# Skill-matrix gap analysis

By Niels Langereis. A Monday-morning tool for finding where AI quietly stopped forming your next senior.

Download the Excel workbook: https://nielslangereis.com/downloads/skill-matrix-gap-analysis.xlsx

## What it is

A competency matrix with one extra move. You take the grid you may already have, skills and behaviours down the side, seniority levels across the top, and you run a gap analysis over every cell, sorting the work into piles: work AI can now do with senior oversight, work AI now runs end to end on its own, and work that always needs a human touch. Most teams run this to find efficiency. The point of this version is to run it for the opposite purpose, to find where the executional reps that used to form a junior have disappeared, and to put the training back exactly there.

## How to use it

- 1. Open the 'Gap analysis (example)' tab to see a worked example for a generic analyst role.
- 2. Copy the 'Blank template' tab and rename it for one of your own roles.
- 3. For each competency, describe what it looks like at execution, tactical and strategic level.
- 4. In 'Who does it now', pick 'AI + oversight', 'AI autonomous' or 'Human'. Set 'Bench (people deep)' to how many people can do it at the strategic level.
- 5. Status flags itself: a human-touch cell only one person deep shows 'At risk'; work AI runs autonomously shows 'Reps gone'. Point the second loop at the 'At risk' rows first.

## Status rule

If who_does_it_now is "AI + oversight": "—" (a person still reviews it). If "AI autonomous": "Reps gone" (AI runs it end to end; the formation that happened in reviewing it has gone too). If "Human": bench <= 1 "At risk"; bench = 2 "Thin"; bench >= 3 "OK".

## How to read it

- Green: AI can now do it with senior oversight. Low formation value; reclaim the junior time.
- Amber: always needs a human touch. Concentrate and protect junior hours here.
- Red "At risk": human-touch judgement you are only one person deep on. Point the second loop here first.
- Sand "Reps gone": work AI now runs end to end with nobody reviewing it. Ask where juniors get these reps now; sometimes they no longer need them, often they get them nowhere.

## Worked example: Analyst (generic, anonymised)

| Category | Competency | Execution | Tactical | Strategic | Who now | Formation action | Bench | Status |
|---|---|---|---|---|---|---|---|---|
| Core analytical work | Data gathering & cleaning | Pull and clean source data | Choose sources, spot anomalies | Design the data model itself | AI autonomous | Low formation value here. Let AI run it and reclaim the junior time for the cells below. | 4 | Reps gone |
| Core analytical work | Standard analysis & variance | Run the standard cuts | Interpret, flag what matters | Decide which questions are worth asking | AI + oversight | Route a sample of interpretations to juniors so the reading skill stays alive. | 3 | — |
| Core analytical work | Drafting commentary | Write the standard write-up | Tailor tone and emphasis | Shape the narrative around the decision | AI + oversight | Juniors edit and critique AI drafts against a rubric, rather than write from scratch. | 3 | — |
| Judgement work | Borderline-case assessment | (mostly gone) | Flag the awkward case | Make the call with no clean answer | Human | Core second-loop cell: shadow the decision, junior rules first, then compares to the system. | 1 | At risk |
| Judgement work | Challenging assumptions | (mostly gone) | Question the inputs | Know which assumption is load-bearing | Human | Surface borderline assumptions on purpose and have juniors adjudicate them. | 2 | Thin |
| Judgement work | Scenario / framework design | (n/a) | Apply an existing framework | Design the evaluation itself | Human | Pair junior with senior on framework reviews; let them see the standard get revised. | 1 | At risk |
| Judgement work | Stakeholder reconciliation | (n/a) | Gather the competing views | Reconcile competing analyses into one defensible call | Human | Rotate juniors through reconciliation rooms, observer first, then participant. | 1 | At risk |
| Judgement work | Final recommendation & accountability | (n/a) | Prepare the recommendation | Own the call and carry the consequence | Human | Graduated ownership: juniors sign the smaller calls first, scope widens as the bench proves out. | 2 | Thin |

## Credit and provenance

The matrix idea is adapted from the one Blendle open-sourced in its employee handbook, built by Roland Grootenboer, their first HR hire (categories: work, drive, fit). The gap-analysis use, the two-pile split and the formation column are the addition. The structural reason the human pile cannot be automated is drawn from Quattrociocchi, Capraro and Perc, 'Epistemological Fault Lines Between Human and Artificial Intelligence' (arXiv:2512.19466, 2025). The instrumentation language, levels of independence and bench strength, follows Amodiovalerio Verde's PAIR model (the-thinking-lens.com).