← Back to the guideThe full worked example: Report training against a union agreement, fairly to both sides
campo.
Field guide / L&D / UC-001 · SC-12
The result

"106% of target"? Ulm is short, Toledo in doubt, Kassel missing.

An L&D Director at a 16,000-person automotive supplier checks HR's draft against the collective agreement and brings the joint training committee a page union and management can both rely on, with any AI chat her company allows.

The resultWhat Katrin brings to the joint training committee. Fictional company.
DRAVIK AUTOMOTIVE · JOINT TRAINING COMMITTEE · NOV 26
Training against the agreement, January to September
For: union and management members of the committee
From: Katrin Vogel

WHERE WE STAND

The agreement gives each production employee 16 paid hours a year; 12 is the pace for nine months. Four plants are on pace. Ulm is short. Toledo is short too unless its 6,800 home hours were paid time. Kassel hasn't reported. We can't yet see hours per person. · Inputs 1, 2, 5

BY PLANT, HOURS PER EMPLOYEE (PACE 12)

Bremen, Leipzig, Spartanburg, Lansing13.0 each
On pace (averages)
Ulm8.0
Night shift offered no slots
Toledo11.0 paid time
15.0 incl. 6,800 home hours
KasselUnknown
Data not received

WHAT WE COMMIT TO

ULM
Training slots inside night-shift hours
Agreed with the Ulm works council before December 31. This is our answer to the October 30 grievance. · Input 4
TOLEDO
Confirm whether the home hours were paid
Not counted until confirmed. Any e-learning counts only when it's done in paid working time. · Inputs 1, 2
ALL PLANTS
Kassel's data, then hours per employee
Kassel by the committee's agreed date. Per-person reporting so we can see who is short of 16. · Inputs 2, 5
Caught on review
1
"106%" leaves out Kassel and counts Toledo's home hours
2
Ulm is at 8 hours against 12, hidden inside the total
3
"Train anytime, anywhere" doesn't fix a night-shift scheduling grievance
UC-001 · SC-12 · Rev A
FOR
L&D Director
SECTOR
Automotive supplier
COMPANY
16,000 people · 7 plants · Germany, US
TASK
Report training against the collective agreements
WRITTEN FOR
ChatGPT · Claude · Gemini · Copilot
AI TEST
Claude, Nov 13, 2026 · 6 of 6 found
Step 0 · The mess

What landed on Katrin's desk

The agreement's clause, a table of hours, HR's upbeat draft, a works council grievance and a note from the LMS team. HR's total is right. What it's adding up is the problem.

AGREEMENT CLAUSE[1]
Training entitlement
“Each production employee is entitled to 16 hours … during paid working time.”
1
Each person; paid time.
HOURS EXPORT[2]
By plant
Kassel: data not yet received. Ulm 9,600 hours for 1,200 people. Toledo includes 6,800 hours at home.
2
Three plants to look at.
HR DRAFT[3]
Committee summary
“106% of our nine-month target!” … “train anytime, anywhere.”
3
Only plants that reported.
GRIEVANCE · ULM[4]
Night shift
“Night-shift employees … are not being offered training slots.”
4
A scheduling problem.
NOTE · LMS TEAM[5]
Reporting
“We can't yet report hours per employee.”
5
The clause is per person.
You're not alone
1 in 3
frontline workers work outside standard hours, including 9% on night shifts and 25% on rotating schedules.
JLL Workforce Preference Barometer 2025 (3,411 frontline workers in 26 markets), published January 2026. Source
Steps 1–3 · How it works

Three steps, one chat

Here the AI checks a report that union and management will both read. The prompt asks it to be fair to both.

1
Open a new chat

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

2
Paste the prompt and the reports

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

3
Decide what you'll commit to

Check the plant figures yourself, then pick the few commitments you can keep and say them plainly.

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

I'm L&D Director at an automotive supplier with plants in Germany and the US. Each quarter I report training against our collective agreements to the joint union-management training committee. The next meeting is November 26. Below are the training clause, hours by plant, a draft summary from HR, a works council grievance and a note from our LMS team.

1
Who you are, who reads the report, and what you were sent.

Please give me:

  1. Are we meeting the agreement? A short, honest answer.
  2. A table by plant: hours per employee against the target so far, and anything that affects whether the hours count.
  3. Does the proposed fix address the grievance? Yes or no, and why.
  4. What I should commit to at the meeting.
2
An honest answer, a plant table, a yes or no on the proposed fix, and commitments. The yes or no makes it test the fix against the grievance.

Union and management will both read this closely, so it must be accurate and fair to both sides. Check that every hour counted meets the agreement's terms, that missing data isn't treated as fine, and that totals don't hide a plant that's behind. Say "Unknown" where we can't tell. Don't add facts that aren't in the inputs.

3
The heavy lifting: every hour must meet the agreement's terms, missing data isn't fine, and no total hides a plant. It found all six problems.
Where to paste
Every tool has one message box. Paste the prompt, add a blank line, paste the reports, send. Exports can be attached as files if your tool allows it.
Safe to try
Practise with the fictional data first. Real grievances and agreement terms are 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 Katrin did: check each point against the source it cites.

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

1. Are we meeting the agreement?

We can't say yes. On the evidence we have, the answer is "not everywhere, and we can't confirm the rest."

  • The agreement gives each production employee 16 hours a year, delivered during paid working time. It is an individual entitlement. We only have plant totals, and the LMS team says it can't yet report hours per employee [5]. So we don't know whether every employee is on track, even at plants whose average looks fine. found
  • Ulm is clearly behind: 8.0 hours per employee against 12 for the period. The works council also says night-shift employees aren't being offered training at all [4]. found
  • Kassel (1,500 employees) has not reported. Its status is Unknown. found
  • Toledo includes 6,800 hours of e-learning done at home in the evenings. The inputs don't say whether those hours were paid. If they weren't, they don't count, and Toledo drops from 15.0 to 11.0 hours per employee, which is below the pace. found
  • HR's "106% of target" is correct arithmetic, but it leaves Kassel out of the denominator, counts the Toledo home hours, and lets the plants that are ahead hide Ulm. I don't think it should go to the committee in its current form. found

Note: the agreement sets 16 hours per calendar year. The 12-hour figure for January to September is HR's pro-rata pace. It is a reasonable checkpoint, but it isn't in the clause. Formally, nobody misses the entitlement until December 31.

2. Hours per employee by plant (January to September)

Pace target: 12 hours per employee (16 × 9/12, HR's figure). Averages are plant totals divided by headcount. They are not per-person figures.

Plant Employees Hours reported Hours per employee vs 12-hour pace What affects whether the hours count
Bremen 2,400 31,200 13.0 +1.0 Average only. Individual coverage Unknown.
Leipzig 1,800 23,400 13.0 +1.0 Average only. Individual coverage Unknown.
Kassel 1,500 Not received Unknown Unknown No data. Must not be treated as on track.
Ulm 1,200 9,600 8.0 −4.0 Behind. Grievance: night shift not offered any slots, so some employees may be at or near zero. Night-shift headcount Unknown.
Spartanburg 2,100 27,300 13.0 +1.0 Average only. Individual coverage Unknown.
Toledo 1,700 25,500 (18,700 without home e-learning) 15.0 as reported; 11.0 without the 6,800 home hours +3.0, or −1.0 6,800 hours were done at home in the evenings. Whether that was paid working time is Unknown. If it wasn't paid, those hours don't meet the clause.
Lansing 1,300 16,900 13.0 +1.0 Average only. Individual coverage Unknown.

Totals, shown three ways:

  • HR's version (6 reporting plants, Toledo as reported): 133,900 ÷ (10,500 × 12 = 126,000) = 106%
  • Without the Toledo home hours: 127,100 ÷ 126,000 = 101% found
  • Without the Toledo home hours, with Kassel's 1,500 employees in the denominator (hours Unknown): 127,100 ÷ (12,000 × 12 = 144,000) = 88% so far, Kassel still to come A: Kassel isn't zero

Even the 101% figure hides Ulm at 67% of pace (8.0 ÷ 12) and possibly Toledo at 92% (11.0 ÷ 12).

3. Does the proposed fix address the grievance?

No.

  • The grievance asks for training scheduled within night-shift hours [4]. The problem is that no slots are offered on the night shift. Fixing that is a scheduling job, and it doesn't need a new delivery channel. found
  • "Mobile e-learning so people can train anytime, anywhere" [3] doesn't commit to any slots within paid night-shift time. In practice it invites more training outside paid hours, which is exactly the problem with Toledo's evening e-learning at home. The union is likely to read it as moving training into employees' own time, and the clause wouldn't count those hours anyway.
  • E-learning could be part of the answer only if it is done during paid working time on the shift. The proposal as written doesn't say that.

4. What to commit to at the meeting B: three commitments

  1. Corrected figures. Withdraw the 106% headline. Show the plant table above, with Kassel marked Unknown and Toledo shown both ways. C: lead with the facts
  2. Kassel data. Commit to getting Kassel's hours and reporting them to the committee by a set date. (No date is in the inputs, so the committee should agree one.)
  3. Toledo home e-learning. Confirm whether the 6,800 hours were paid working time. Until that's confirmed, don't count them towards the agreement. If they weren't paid, say how the company will handle it.
  4. Ulm night shift. Respond to the October 30 grievance by scheduling training within night-shift hours. Agree with the Ulm works council how many night-shift employees are affected (Unknown today) and a plan to close Ulm's gap before December 31.
  5. Per-employee reporting. Ask the LMS team to report hours per employee, so the committee can see who is short of 16 hours, not just plant averages. Agree a target date with them.
  6. Rework the mobile e-learning proposal. Any e-learning, mobile or otherwise, must be done and recorded during paid working time to count under the agreement. Bring it back only on that basis.
  7. Year-end check. Report at the next meeting on whether each employee will reach 16 hours by December 31, by plant and by shift.

Unknowns to state openly: Kassel's hours; whether Toledo's home hours were paid; Ulm's night-shift headcount and hours; how hours are spread across individual employees at every plant.

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

What Katrin changed, and why

Claude found every planted problem and was careful not to assume Toledo's home hours were unpaid. Katrin removed one figure that treats missing data as zero, and shaped the rest for a room with two sides.

What Claude found: 6 of 6 planted problems
One run
✓Kassel left out of 106%
✓Toledo's home hours may not count
✓Ulm behind inside the total
✓Entitlement is per employee
✓Fix doesn't match the grievance
✓106% shrinks to 101%
Three edits, one catch and two judgment calls
A
CLAUDE WROTE
“With Kassel's 1,500 employees in the denominator (hours Unknown) … 88% so far”
KATRIN'S VERSION
Removed. Kassel shows as Unknown, not as zero.
WHY · ACCURACY
Dividing by Kassel's people without their hours counts them as having had no training. The prompt asked not to treat missing data that way.
B
CLAUDE WROTE
Seven commitments for the meeting.
KATRIN'S VERSION
Three: Ulm night-shift slots, Toledo's paid-time check, then Kassel's data and per-person reporting.
WHY · LENGTH
A joint committee will hold her to every promise. Three she can keep beat seven she can't track.
C
CLAUDE WROTE
“Withdraw the 106% headline.”
KATRIN'S VERSION
“Four plants are on pace. Ulm is short. Toledo is short too unless its home hours were paid time. Kassel hasn't reported.”
WHY · AUDIENCE
Union and management both read it. Plain facts, good and bad, land better than announcing that HR's number is withdrawn.
What only you can do
Read the room, promise only what you can deliver, and give the union a direct answer to its grievance. The AI checked the hours. You keep the trust.
Your turn

Make it yours

Use it whenever you report against an agreement, policy or commitment to people who will check your work: unions, works councils, regulators, customers.

Your prompt: fill in the brackets

I'm the [your role] at a [industry] company with [sites] in [countries]. Each [period] I report [what] against [the agreement or commitment] to [the committee or audience]. The next meeting is [date]. Below are [the clause, the data by site, any draft summary, complaints and system notes].

Please give me: 1. Are we meeting [the agreement]? A short, honest answer. 2. A table by [site]: [the measure] against the target so far, and anything that affects whether it counts. 3. Does the proposed fix address [the complaint]? Yes or no, and why. 4. What I should commit to at the meeting.

[Both sides] will read this closely, so it must be accurate and fair to both sides. Check that every [unit] counted meets the agreement's terms, that missing data isn't treated as fine, and that totals don't hide a [site] that's behind. Say "Unknown" where we can't tell. Don't add facts that aren't in the inputs.

Five checks before you send
1
Every hour or unit counted meets the agreement's exact terms.
2
Missing sites show as Unknown, not zero and not fine.
3
No total hides a site that's behind.
4
The proposed fix answers the actual complaint.
5
You can keep every commitment you make.
Make it repeatable (optional)
Save the prompt once and reuse it every quarter: 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 data
Grievances, agreement terms and employee records are confidential. Use only the AI tool your company approves, and follow its rules.
What this example shows
One run of this prompt in Claude found all six problems planted in fictional reports 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. Dravik Automotive and its people are fictional.
Appendix

Copy-ready: the prompt and the five inputs

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

The prompt

I'm L&D Director at an automotive supplier with plants in Germany and the US. Each quarter I report training against our collective agreements to the joint union-management training committee. The next meeting is November 26. Below are the training clause, hours by plant, a draft summary from HR, a works council grievance and a note from our LMS team.

Please give me:

  1. Are we meeting the agreement? A short, honest answer.
  2. A table by plant: hours per employee against the target so far, and anything that affects whether the hours count.
  3. Does the proposed fix address the grievance? Yes or no, and why.
  4. What I should commit to at the meeting.

Union and management will both read this closely, so it must be accurate and fair to both sides. Check that every hour counted meets the agreement's terms, that missing data isn't treated as fine, and that totals don't hide a plant that's behind. Say "Unknown" where we can't tell. Don't add facts that aren't in the inputs.

[1] Training clause from the collective agreements (same wording in Germany and the US)

Each production employee is entitled to 16 hours of training per calendar year, delivered during paid working time.

[2] Training hours by plant, January to September
Plant Production employees Training hours delivered
Bremen 2,400 31,200
Leipzig 1,800 23,400
Kassel 1,500 data not yet received
Ulm 1,200 9,600
Spartanburg 2,100 27,300
Toledo 1,700 25,500
Lansing 1,300 16,900

Toledo note: includes 6,800 hours of e-learning completed at home in the evenings.

[3] Draft summary from the HR business partner team

We delivered 133,900 training hours, 106% of our nine-month target! (Target: 12 hours per employee for January to September, at the plants that reported.) Proposal for the committee: roll out mobile e-learning so people can train anytime, anywhere.

[4] Grievance from the Ulm works council, filed October 30

Night-shift employees at Ulm are not being offered training slots. All courses are scheduled on day shift. We ask the company to schedule training within night-shift hours.

[5] Note from the LMS team

We record training hours by plant. We can't yet report hours per employee.