Measuring Time Saved by AI: A Simple Method

Key Takeaways

• Measure the same task before and after, on real cases, with the same start and end points.
• Count the time spent checking and correcting as well. Otherwise you're measuring the demo instead of the work.
• A simple log kept for a few weeks by the people who do the work is enough.
• Multiply the difference per case by how often the task comes up, and give a range rather than a single figure.
• Watch the quality: if errors or follow-up questions increase, the time has only moved elsewhere.

To measure how much time AI saves, time the same task before and after on real cases. Count everything: the minutes the AI takes off your hands and the minutes you spend preparing, checking and correcting. Then multiply the difference per case by how often the task comes up. A simple log kept for a few weeks by the people who do the work is enough for a reliable picture.

Why isn't gut feeling enough?

Ask a team after their first month with an AI assistant whether it saves time, and you'll get an impression but no figure. That impression can be wrong in both directions. A draft that appears in seconds feels very fast, even if fixing it afterwards takes a quarter of an hour. On the other hand, savings are often spread over many small moments that nobody consciously notices. Anyone deciding on licences, a rollout or the next step needs figures from real cases. They don't have to be exact, but they do have to be traceable.

Which task should you measure?

Don't try to measure “AI in general”. Measure one specific task that AI changes. Good candidates have these features:

• They come up often, at least several times a week.
• They have a clear start and a clear end.
• The cases are similar enough to compare.
• The same people or the same role do the task before and after.
• The AI supports this one task, rather than half the business at once.

An example: a fiduciary firm answers routine client questions by email, about deadlines, missing receipts or where the tax return stands. Now an assistant drafts the replies, and the person handling them checks and sends them. The task starts when the email is opened and ends when the reply is sent. That's all a clean measurement needs. Our guide to client requests in a fiduciary firm describes how requests like these can be automated.

How do you measure the baseline?

1. Fix the start and end.Write down in one sentence when the task begins and when it ends. Everyone who measures sticks to that definition, or you'll end up comparing apples with oranges.
2. Choose how to measure.The simplest way is a log in which each person notes the start, the end and a short comment on difficulty for every case. Timestamps from your systems help if they reflect the work itself rather than just waiting time.
3. Capture enough cases.A few dozen cases over two to three weeks give a usable picture. Record easy and difficult cases as they come, and don't leave one out because it was a pain.
4. Note the volume.Record how often the task comes up each week or month. Without that figure, you can't scale up the saving per case.
5. Record the quality.Note how often a case needs rework or comes back with a follow-up question. That's your comparison for later.

This measurement doesn't have to be perfect, but it has to happen before you start. Afterwards, nobody remembers reliably how long things used to take. If you want to capture a whole process rather than a single task, see our five-step workflow audit.

How do you measure with AI without fooling yourself?

Once AI is in use, measure the same task with the same definition, the same people and a similar mix of cases. Count everything that belongs to the task:

• The time spent preparing the prompt or the input.
• Waiting time, if someone waits for the result instead of doing something else.
• Checking and correcting the draft.
• Rework that comes back later, for example because a client spotted a mistake.

Give the team time to settle in first. In the early days much takes longer, because everyone is still experimenting. Measure after that, or measure throughout and show the trend. And don't only measure the enthusiasts: if only the AI champions take part, you'll see the best case, which may be far from an ordinary day.

How do you calculate the time saved?

The arithmetic is simple. Take the typical time per case before, subtract the typical time per case with AI, checking included, and multiply the difference by the number of cases per month. That gives you the hours saved each month. From that, deduct the new effort that doesn't fall per case: maintaining templates, spot-checking results, showing new people how it works.

• Use the middle value, the median, rather than the average when a few cases take extremely long. Otherwise those outliers distort the result.
• Give a range: what do easy cases show, and what do difficult ones show? An honest range is more convincing than a round number.
• Compare the quality. If errors or follow-up questions increase, the time hasn't been saved, only moved.
• Write down how you measured. Whoever reads the figure later will want to know where it came from.

Count the minutes spent checking. Otherwise you're measuring the demo instead of the work.

What happens to the time you gain?

Time saved disappears quickly into the daily routine if nobody decides what it's for. So write down what the hours should go to: more time for client conversations, less overtime in the busy season, faster replies to customers. Whether that becomes a saving in francs or freed-up capacity is a separate question, covered in detail in how to build a business case for AI. Set the time you measured against the running costs as well. What AI costs an SME explains which costs to include.

Which mistakes distort the measurement?

• Timing only the AI and forgetting the checking.
• Comparing easy cases with AI against hard cases without it.
• Measuring in the first flush of enthusiasm, or only among people who like trying new things anyway.
• Changing the process at the same time, such as introducing a new template, and crediting the whole improvement to AI.
• Counting seconds on tasks that happen twice a year.
• Turning the measurement into a performance check on individuals. People will then make the numbers look good. Make it clear that you're measuring the task and leaving the person out of it.

What belongs in a one-page report?

Leadership decides best from a single page. Put these points on it:

• The task, with its start and end, and the period you measured.
• How many cases you recorded, before and after.
• The typical time per case before and after, with the range from easy to difficult cases.
• The hours saved per month, after deducting the new effort.
• How quality has developed: rework, follow-up questions, complaints.
• What the team says about working with the tool, and what should happen to the time gained.

If you want to capture several tasks at once, a workflow audit provides the baseline figures. Measuring and scaling is also the last stage of our process: what works gets extended, and what doesn't gets stopped. That's a good outcome too.