with the gaps highlighted, and fills in what genuinely requires professional knowledge. The finished document is generated in one click from the case data, together with a justification prepared and ready to edit.

The problem

Data on thousands of parties arrived in the form it most often arrives in: as a PDF. The work that followed needed a table, and between the one and the other stood a task everybody recognises and nobody plans for. At over 4,000 entries and 20–30 seconds per row, that came to between 22 and 34 hours, several days of one person’s work. It is the worst kind of task: too big to do off the cuff, and too much of a one-off for anybody to think about automating it.

What we built

We wrote a one-off extraction tool with an automatic completeness check: it pulls out the data, verifies that the number of entries matches, and shows the discrepancies for a person to settle. A job for one occasion that paid for itself anyway.