Auditing Your Tasks: Where AI Really Saves You Time

Most people start with the tool and look for something to do with it. A better method is to start with your work, list your tasks, score them and only then pick the few where AI is worth the effort. In this lesson you will build a task inventory, apply four simple criteria and choose your first three use cases.

In the previous lesson you split your job into tasks and spotted a few candidates. Now we turn that intuition into a method, so that your choices are based on evidence rather than enthusiasm.

Step 1: make an inventory

Nadia, the HR officer, thinks she has "no time for AI". We ask her to track her work for one week with a very simple table: the task, how often she does it, roughly how long it takes each time, and what annoys her about it. At the end of the week she has about thirty lines. The point is not to be perfect; it is to see her work as it is, not as she remembers it.

Some practical tips for your own inventory:

• Use a notebook, a note on your phone or a spreadsheet. Fill it in at the end of each day, not from memory at the end of the month.
• Be specific: "answer candidates who have not been selected" is better than "recruitment".
• Write down time estimates in ranges (10 to 15 minutes). You will compare them later.
• Include the small, boring tasks. They are often the best candidates.

Step 2: score each task with four criteria

For each task, answer four questions. You can use a simple scale from 1 (low) to 3 (high).

Criterion	Question	Good sign for AI
Repetitiveness	Do I do this often, in similar forms?	High score
Textual nature	Is the input and output mostly text or structured information?	High score
Tolerance to error	If the draft contains a mistake, can I catch it, and is the damage limited?	High tolerance
Data sensitivity	Does it involve personal, confidential or regulated information?	Low sensitivity (or data that can be anonymized)

The first two criteria tell you whether AI is capable of helping. The last two tell you whether it is safe and reasonable to use it. A task can be extremely repetitive and textual, such as writing employment contracts, and still be a bad candidate because the tolerance to error is low and the data is sensitive.

Step 3: the value, effort and risk matrix

Scores alone do not tell you what to do first. Add three simple judgments for each shortlisted task:

• Value: how much time or frustration would I save if this worked well? (Low, medium, high.)
• Effort: how much work to set it up, test it and learn to use it? (Low, medium, high.)
• Risk: what happens if it goes wrong, considering accuracy, confidentiality, fairness and reputation? (Low, medium, high.)

Your first use cases should be high value, low effort, low risk. Put the rest in a "later" list, and mark clearly the tasks that are "not with AI, or not yet".

Task (Nadia)	Value	Effort	Risk	Decision
Polite replies to unsuccessful candidates (general template)	Medium	Low	Low	Start now
Drafting job ads from a short brief	High	Low	Medium (wording and bias)	Start, with review
Summarizing internal policy documents for newcomers	Medium	Medium	Low to medium	Later
Ranking candidates from their CVs	Looks high	Looks low	High	Not delegated to AI

The last line is a good example of why the matrix exists. "Ranking candidates" feels attractive on value and effort, but the risk is high, and we will see in chapter 4 why decisions about people should not be handed to a tool.

Step 4: choose three first use cases

Resist the temptation to start ten things at once. Pick three, ideally:

1. One very simple "quick win" you can try this week.
2. One that saves meaningful time on a recurring task.
3. One that stretches you a bit, for example a task with a longer document.

For each, write a one-line goal ("cut the time spent on routine supplier replies, without any drop in quality") and the way you will check the result. This is the beginning of your measurement plan, which you will develop in chapter 5.

Be realistic about time

AI does not delete a task; it moves effort. You spend less time on the blank page and more on giving context, reading and correcting. If the reading and correcting takes as long as writing from scratch, there is no gain. That is why we track time before and after, with an honest estimate. We will not promise you any particular gain here: it depends on you, your tasks and your tools. Your own measurements will tell you.

A ready-to-use prompt: scoring your tasks

If you have a long list, an assistant can help you apply the grid, as long as you describe tasks generically.

I am going to give you a list of tasks from my job as [job title]. Evaluate each task with these criteria, from 1 (low) to 3 (high):
- Repetitiveness
- Textual nature (input and output are text or structured data)
- Tolerance to error (a mistake can be caught and is not serious)
- Data sensitivity (3 = very sensitive, personal or confidential)

Then give me a table with a short comment for each task, and propose the 3 best candidates for a first AI use case (high value, low effort, low risk) and the 3 tasks I should keep without AI.
Tasks:
[paste your list, with no names or confidential information]

If a task is ambiguous, ask me a clarifying question instead of guessing.

Bad use versus good use

Bad use	Good use
Picking the most impressive-sounding use case ("AI chatbot for the whole company").	Picking a modest, frequent task where you can see the result in a week.
Estimating gains from a vague feeling.	Timing the task a few times before, then again after.
Pasting your real list of customer names to "score" them.	Describing the tasks in general terms without any personal data.

Your turn: score and choose

Exercise. Take your list from the previous lesson (or build a new one of about fifteen tasks). Score every task from 1 to 3 on the four criteria, add value, effort and risk, then select three first use cases, one task for the "later" list and one task you keep without AI. For each of the three, write a one-line goal and how you will check quality.

See a solution

Here is how Julien, the accountant, might do it. Three first use cases: (1) rephrasing standard reminders to clients for missing documents (repetitive, textual, low risk, data removed from the prompt); (2) drafting a clear explanation of a regulatory change for clients from a text he provides (textual, medium effort, review by him); (3) turning his rough notes after a client meeting into a structured summary (medium sensitivity, so he uses only the approved tool).

"Later": automating the classification of incoming documents. "Not with AI": deciding how to treat an accounting entry or giving tax advice, which require professional judgment and accountability.

Goals and checks, for example: "reduce the time to prepare reminders, checked by comparing the time for 10 reminders before and after, and by re-reading all of them before sending".