AI for business · Decision tool

The AI task audit: assist, automate or leave it alone?

Beginner · Published 2026-10-02 · Reviewed 2026-10-05

Tools
Multi-tool
Audience
Managers and small-business operators
Topic
AI for business
Colourful sticky notes arranged for planning and brainstorming
Colourful sticky notes arranged for planning and brainstorming

You’ll finish with: a decision record for assistance, automation or no AI use.

Some work benefits from a draft. Some work follows a fixed rule. Some work needs a person to make the call. Sorting those first prevents a very efficient solution to the wrong problem.

Set up

Allow 25 minutes. List five recurring tasks or use the sample. Keep confidential task details out of an unapproved account.

Download this guide’s practice sheet ↓

Try this example

FICTIONAL PRACTICE MATERIAL
Tasks: summarise internal notes; copy approved reference numbers into a table; decide a refund exception; draft a public event reminder; remove duplicate files from a shared drive.

Do the task

  1. For each task identify whether the rules are fixed and how a wrong result would be found.
  2. Classify it as human-led, AI-assisted or a candidate for tested automation.
  3. Choose one low-consequence task for a manual trial before connecting tools.

Your reusable prompt

Classify these tasks and explain the reason for each. Include required review, access needed and a first safe test. Treat irreversible file removal and refund exceptions separately from drafting.

PRACTICE INPUT
Tasks: summarise internal notes; copy approved reference numbers into a table; decide a refund exception; draft a public event reminder; remove duplicate files from a shared drive.

Refine your result

Offer a non-AI alternative for the reference-number task. Do not assume every repetitive activity needs a language model.

Check the result

A simple rule-based transfer may suit reference numbers. Refund exceptions remain human-led. Duplicate-file detection can propose a list without deleting anything.

Save and reuse

Save your task map and the one experiment you will actually run. Leave the other tasks as decisions, not an automation backlog.