Measure more than time saved

Time saved is the easiest number to get wrong

Time saved is the number everyone asks for, and the easiest one to get wrong. People remember the fast drafts and forget the minutes spent checking and fixing. A tool can save ten minutes of typing and add twelve minutes of checking.

A tool can also save time while quietly letting errors through, or save time for the one person who uses it while everyone else carries on the old way. So measure time honestly, as minutes per task including review, compared with the old way. Then look at four more numbers.

Four numbers beyond time

Together they answer four questions: is the output good, is the check catching things, are people using it, and is the business better off?

• Overrides: how often a person edited or threw away the output instead of using it as it was. A high rate means the tool isn’t doing the job yet, however fast it is. Watch the trend, too: you want it falling as people learn what to give it.
• Errors caught at review: what the approver fixed, and how serious it was. Count red-line catches separately. These are the mistakes that would have gone out without a person checking.
• Adoption: of all the tasks that came in, how many went through the tool. If people drift back to the old way in week two, find out why before anything else.
• The business number: the result the workflow exists for. For quote replies, that’s same-day responses or quotes that become booked jobs. For a bakery’s wholesale orders, it’s orders entered right the first time. For invoice reminders, it’s days until paid.

Get a baseline, and set the bar first

A number means little without a baseline: the same measure for the old way, before the tool. Collect time per task (an estimate from the people who do it is fine, if you label it as one), how many tasks came in, and last month’s business number.

Then write down, before the pilot ends, the result you need on each measure to keep the tool. For example, and yours will differ: fewer than a third of outputs overridden by week two, no red-line errors past review, most tasks going through the tool, and the business number no worse than before. Writing it first stops you picking the flattering numbers afterward.

Let formulas do the counting

Count with spreadsheet formulas, not by asking a chatbot to total your log. Assistants can get arithmetic wrong and sound sure about it, while a formula shows its work and every number traces back to a cell. Use AI for what formulas can’t do: writing the formulas for you, and grouping the approvers’ notes into themes. Then check each formula against a few rows you count by hand. If a number looks too good, check the rows behind it before you celebrate.