← Back to the exerciseThe full worked example: Decide whether one division's programme should go group-wide
campo.
Field guide / L&D / UC-001 · SC-07
The result

"Roll it out group-wide?" Not yet. Pilot it.

A Chief Learning Officer at a five-division industrial group makes the divisions' numbers comparable and tests the case for scaling one division's leadership academy, with any AI chat her company allows.

The resultWhat Sofia gives the CEO. Fictional company.
VARROW HOLDINGS · CEO QUARTERLY REVIEW
Learning across the group · Q4
For: the CEO
QBR, Dec 8 · From: Sofia Lindqvist, CLO

YOUR TWO QUESTIONS

BEST DIVISION
Rail logs the most learning per person. No division measures impact yet.
Rail 19.6 hours per employee; Energy and Automation 18. Fluid Controls counts course completions, not hours, and Specialty Materials hasn't reported. · Inputs 2, 3
ACADEMY
Not group-wide yet. Pilot it in two divisions next year, with results agreed up front.
Its evidence today is satisfaction (4.8 out of 5). Energy runs the pilot with us, and we compare participants with similar managers who didn't attend. · Inputs 1, 3

CORRECTIONS TO THE DRAFT SCORECARDInput 2

11.9 hoursWrong
18.5 for the 71,000 who report hours
Energy leadsWrong
Rail is 19.6
Fluid ControlsNot comparable
Completions not hours
Specialty MaterialsMissing
No report yet

WHAT I NEED FROM YOU

Back the two-division pilot, and ask each division president to name the leadership gap they most need closed. · Input 1
Caught on review
1
"Energy leads the group" is Rail, at 19.6 per person
2
"11.9 hours per employee" is 18.5 for those who report
3
"Promoted at 22%" is nominated high-potentials, not proof
UC-001 · SC-07 · Rev A
FOR
Chief Learning Officer
SECTOR
Diversified industrial
COMPANY
110,000 people · 5 divisions · 30 countries
TASK
Answer the CEO on a group-wide rollout
WRITTEN FOR
ChatGPT · Claude · Gemini · Copilot
AI TEST
Claude, Nov 13, 2026 · 6 of 6 found
Step 0 · The mess

What landed on Sofia's desk

A CEO question, a confident scorecard and five divisional summaries in five different shapes. One division wants its programme rolled out everywhere. The numbers can't answer that yet.

EMAIL · CHIEF OF STAFF[1]
QBR Dec 8, learning
“Which division is doing best on learning, and whether we should roll out Energy Systems' leadership academy across the whole group.”
1
Two questions. One is a big decision.
DRAFT SCORECARD · ANALYST[2]
Group learning
“11.9 hours per employee (1.31M ÷ 110,000)… Energy Systems leads the group.”
2
Divided by people who didn't report hours.
DIVISIONAL SUMMARIES[3]
Five divisions
Hours per person, a completion rate, a total, a promotion rate, and one summary not received.
3
Rail's total hides the highest rate.
You're not alone
56%
of organisations say they can measure the business impact of their learning, though 97% want to.
Watershed, Measuring the Business Impact of Learning (annual survey of L&D professionals). Source
Steps 1–3 · How it works

Three steps, one chat

Here the AI makes unlike numbers comparable, then tests the decision on the table.

1
Open a new chat

Use the AI tool your company approves: ChatGPT, Claude, Gemini or Copilot.

2
Paste the prompt and the summaries

Copy the prompt and the three fictional inputs from the appendix into one message.

3
Answer the question you were asked

Check one division's number yourself. Then give your leader a decision, not just a no.

The prompt, part by part
Copy-ready version in the appendix

I'm the Chief Learning Officer of a diversified industrial group with about 110,000 people in five divisions. Our CEO's quarterly review is on December 8. He wants one page: which division is doing best on learning, and should we roll out one division's leadership academy across the whole group? Below are his chief of staff's email, my analyst's draft scorecard and the five divisional summaries.

1
Who you are and the two questions your CEO asked, in his words.

Please give me:

  1. A comparison table: each division, what it reported, and whether it can fairly be compared with the others.
  2. Corrections to the analyst's draft.
  3. My answer on the academy rollout, and what evidence I would need first.
2
A comparison table, corrections, and your answer on the decision. Three short outputs, one page.

Be skeptical. Check that numbers measure the same thing before you compare or add them, check every average against what it was divided by, and ask whether the evidence actually supports the decision being asked for. If something is missing, say so. Keep it short enough for one page.

3
The heavy lifting: same measure before comparing, check every average's denominator, and ask whether the evidence supports the decision. It found all six problems.
Where to paste
Every tool has one message box. Paste the prompt, add a blank line, paste the inputs, send. Real summaries can be attached as files if your tool allows it.
Safe to try
Practise with the fictional summaries first. Real divisional data may be confidential: use only your company's approved tool.
The AI's first answer

What Claude actually wrote

Exactly as it came back, not edited. Read it the way Sofia did: check each point against the source it cites.

UNEDITED
Claude · fresh session, no other context · Nov 13, 2026 · one run

One-page brief for the CEO QBR on December 8: Learning across the group

1. Comparison table

Division People What it reported Hours per employee Comparable?
Energy Systems 32,000 18 hrs/employee; Academy: 1,240 managers, satisfaction 4.8/5, "best in group" 18.0 (reported) Partly. The hours can be compared only if "learning hour" means the same thing in every division. That isn't confirmed. "Best in group" is Energy's own label and has no supporting data.
Fluid Controls 26,000 94% of mandatory courses completed. Hours not tracked. Unknown No. A completion rate is a different measure from hours, and it covers mandatory courses only. found
Rail & Mobility 21,000 412,000 total hours 19.6 (412,000 ÷ 21,000, my calculation) Yes, on hours, if the definitions match. It reported a total, not a per-person figure.
Industrial Automation 18,000 18 hrs/employee. Leadership grads promoted at 22% vs 9% for other managers. 18.0 (reported) Partly, same caveat on definitions as Energy. The promotion figures are not comparable (see section 3).
Specialty Materials 13,000 Nothing received (ERP migration) Unknown No data. found

Comparable hours data covers 3 of 5 divisions: 71,000 of 110,000 people (65%).

2. Corrections to the analyst's draft

  • "Group learning hours: 1.31 million." This is a three-division total, not a group total: Energy 576,000 (18 × 32,000) + Automation 324,000 (18 × 18,000) + Rail 412,000 = 1,312,000. It leaves out Fluid Controls and Specialty Materials, about 39,000 people. Label it "3 of 5 divisions".
  • "11.9 hours per employee (1.31M ÷ 110,000)." Wrong denominator. The hours come from 71,000 people, but the analyst divided by all 110,000. The correct figure is 1,312,000 ÷ 71,000 = 18.5 hours per employee for the divisions that report hours. We have no group-wide figure. found
  • "Energy Systems leads the group at 18 hours." Not correct. Rail & Mobility works out to 19.6 hours per employee, the highest of the three. Energy and Automation are tied at 18. And even the three-way ranking only holds if each division counts a "learning hour" the same way (for example, whether compliance modules are included). That has not been checked. found
  • "Recommendation: roll out the Energy Leadership Academy group-wide." Nothing in the draft supports this. Learning hours measure how much time people spend learning, not results, and nothing in the data links the academy to any business outcome.
  • The CEO's question, "which division is doing best on learning?", can't be answered from these inputs. Two divisions have no comparable data, and the data we do have measures activity, not impact. The honest answer: "Rail logs the most hours per person among the three divisions that track hours. Nobody can yet say which division is learning most effectively."

3. My answer on the academy rollout

Recommendation: not yet. Don't approve a group-wide rollout on December 8. A: reframed

Why not:

  • The academy's only evidence is satisfaction (4.8/5). That tells us participants liked it. It doesn't show that they lead better or that their teams perform better. We also don't know how many of Energy's managers the 1,240 represent, what it cost, or how many people answered the survey. found
  • The strongest outcome data in the pack belongs to a different programme. Automation's 22% vs 9% promotion rate is the only result-type evidence, and it's Automation's programme, not Energy's. It also doesn't show the programme causes promotions. Participants come from the high-potential pool, so they were already more likely to be promoted. Comparing them with all other managers compares two different groups. found
  • Nobody has stated the problem. A group-wide leadership academy is a solution. Before choosing it, we should know what leadership gap it is meant to close, in which divisions, and whether training is the right fix for that gap rather than, for example, role design, spans of control or the promotion process. Energy pushing hard is not evidence of need.

Evidence I'd want before deciding: B: cut

  1. A clear problem statement: which leadership outcomes are weak, where, and how we know.
  2. For Energy's academy: participation as a share of eligible managers, cost per participant, and at least one outcome measure (team engagement or retention, 360 feedback, internal promotion), compared against a similar group of non-participants.
  3. For Automation's programme: promotion rates compared with high-potentials who did not attend, not with all managers.
  4. One shared definition of a "learning hour", plus Fluid Controls hours and Specialty Materials data, so a group view becomes possible.
  5. If the case still looks promising, a controlled pilot in one or two other divisions with outcomes agreed in advance, reviewed before any group-wide commitment.

What to tell the CEO in one line: "We can't yet name a best division or justify a group-wide academy. Here's what we'll have by [date] to decide properly." C: answered

foundcaught a planted problemA–CSofia changes it; see the next section
The human part

What Sofia changed, and why

Claude found every planted problem and named the bigger one: nobody had said what the academy was for. Sofia's edits are about giving a CEO a decision and keeping a division on side.

What Claude found: 6 of 6 planted problems
One run
✓Average divided by the wrong headcount
✓Rail's total hid the top rate
✓Completions aren't hours
✓One division missing
✓Academy evidence is satisfaction only
✓Promotions compare different groups
Three edits, all judgment calls
A
CLAUDE WROTE
“Recommendation: not yet. Don't approve a group-wide rollout on December 8.”
SOFIA'S VERSION
“Not group-wide yet. Pilot it in two divisions next year, with results agreed up front.” Energy runs the pilot with Sofia's team.
WHY · AUDIENCE
The same caution, but the CEO gets a decision and Energy gets a role instead of a refusal.
B
CLAUDE WROTE
Five pieces of evidence to collect before deciding.
SOFIA'S VERSION
Two: the leadership gap each division needs closed, and a comparison with similar managers who didn't attend.
WHY · LENGTH
The CEO is approving a pilot, not a research programme.
C
CLAUDE WROTE
“We can't yet name a best division or justify a group-wide academy.”
SOFIA'S VERSION
“Rail logs the most learning per person. No division measures impact yet.”
WHY · AUDIENCE
Answer the question that was asked with what is known, then state the limit. Leading with “can't” sounds like no answer.
What only you can do
Turn a no into a decision a CEO can take, keep the division that's pushing on side, and decide what evidence is worth collecting. The AI tested the case. You make the call.
Your turn

Make it yours

Use it whenever someone wants to scale a programme, a pilot or a practice from one part of the business to all of it.

Your prompt: fill in the brackets

I'm the [your role] of [your organization] with about [number] people in [number] divisions or regions. [Your leader] wants [one page / an answer] on [date]: [their question], and should we [the proposed rollout]? Below are [their request, any scorecard, and each unit's summary].

Please give me: 1. A comparison table (each unit, what it reported, and whether it can fairly be compared with the others); 2. Corrections to [the scorecard or draft]; 3. My answer on [the rollout], and what evidence I would need first.

Be skeptical. Check that numbers measure the same thing before you compare or add them, check every average against what it was divided by, and ask whether the evidence actually supports the decision being asked for. If something is missing, say so. Keep it short enough for one page.

Five checks before you send
1
Every unit in your table is measured the same way, or flagged.
2
Each average is divided by the people it actually counts.
3
Satisfaction scores aren't presented as results.
4
Comparisons between groups ask who was chosen for which group.
5
Your answer to the decision is a step forward, not just a no.
Make it repeatable (optional)
Save the prompt once and reuse it for any rollout decision: a custom GPT in ChatGPT, a project in Claude, a Gem in Gemini, or an agent in Copilot. Some companies switch these off; copy and paste always works.
With real division data
Divisional results may be confidential. Use only the AI tool your company approves.
What this example shows
One run of this prompt in Claude found all six problems planted in fictional summaries and invented nothing.
What it doesn't show
That every tool, setting or run will do the same, or how much time it saves you. Varrow Holdings and its people are fictional.
Appendix

Copy-ready: the prompt and the three inputs

Copy the prompt, then all three inputs, into one message in a new chat. Everything here is fictional and safe to paste.

The prompt

I'm the Chief Learning Officer of a diversified industrial group with about 110,000 people in five divisions. Our CEO's quarterly review is on December 8. He wants one page: which division is doing best on learning, and should we roll out one division's leadership academy across the whole group? Below are his chief of staff's email, my analyst's draft scorecard and the five divisional summaries.

Please give me:

  1. A comparison table: each division, what it reported, and whether it can fairly be compared with the others.
  2. Corrections to the analyst's draft.
  3. My answer on the academy rollout, and what evidence I would need first.

Be skeptical. Check that numbers measure the same thing before you compare or add them, check every average against what it was divided by, and ask whether the evidence actually supports the decision being asked for. If something is missing, say so. Keep it short enough for one page.

[1] Email from Paul Okonkwo, Chief of Staff to the CEO

Subject: QBR Dec 8, learning

Sofia, one page please. The CEO wants to know which division is doing best on learning, and whether we should roll out Energy Systems' leadership academy across the whole group next year. Energy is pushing hard for it.

[2] Draft scorecard from the group L&D analyst
  • Group learning hours: 1.31 million
  • 11.9 hours per employee (1.31M ÷ 110,000)
  • Energy Systems leads the group at 18 hours per employee
  • Recommendation: roll out the Energy Leadership Academy group-wide
[3] Divisional summaries (as received)

Energy Systems (32,000 people): 18 learning hours per employee. Leadership Academy: 1,240 managers this year, satisfaction 4.8 out of 5, "best in group".

Fluid Controls (26,000): 94% of mandatory courses completed. Learning hours not tracked.

Rail & Mobility (21,000): 412,000 learning hours in total.

Industrial Automation (18,000): 18 learning hours per employee. Leadership programme graduates promoted at 22% this year, against 9% for other managers. Participants are nominated from the high-potential pool.

Specialty Materials (13,000): summary not received (ERP migration).