Skip to content
Search
AppliedPlaybooks

Which Tasks Can AI Do at Work? Score Them, Pick Three Pilots

Difficulty
Beginner
Time
2 hours with your team

Before you start

  • A list of the recurring tasks your team does (each person can write theirs in 15 minutes)
  • A spreadsheet
  • An AI chat assistant your company has approved, for the scoring discussion only
  • Your team's AI use policy, or the decision to write one

Which tasks can AI do at work, and which shouldn't it touch? Research gives a useful warning. In a preregistered field experiment with 758 consultants at Boston Consulting Group, people using AI on tasks inside its capabilities completed 12.2% more tasks, 25.1% faster and at higher quality, but on a task chosen to be outside those capabilities they were 19% less likely to reach a correct answer (Dell'Acqua and colleagues, "Navigating the Jagged Technological Frontier", 2023). The authors call this uneven edge a jagged frontier: tasks that look equally hard can fall on different sides of it. So the safe way to start is to score your own tasks and pilot the ones where a mistake is cheap and easy to spot.

Step 1: List the recurring tasks

Ask each person to list the tasks they do at least monthly, one line each, as a verb and an object: "write the weekly status update", "answer refund questions", "reconcile the card account". Collect them in one spreadsheet with a column for who does it.

Step 2: Score each task

Add three columns, each scored 1 to 3. Score as a team; you know your work better than any AI does.

  • Frequency: 1 = monthly, 2 = weekly, 3 = daily.
  • Risk if wrong: 1 = easy to fix and nobody outside sees it, 2 = embarrassing or costly, 3 = harms a customer, breaks a law or a contract, or affects someone's job or money.
  • Checkable: 3 = you can check the output quickly against a source you trust, 2 = you can check it with some work, 1 = hard to tell if it's right.

Step 3: Ask AI to challenge your scores

Paste only the task names and your scores, not the work itself.

Here is a list of my team's recurring tasks with scores for frequency,
risk if wrong, and how checkable the output is (1 to 3). For each task:
- Say whether you'd change any score and why, in one sentence.
- Name the most likely way an AI draft would go wrong on this task.
Don't recommend tools. Don't rank the tasks; I'll do that.

[paste table]

Treat the answer as a second opinion. Change a score only if the reason convinces you.

Step 4: Pick three pilots and a never list

First set aside every task with a risk of 3 or a checkable score of 1. Sort what's left by frequency, highest first, and break ties by checkable, highest first. Pick three tasks from the top as pilots. Good first pilots are usually drafts a person already reviews: meeting summaries, first drafts of routine emails, turning notes into a checklist.

Put every task with a risk of 3 on the never list for now, along with any task where the output is hard to check. "Never" can be revisited; write down the date you'll look again.

Step 5: Run each pilot for two weeks

For each pilot, write down: the prompt, who checks the output, what they check it against, and how long the task took before. After two weeks, compare time and the number of mistakes caught in review. Keep what worked; drop what didn't.

What to do next

Write the pilots and the never list into your team's AI use policy. For pilot ideas with a check step already built in, start with meeting notes or turning a process into a checklist.

Rather have someone set it up for you?

That's a fair call. LTFI, the agentic business operating system from Kief Studio, can set up the work these guides describe and run it with you. No website yet? Kief Studio will build you one for free.

Go further on kief.studio