Metrickle

Protect your key flows

Last updated

A protected flow is a task your users come to do, like signing up or checking out, that Metrickle checks after every release. If a release means some people stop getting through, on one browser, one device type or with one accessibility setting, you find out with the step that broke, the money at stake and the recordings that show it.

In the dashboard, flows live on each app's Tasks page. A saved task is a draft until a person protects it. Only protected tasks are checked every hour, open cases and send alerts.

Set up a task

A task has a start step, a success step and a time window. An attempt succeeds when the same person reaches the success step within the window after reaching the start step.

To set up a task:

  1. Open your app and select Tasks.
  2. Under Saved tasks, select New.
  3. Type a name in the Task name field, such as "Checkout".
  4. Under Starts when a user reaches, choose the start step: a page path or an event, such as /checkout or checkout_started.
  5. Under Succeeds when they reach, choose the success step, such as order_placed.
  6. Choose how long people have, from within 5 minutes to within 7 days.
  7. If a success can be cancelled later, turn on Can be undone and choose the Undo step, such as a subscription_cancelled event. See Net success.
  8. Select Save and protect. If you don't have permission to protect tasks, the button reads Save draft.

Page steps match a path, and a trailing * matches anything after it. The same step editor is used for funnels, so funnels and goals explains the options in more detail.

When to alert

Below the steps, three settings decide when a group counts as falling behind. The defaults suit most flows.

SettingDefaultWhat it means
Success gap10 pointsA group finishes this many percentage points less often than the people it's compared with
Slowest 10% take1.5× as longFor an accessibility setting, the 90th-percentile time against people using no settings
Wait for30 attempts per groupA group needs this many attempts before it's judged, so one bad session can't fail a release

Drafts and protected tasks

Anyone who can draft tasks can save, change and discard drafts. A draft shows the Draft badge and the note "We won't alert you about this task until a person protects it." To protect it, open the draft and select Protect this task.

Protecting is always a person's decision:

  • It needs the "Protect tasks, and change or delete protected ones" permission. See team and access.
  • AI assistants and API tokens can draft tasks and change drafts, but they can never protect a task, or change or delete a protected one. A draft made by an assistant says "Drafted by an assistant."
  • Once a task is protected, the verdict card shows who protected it and when.

If you can't change a task, you can still try different steps on screen to see their results. You just can't save them.

How many you can protect

The Free plan protects one flow at a time, across every app in the workspace. Every other plan protects as many as you like. Drafts don't count towards the limit. If you're at the limit, saving or protecting another task says so, and you can see your usage under Protected flows on the Plan & billing page. See pricing for what each plan includes.

Read the result

The card at the top of Tasks gives the result for the selected task since the app's latest release. If there's been no release in the last 14 days, it covers the last 7 days instead. See releases for how releases are recorded.

Each result is one of four words, always shown with an icon, never colour alone:

ResultWhat it means
No gapEvery group with enough attempts finishes about as often, and as fast, as the people it's compared with
Early signalA group is past the limit, but there isn't enough usage yet to be sure it isn't noise. It's never shown as green
Gap foundA group is clearly behind. The card names who, by how much, and the step where they fall behind
Needs more usageToo few attempts from people using no accessibility settings to compare anything yet

Two comparisons feed the result:

  • Accessibility settings. People using each setting, such as keyboard navigation or large text, against people using none. See accessibility breakdowns.
  • Browsers and devices. Each browser, operating system and device type against the others of its kind, so a release that breaks Safari for everyone still shows up.

The tables under the result list every group with its status, attempts, success rate and the page where it abandoned most. A group below the attempts threshold shows Too few attempts and isn't judged.

Protected tasks are re-checked every hour. When one changes to Gap found, Metrickle alerts the app's members and any Slack, Teams or webhook destination set up on the Integrations page, naming the release the window started at. Portfolio shows every protected flow's latest result across your apps, under "Can everyone finish your key tasks?".

Net success

A completion only counts for real after seven days. Until then it's provisional, and it's taken back if the same person does any of these within the seven days:

  • gets a refund, sent as an event with negative revenue or from a revenue integration (see revenue)
  • reaches the task's Undo step, if it has one
  • sends a problem report through the feedback widget (ideas don't count). A report marked as an accessibility barrier counts against the setting that was on when it was sent

So Net success is completions that still stand, over attempts. The result card shows the rate before anything was taken back alongside it, and the revenue kept against the revenue taken. Below the results, Revoked and provisional completions shows what took completions back and lists provisional ones Likely to be revoked, based on signals such as thrashing coupon codes or revisiting pricing.

Cases

When a release breaks a protected task, Metrickle opens a case for it. You can also open one by hand: select Open case under the task on the Tasks page. A task has at most one active case at a time.

Find cases on the app's Cases page. Each case names:

  • the losing group, such as keyboard navigation users or Safari users, against everyone else
  • how attempts failed: Never started, Wrong path, Stuck or Done but expensive, and the evidence, such as a dead click or a form error
  • the step and the element, such as the "Pay" button
  • how many attempts it affected, and the money at stake: the value of each lost attempt, or the revenue undone
  • up to five Recordings that show it, each starting just before the moment

Recordings need session replay, which is on Growth, Pro, Agency and Enterprise. They're masked, only exist where the user consented, and need the "watch session replays" permission to view. Without them, the case stands on its numbers. See heatmaps and replay.

Work a case

A case moves from Open to Fix drafted to Verifying, and ends as Closed: moved or Dismissed.

  1. Select Draft fix, give it a title, and describe the cause and the change. Say how to test it the way the losing group uses it. Metrickle attaches the recordings and a link to the case.
  2. When the change is live, select Mark fix shipped and set the time it shipped. Verification measures from that moment.
  3. Metrickle re-checks every hour. You can also select Verify now.

A case closes only when the task improves for the same group, measured on the task as it was when the case opened. It needs at least 20 attempts after the fix ships and a clear improvement. Nobody can close a case by hand.

Other actions on an active case:

  • Create issue sends it to Jira, Linear, GitHub or another tracker set up on Integrations.
  • Hand to an assistant gives you a prompt to paste into an AI assistant connected through Connect AI. The assistant can draft the fix, mark it shipped and verify it, but can't close or dismiss the case. See AI assistants over MCP.
  • Dismiss is for a case that isn't a real problem. You give a reason, and it doesn't count as fixed. Only a person with permission to dismiss cases can do this.

Issues created from cases also appear on the app's Fixes page, which compares the metric before and after each one is closed.