How Can a CEO Use AI to Save Time Every Week?
A CEO saves real time with AI by handing it the prep and follow-up around decisions, not the decisions themselves. Audit one normal week of your calendar and inbox, pick the three recurring tasks that cost you the most hours and need the least of your judgment, and give each one a fixed format, a schedule and a person who checks it. Then time the new version against the old one.
That sounds obvious. Most CEOs still skip it. They open a chatbot when they remember to, get a decent answer, and go back to the old way by Thursday. The average worker who uses generative AI saves about 5.4% of their work hours, roughly 2.2 hours in a 40-hour week, according to the St. Louis Fed. A CEO's week is longer and has far more prep and paperwork in it, so the ceiling is higher. You only reach it if you choose the tasks on purpose.
This post covers where a CEO's week actually goes, which tasks AI should and should not take, a six-step method to find your first three, and why your own week is the best AI pilot your company will run.
Where does a CEO's week actually go?
The best data still comes from Michael Porter and Nitin Nohria, who tracked 27 CEOs around the clock for 13 weeks, about 60,000 hours in total. Their CEOs worked an average of 62.5 hours a week and spent 72% of their work time in meetings. Only 28% of their work time was spent alone, and 75% of it was scheduled in advance.
Two things follow from that. First, you cannot hand the meetings themselves to AI. They are where you decide, persuade and listen, and that is the job. Second, almost every meeting has a before and an after. Someone has to read the background, work out what you want from the room, write down what was decided and chase the owners. That work is predictable, it repeats every week, and without help much of it lands on you, in the 28% of time that was meant for thinking.
The inbox makes it worse. Microsoft's Work Trend Index found that the average worker receives 117 emails a day and is interrupted every two minutes by a meeting, email or notification.
Which CEO tasks should AI take, and which should it never touch?
Sort every recurring task in your week into one of three groups. This is the most useful five minutes in the whole exercise.
- Only you. Decisions, hard conversations, hiring calls, board judgment, the relationship with your top customers. AI can prepare you for these. It should never do them.
- Draft for you. Work where you own the final version but should not write the first one: the board update, the investor note, the all-hands email, a reply to a difficult customer, a one-page brief before a big meeting. AI writes, you edit.
- Run without you. Work with a fixed format that nobody needs you to touch: meeting notes, the list of decisions and owners after each meeting, sorting the inbox into "needs you", "FYI" and "someone else", the Monday digest of numbers against plan.
Most CEOs put far too much in the first group. "Only I can write the board update" usually means "only I can decide what the board update says." Those are different jobs. The second is yours. The first is a draft.
One rule applies to all three groups: nothing goes out under your name without you reading it. AI drafts, you send. That rule costs a minute per message and saves you the one email you would not have sent.
How do you find your first three AI tasks?
Here is the method. It takes about two hours spread over a week, and it turns "I should use AI more" into a short list you can act on.
- Log one normal week. Export your calendar. For each meeting, write two numbers next to it: minutes spent preparing and minutes spent on follow-up. Then list every piece of writing you did that week and how long it took. Estimates are fine for this first pass. Leave out a week with a board meeting or an offsite, unless that is what a normal week looks like for you.
- Keep only what repeats. Cross out anything that happens less than once a month. AI pays back on repetition. A one-off task is rarely worth the setup.
- Score each task on three things. Hours per month. Judgment, from 1 (a format anyone could follow) to 3 (only you know what good looks like). Risk if it is wrong, from 1 (you would catch it and shrug) to 3 (it reaches a customer, an investor or the board). Divide the hours by judgment times risk. The highest scores are your best first targets: lots of hours, little judgment, low risk.
- Write a one-paragraph spec for the top three. For each one: what goes in (calendar, email thread, CRM record, last week's notes), what comes out (a five-bullet brief, a draft email, a table), when it runs (every weekday at 7am, 30 minutes after each meeting) and who checks it. If you cannot write the spec, the task is not ready for AI. That is the "What" step, and it is where most AI projects in a company go wrong too.
- Decide how to run each one. Some tasks work with an assistant you already pay for. Some need it configured: connected to your email, calendar and notes, with saved instructions. A few need a small custom workflow. Start with the cheapest option that meets the spec. We cover the trade-offs in Build or Buy AI? How a CEO Should Decide.
- Run it for two weeks and time it. Compare the new minutes per task with your baseline from step one. Then decide, in advance, where the saved hours go. An hour saved that quietly fills with another meeting is not saved. The same logic applies to company-wide AI, and we go deeper on it in How to Measure AI ROI: A CEO's Guide.
What does an AI-assisted CEO week look like?
Once the first three tasks are running, a typical week starts to look like this:
- Monday morning: a one-page digest of last week's numbers against plan, with the three biggest gaps flagged. You read it before your leadership meeting instead of building it during the meeting.
- Before each external meeting: a short brief with who is in the room, your last interaction with them, open issues and the one outcome you want.
- Within an hour of each meeting: notes, decisions and owners, ready for you to approve and send.
- Every day: your inbox sorted into "needs you", "FYI" and "someone else", with draft replies for the routine ones waiting for review.
- Thursday: a first draft of your investor or board update, built from the week's notes, so Friday is for editing, not writing.
Notice what is missing from that list: you sitting in a chat window typing prompts. The difference between "using AI" and having AI save you time is that the outputs arrive on a schedule, in the same format each time, without you asking. We describe the full version of this in our post on the AI chief of staff.
Why is your own week the best place to start with AI?
Most companies start AI in a department and hope the CEO catches up. We think that is backwards, for three reasons.
It is the only pilot you fully control. No change management, no IT ticket, no team to persuade. If the morning brief is wrong, you fix the spec the same day.
It teaches you what AI is good and bad at. A CEO who has personally watched a draft go from useless to good after three rounds of better instructions makes better calls about which company projects to fund. A CEO who has not is guessing.
It shows the company what "Operating" looks like. In our four-stage AI maturity model, companies move from Scattered to Experimenting to Operating to Compounding. The jump from Experimenting to Operating is the hard one: AI stops being something people try and becomes part of how a workflow runs, with a set format, a schedule and an owner. When your own week works that way, you have a working example to point to. Your Monday digest becomes the template for your sales leader's pipeline review. The spec you wrote in step four is the same document your team will need for every company workflow.
What mistakes waste the time AI saves?
- Prompting on demand instead of on a schedule. If you have to remember to ask, you will stop asking within a month.
- No baseline. Without the week-one log, you cannot tell a real saving from a nice feeling.
- Starting with the highest-stakes task. The board deck is tempting. Start with meeting follow-up, where a mistake costs you a correction, not your credibility.
- Letting the calendar refill. Microsoft found that meetings after 8pm are up 16% year over year. Calendars fill whatever space you leave. Block the hours you win back before someone else books them.
What should you do this week?
Start the log on Monday. By Friday you will have a list of recurring tasks with real minutes next to them, and the three that should go first will be obvious. Two weeks after that, you will know exactly how many hours came back and where they went.
If you want a second pair of eyes on where AI fits in your own week and in your company, Journey offers a free CEO AI audit. Either way, start with your own calendar. It is the one place where you can make AI work by next Friday.
Frequently Asked Questions
It depends on how much of your week is prep, follow-up and routine writing. The St. Louis Fed found that the average generative AI user saves about 5.4% of work hours, roughly 2.2 hours in a 40-hour week. A CEO who picks recurring tasks on purpose and runs them on a schedule can aim higher. The only number that counts is the one you measure against your own baseline.
Start with recurring tasks that take hours, need little of your judgment and carry low risk if they are wrong. Meeting notes and follow-ups, inbox sorting, pre-meeting briefs and a weekly metrics digest are common first picks. Keep decisions and hard conversations for yourself, and never let anything reach the board without your review.
No, not without review. Let AI draft replies and sort the inbox, but read everything before it goes out under your name. The review takes a minute per message and protects relationships with customers, investors and your team. Routine replies can be approved in seconds once the drafts are consistently good.
Usually not at first. Many CEO tasks work with an assistant you already pay for, once it is connected to your email, calendar and notes and given saved instructions. Build something custom only when a task matters enough and the off-the-shelf version cannot meet the spec you wrote for it.
Log a normal week before you start, with minutes for each recurring task. Run the AI version for two weeks and time the same tasks again. Then check where the freed hours went. If they filled up with more meetings, you moved work around without gaining anything. Decide in advance what the saved time is for.
