I self-host analytics and my customers prefer it. Notes from 18 months of Eigenstats

Selling point I did not expect: "our pageviews never leave your server" closes deals against the big platforms. Privacy is a feature with a price tag. 300 instances running, most on a 4-euro VPS next to the app.

Technical notes: columnar storage for events (Parquet + DuckDB for queries, Postgres for metadata), aggressive 1-minute rollups, raw events deleted after 30 days unless you pay for retention. The dashboard is one page and loads in 300ms because there is exactly one page.

Hardest part: bots. A third of "traffic" is junk and every customer discovers this during the trial. Built-in bot filtering is now the demo moment, not the charts.

The bot-filtering-as-demo observation is gold. Same with SEO: the moment a client sees their real numbers versus bot-inflated ones, trust flips instantly. What lists do you use?

Three public bot lists plus behavioral rules (datacenter ASNs, inhuman regularity, no mouse ever). Customers can toggle each layer and watch the numbers move. Transparency about the sausage sells the sausage.

Marketing-brain question: how do you price against free? "Self-hosted, 4 euros" versus "Google, zero euros" is a hard pitch.

I do not pitch against free. I pitch against the analytics meeting where nobody trusts the numbers. Price is per-site flat, no event metering, because metering pageviews punishes success and everyone hates it.

Self-hosted means the customer patches. How many instances run ancient versions with known issues?

Too many. Solved with auto-update for patch versions (opt-out, announced) and a dashboard nag for minors. The nag is politely relentless.