Learning and development · working model

Training Needs Model

A fake company of 1,470 people runs for a year with problems hidden in it. The model watches each team's figures, opens a row when something moves, checks the skill, and proposes a cause. Every number below is worked out in the page, and every step shows its working.

How a Need Moves Through the Model

1. WatchEach team's monthly figure against its own normal, adjusted for what the whole department did.

2. Open a rowWhen a team is well above expected for enough months in a row.

3. CheckThe team takes six questions on the skill linked to that figure.

4. DecideAbove the error bar is a skill gap. Otherwise the row goes back to the department.

5. Read againAfter practice in months 7 to 9, the same check and figure in month 12.

Results are shown by team, never by person. A team under the smallest-group size is merged into its whole role. A simulated L&D lead accepts every proposal, which a real person would not always do.

Settings

Company

Each number builds a different random company with the same six planted problems.

Alarm bar: wobbles How far above expected a month must be. One wobble is the normal month-to-month variation for that team.

Months in a row: One month lets single bad months through.

Check error bar: % Most staff get 12% to 22% of check questions wrong.

Smallest group shown: people Smaller teams are merged into their whole role.

Design choices Adjust for the whole department (stops a company-wide change looking like a training need) Improvement: read teams under ten people over three months Improvement: no decision when under 60% took the check

This Company's Year

Needs register

Look at Any Team

Pick any team and figure to see why a row did or did not open.

Cleaning the Messy Data

The monthly figures arrive the way real exports do: team names written more than one way, measure names spelled differently, rows copied twice and months missing. The model cleans them before anything else.

Marking It Against the Hidden Answers

Six problems were planted where the model could not see them. Reveal them to see what it got right.

What This Shows and What It Does Not

The model's logic and privacy rules behave as designed on messy data, and the settings show what each rule costs. Whether the model leads to better training decisions in a real company has not been tested. The staff and roles come from IBM's fictional HR attrition dataset, in which 85% of staff are rated 3 and the rest 4, so the performance rating can point at no skill. The monthly figures and check answers are generated in this page.