What Umzi Labs is for.

The thing we are trying to be good at

There is a specific failure that keeps happening in software, and it is not a failure of engineering skill. A team is asked to build something. The brief contains an assumption that has never been tested, usually a quiet one: that the data will be clean enough, that users will tolerate the extra step, that the model will be accurate enough at the volumes involved, that the third party system will return what its documentation says it returns. The team builds. Six months later the assumption turns out to be false, and by then the architecture has grown around it.

The cost of that is not the six months. It is that the organisation now has a working system built on a wrong premise, which is much harder to abandon than a proposal on a page.

Umzi Labs exists to do the part that gets skipped: take the assumption out of the brief, state it in a form where it could be shown false, and spend a bounded amount of time and money finding out. Sometimes the finding is that the assumption holds and you should build with confidence. Sometimes it is that the assumption does not hold and the product needs a different shape. Both of those are worth paying for. Only one of them is comfortable to deliver.

Why research and engineering in the same company

Research that never becomes a product tends to answer questions nobody was going to act on. Engineering that never questions its brief tends to build the wrong thing well. Keeping both under one roof is a constraint on each: the research has to produce something a build can act on, and the build has to be able to say what evidence it rests on.

In practice that means the same people who investigate a question are the people who would build the result. There is no handover document between a research team and a delivery team, because there are not two teams.

How we would like to be judged

Not on this website. A company's website is the least reliable evidence available about it, because every word of it is written by the company. What we can offer instead is a small, bounded first piece of work with a written output, which is a real sample of how we think and costs you very little to find out.

If a first engagement goes badly, you will have spent a fixed amount and hold a document explaining what was tried. If it goes well, you have a working relationship and evidence for a decision. That asymmetry is deliberate. It is the offer we would want if we were the ones commissioning the work.