The same survey runs three times, in week one, at day 45, and in the final week, so the change is the measurement.
Every figure on this page is made up. It exists to show the format, not to report anyone's results. There is no real firm behind it and no client data on it. I will not put a number in front of you that I cannot trace.
The headline
Four hours a week came back.
4 hrs
back per person per week, week 1 to final
83%
weekly active, was 25% in week 1
8%
shadow AI, was 42% in week 1
4.1
confidence out of 5, was 2.4 in week 1
Each person estimates the hours a normal week takes on five kinds of work.
Week 1Day 45Final
Status reports and client updates
2.5
1.5
Reconstructing a project from email
2.0
1.0
Meeting minutes and action items
1.5
0.75
Bid leveling and budget checks
1.5
1.0
Formatting documents to the template
1.0
0.25
Week 1
8.5
Day 45
6.25
Final
4.5
Status report drafts and email catch-up fell the most, and both were built in the first month. Across a team of twelve, the final-wave difference is 48 hours a week.
Sample · hours per person per week, three survey waves
Sample: hours per person per week, week 1 and final wave
Where the week goes
Week 1
Final
Status reports and client updates
2.5
1.5
Reconstructing a project from email
2.0
1.0
Meeting minutes and action items
1.5
0.75
Bid leveling and budget checks
1.5
1.0
Formatting documents to the template
1.0
0.25
Sample: total hours per person per week, three waves
Total
Week 1
Day 45
Final
Per person, per week
8.5
6.25
4.5
Sample figures, and self-estimates. In a live engagement they sit beside at least one measured artifact time, usually the flagship report's turnaround.
Adoption and judgment
The team moved onto sanctioned tools.
Adoption
Week 1Final
Weekly active25%83%Using the tools daily or a few times a week.
Shadow AI42%8%Consumer tools used for work. It should fall as sanctioned tools land, and when it does not, that is the finding.
Sample · responded: 12 of 12 in week 1 and the final week
Sample: adoption, week 1 and final wave
Measure
Week 1
Final
Weekly active, daily or a few times a week
25%
83%
Shadow AI, consumer tools used for work
42%
8%
Responded
12 of 12
12 of 12
Confidence
People who can spot a wrong answer are the reason any of this is safe to use on client work.
Week 1Final
I could get a useful first draft out of an AI tool today
152.64.3
I can tell when an AI answer is wrong or invented
152.24.0
I know which tasks I should not hand to an AI tool
152.44.1
Sample · confidence, average of 1 to 5
Sample: confidence, average of 1 to 5, week 1 and final wave
Statement
Week 1
Final
I could get a useful first draft out of an AI tool today
2.6
4.3
I can tell when an AI answer is wrong or invented
2.2
4.0
I know which tasks I should not hand to an AI tool
2.4
4.1
The rules
How the numbers are handled.
Nobody is identifiedIndividual answers stay with me. No group smaller than four people is ever reported on its own.
Misses get reportedWorkflows that did not earn their keep show up in the report too.
Your version has your numbers in it.
The format is the same. The figures come from your team, in week one, before anything is built.