30-Day AI Workflow Review Checklist

This checklist is designed to be completed in a single sitting at the end of your first 30 days running AI workflows in DAVE. You will need access to the Reports page, which is available to Admin and Reporting roles. If your role does not include either, ask your workspace Admin to grant you Reporting access before you begin.

Work through each section in order. For each item, record the current value from the Reports summary cards, note whether it is trending in the right direction, and decide whether it needs action in month two. The goal is not to hit a specific number: it is to have a number, understand what is driving it, and know what you would change.

For background on what each metric means and how to read it, see the companion article: What to measure in the first 30 days of AI workflow automation. For a walkthrough of the Reports page itself, see the companion guide: Read your reports and see what your workflows are actually doing.

Open the Reports page and read the Total Runs card (blue) and derive your Completion Rate from the instances that reached completed status.

• Record your Total Runs count for the 30-day period. Use the date range filter to scope the window if needed.
• Calculate your Completion Rate: completed instances divided by total started, expressed as a percentage. Note it here: ____%
• Is the Completion Rate stable or improving across the four weeks? (Declining in week four is a signal to investigate before adding volume.)
• Are there specific workflows with a materially lower completion rate than others? If so, name them and flag for month-two investigation.
• Is Total Runs volume in line with what you expected when you designed the workflow? If it is significantly lower, the workflow may not be triggering as intended. If it is significantly higher, confirm your provider capacity can sustain the pace.

Month-two action if Completion Rate is below your target: review the workflow graph for paths that cannot complete under certain input conditions, and check whether any agents are failing silently rather than routing to a Human Review node.

Read the Avg Latency card (purple) on the Reports page. This is the mean elapsed time between an instance's start and completion across the 30-day period.

• Record your Avg Latency for the full 30-day period: ____
• Compare week one Avg Latency against week four Avg Latency using the date range filter. Is it stable, improving, or degrading?
• Is the latency acceptable for the use case this workflow serves? (A content-generation workflow running overnight has a different latency requirement than one that responds to a live user request.)
• Correlate Avg Latency with Pending Tasks. If both are elevated at the same time, the bottleneck is a Human Review node waiting for a person, not an AI call taking too long.
• If Avg Latency is higher than expected and Pending Tasks is low, the bottleneck is likely an AI provider call. Note which agents are involved and whether the provider has a rate limit or timeout issue.

Note: the Latency per Agent trend panel is listed as coming soon in DAVE's Reports page. Until it ships, use instance-level investigation to identify which specific agent or node is contributing most to latency. The date range filter can help you isolate periods when latency spiked.

Month-two action if latency is degrading: check provider health, review whether Human Review node criteria are triggering too broadly, and consider whether any workflow steps can run in parallel rather than sequentially.

Read the Failure Rate card (red) on the Reports page. This is the proportion of workflow instances that ended in a failed state rather than a completed one.

• Record your Failure Rate for the full 30-day period: ____%
• Plot the trend across the four weeks using the date range filter. Is the rate stable, declining (good), or rising (investigate)?
• Did you observe any sudden spikes in Failure Rate? A sudden spike typically indicates a provider outage or rate limit rather than a workflow design issue.
• Did you observe a gradual climb in Failure Rate? A gradual climb typically indicates a misconfigured agent, a prompt that degrades with certain input types, or a workflow design gap.
• Are failures concentrated in one workflow or spread across all of them? Concentrated failures point to a specific configuration issue; spread failures point to a systemic issue such as a shared provider problem.
• Have you reviewed the audit log for failed instances to identify the specific node or agent where failures occur? The audit log is a chronological record of significant system actions and is the primary tool for diagnosing failure causes in DAVE.

Note: the Failure Rates trend panel (a breakdown of failure reasons by workflow) is listed as coming soon in DAVE's Reports page. Until it ships, use the audit log and instance-level review to identify failure causes.

Month-two action if Failure Rate is above your acceptable threshold: prioritise the highest-volume failing workflow first, review the agent configuration and prompt, and confirm the provider connection is healthy using the Fetch Models function in the Providers page.

Read the Pending Tasks card (amber) on the Reports page. This is the count of tasks created but not yet completed, typically because a Human Review node is waiting for a person to act.

• Record your current Pending Tasks count: ____
• Is the Pending Tasks count growing, stable, or declining over the 30-day period?
• Compare Pending Tasks to Total Runs. A high ratio (many pending tasks relative to runs started) means human review is a bottleneck.
• Are pending tasks clearing within your expected SLA? (The Task SLAs trend panel is listed as coming soon in DAVE's Reports page. Until it ships, track this manually by noting the oldest open task date.)
• Do you have enough reviewers assigned to the workflows generating the most tasks? If not, note which workflows need additional reviewer capacity in month two.
• Are any Human Review nodes triggering too broadly, sending items for human review that could safely be handled automatically? If so, the node criteria may need tightening.
• Are any Human Review nodes triggering too narrowly, letting items through that should have been reviewed? If so, the node criteria may need broadening.

Month-two action if Pending Tasks is accumulating: either add reviewer capacity, tighten the Human Review node criteria to reduce unnecessary review triggers, or both. Do not simply increase workflow volume until the review backlog is under control.

Before closing the review, confirm that the right people have the right access to Reports, and write down your top three month-two priorities while the data is in front of you.

• Confirm that everyone who needs to see Reports data has either Admin or Reporting role access. DAVE has six default roles: Admin, Create, Curate, Use, Reporting, and Financial. Only Admin and Reporting can access the Reports page.
• Confirm that nobody has Admin access purely to view reports. If they only need to read metrics, Reporting role is sufficient and carries fewer permissions.
• If the elevated_reporting feature flag is enabled on your tenant, confirm that advanced analytics access is limited to the people who need it.
• If PDF export is enabled (pdf_export feature flag), confirm you know how to export a report for stakeholders who do not have DAVE access.

Write your top three month-two priorities here before you close this checklist:

1. ____
2. ____
3. ____

Common month-two priorities based on typical 30-day findings: reducing Failure Rate in a specific workflow, adding reviewer capacity for a high-volume Human Review node, tightening a Routing node's branching logic, or expanding to a second workflow now that the first is stable.

For a full walkthrough of how to read each metric and act on what you find, see: Read your reports and see what your workflows are actually doing.