Can any integration become a KPI evidence source?

Any connected integration can participate once it publishes a numeric metric descriptor and samples Fleet can evaluate. The mapping screen lists what is actually available; it does not invent metrics for integrations that only expose other kinds of data.

Can I create a KPI before the integration is ready?

Yes. Create a manual KPI with a current value, unit, target, and update cadence. Record new observations in Fleet until an automated metric is available.

What happens when evidence is stale?

The KPI moves to Evidence problems instead of retaining an old healthy or unhealthy verdict. Activation and intervention are blocked until current, sufficient evidence is available.

Can everyone change KPI mappings?

No. Signed-in users can inspect the queue and evidence, while changes and interventions require operator access with integration-mapping permission.

How to Turn Product Metrics into KPI-Driven Workflows

Name: Fleet
Author: Fleet

A dashboard can tell you that traffic is soft, activation is low, or users are not returning. It rarely closes the next gap: deciding which signal matters now and turning that evidence into work. The result is a familiar loop of opening analytics, interpreting the chart, inventing a response, and manually moving that response into a tracker.

Fleet's KPI Control Tower keeps that loop together. It maps a numeric metric or manual observation to a condition, ranks outcomes that need a decision, and starts a saved workflow from the exact evidence that triggered the response. This guide builds one complete KPI from evidence to intervention.

Before you start

A Fleet dashboard account connected to your self-hosted worker
Operator access with permission to manage integration mappings
A connected integration that publishes the numeric metric you need, or a current value you can record manually
A saved intervention workflow if you want one-click response; you can also create it later from the KPI

Start from the outcome, not the dashboard

Open KPIs in Fleet and choose Map a KPI. Write the outcome as a condition you can evaluate repeatedly. Activation rate should stay at or above 30%. Error rate should stay at or below 10%. Weekly returning users should stay at or above 100. The condition needs a direction, a number, and evidence that can be measured on a cadence.

Choose integration or manual evidence

For automated evidence, choose a numeric metric published by a connected integration and fill any dimensions that metric requires, such as project, environment, or repository. If the metric is not available yet, go to Manage KPIs → Create manual KPI and enter the current value, unit, and update cadence. When the metric becomes available, select that manual definition and choose Map integration evidence to keep the KPI identity while automating its source.

Define the condition and evidence window

Choose At least for a floor such as signups or activation, and At most for a ceiling such as error rate. Enter the target. For integration evidence, choose the aggregation and window Fleet should evaluate. Add a deadline only when the target is paced toward a date. Read the plain-language check before continuing; it restates the rule Fleet will enforce.

Set ownership, priority, and the default response

Scope the KPI to the company, a product, or a repository. This controls ownership and filtering in the Control Tower; it does not duplicate the definition. Set the priority that should break ties in the queue. If a saved workflow already represents the right response, select it as the default intervention. Otherwise leave it open and Fleet will offer to create an intervention workflow when the KPI needs attention.

Preview against current evidence, then activate

Choose Preview with current evidence. Fleet runs the real evaluator and shows the value, target, verdict, and any evidence problem. Activation remains disabled when samples are missing, stale, invalid, below the required coverage, or changed after preview. Repair the evidence and preview again. When the result is trustworthy, activate the KPI; only active definitions appear in the daily queue.

Triage the Control Tower

Return to KPIs. Work the queue from the top: repair Evidence problems, make decisions on Needs attention, and expand Healthy KPIs only when you need reassurance or context. Search across outcome, evidence, and owner, or filter by scope. Selecting a KPI opens its evidence rail with current value, target, trend, freshness, evaluation time, and response state.

Start the intervention from the evidence

For a breached or at-risk KPI, start the bound workflow from the evidence rail. Fleet passes the KPI binding and current evidence into the run. If no workflow is bound, choose Create intervention workflow and build the response on the workflow canvas. Fleet shows an active intervention instead of launching a duplicate response against the same KPI.

Manage the definition without cluttering triage

Use Manage KPIs for setup and lifecycle changes: edit, pause, resume, archive, delete, or record a new manual observation. Editing an active KPI creates a draft revision; the active definition continues evaluating until the draft passes preview and you activate it. This keeps half-finished setup out of the daily decision queue.

Common pitfalls

Do not assume every connected integration automatically exposes KPI evidence. The mapping screen only lists numeric metric descriptors the integration actually publishes.
Use At most for ceilings such as error rate. Reversing the operator makes a bad increase look healthy.
Do not automate from stale or incomplete data. Fleet blocks activation and intervention when the evidence cannot support a current verdict.
Editing an active KPI does not silently replace it. Preview and activate the draft revision when you are ready for the new definition to take over.
Keep the default intervention specific enough to act on the evidence. A generic 'investigate KPI' workflow recreates the same interpretation work you were trying to remove.

When Fleet is the right tool

Use the Control Tower when the hard part is no longer collecting charts but consistently turning those charts into decisions and governed work. It is especially useful for founders and product leaders who move between acquisition, activation, retention, reliability, and delivery signals during the same week. If you have no repeatable metric yet, begin with a manual KPI; if you have a metric but no response pattern, start with triage and create the workflow from the first real breach.