Join the Waitlist
Guide · Measurement

Read Your Reports and See What Your AI Workflows Are Actually Doing

By DAVE by voolama4 steps8 min readAI

Summary

DAVE's Dashboard and Reports pages give you a live read on every workflow running in your tenant. This guide walks you through each metric, explains what it means in practice, and shows you how to spot problems before they become incidents.

The steps

  1. How do I open the Dashboard and read the four summary cards?

    Log in to DAVE. The Dashboard is your home screen. Four cards load automatically: Active Agents, Workflows, Running Instances, and Pending Tasks. If any card shows a dash, no data exists yet for that metric. If counts look stale, refresh the page.

    Do this
    • Note the Pending Tasks count. Write it down or screenshot it so you have a baseline for next week.
    • Compare Running Instances to Pending Tasks. A high ratio of pending to running suggests a human-review bottleneck.
    • Check the Recent Instances panel (lower-left) for any instances that appear stuck or unexpectedly old.
    Example

    You log in on Monday morning and see: Active Agents 4, Workflows 6, Running Instances 12, Pending Tasks 9. Nine pending tasks against twelve running instances is a high ratio. You open the Pending Tasks panel and find that eight of the nine tasks are waiting on the same reviewer, who is out of office. You reassign them before the day starts.

    Best practice
    • Make the Dashboard the first tab you open each working day, not an occasional check.
    • Treat a rising Pending Tasks count as a bottleneck signal, not a normal background number.
    • If all cards show a dash, create an agent, build a workflow, and launch a test instance to confirm the pipeline is working end to end.

  2. How do I navigate to Reports and read the metric cards?

    The Reports page requires the Admin or Reporting role. Navigate there from the main menu. Four cards appear: Total Runs, Avg Latency, Failure Rate, and Pending Tasks. These give you rates and averages, not just counts.

    Do this
    • Read the Failure Rate card first. Any value above your established baseline needs investigation before you look at anything else.
    • Read Avg Latency. Note whether it has changed since your last visit.
    • Correlate Failure Rate and Avg Latency. Rising latency with a stable failure rate usually means slowness, not breakage. Rising failure rate with stable latency usually means a configuration or provider issue.
    Example

    Failure Rate shows 14%, up from 6% last week. Avg Latency is unchanged. You check the audit log and find that a workflow was edited three days ago. The failure spike started the same day. You review the edit, find a misconfigured agent node, and revert it. Failure Rate returns to baseline within an hour.

    Best practice
    • Establish a baseline Failure Rate in your first week of production use. Any deviation from that baseline is worth investigating.
    • If you cannot access the Reports page and expect to, ask your Admin to verify your role includes Reporting.
    • If advanced analytics are not visible, ask your Admin whether the elevated_reporting feature flag is enabled for your tenant.

  3. How do I apply a date-range filter to compare periods?

    The date-range filter narrows report data to a specific time window. Use it to make week-over-week or month-over-month comparisons. A single-session view hides trends that only appear over time.

    Do this
    • Set the filter to the current week and note all four card values.
    • Set the filter to the previous week and compare. Look for changes in Failure Rate and Avg Latency first.
    • If you made workflow changes during the period, correlate the change date with any metric shifts.
    Example

    You filter to the past seven days and see Avg Latency at 4.2 seconds. You filter to the seven days before that and see 2.8 seconds. You check whether a new AI provider was configured or a prompt was significantly lengthened during the gap. Finding the cause takes five minutes; fixing it takes one workflow edit.

    Best practice
    • Run a week-over-week comparison every Monday as part of your standing review.
    • Save screenshots or notes from each comparison so you can build a trend picture over months, not just weeks.
    • Use the filter before and after any significant workflow change to measure its impact.

  4. What should I do when metrics look wrong?

    Most metric anomalies have a small set of root causes. Matching the symptom to the cause quickly is a skill that saves hours of investigation.

    Do this
    • If Reports shows no data or all dashes: no workflow instances have completed yet. Run a test instance end to end.
    • If Failure Rate is rising: check for a recent workflow edit, a misconfigured agent, or a provider outage. The audit log is your first stop.
    • If Avg Latency is rising alongside Pending Tasks: human-review steps are likely the bottleneck. Check reviewer availability and task assignment.
    • If you see a 403 Forbidden on the Reports page: your role does not include Admin or Reporting. Contact your Admin.
    • If PDF export is not visible: the pdf_export feature flag is disabled for your tenant.
    Example

    Total Runs is 0 and all cards show dashes. You have built three workflows but never launched an instance. You start a test instance on your simplest workflow, let it run to completion, then return to Reports. The cards now show real values. You have confirmed the measurement pipeline is working.

    Best practice
    • Treat the audit log as your first diagnostic tool, not a last resort. It records significant system actions and is the fastest way to correlate a metric change with a system event.
    • Keep a short written log of workflow changes and the dates they were made. Correlating metric shifts with change dates is much faster when you have that record.
    • If audit log entries are missing for actions you expect to see, note that audit logging is being progressively expanded, with high-priority actions logged first.

Why does measuring your AI workflows matter?

Shipping an AI workflow is the beginning, not the end. Once instances are running, three questions matter every day: Are they completing? How fast? And where are they stalling?

Without answers, a rising failure rate or a growing backlog of pending human-review tasks can go unnoticed for days. DAVE surfaces the answers in two places: the Dashboard (your operational snapshot) and the Reports page (your analytical layer). Together they form the measurement foundation for the next two pieces in this cluster, which cover interpreting failure patterns and building a weekly review habit.

This guide covers both surfaces in the order you should use them: Dashboard first, Reports second.

What do the four Dashboard cards tell you?

The Dashboard is DAVE's home screen. It loads four summary cards when you navigate to it. If counts are not current, refresh the page or navigate away and return.

  • Active Agents (blue): the number of agents whose status is not retired or disabled. A drop here can mean an agent was accidentally retired.
  • Workflows (purple): the total number of workflow definitions in your tenant. This is a count of definitions, not running instances.
  • Running Instances (green): workflow instances currently executing. A number that stays high and does not move suggests instances are stuck.
  • Pending Tasks (amber): tasks created but not yet completed, typically waiting on human review or user input. This is your primary bottleneck indicator.

A practical comparison worth keeping in mind: if Running Instances is high but Pending Tasks is low, workflows are progressing smoothly. If both are high together, human-review steps are likely the constraint.

Below the cards, the Recent Instances panel (lower-left) lists the most recently created or updated instances. The Pending Tasks panel (lower-right) lists tasks awaiting action. Both panels show an empty-state message until your first workflow instance runs.

Role note: all six default roles can view the Dashboard. Clicking through to a destination page enforces that page's own role restrictions.

What does the Reports page show, and who can access it?

The Reports page is accessible to users with the Admin or Reporting role. All other roles are restricted. Navigate there after the Dashboard when you want to move from counts to rates.

Four summary cards appear at the top:

  • Total Runs (blue): the number of workflow instances that have been started. On a new tenant, this shows a dash until the first instance completes.
  • Avg Latency (purple): the mean elapsed time between an instance's start and completion timestamps. Rising latency with stable run counts often points to a slow AI provider or an overloaded agent.
  • Failure Rate (red): the percentage of instances that did not reach a completed status. Watch this card closely: a rising rate can indicate a misconfigured agent, a provider outage, or a workflow design issue.
  • Pending Tasks (amber): mirrors the Dashboard card. Seeing it here alongside latency lets you correlate the two: if both climb together, human-review steps are the bottleneck.

Use the date-range filter to narrow data to a specific window. Week-over-week and month-over-month comparisons are the most useful starting points for spotting trends that a single-session view would miss.

What trend panels are coming, and how should you prepare?

Four trend panels are labeled Coming Soon in the current release. Knowing what they will show helps you structure your workflows and naming conventions now, so the data is clean when the panels go live.

  1. Runs over Time: a line chart of workflow runs per day or week. Useful for spotting usage spikes and correlating them with external events.
  2. Failure Rates: a breakdown of failure reasons by workflow. This will make it possible to see whether failures cluster on one workflow or spread across many.
  3. Latency per Agent: an average execution-time histogram per agent. Useful for identifying which agents are slow and whether the cause is model choice, prompt length, or provider latency.
  4. Task SLAs: SLA compliance and overdue-task metrics. If your workflows include human-review steps with time expectations, this panel will be the primary accountability surface.

No action is needed to enable these panels when they ship. They will populate from the same instance and task data that already drives the summary cards.

Which feature flags and roles affect what you can see?

Two feature flags affect what you see on the Reports page:

  • elevated_reporting: unlocks advanced analytics, including deeper drill-downs and additional metric breakdowns. If you expect to see these and they are not visible, ask your Admin to check whether this flag is enabled for your tenant.
  • pdf_export: enables the PDF export button. Until this flag is on, the export option does not appear in the UI. PDF export is planned; the flag controls rollout.

If you see a 403 Forbidden error on the Reports page, your role does not include Admin or Reporting. DAVE ships with six default roles (Admin, Create, Curate, Use, Reporting, Financial), and roles are editable, so your Admin can adjust permissions without waiting for a product change.

Audit log note: the audit log records significant system actions for compliance and debugging. Audit logging is being progressively expanded, with high-priority actions logged first. If you expect an entry and do not see it, that expansion is ongoing.

How do you build a weekly workflow review habit?

The help documentation recommends checking reports weekly. Here is a minimal routine that takes under ten minutes and catches most issues before they escalate.

  1. Open the Dashboard. Note the Pending Tasks count. If it is higher than last week, find out why before moving on.
  2. Go to Reports. Check Failure Rate. Any increase since your last review needs an explanation. Common causes: a misconfigured agent, a provider outage, or a workflow design issue introduced by a recent edit.
  3. Check Avg Latency. If latency is rising alongside Pending Tasks, human-review steps are likely the constraint. If latency is rising but Pending Tasks is stable, look at your AI provider response times.
  4. Apply a date filter. Compare this week to last week. A single-session view hides trends that only appear over time.
  5. Note anything unusual. Even if you cannot act on it immediately, a written note creates a baseline for next week's comparison.

The next two pieces in this cluster go deeper: one covers interpreting failure patterns and knowing when a failure rate is a signal versus noise, and the other builds this checklist into a structured weekly review template you can share with your team.