Most advice about AI and business has never been tested by the person giving it.

What Ovriko is

Ovriko is an experiment-driven media and business publication covering the intersection of AI, business and money. Its output is experiments and the writing around them: what was tried, in what conditions, at what cost, and what the numbers said afterwards.

It is not a consultancy, a software product or a newsletter about other people's announcements.

Why it exists

There is no shortage of claims about what AI does to a business. There is a shortage of anyone showing the working: the setup, the hours, the invoice, the thing that broke in week three, and the honest answer to whether it was worth keeping.

Ovriko publishes that working. A result that fails is as useful as one that succeeds, and considerably rarer to find written down.

What it tests

Whether AI systems can do work a business currently pays for. Which offers, prices and channels produce revenue rather than activity. And what any of it actually costs once setup, supervision and the hours nobody logs are counted in.

How experiments are approached

Every experiment starts with a specific claim and a threshold for what would count as a pass, both written down before any money is spent. What changes and what deliberately stays the same are defined up front, so the outcome can be attributed instead of assumed. Metrics and the cost model are fixed in advance.

The result is published against the original threshold, with the method in enough detail to be repeated and challenged.

Observation is not evidence

Both belong on the site. They do not get to look the same.

Observation

Something noticed while working: a pattern, a suspicion, a question worth testing. Useful, but it proves nothing on its own. Published as commentary and labelled as commentary.

Experiment

A claim with a threshold, a test and control setup, a cost model and a measured outcome, all defined before the work starts. Published with its method and its numbers, and with a verdict attached.

The commitment

Numbers get published with their source. Costs get published in full. Experiments that fail get published with the same prominence as the ones that work, and a result that does not earn its cost gets said plainly.