Before you add AI to a task, run this 15-minute audit.

It is tempting to start with a tool. Start with the task instead. A workflow is only an improvement if it produces a useful result with less total effort—or a meaningfully better result for an acceptable cost.

Minute 1–4: define the task

Write down one repeatable task in a single sentence. For example: “Turn my notes into a first draft of a weekly project update.” Avoid vague goals like “be more productive.”

• What goes in, and what should come out?
• How often does the task happen?
• What makes the result acceptable?
• Which information is sensitive and must not be uploaded to an unapproved service?

Minute 5–8: record the baseline

Complete the task the way you normally would. Record the time, the quality of the result and any recurring frustrations. If possible, repeat it on a second similar example; one run can be unusually easy or difficult.

Minute 9–12: choose a small, low-risk test

Use non-sensitive sample information first. Give the tool clear instructions, a sample of the desired format and the constraints it must follow. Keep the original result so you can compare fairly.

• Check important facts against the source.
• Look for missing steps, invented details and formatting errors.
• Count the time needed to review and correct the output.
• Do not automate an action that needs human approval until you have tested its failure modes.

Minute 13–15: decide with a simple scorecard

Use the same criteria every time. You do not need a complicated spreadsheet; a note with the following fields is enough.

• Total time: setup + generation + checking + correction + maintenance.
• Quality: did the result meet your pre-written requirements?
• Reliability: did it work on more than one example?
• Risk: were there privacy, accuracy, licensing or security concerns?
• Cost: does the free tier or subscription fit the actual workload?

Keep it, change it, or stop?

Keep a workflow only when the result is good enough, the risks are manageable and the full process makes sense. If it saves time generating a draft but adds more time to fact-check it, that is important evidence—not a failure to make AI work.

Write down the date, tool and plan used, prompt or task, examples tested and limitations. This makes future comparisons more useful and helps you notice when a tool changes.

Your turn

Choose one repetitive, low-risk task you do every week. Run this audit before paying for a new tool. If you want to compare a specific workflow, start with one of the CurioLab experiment plans.

Editorial note: this is general workflow guidance, not a review of a specific product. Always follow your organisation’s data-handling and security rules.