Which work tasks can I automate with AI?

The fastest way to waste time with AI automation is to automate the wrong task. The fastest way to save time is to spend one hour auditing your week first. This tutorial walks you through that audit: how to list your recurring tasks, score them against five criteria, and pick one clear winner to automate this week.

How to determine which workflows to automate with AI

Professionals can deeply benefit from taking time to thoughtfully assess where AI automation makes sense in their current workload. No AI tools are required for the audit itself. Here’s all you need to get started:

• 30 to 60 minutes
• Your calendar and sent folder from the last two weeks
• A blank spreadsheet or document

Step 1: List your recurring tasks

Open your calendar and sent items from the past two weeks. Write down every task you did more than once. Remember to be concrete. Simply listing “email” is too general. Break it down into more specific tasks, for example "reply to inbound demo requests" and "summarize the Monday ops meeting and send action items."

Most people find 15 to 25 recurring tasks. Include the small ones. Ten minutes a day adds up quickly.

For each task, note three things: how often it happens, roughly how long it takes each time, and what the input and output look like.

Step 2: Score each task against five criteria

Give each task a score from 1 to 5 on each of these:

1. Frequency and time. Daily tasks that take substantial minutes score high. Quarterly tasks score low, even if painful.
2. Repetitiveness. Does the task follow the same basic shape every time? Same inputs, same steps, same kind of output?
3. Language work. Does it mostly involve reading, writing, summarizing, sorting, or extracting information? That's where current AI models are strongest. Physical tasks and relationship work score low.
4. Error tolerance. If the AI got one instance wrong and you caught it in review, would that be fine? Tasks where a single miss is a real problem score low.
5. Clear success criteria. Could you explain to a new hire exactly what a good result looks like? If you can't describe it, you can't prompt for it.

Add the scores. Anything at 20 or above is a strong candidate. Anything below 12 should stay manual for now.

Step 3: Screen for red flags

Before you commit, check your top candidates against these disqualifiers:

• Judgment tasks. Hiring decisions, legal interpretations, pricing exceptions, and anything a regulator cares about stay with humans.
• Irreversible actions. Sending money, deleting records, or contacting customers with no review step.
• Sensitive data. If the task involves data you can't send to a third-party AI vendor under your company's policies, stop and check with whoever owns that policy first.
• Broken processes. If the manual version is chaotic, fix the process first. Automation makes a process faster, not better.

A red flag doesn't always kill a candidate. Often it means you automate part of the task, like drafting, and keep the send action human.

Step 4: Pick one and define the win

Choose the single highest-scoring task with no unresolved red flags.

Before you begin, write one sentence defining success. For example: "The workflow drafts a summary of each sales call within ten minutes of the meeting ending, and I edit and send it in under five minutes instead of writing it in twenty."

This definition becomes your baseline for measuring the payoff later. When you’re ready, our guide on how to measure the value of AI automation will come in handy.

Step 5: Sanity-check with a manual dry run

Before building the workflow, do the task once using an AI assistant by hand. Paste a real example into ChatGPT, Claude, or your company's approved assistant with the instructions you'd give the workflow. If the output is close to usable, you have a green light. If it's far off after a few prompt revisions, the task may need to be split into smaller steps, or it may not be a fit yet.

This quick check may save hours of building the wrong systems.

Final AI workflow audit results

At the end of this process you should have: a scored list of your recurring tasks, one selected task with no red flags, a one-sentence success definition, and a confirmed dry run. That's everything you need to start building. The next tutorial in this series covers how to build an AI workflow without coding.

Troubleshooting AI automation tasks

What if every task scored low? Your work may be judgment-heavy or relationship-heavy. Get more granular: even judgment tasks usually contain a prep step, like gathering and summarizing background, that scores well on its own.

What if too many tasks scored high? Prioritize by time saved per week. When two tie, pick the one with the most tolerant error profile.

My top task involves customer contact. Split it. Automate the draft, keep the review and send with a person.

For more practical AI techniques you can apply at work, subscribe to The Rundown newsletter.