Some AI wins won't fit in a spreadsheet, and that's fine.
Six months into using AI, most teams can point to one or two results with a number attached. A quote goes out a day sooner, or the backlog of unread intake forms is gone. They can also point to a longer list of things that clearly got better and resist counting: the new coordinator found her footing faster, the Monday handoff stopped dropping tasks, and people who were nervous about the tools now trade prompts at lunch. When the quarterly review comes around, the second list usually gets left out, or it gets dressed up as a number it can't support.
Two kinds of value, two kinds of question.
The countable wins answer a narrow question: is this particular use of AI working? You pick one workflow, measure it before and after, and report the change. That is the right standard for anything with its own budget line, and it is the standard finance will hold you to.
The harder-to-count wins answer a broader one, which is whether the organization is getting better at using these tools at all. That question matters because the countable wins come out of it. A team that has spent a few months trying things, failing at some and talking about it, is the team that eventually spots the workflow worth measuring. A team that only ever ran the one approved pilot rarely finds the second.
Why the soft wins get dropped.
Leaders leave the soft wins out because they are afraid of sounding vague in a room that wants figures. That fear is reasonable. A slide that says "morale is up" invites an eye roll, and the person presenting knows it.
So the soft wins either vanish or get converted. Someone estimates that better onboarding saved "about forty hours," multiplies by an hourly rate, and adds the result to the total. Now a guess sits in the same column as a measured result, and the first time a board member or finance lead pulls on it, the whole report loses credit, including the parts that were solid.
A guess placed next to a measurement makes the measurement look like a guess too.
Report them side by side, and keep them separate.
The fix is a report with two columns that never borrow from each other. The left column holds what you measured: a named workflow, the number before, the number after, and where the number came from. The right column holds what you observed: specific changes, described plainly, with who saw them and when.
The right column has its own standard of evidence, and it is stricter than people expect. "The team is more confident with AI" is too loose to count. "In July, three people on the front desk started drafting their own reply templates without being asked, and the shared folder now has eleven of them" can be checked by anyone who opens the folder. It has a date and a source, and it makes no claim about money.
What counts as a good observation.
Good observations tend to describe what people did. Someone did something they didn't do before, a recurring problem stopped recurring, or a question that used to land on one overloaded person now gets answered by a teammate. A school district office might note that the new hire wrote her first board memo draft in a week instead of the usual month of shadowing. A clinic might note that the referral follow-up list hasn't had a forgotten patient since the summary step was added, which is a fact even before anyone calculates what a forgotten referral costs.
Quotes from staff are fine in this column, as long as they are short and named by role. Surveys are fine too, if you ran one; report what people said without turning it into a satisfaction score that implies more precision than a dozen responses can carry. What doesn't belong here is a forecast. "This will save us a hire next year" is a claim about the future, and it should wait until it can move to the left column.
Promote observations when they earn it.
The two columns aren't permanent homes. Some observations will turn out to point at something you can measure. If the front desk keeps building reply templates, at some point it becomes worth tracking response time on the inbox they cover, and that item moves left with a proper baseline.
Other observations stay on the right for good, and that is acceptable. Fewer dropped balls in a weekly handoff may never become a clean figure, because the dropped balls were never counted in the first place. Reporting them honestly, quarter after quarter, builds a record that a reader can follow. Over a year, a director can see that the right column keeps feeding the left one, which is the strongest case for continuing to give people room to experiment.
A two-column report you can fill in this week.
Use one page per quarter. Keep the columns side by side so nobody reads one without seeing the other.
Left column: what we measured
- Name the workflow in one line (for example, "drafting replies to permit questions").
- Give the number before AI, with the date range and the system it came from.
- Give the number after, from the same system and the same kind of date range.
- State the difference, and stop there. Don't convert it to dollars unless finance agrees with the method.
Right column: what we observed
- Describe one change in behavior or outcome, in a sentence a skeptic could check.
- Say who noticed it, by role, and roughly when.
- Point to the evidence: a folder, a thread, a log, a calendar, a named person to ask.
- Mark it as a candidate to measure next quarter, or as one that will stay descriptive.
At the bottom, add one line that lists any right-column items that moved left since the last report. If that line has an entry most quarters, you have your answer to whether the experimenting is paying off.
