AI in a small company that bids for public tenders

A sports flooring company that works mostly on public tenders. Years of emails, PDFs, photos, quotes and tender documents spread across different people and places.

The information exists. The problem is finding it when it's needed.

I didn't start by tidying up

Tidying everything would have taken months before solving a single visible problem. The proposal was to start the other way round: show value on real work before building the infrastructure.

A real specification as a first test

Before proposing any change in the company, I tested the approach on a real public tender: replacing the artificial turf of a municipal pitch in Seville. The administrative specification ran to 38 pages.

I analysed it with Claude and reduced it to a one-page sheet: key data, required solvency, scoring criteria, deadline and fit with the company. It's the information needed to decide whether bidding is worth it.

Start with the present, not the archive

The test solved one part of the process. The next phase is doing it with a real tender and a real deadline, end to end.

After that comes the infrastructure to repeat it: controlling where new information comes in, solving recurring tasks with AI (analysing specifications, preparing technical proposals, reviewing contracts or generating presentations) and, last, making the archive searchable.

I don't need to tidy the whole archive before starting. I need it to be searchable.

Order comes as a consequence of the work. Each solved task leaves what it touched in its place.

Where it stands

No tender has been prepared end to end with this method yet. That's the next phase.