Docs
Your weekly readout
Each week, Torque Lift tells you what moved and why, in plain words, with the numbers behind it. It is the result you came for, so it comes whether or not an experiment won.
When it arrives
Your first readout comes from day 14, once your first full week has closed. After that you get one each week, by email with an HTML and a PDF copy, and in your console.
If a week's data hasn't closed yet, Torque Lift waits a few days for it rather than send half a week. During pilots, a person on our team reads each readout before it goes out.
What is in it
It opens with the week in a few plain sentences, for the reader who reads nothing else. Then, in order:
| Section | What it tells you |
|---|---|
| Growth accounting | Your active users, and how many were new, retained, resurrected and churned |
| Activation by cohort | Of the people who first came in each week, the share still active in the weeks after |
| Season or break? | Whether a drop looks like the calendar, such as a holiday, before you read it as a problem |
| Notes from the team | The business notes your team wrote for the week, shown as a lead, never as a cause |
| The test | Your running experiment: what it measures, when it started, and when it is read |
| Recommended change | The next experiment, with its hypothesis, metric and read day |
| Breaks | Any metric that fell sharply and stayed down, found by the nightly scan |
| Ranked opportunities | The biggest ways to add weekly active users, as leads to test or check |
| What Torque Lift handled | Pipeline breaks fixed, events kept out of the numbers, personal data removed, traffic split checks run |
| Instrumentation health | Events outside your plan, event volume, and what was filtered |
| Pull requests | What Torque Lift opened for you, and where each stands |
| Definitions and sources | How each number is defined, and the saved query behind each figure |
A section is left out when its data could not be read, and a section with nothing to report says so in a line.
How to read growth accounting
A person is active in a week when they have at least one event your tracking plan counts as activity. Every active person is in exactly one of these groups, and every person who was active last week is either retained or churned.
| Group | Means |
|---|---|
| New | Active this week, and their first event ever was this week |
| Retained | Active this week and the week before |
| Resurrected | Active this week, not the week before, and seen in an earlier week |
| Churned | Active the week before and not this week, shown as a positive count |
- Active = new + retained + resurrected. Last week's active = retained + churned.
- Quick ratio is (new + resurrected) over churned. Above 1.0 the active base grew. Below 1.0 it shrank.
- Retention is retained over last week's active: the share who came back.
A week is a calendar week in your time zone, from the week start in your settings. Your console shows the same groups by day, week or month.
How the readout stays honest
- Every experiment is reported, including the ones that lost or failed a check. Experiments says how each is fixed before it starts.
- A rate that can't be measured yet reads n/a. It is never filled in with a guess.
- Leads are called leads. A ranked opportunity or a note from your team is something to check, never a proven cause.
- Every figure names the saved query behind it, so a number can be traced.