Usage Data

When you use an app, you leave traces. The app can record when you opened it, which buttons you pressed, and when you closed it again. Such records of user behavior are called usage data. They don’t describe who you are, but what you do. So this isn’t about your name or your address, but about your path through a product. Almost every modern program collects such data in some form.

Why companies would be working in the dark without it

Without usage data, a company would have to guess what its customers want. With it, it can see exactly where people drop off. One example: in a sign-up process with five steps, half of users abandon it at the third step. That’s a strong signal that something there is unclear or too complicated. Such insights would hardly be obtainable through surveys, because hardly anyone reports having given up.

For publicly traded tech companies, usage data is also a source of revenue. Advertising sells for more when it matches the interests of the audience. It is precisely these interests that are read from behavior. That’s why usage data regularly appears in quarterly reports, usually packaged in metrics such as daily active users or average time spent.

At the same time, it’s a legal risk. The European General Data Protection Regulation, GDPR for short, requires consent for much of this data. That’s why these banners with consent questions appear when you open websites. Those who fail to comply can face fines in the millions.

From click to analysis

Technically, everything begins with an occurrence, referred to in technical jargon as an event. An event is a single logged action, such as “video started” or “search submitted”. Each event comes with additional details: timestamp, device type, app version, sometimes rough region. These entries are collected and sent to a server, often bundled and not immediately.

On the server, they end up in a large database. There they are aggregated, so that patterns across millions of people can be recognized rather than individual persons. You can think of it like a traffic count: what’s interesting is that particularly many cars drive at eight o’clock, not exactly which car. For such analyses, names and addresses are often removed or replaced with a random identification number. This replacement is called pseudonymization, but it is not a perfect protection, because individuals can sometimes be re-identified from their behavior.

A common misconception is that usage data always contains content. The distinction matters: the fact that you typed a message at 8 p.m. is usage data. What is written in the message is content data. Many services collect the former extensively and the latter not at all.

Usage data in apps, news, and AI training

They are most clearly felt in recommendations. The home screen of a streaming service or the suggestions in a video app arise almost entirely from the audience’s behavior. Autocorrect on your phone also improves by observing which corrections you accept and which you undo.

In the debate about artificial intelligence, usage data is particularly contentious. If you ask a chatbot a question and then rate it with a thumbs up or down, that’s valuable feedback. Providers use such signals to fine-tune their models. That’s why many programs now have a switch that lets you opt out of this use.

In business news, you often encounter the term in connection with penalties and disputes. Regulators check whether companies are collecting more than necessary or improperly forcing consent. For investors, usage data is thus both: the core of the business model and a possible source of costly proceedings.