loggie
Start free

Amplitude · Workflow guide

Amplitude weekly metric review with an AI agent

Have your AI agent read saved Amplitude charts through Loggie and write a weekly product review that flags real changes and the next question for each.

Who this is for: Product teams with established Amplitude charts who want less dashboard touring and better follow-up questions.

Before you start

  • An Amplitude project connection in Loggie with authorized analytics read access and the correct regional API host.
  • An agent granted access to saved-chart results, without event ingestion or configuration changes.
  • A short list of saved chart IDs, their definitions, timezone, cohort filters, and agreed comparison periods.

In your Loggie workspace, connect Amplitude, create an agent identity, and review its access. On the agent’s machine, use Node.js 20 or later and initialize the CLI:

npm install -g @loggie-ai/cli
loggie init
loggie discover

Enter the agent’s Loggie key through the initialization prompt. Keep provider credentials in the connection setup, not in the agent prompt. Use the connection slug returned by discovery in the examples below.

Step-by-step workflow

  1. Use the charts the team already trusts

    A recurring metric review becomes unreliable when the agent chooses a new definition each week. Pick a few existing charts with explicit definitions: the activation event, the eligible population, the conversion window, and any exclusions.

    Choose the decision each chart should inform. A decline in activation might prompt an onboarding investigation; a retention shift might prompt a cohort review. If no action would follow a change, the metric probably does not belong in the weekly brief.

  2. Read a saved chart through Loggie

    Connect Amplitude in Loggie, grant the reporting agent access, and initialize the CLI. Discover your actual slug; amplitude below is illustrative. Get the chart ID from the saved chart and inspect the endpoint available to your connection.

    The saved chart determines the query semantics. Review its date range and filters before comparing runs. If you override dates using supported endpoint parameters, record those exact parameters. Check each chart’s timezone and cohort separately.

    loggie discover amplitude --endpoint '/api/3/chart/{chart_id}/query' --method GET
loggie call amplitude GET /api/3/chart/CHART_ID/query --pretty

  3. Explain the change the data actually supports

    Have the agent preserve the chart name, ID, definition, reporting period, and fetch time with every result. Calculate absolute and relative movement where the denominator is valid. A rate change from 20% to 25% is five percentage points and a 25% relative increase; label which you mean.

    When a change deserves investigation, propose the next segment or cohort to inspect. Attribute a cause only when the analysis supports it. Tracking changes, acquisition mix, and incomplete observation windows are concrete alternatives to check.

  4. Make the review brief a decision aid

    Limit the headline section to a few changes with a chart link, the two period values, and one next question. Put stable metrics in a compact appendix so the team can confirm coverage without reading a wall of unchanged numbers.

    Run the report on the agent’s existing scheduler after the reporting period is complete. Keep the previous output and definitions to spot a changed chart before interpreting a changed number. No event upload or analytics configuration write is needed.

Amplitude agent prompt

Adapt this prompt to your connected account and the scope you have approved.

Read the approved saved Amplitude charts. Preserve each definition, cohort, timezone, period, and fetch time. Compare matched completed periods, distinguish percentage points from relative change, and flag incomplete windows. Return the three changes most worth investigating with source chart links and a next question. Do not invent causes or alter tracking.

Example Amplitude report

Example output with invented data.

Weekly product review: last complete week vs the week before (ET)

1. Activation: 20.0% → 25.0% (+5.0 pts, +25% relative)
   Chart: "Signup → first project"
   Ask: is the lift present for both paid and organic signups?
2. Week-1 retention: 41% → 38% (−3 pts)
   Ask: did the onboarding email change reach every cohort?
3. Exports per active team: 3.2 → 3.3, stable

Causes not established. 6 unchanged charts in the appendix.

Success checks

  • Match one result with the saved chart in Amplitude, including timezone and cohort filters.
  • Manually verify the rate arithmetic and a zero-denominator case.
  • Confirm the report identifies changed chart definitions instead of silently comparing different metrics.

Pitfalls

  • Incomplete retention windows cannot be compared with fully observed cohorts as though they were equivalent.
  • A successful read of a chart is not proof that every analytics endpoint or plan feature is available.
  • Event or property names may contain sensitive product details; share only the audience-approved metric definitions.
  • This endpoint uses /api/3 and needs an origin-root connection base such as https://amplitude.com or its regional equivalent. Call paths are appended to the base.

Sources