How does use decide the next release?
The delivery practice · Checked 20 Oct 2026
We build usage tracking into every workflow we ship. When people abandon a step, finish the work in a spreadsheet, or approve in email, the gap becomes a signal. An agent drafts what it means with the evidence, the fixes are ranked by the number they should move, and a person decides which ones the software factory builds.
We record only events that could change the next release. Agents prepare the list of fixes, and a person decides what gets built.
Chapter 1 · adoption and valueWe answer for two numbers: adoption and value
Software that people work around moves no business number. So we measure both from the first release, against a baseline taken before it.
| Number | How it's counted |
|---|---|
| Adoption | The share of the people the workflow is for who finish their work in it each week, instead of around it. |
| Value | The business number agreed at the start, such as hours per review or days to close, against the baseline. |
Chapter 2 · four signalsFour signals show where the software and the work have split
| The signal | What it usually means | The fix we try first |
|---|---|---|
| People abandon the same step | The form asks for something people don't have yet. | Move the field to the step where people have it. |
| A report is exported and finished elsewhere | The report is missing the column that matters. | Add the column, and stop the export. |
| Approvals happen in email | The process map and the real work have split, so every number after that step is wrong. | Bring the approval into the software, where the record is. |
| Two teams use a feature differently | One team uses it daily and another never opens it. | Find out whether it's a fit gap or a training gap, and fix that one. |
Chapter 3 · the event planWe name the events before the workflow ships
Every workflow ships with a short event plan. An event earns a place only if its reading could change what we build next, and the plan also lists what we never record.
workflow: claims intake baseline: 18 minutes per intake, measured before the release record: - intake.step_viewed [step] - intake.step_abandoned [step, last_field] - intake.completed [minutes] - report.exported [report, format] never_record: [keystrokes, screen contents, a person's speed] reported_by: team and workflow, never by person
Chapter 4 · from signal to fixA signal becomes a fix in five steps
- Recorded. A signal appears, such as intakes stalling at the same step.
- Read. An agent drafts the likely reason, with the evidence attached.
- Ranked. The proposed fix is ordered by how far it should move the number.
- Decided. A person approves it, or doesn't, on the weekly steering call.
- Built. The fix ships through the same factory, signed by an engineer.
fig 1 · signal-to-fixChapter 5 · the ranked listThe list ranks each fix by the number it should move
| Proposed fix | The signal behind it | Number it should move | Decision |
|---|---|---|---|
| Ask for the repair estimate at step five, not step four | 31% of intakes stop at step four | Intakes finished in the software | Approved |
| Add the reserve column to the monthly report | 12 of 14 analysts export it every month | Analyst hours per month | Approved |
| Bring large-loss approvals into the app | 40% of approvals arrive by email | Days from intake to approval | Deferred to next quarter |
A deferred fix stays on the list with its reason, so it comes back with its evidence.
Chapter 6 · who does whatAgents prepare the list, and a person decides
| What agents do | What agents never do |
|---|---|
| Watch usage for breaks in the pattern | Decide what gets built |
| Draft what a reading means | Change the software on their own |
| Gather the evidence behind it | Rank fixes by who complained loudest |
| Propose fixes with a first estimate | Grade the people using the software |
Questions reviewers ask
Do you track individual people?
No. We report by team and by workflow, never by person, and we never record keystrokes or screen contents.
Who decides what gets built?
A person on your side, on the weekly steering call. Agents prepare the list and the evidence, and never decide.
What if the fix is training, not software?
Then we say so. When two teams use the same feature differently, we check whether it's a fit gap or a training gap before anything is built.
How soon do the first signals arrive?
From the first release, because the event plan ships with the workflow. The baseline is taken before that release.