Last updated: 2 November 2026
To decide what to automate with AI, audit one week of your work, score each repeated task on four questions and sort it into automate, AI assists or stays human. Automate tasks with clear rules and low risk. Let AI assist where judgement is needed. Keep sensitive and relationship-heavy work with a person.
This guide gives you the audit table, the scoring questions and a decision matrix you can copy into a spreadsheet. It takes about a week to collect the data and an hour to sort it.
How do you run a one-week task audit?
For five working days, write down every task you repeat. Note how often it happens, how long it takes and a few facts about it. Do not try to judge the tasks while you collect them. Copy this table into a spreadsheet and add a row for each task.
| Task | How often | Minutes each time | Rules clear? (yes or no) | Sensitive data? (yes or no) | Needs judgement or a relationship? (yes or no) | Decision |
|---|---|---|---|---|---|---|
| Example: sort the morning inbox | Daily |
How do you score each task?
Ask four questions about every row.
| Question | If yes | If no |
|---|---|---|
| Does it repeat at least weekly? | It is a candidate | Leave it for now |
| Can you write the rules down? | AI can do the routine steps | AI can at most assist |
| Does it involve sensitive personal data? | Keep it with a person, or use approved tools with care | Lower risk |
| Does it need judgement or a relationship? | A person decides, and AI assists | AI can do the task, with checks |
What is the decision matrix?
Use the answers to place each task in one of three groups.
| Rules clear? | Judgement needed? | Sensitive data? | Decision |
|---|---|---|---|
| Yes | No | No | Automate, and have a person check a sample of the output |
| Yes | Yes | No | AI assists: AI drafts or prepares, and a person decides |
| No | Any | No | AI assists with research or first drafts, or write down the rules first |
| Any | Any | Yes | Stays human, unless you have checked that the tool and the data handling are appropriate |
How do roles work when AI does part of the task?
A RACI chart names who is responsible, accountable, consulted and informed. With AI in the picture, one rule holds: AI can be responsible for doing the work, but a named person stays accountable for the result. In our class, we use a RACI and a decision matrix to decide what to automate, what AI should assist with and what stays human.
| Role | Who | Example: drafting purchase requisitions |
|---|---|---|
| Responsible | The AI agent | Drafts the requisition from the request |
| Accountable | A named person | Approves it and answers for it |
| Consulted | Finance | Confirms the budget code when asked |
| Informed | The requester | Receives the approved requisition |
What do sorted tasks look like?
These are illustrations. Your own answers depend on your rules, your risks and the tools you use.
| Task | Likely decision | Why |
|---|---|---|
| Sorting the daily inbox | Automate, with sample checks | Clear rules, low risk |
| Summarising meeting notes | Automate, with sample checks | Clear format, and a person can verify against the notes |
| Drafting purchase requisitions | Automate, with sign-off | Clear rules, but money is involved |
| Answering leave questions from the policy | Automate, with escalation | Clear source, and sensitive cases go to a person |
| Drafting a routine reply to a client | AI assists | A person checks tone and facts |
| Choosing a supplier | AI assists | AI compares options, and a person decides |
| Handling a customer complaint | Stays human | It needs judgement and the relationship |
| Giving feedback to a team member | Stays human | It is personal and needs judgement |
How do you turn the result into action?
- Rank the Automate rows by minutes per week saved.
- Pick the top one that is low risk.
- Build or test it alongside your manual process for a week.
- Compare the time taken before and after, and check the quality.
- Repeat with the next row.
What mistakes should you avoid?
- Starting with the most exciting task instead of the most repetitive one.
- Skipping the rules. If you cannot write the steps down, an agent cannot follow them reliably.
- Leaving no one accountable for the result.
- Automating a broken process. Fix the process first.
- Sending personal data to a tool you have not checked.
Frequently asked questions
What tasks should I automate first with AI?
Start with tasks that repeat at least weekly, follow clear rules and carry low risk if they go wrong, such as sorting an inbox, summarising meetings or drafting routine replies.
Which tasks should not be automated with AI?
Keep tasks that involve sensitive personal data, difficult conversations, final decisions or important client relationships with a person. AI can prepare for these tasks, but a person should handle them.
How long should a task audit run?
One working week is enough for a first pass. If your work varies by month, such as month-end reporting, add those tasks from memory or from your calendar.
How do I know if automation is saving time?
Record the minutes the task took before the change and after it, over the same length of time. Also check the quality of the output, because a faster task with more errors saves nothing.
What is a RACI chart?
A RACI chart lists who is responsible, accountable, consulted and informed for a task. When AI does part of the work, name a person as accountable for the result.
Where does this guide fit in our AI productivity series?
This is the planning step in our complete guide to improving productivity at work with AI. Once you have your list, 5 AI agents a Singapore SME can build in a day shows what to build first, and What Is an AI Agent? explains the terms. If your tasks involve personal data, read our PDPA checklist for using AI on customer data, which is practical guidance and is not legal advice.
Get a framework for what to automate next
QD Academy’s one-day AI-Powered Productivity: Strategies for the Modern Workplace course ends with a productivity framework that includes a weekly audit and a decision matrix for what to automate next, along with a working AI agent built on a real process from your job. No coding or AI experience is needed. Check the course page for the latest dates, fees and funding.