trysonex.com

Analytics products start with a feature list and then work out how much data they need to support it, which is how you end up following people around the internet. Starting from the other end changes the product: decide what you will never collect, then find out what is still answerable.

No cookies, no cross-site identifiers, no per-person profile. A site owner gets pages, referrers, countries and trends, and nothing that could be joined back to an individual. The constraint came first and the feature set was designed to fit inside it, which is the opposite of how most analytics products are built.

Counting unique visitors without identifying anyone is the whole engineering problem. The answer was a daily rotating salt over a coarse fingerprint, hashed and discarded, which makes yesterday's visitors permanently unlinkable to today's. It costs some accuracy at the margins. That trade was the product.

It costs accuracy at the margins. Someone who changes network mid-session counts twice. Someone on a shared corporate IP with an identical browser might count once. A number that is roughly right and structurally incapable of being abused beats a number that is exactly right because it followed someone home.

There is very little to store, so it runs on a rounding error of infrastructure, and there is no database of personal data to secure, breach, subpoena or migrate. The constraint that looked like a limitation is the cheapest thing about it.

Live, and currently in acquisition.