What to Measure in the First 30 Days of AI Workflow Automation

Measure four things from day one: how many workflow instances are completing (Total Runs and Completion Rate), how fast they are completing (Avg Latency and Average Duration), how often they are failing (Failure Rate), and how many are stalling at a human decision point (Pending Tasks). In DAVE, all four are visible on the Reports summary cards the moment your first instance finishes. You do not need to build a dashboard or configure analytics separately. Open Reports from the main navigation and the numbers are there.

These four metrics answer the only question that matters in month one: is the workflow doing what I designed it to do, reliably and at the speed I need? Everything else, cost optimisation, model tuning, prompt refinement, is a month-two conversation. Get the baseline first.

Total Runs is the count of workflow instances that have been started. On its own it tells you volume. Paired with Completion Rate (the percentage of started instances that reach a completed status), it tells you reliability. A high run count with a low completion rate means something is breaking or stalling before the workflow finishes. A low run count with a high completion rate means the workflow is solid but underused.

In the first week, you want to see the completion rate climb toward a stable level as you fix configuration issues. Do not benchmark against an external standard in week one: benchmark against your own prior day. If Tuesday's completion rate is higher than Monday's, you are iterating correctly.

Where to find it in DAVE: the Total Runs card is the first summary card on the Reports page, accessible to Admin and Reporting roles. Completion Rate is a defined metric in the reports system, representing the percentage of instances reaching completed status out of all started.

Avg Latency is the mean time from when an instance starts to when it completes. Average Duration is the mean elapsed time between an instance's startedAt and completedAt timestamps. In practice these tell you the same thing from two angles: how long a typical run takes, and whether that is getting better or worse over time.

In the first 30 days, use latency as a design signal rather than a performance target. If a workflow that should take seconds is taking minutes, the bottleneck is almost always one of two things: an AI provider call that is slow or timing out, or a Human Review node waiting for a person to act. DAVE's coming Latency per Agent trend panel will break this down by agent once it ships. Until then, correlate Avg Latency with Pending Tasks: if both are rising together, the bottleneck is the human review step, not the AI call.

The practical move in week two is to set a mental baseline for your workflow's expected latency, then check whether the Avg Latency card is above or below it each time you open Reports. A rising trend that is not explained by volume growth is a signal to investigate.

Failure Rate is the proportion of workflow instances that end in a failed state rather than a completed one. In the first 30 days, some failure rate is expected: you are still tuning prompts, configuring providers, and learning where your workflow design has gaps. The number to watch is the trend, not the absolute value.

A rising Failure Rate after the first week of stabilisation points to one of three causes, in order of likelihood:

• A misconfigured agent. The agent's prompt, model selection, or provider connection has a problem that only surfaces at scale or with certain inputs. Check the agent configuration and test with edge-case inputs.
• A provider outage or rate limit. If the failure rate spikes suddenly rather than climbing gradually, the AI provider is likely throttling or unavailable. Check your provider's status page and review whether you need to distribute load across multiple providers in DAVE.
• A workflow design issue. The workflow graph has a path that cannot complete under certain conditions. Review the Routing nodes and any conditional branches for logic gaps.

DAVE's coming Failure Rates trend panel will break failures down by workflow and by reason. Until it ships, use the Failure Rate card as an early-warning indicator and investigate at the instance level when it moves.

Pending Tasks is the count of tasks created but not yet completed, typically because a Human Review node is waiting for a person to act. This is the metric most teams underestimate in month one.

A high Pending Tasks count relative to Total Runs means your human review steps are a bottleneck. The workflow is generating work faster than reviewers are processing it. This is not a failure of the AI workflow: it is a signal that the review process needs attention, whether that means adding reviewers, adjusting the criteria that trigger a Human Review node, or redesigning which decisions actually need a human in the loop.

The most useful diagnostic in DAVE is to correlate Pending Tasks with Avg Latency. If both are climbing together, the human review step is slowing the entire workflow, not just accumulating a queue. If Pending Tasks is high but Avg Latency is stable, the queue is growing but completed instances are still finishing on time, which means you have a capacity problem rather than a design problem.

For a deeper look at how to read and act on these metrics, see the companion guide: Read your reports and see what your workflows are actually doing.

The goal of the first 30 days is not to optimise: it is to establish a baseline and build the habit of reading the numbers. Here is a practical rhythm that works for most teams starting with DAVE:

1. Day 1 to 3: confirm the plumbing works. Run a small number of instances deliberately. Check that Total Runs increments, that Completion Rate is above zero, and that Failure Rate is not immediately high. If it is, fix the configuration before running more volume.
2. Week 1: set your latency baseline. After 20 to 30 completed instances, the Avg Latency figure is meaningful. Note it. This is your week-one benchmark for every comparison that follows.
3. Week 2: watch the trends, not the absolutes. Open Reports at the start of each working day. You are looking for Failure Rate rising, Avg Latency rising without a volume explanation, or Pending Tasks accumulating faster than they are being cleared.
4. Week 3: use date range filtering. DAVE's date range filter lets you narrow the report window to a specific period. Compare week two against week one. If completion rate is up and failure rate is down, the workflow is stabilising. If not, investigate before adding more volume.
5. Week 4: make a design decision. By the end of month one you should know whether the workflow's latency is acceptable, whether the failure rate is at a level you can manage, and whether your human review steps are appropriately scoped. Use that knowledge to decide what to change in month two.

The Reports page in DAVE is accessible to Admin and Reporting roles. If your team includes people who need to see the numbers without needing to configure anything, assign them the Reporting role and they can open Reports directly without access to workflow configuration or agent settings.