Six months ago I helped a small team run a study. The setup was simple: 6,000 Russian households, a domestic AI assistant installed on their phone, two months of usage data, three surveys. The thing we were measuring was not "do they use it". We already knew they would. The thing we were measuring was: what changes in how they run their lives.

I am writing this now because the embargo lifted, and because the answer is more interesting than the marketing.

What we measured

Three things. Frequency: how often do they open the assistant. Range: what kinds of tasks they use it for. Outcome: self-reported change in how they handle routine problems — paperwork, scheduling, school stuff for kids, health admin.

The first two are easy to count. The third is what the study was about.

What changed

The biggest reported change was not productivity. People did not say "I save 2 hours a day". They said "I start things I used to put off".

The pattern was consistent. People who used the assistant reported being more willing to begin difficult tasks — calling an official, drafting a complaint, asking a doctor for a clarification. The assistant did not do these things for them. It lowered the cost of starting.

The smaller reported change was in routine paperwork. Forms, applications, small bureaucratic tasks. The assistant reduced the time, but more importantly it reduced the emotional cost. People stopped dreading paperwork as much.

What did not change

Daily productivity. Time saved on routine writing. Cost of living. None of these moved meaningfully.

This is the part I think is important. The AI did not make people more "productive" in the way the launch posts claim. It made them more willing to do the things they were already avoiding. That is a different value proposition, and a more honest one.

What I take from it

I am changing the way I describe what I do. I no longer say "AI makes you faster". I say "AI lowers the cost of starting". For my customers — people who have been meaning to do something for months and haven't — that is the value they actually feel.

If you are building AI products, especially consumer ones, the metric you should be watching is not "tasks completed per week". It is "tasks started that were previously avoided". They are not the same number. The first is easy to fake. The second is the real signal.