Estimates from daily history

For older releases, Needlr counts from a summary of each day instead of every single event, so the verdict is a little more careful.

Needlr keeps two kinds of memory of your website. For releases from more than about a month ago, a before vs after impact check is counted from the daily summaries. That's a tiny bit less exact, so Needlr makes the verdict a little more careful.

Think of a school register

A teacher writes down everything that happens in class: who came, who was late, who answered. After three months, the notebook is thrown away.

But every day, the teacher also writes one line in the register: how many children came. That line is kept all year. You can still compare March with April, even after the notebook is gone.

Two kinds of memory

• Every single event, for 90 days. Every visit and every time someone did your KPI, in full detail.
• A summary of each day, for 13 months. For every KPI Needlr counts: how many different people visited that day, and how many did the KPI. It's kept on every account.

A summary doesn't keep names or visitor IDs. It's a small anonymous fingerprint of the day. Needlr can merge the fingerprints of many days, and tell how many different people came in all of them, without knowing who they were. Someone who visited every day still counts once.

When Needlr uses the summaries

A before vs after impact check looks at up to 4 weeks after the release, 4 weeks before it, and 4 more weeks before that, to check if the KPI was already moving (trend check).

If any of those days is older than the full event log, the whole impact check is counted from the daily summaries. In practice: releases from more than about a month ago.

Summaries are only for KPIs Needlr counts itself. A KPI from Google Analytics uses Google Analytics' own daily numbers (GA4).

How exact it is

• Small crowds are counted exactly. With up to a few thousand people in a window, the summary still knows the exact number.
• Big crowds are an estimate, accurate to about 1%. Merging many days of a big crowd gives a very close guess, not an exact count.
• Needlr adds that 1% to the range around the result. The range gets a tiny bit wider, so the verdict is a little more careful, never bolder (the range).

The check page tells you: Older than 90 days: counted from daily summaries, accurate to about 1%, so the verdict is a little more cautious.

How far back you can go

• Every account can measure releases up to 395 days back, the full 13 months.
• Needlr also needs 4 weeks of the KPI from before the release. In step 3, Needlr shows History from a date, and The earliest release it can measure is a date (more).

Common mistakes

• Thinking “estimate” means “guess”. It's within about 1%, and Needlr already accounts for it in the verdict.
• Making a KPI the day you need it. Summaries only start when Needlr starts counting a KPI. Make your main KPIs early (how).
• A release more than 13 months old. Needlr's daily summaries don't reach that far back. Step 3 says the earliest release your KPI can measure.

Words to know

Daily summary

One small record per day: how many different people visited, and how many did each KPI. No names.

Event

One thing a visitor did: a visit, a click, a form sent. Kept in full for 90 days.

Estimate

A very close count, here within about 1%, instead of an exact one.