Skill Matrices and Capability Models: From Job Descriptions to a Real Growth System

A capability model is a structured, shared answer to two questions: 'what does great look like at each level of this role?' and 'how do we grow into the next one?'. Without it, promotions feel political, hiring loops are inconsistent, and learning plans are wishful thinking. With it, growth becomes a system.

Why capability models exist

Calibration — same word, same meaning across managers (especially for 'senior', 'staff', 'principal').
Hiring — interview rubrics map directly to capabilities; no more 'I liked them' debriefs.
Promotion — evidence-based, not popularity-based; reduces bias measurably.
Learning — every person knows the 2–3 capabilities they're growing this cycle.
Talent reviews — succession and bench strength become legible.

Anatomy of a good model

Five components every model needs

1
Role families
Group similar roles (IC engineering, design, product, sales, etc.). Avoid one mega-model.
2
Levels
5–8 levels typically. Each level has a 1-sentence definition that fits on a slide.
3
Capabilities
8–15 named capabilities per family. Outcome-led ('ships systems that others build on') not task-led ('writes code').
4
Behavioral anchors
For each capability × level, 1–2 concrete behaviors observable at work.
5
Evidence types
What proof counts — work samples, peer feedback, business outcomes, written artifacts.

The single best book on this

Camille Fournier's The Manager's Path and the open-source 'Progression.fyi' library (Monzo, Medium, Songkick ladders) are the most-cited references in modern engineering ladder design.

Building yours in 30 days

Week 1 — Pick one role family and interview your top 3 performers + 2 weakest. Look for what differentiates.
Week 2 — Draft 8–12 capabilities. Test each against 'would I cut someone with a 1, promote someone with a 5?'
Week 3 — Write behavioral anchors per level. Pressure-test with 3 managers and 3 ICs.
Week 4 — Calibrate: rate 6 real employees against the draft. If two managers disagree by >1 level, the anchors are unclear — rewrite.

Rating scales that hold up

Two scales that work — pick one and stick to it

Scale	Levels	When to use	Risk
5-point proficiency	Awareness · Working · Practicing · Leading · Coaching	Most cross-functional models	Inflation toward 3
Below / At / Above level	Three buckets per capability	Performance reviews tied to a ladder	Manager bias without anchors

Avoid 9-box without training

9-box grids (performance × potential) are popular but routinely abused without calibration. Use them only with trained facilitators and explicit definitions of 'potential' — otherwise you encode bias.

Where the model plugs in

Job descriptions — derived from the model, not written separately.
Interview rubrics — each loop covers a subset of capabilities; debriefs use the same vocabulary.
Onboarding — new hires get a self-assessment in week 2 and again at month 3.
1:1s — once a quarter, the 1:1 is structured around the model.
Performance reviews — ratings are evidence against anchors, not vibes.
Promotions — packets reference behaviors at the next level, not titles or tenure.

Worked examples by function

Examples adapted from public ladders (Monzo Engineering, CircleCI, Spotify, Etsy, Khan Academy).

Engineering — one capability across levels

Level	Capability: 'Designs systems'
L3 Engineer	Designs a single service with guidance; reviews their own design doc.
L4 Senior	Designs systems spanning 2–3 services; runs the design review.
L5 Staff	Designs cross-team systems; identifies what NOT to build; mentors L4s.
L6 Principal	Sets architectural direction for a product line; predicts second-order effects.

People manager — one capability across levels

Level	Capability: 'Develops people'
M1 Manager	Runs weekly 1:1s; gives direct feedback; identifies growth areas.
M2 Sr Manager	Designs stretch assignments; owns promotion packets for 2+ reports.
M3 Director	Builds the bench: 1–2 reports promoted into management this year.
M4 VP	Designs the org's leadership pipeline; coaches managers of managers.

Keeping it alive

Versioned and dated. v2.1, last reviewed 2026-Q1. Stale ladders silently lose credibility.
Owned by a person, not 'HR'. A working group of 3–5 senior practitioners per family.
Re-calibrated every 6 months with real cases from the last cycle.
Public to all employees — internal Notion / wiki, not a locked HR file.

Anti-patterns

Task lists — 'writes unit tests' is a behavior, not a capability.
One model for everyone — sales and engineering need different families.
No anchors — 'shows leadership' means nothing without a behavior.
Set and forget — ladders rot the moment the business model shifts.
Used only at review time — should be a living tool in 1:1s and hiring.

References

Progression.fyi — open library of career ladders — Curated by Progression
Monzo Engineering Progression Framework — Monzo
The Manager's Path — Camille Fournier — O'Reilly
Radical Candor — Kim Scott on growth conversations — Radical Candor
Spotify Squad Health Check — Spotify Engineering

Written by Pawan Joshi. Sources cited inline. Last updated 2026-05-17.