Start with your week, not with the tool

You will be able to choose where to use AI based on how your time is spent rather than on what a tool can do.

Most people start using AI at work the same way. Someone shows them a clever demo, they open an assistant, and they try it on whatever is in front of them that afternoon. Sometimes it helps. Often it produces something bland, or wrong, or slower to fix than to write. After a few weeks they decide AI is useful for some things, without being able to say which things, and their working week looks much as it did before.

This course starts from the other end. Before you choose a tool or a technique, you look at where your hours actually go. A saving only matters if it lands on a task that takes real time, and you can only know it is real if you measured the task before you changed it.

Take a typical week for an operations executive at a mid-sized firm. Monday morning goes on a weekly report that pulls figures from three spreadsheets. Two hours a day disappear into email. There are six or seven meetings, each followed by notes nobody enjoys writing. Once a month there is a deck for management. In between there are one-off tasks, such as researching a new supplier or learning a system the company has just bought.

Looked at this way, the candidates for AI help sort themselves. Email is frequent, mostly routine and quick to check, which makes it a strong candidate. Meeting notes are frequent and tedious, but they carry decisions and names that must be right. The weekly report is frequent, but an error in a figure that management relies on is costly. The supplier research is occasional and depends on current facts, which is exactly where assistants are most likely to invent things.

Three questions do most of the sorting. How often does the task come up? A prompt or template you reuse every week pays back far faster than one you use twice a year. How quickly can you check the output? A draft email takes a minute to read, while a reconciliation you would have to redo by hand saves nothing. What does an error cost? A clumsy internal note costs very little, while a wrong number in a client proposal costs a lot.

There is also research behind the caution. In 2023, a research team including Harvard Business School academics ran a field experiment with consultants at Boston Consulting Group, published as Navigating the Jagged Technological Frontier. On tasks that suited the model, consultants using GPT-4 finished more work, faster and at higher quality. On a task chosen to fall outside what the model did well, consultants using it were more likely to reach a wrong answer than those working without it. The authors described the boundary as jagged because tasks that look equally hard to a person can fall on either side of it. The practical lesson for you is that you find your own boundary by testing on tasks you can check.

There is one more filter before anything else: data. Some of your most time-consuming tasks will involve client names, salaries, contracts or other information you may not be allowed to paste into an assistant. Module 2 covers how to handle that. For now, simply mark those tasks.

By the end of this module you will have three things: a list of your recurring tasks, the three to five you will work on through the course, and a baseline log of how long they take now. The rest of the course takes each kind of task in turn, then comes back to your log at the end so you can see, in minutes rather than impressions, what changed.

Your task for this lesson: write down the ten tasks that took most of your time last week, with a rough number of hours next to each. Do not judge them yet. You will score them in the next two lessons.

Write down the ten tasks that took most of your time last week, with a rough number of hours for each.

Course

• AI at work: writing, research, spreadsheets and meetings