Skills-Based Hiring: Building a Skills Taxonomy

Most hiring processes in India still screen on proxies — college tier, employer brand, years of experience — and then interview for something else entirely. Skills-based hiring closes that gap by making the actual requirement the actual screen. It is not a slogan about banning degrees; it is a discipline that starts with a skills taxonomy, runs through assessment design, and ends in how you pay and promote people. This guide covers how to build it, including the parts that are genuinely hard.

What Skills-Based Hiring Actually Means

Skills-based hiring defines a role by the capabilities needed to do it, assesses candidates against those capabilities, and decides on that evidence. Degree, tenure, employer and title become possible indicators of a skill rather than substitutes for one.

The shift is subtle but it changes the whole funnel. Instead of "does this person have five years in payroll at a large company," you ask "can this person reconcile an attendance file against a muster roll, compute statutory deductions correctly, and explain a payslip discrepancy in writing to an upset employee." Those are observable.

What It Does Not Mean

Most resistance comes from a caricature of the idea. Being clear about the boundaries shortens the conversation with hiring managers.

• Not banning degrees. A degree has to earn its place by naming the skill it evidences. For a chartered accountant leading statutory audit prep, the qualification is the requirement. For a support associate, "any graduate" is a filter with no defensible link to performance.
• Not ignoring experience. Experience remains one of your strongest evidence sources. You read it for what it demonstrates instead of counting years.
• Not six-hour tests for everyone. Badly sized assessments are the most common way this fails. Depth should scale with role stakes.
• Not lowering the bar. Done well it raises the bar, because you stop accepting a brand name in place of demonstrated capability.
• Not abandoning judgement. You are replacing unexamined gut feel with recorded, comparable observations.

The one-line version worth repeating internally: make the requirement the requirement. If a criterion cannot be tied to something the person will do in year one, it is a preference — and preferences should not silently eliminate people.

Why Indian Employers Are Moving This Way

The pressure is practical, not ideological.

• Tool stacks change faster than resumes describe them. Screening for "3 years of tool X" selects for people whose employer happened to buy X, not for people who learn a tool in a fortnight.
• AI is reshuffling the task mix inside jobs. When drafting, summarising and first-pass code get faster, a role's residual value concentrates in judgement, verification, edge cases and stakeholder handling — exactly what resumes describe worst.
• College quality varies enormously. Institution tier is a noisy signal that often tracks a family's ability to pay for coaching and relocation rather than the candidate's capability.
• Tier-2 and tier-3 pools are now reachable. Hybrid models and capability-centre expansion make hiring viable in Coimbatore, Indore, Bhubaneswar, Kochi and Nagpur. Those pools are deep and under-screened by brand filters; a skills screen is how you access them.
• Career-break candidates are a mispriced pool. Returnships, relocations, a venture that did not work — traditional screens punish the gap. A skills assessment prices the person, not the calendar.
• Internal mobility is under pressure. Titles are inconsistent across teams, so you need a shared vocabulary to match someone in one function to an opening in another.
• Mis-hiring costs more than the visible bill. Beyond recruiter time sit the manager's attention, the delayed project, the effect on the team and the awkward exit.

The Honest Counterargument

A guide that only sells the idea is not useful. There are real cases where credential and experience screens carry information.

• Regulated and licensed roles. Statutory sign-off, certain medical and safety functions, roles where a client contract specifies qualifications — here the credential is the requirement. Do not waste effort trying to skills-ify it.
• Credentials as compressed evidence. A long, examined, standardised programme evidences sustained effort and a body of knowledge. That is one piece of evidence with known limits, not a meaningless one.
• High-volume funnels with thin margins. With several thousand applications for ten seats you need some cheap early filter. Usually a short job-relevant screen beats a degree filter — but the constraint is real, so design for it.
• Roles where signal-to-cost is poor. For senior leadership or deeply relational work that only becomes visible over quarters, a work sample is a weak simulation. Structured referencing and a scoped trial engagement usually beat an artificial exercise.
• Assessment fatigue is a real cost. Every extra stage loses candidates, disproportionately the strong ones with options. A stage that adds three days and improves prediction slightly is a net loss.

The practical resolution is a tiering rule: roles above a defined stakes threshold get full skills assessment; high-volume entry roles get a short standardised screen plus a structured interview; regulated roles keep the mandatory credential plus a skills assessment for the discretionary part.

The Foundation: A Skills Taxonomy

A skills taxonomy is a controlled vocabulary of the capabilities your organisation cares about, organised into groups, with defined proficiency levels and accepted evidence types. It lets a job description, an interview scorecard, an internal job posting and a learning plan all refer to the same thing.

Without it, "communication" means five different things to five managers and your capability mapping is unusable. With it, you can ask "how many people are at Practising or above on statutory payroll computation, and where are they" and get an answer.

The Layers, and What Belongs in Each

The most common early mistake is mixing categories — putting "MBA", "ownership", "Excel" and "handles escalations well" in one list. They need different evidence.

Two distinctions save the most arguments later.

Skills vs competencies. "Writes clear payroll explanation emails" is a skill — capability to perform a task. "Stays calm and factual when an employee is angry" is a competency — behavioural consistency. The first is assessed with a sample, the second by probing several past situations.

Knowledge vs skill. Knowing the current professional tax slabs is knowledge; running a correct monthly payroll cycle is a skill. If your assessment tests only knowledge, you are testing the most trainable thing in the room.

The Anatomy of One Entry

Keep each entry short enough that a busy hiring manager reads it: a stable ID and plain-language name (PAY-014 · Statutory deduction computation), the skill group, a one-line definition, one observable sentence per proficiency level, accepted evidence types, related skills for mobility matching, an owner, a last-reviewed date, and a status.

That last trio matters more than it looks. A taxonomy without owners and dates goes stale within a year and quietly loses the organisation's trust.

How to Build a Taxonomy Without a Two-Year Project

The failure pattern is a committee describing every skill in the company before anyone uses anything. Build it the other way: start from what you actually hire and staff for, cap the size, version it.

Estimates below assume a fictional 200-person company — Kanira Logistics, with operations, technology, finance, sales and support — and one HR lead spending part of their week, borrowing manager time in short blocks.

1. Scope to live demand (~4 hours). Pull every requisition from the last 12 months and every role expected in the next 6. At 200 people that is typically 25 to 40 distinct roles, not 200. Add roles carrying succession risk; everything else waits.
2. Harvest from live job descriptions (~8 hours). Extract the verb-object phrases describing real work — "reconcile vendor invoices", "handle escalated tickets" — and ignore the self-starter boilerplate. You will get 400 to 700 messy phrases, which is normal.
3. Interview a few practitioners (~10 hours). One strong performer and one manager per function, 45 minutes each. Ask what a new joiner struggles with in month one, what separates your best person from your average one, and what you had to teach someone last quarter. This surfaces the exception-handling skills that never reach a JD.
4. Cluster and deduplicate (~12 hours). "Excel modelling", "advanced Excel" and "spreadsheet analysis" become one skill. Resist splitting by tool unless the tool genuinely changes the capability. Roll skills into 8 to 15 groups aligned to how work happens, not to the org chart.
5. Cap the vocabulary (~2 hours, plus discipline forever). Set the ceiling first: 120 to 180 skills for a 200-person company. Below 80 you lose precision; above 250 nobody can navigate it. A skill earns its place only if it appears in two roles, or is business-critical in one.
6. Write anchors for the top skills only (~16 hours). Full five-level anchors for the 30 to 40 skills in your most-hired and highest-risk roles; everything else uses the generic scale until a role needs more. This alone cuts weeks off the project.
7. Pilot on three roles (~8 hours). Build complete profiles for three live openings, run them, and collect the complaints. You will find missing skills, overlapping ones, and anchors two managers read differently. That is the point.
8. Version, publish, assign owners (~4 hours). Publish v1.0 with a date. Give each skill group an owner — a senior person in that function, not HR — with a light quarterly review, a full annual pass and a visible changelog.

Total: roughly 60 to 70 hours over 8 to 10 weeks, about 20 hours of it manager time. A real project, but not a transformation programme — and usable by week four.

You do not have to invent skill names from nothing. Public occupational frameworks, sector skill council role definitions and well-written job ads from adjacent industries are all legitimate starting vocabulary. Borrow the words, then rewrite the anchors against your own standard.

Designing a Proficiency Scale People Score Consistently

Five levels is the practical sweet spot. Three is too coarse to support pay and development decisions; seven invites false precision and endless argument about the middle.

How to Write Anchors Two Managers Score the Same

The scale above is generic; the value comes from skill-specific anchors underneath it.

1. Describe observable behaviour, not internal states. "Has a strong grasp of PF rules" is unscoreable. "Computes PF on the correct wage components including for employees crossing the wage ceiling mid-year, and can show the working" is scoreable.
2. Vary one dimension at a time — scope, then supervision, then exception difficulty, then effect on others. If level 4 introduces a completely different activity from level 3, the scale has broken.
3. Include a disqualifier. One sentence per level on what knocks someone down. This is what stops the generous and the harsh scorer drifting apart.
4. Calibrate on real people first. Take three current employees whose capability everyone agrees on, have two managers score them independently, and compare. More than one level apart means the anchor is at fault, not the manager.

Run that calibration once per skill group. It takes about 90 minutes and it is the highest-return hour in the whole build.

A Worked Example: Payroll Executive at Kanira Logistics

Role: Payroll Executive · Function: Finance & HR Operations · Scale: ~600 employees across 4 states, monthly cycle plus off-cycle runs.

Note what the profile does not say: no commerce degree, no years threshold, no company-size requirement. A candidate from a hospital payroll team, one from an outsourcing firm, and one returning from a three-year caregiving break who previously did accounts payable are all assessed on the same evidence.

Note what it says clearly: the two non-negotiables are reconciliation ability and written communication under pressure, because those cannot be taught in a quarter. Demanding level 4 on everything just recreates the unicorn JD in new language.

Rewriting Job Descriptions and Job Ads

The JD is where a taxonomy becomes real or stays a spreadsheet. Two documents are involved: the internal JD, which is the contract between HR and the hiring manager about what gets assessed, and the external ad, whose only job is to get the right people to apply.

Turning Requirements into Capability Statements

• "5+ years of experience" → "Can independently run a monthly payroll cycle end-to-end for a multi-state employee base, including exceptions"
• "B.Tech in Computer Science" → "Can reason about time and space complexity when choosing a data structure, and explain the trade-off to a non-specialist"
• "Must be from a product company" → "Has shipped and maintained software others depended on, and can describe a production incident they handled"
• "Excellent communication skills" → "Can write a clear, non-defensive explanation of a billing error to a customer in under 200 words"
• "Team player" → "Gives and receives review feedback without escalation, and can describe changing their approach based on a colleague's input"

Language That Widens the Funnel

Wording changes who applies, and the effect is asymmetric — phrasing that reassures a confident applicant deters a well-qualified cautious one.

• Split requirements into "you'll need" (3 to 5 genuinely non-negotiable items) and "helps, but we'll teach you." Long undifferentiated lists are read as a checklist to be met in full.
• Say what you will not screen on: "We don't filter on college, and we don't mind career breaks."
• Publish the process: stages, what each involves, expected time, and payment for any long exercise.
• Publish a salary band. Where band opacity is the norm, this is one of the strongest widening levers available and it prevents late-stage collapse.
• Describe the first 90 days concretely. Candidates self-select accurately when they can picture the work.
• Drop "rockstar" and "ninja" — they narrow to a self-presenting personality type without adding predictive signal.

What to Do About ATS Keyword Matching

This recruiter objection is legitimate: if sourcing depends on keyword matching, capability statements can hurt discoverability.

• Keep a plain-noun skills block alongside the capability statements: "PF, ESI, professional tax, Form 16, attendance reconciliation, Excel." It reads fine to a human and matches to a machine.
• Match on skill tags, not free text. With structured skill fields on the requisition, search runs against the taxonomy rather than whatever words the candidate happened to use.
• Build a synonym list inside the taxonomy — "F&F", "full and final", "exit settlement" for one skill. Five minutes per skill, pays back on every search.
• Audit inherited auto-rejection rules for degree fields, gap length and years-of-experience thresholds before rewriting a single JD, or the new ads will feed the old filter.

Before and After

Before: > Senior Payroll Executive. B.Com/M.Com required, MBA preferred. 5–7 years in payroll processing, preferably in a large MNC. Hands-on experience with leading payroll software. Excellent communication and interpersonal skills. Team player with strong attention to detail. Immediate joiners preferred.

After: > Payroll Executive — Kanira Logistics, Pune (hybrid, 3 days in office) > Band: ₹X–₹Y per annum, published because you should know before you apply. > > You'll own the monthly cycle for about 600 employees across four states, alongside our Payroll Manager. In your first 90 days you'll run two full cycles under review and take over the statutory filing calendar. > > You'll need to be able to: > - Reconcile a messy attendance and leave file against payroll input, and find the errors before finance does > - Compute PF, ESI, professional tax and LWF correctly, including the awkward cases — mid-year ceiling crossings, LOP months, mid-month joiners > - Write a calm, clear explanation of a payslip discrepancy to an employee who is upset > - Work in spreadsheets where lookups, pivots and error-checking are routine > > Helps, but we'll teach you: our payroll platform, gratuity and leave encashment, multi-state nuances, report building. > > How we'll assess it: a 30-minute conversation, a 90-minute paid reconciliation exercise on anonymised sample data, a 60-minute discussion of it, and 45 minutes with the Payroll Manager — four to five hours over two weeks. > > What we don't screen on: college, degree stream, employer brand or career gaps. If you've done this work anywhere — outsourcing firm, hospital, factory, startup — we're interested. > > Skills: payroll processing · PF · ESI · professional tax · TDS · Form 16 · full and final settlement · attendance reconciliation · Excel

Assessment Design: Where This Lives or Dies

A taxonomy with weak assessment is decoration. The goal is maximum signal per hour of candidate time, and the second half of that phrase matters as much as the first.

Rules That Make Assessments Work

• Size against the job's stakes, not the manager's curiosity. Total candidate time across all stages should stay under roughly one working day for a mid-level role.
• Pay for anything over two hours. It is fair, and it works in your favour — candidates take paid work seriously, and paying forces you to keep the task tight.
• Use anonymised real work, not invented puzzles. The best reconciliation exercise is a real file from last quarter with identifiers stripped. An hour to prepare, and it predicts far better than a synthetic case.
• Score against the anchors, in writing, before discussion. Scorecards submitted within 30 minutes and locked before the panel talks. The most effective anti-bias mechanism available, and it costs nothing.
• Offer an alternate path. Someone working 60-hour weeks who cannot do a take-home should get a longer live session. Otherwise your "skills-based" process quietly selects for people with free time.

Building the Scorecard and Coverage Matrix

Map the skills profile against your stage plan so every must-have is assessed at least once, the critical ones are corroborated by two different people using different methods, and nothing is assessed four times by accident.

Coverage matrix — Payroll Executive, Kanira Logistics. P = primary (drives the score), S = secondary corroboration.

Three checks on any matrix you build:

• No must-have skill has an empty row. If one does, add it to a stage or drop it — it was not really a must-have.
• The highest-risk skills carry at least two marks from different assessors. Single-assessor decisions on critical skills are where expensive mistakes come from.
• No stage carries more than three or four primary skills. An interviewer assessing six things assesses none of them well.

Each cell should map to specific prompts in the interview kit — questions, expected evidence, anchor reminders, a scoring box per skill. That kit is what keeps the process repeatable when your best interviewer is on leave.

Reducing Bias Without Turning Hiring Into a Ritual

Skills-based hiring reduces bias by construction, but bias re-enters through assessment design and through the debrief.

• Ask every candidate for a role the same core questions. Follow-ups can vary; the core set cannot.
• Score independently before the panel talks. The first person to speak anchors everyone else.
• Debrief on evidence, not impressions. The facilitator asks "what did you see that led to that score" and moves on when the answer is "just a feeling."
• Diversify panels without tokenising. Making one person the designated diversity voice in every panel is unfair and ineffective.
• Name your proxy filters and ban them explicitly: college tier, employer brand, English accent where the job does not need it, marital status, gender assumptions about travel or shifts, career gaps, hometown, surname inference, age implied by graduation year. Some are legally risky; all are predictively weak.
• Separate language proficiency from communication skill. If the job needs written English at a defined level, assess written English rather than letting spoken fluency stand in.
• Audit pass-rates by stage, quarterly, split by dimensions you can legitimately collect. A stage where one group's pass-rate falls off a cliff is measuring something other than skill.

Handle candidate data carefully throughout: collect only what you need, tell candidates why, keep it for a defined period, and store assessment records with access controls. India's data protection regime has moved toward explicit consent and purpose limitation, and hiring data attracts exactly that scrutiny.

Using AI Tools Responsibly

AI is genuinely useful here, and it is also the fastest way to build a discriminatory process at scale without noticing. The split runs between drafting and structuring, where it helps, and deciding, where it should not be in charge.

Where it helps: rewriting JDs into capability language for human review; extracting candidate skills into your taxonomy's vocabulary as a suggestion; drafting interview questions mapped to a skill and its anchors; scheduling; note-taking with consent; summarising panel scores into a comparison without inventing a recommendation; flagging exclusionary language in a JD.

Where a human must stay in charge: the hire/no-hire decision and the scoring behind it; any adverse decision communicated to a candidate; automatic ranking or filtering out of the funnel; interpreting a career gap, a non-standard background or an accommodation request; anything inferring protected characteristics.

A short governance checklist:

1. Write a one-sentence purpose statement per tool. If you cannot, do not use it.
2. No automated rejection — machines sort and suggest, humans reject.
3. Disclose to candidates, plainly, where AI is used.
4. Ask consent before any recording or transcription, and offer a no-recording option.
5. Know where candidate data goes, whether the vendor retains it, and whether it trains anything. Get it in writing.
6. Every model-suggested skill tag is confirmed by a person before it affects a decision.
7. Run the quarterly pass-rate audit specifically on any stage where a tool is involved.
8. Name one owner who answers for AI use in hiring, and log changes to tools and prompts.

Making It Work Internally: Capability Mapping and Mobility

The taxonomy is worth more inside the company than outside it. Once employees carry skill tags, internal mobility stops depending on who knows whom.

Getting Employee Skills Data Without a Survey Marathon

1. Seed from role. Everyone starts with their current role's profile at target level — a defensible default that removes most of the data entry.
2. Self-assessment against anchors. Employees adjust their levels and add skills the profile missed. Twenty minutes, no more.
3. Manager validation. Confirm or adjust, with a required comment on any change of more than one level. A conversation, not an approval workflow.
4. Attach evidence where it matters. For skills that gate pay or mobility, link a project, an assessment record or a certification.
5. Refresh at a natural moment. Hang it off appraisal or a mid-year check-in rather than creating a new event nobody wants.

Be honest about what the data is for. If employees suspect it feeds a redundancy list, you will get uniformly average self-assessments and the exercise is worthless.

A Skills Gap Heat Map

Finance & HR Operations at Kanira Logistics, 16 people:

Each row implies a different action. Full and final settlement is a person-dependency — cross-training this quarter, not a hire. Report building is a genuine team-wide gap — a learning programme or a hire. Attendance reconciliation is one short, possibly solved by a lateral move.

Internal Job Posting Matched on Skills

With profiles on both sides, an opening can be matched against the employee base as full matches, near matches (within one level on one or two must-haves), and adjacent matches (strong on transferable skills, missing a teachable one).

The near-match list is where internal mobility actually happens, and it is invisible without a taxonomy — because on paper, someone in customer operations does not look like a payroll candidate even when they share six of the eight required skills.

Two rules keep it honest. Internal candidates go through the same assessment and scorecard as external ones. And a near match generates a development plan, not just a rejection: "you're one level short on statutory computation; here's what closing that looks like, and the next opening is likely in two quarters."

Compensation Implications

This is where skills-based hiring meets finance, and where it most often stalls. If you hire on skills but pay on titles and pedigree, the two systems will fight and pay will win.

• Anchor bands to the role's skill profile, not the candidate's history. This is what makes it defensible to pay a returner and a continuously-employed candidate the same for the same role.
• Stop using last-drawn salary as the anchor. Salary history perpetuates whatever mispricing the candidate carried in, and disproportionately penalises women, career-break candidates and people from smaller cities.
• Make skill premiums explicit. Handle scarce skills as a documented premium on top of the band rather than by inflating the band, so they can be reviewed when the market moves.
• Protect internal equity actively. Before any offer above band midpoint, check what employees at the same skill profile earn. If it breaks the pattern, adjust the offer or budget the correction for incumbents.
• Pay for demonstrated level, not claimed level. Money attached to proficiency needs evidence and a validation step, or level inflation is guaranteed within two cycles.
• Keep paid levels few. The band is set by the role profile; only two or three critical skills carry a level-linked premium.

Measuring Whether It Works

Most of the answer is available from ATS exports and a spreadsheet.

Keep a simple running sheet of every hire — role, process used, source, 6-month and 12-month outcome. Eighteen months of that sheet is worth more than any dashboard you could buy today. Do not expect statistical significance at small scale; you are looking for direction and obvious breakages.

A 90-Day Rollout Plan

Pilot one function. A visible success in one function creates more adoption than a mandate across five.

Run one thing in parallel from week 6: seed employee skill profiles for the pilot function, so internal candidates are matched on the same basis as external ones. That step converts the project from a recruiting exercise into a capability system.

Common Failure Modes

• Taxonomy bloat. Six hundred skills, half near-duplicates, nobody able to find anything. Cure: the cap, enforced by an owner with authority to say no.
• Assessment inflation. Each manager adds a stage; nine months later candidates give you two days and drop out at stage three. Track candidate hours per hire as a budget.
• Managers reverting to gut feel. Scorecards filled in after the debrief to match a conclusion already reached. Fix it with locked pre-submission in the system, not reminder emails.
• No evidence storage. Scores exist but work samples, notes and reasoning live in email. Six months later nobody can explain a decision or reuse a good exercise.
• Stale skills data. Profiles captured once, never refreshed, actively misleading. Show a last-reviewed date everywhere.
• Treating it as a recruiting-only project. If pay, promotion and learning still run on titles and tenure, the taxonomy dies with the next change of recruiting leadership.
• Perfectionism before first use. v0.9 in use beats v3.0 in draft.
• Ignoring the candidate side. A rigorous but opaque, slow process loses good people. Publish stages, hold your timelines, give a reason when you say no.
• No feedback loop. If nobody compares 6-month performance against interview scores, you will never find the stage that predicts nothing.

How an HRMS and ATS Support This

Skills-based hiring generates structured data at every step, and that data needs somewhere to live. Spreadsheets work for one pilot function; they collapse around the third function or the second year.

• Structured requisitions — skills, target levels and must/nice flags as fields, not prose in an attachment. This is what makes matching and reporting possible later.
• Skill tags on employee records — the same vocabulary as candidates, with level, evidence link, source (self / manager / assessment) and last-reviewed date.
• Scorecard capture in the workflow — interviewers score against the role's skills in-system, with submission locked before the debrief opens.
• Evidence storage with access control — work samples, assessment records and notes attached to the record, with a retention policy.
• An internal job board matched on skills — full, near and adjacent matches surfaced to employees, flowing into the same pipeline as external applications.
• Reporting that answers real questions — pass-rates by stage and segment, time-to-fill by role family, internal fill rate, gap heat maps by team.
• Learning links — gaps in a profile connected to a development plan, so the heat map produces action rather than a slide.

CozyHR is built around this kind of structured people data: requisitions with skill fields, employee skill profiles, scorecards captured inside the hiring workflow, internal job posting, and reporting that runs off the same records as payroll and performance. A single-record approach is what stops a skills taxonomy from becoming a parallel spreadsheet nobody maintains.

Frequently Asked Questions

Does skills-based hiring mean removing degree requirements from every job?

No. Remove them where you cannot articulate which skill the degree evidences, and keep them where they are genuinely required — regulated roles, licensed roles, and roles where a statute or client contract specifies the qualification. If you can describe the underlying capability and assess it directly, assess it directly.

How long does it take to build a usable skills taxonomy for a 200-person company?

Roughly 60 to 70 hours of focused work over 8 to 10 weeks, if you scope to live hiring demand rather than the whole organisation. You should have something usable on three pilot roles by week five. The variable is not the analysis — it is how fast you can get 45 minutes each from a dozen busy practitioners.

How many skills should a taxonomy contain?

For a 200-person company, 120 to 180 skills in 8 to 15 groups. A role profile should list 8 to 14 skills, of which no more than 6 or 7 are must-haves. If your role profiles routinely run past 20 skills, the taxonomy is too granular and matching will produce noise.

Won't skills assessments slow hiring down and lose candidates?

They add time in the first quarter, then usually save it, because you stop running three rounds of unstructured interviews that end in disagreement. Keep total candidate time under about one working day for mid-level roles, publish the process up front, pay for anything over two hours, and hold your timelines. Most drop-off comes from silence and slippage, not from being assessed.

How do we handle career-break and returnship candidates fairly?

Assess them on the same profile as everyone else, and remove gap length from every screening step including automated ATS rules. Distinguish knowledge that refreshes quickly — current statutory rates, a specific tool — from skill that does not, like reconciliation ability or written clarity. Offer a live alternative to any take-home, since unpaid evening work is unequally accessible to people with caring responsibilities.

What is the difference between a skills taxonomy and a job architecture?

A job architecture is the structure of jobs — families, levels, titles, bands. A skills taxonomy is the vocabulary of capabilities. They meet at the role profile: the architecture says what the job is and where it sits, the taxonomy says what it takes to do it. Most companies should build the taxonomy first, because architecture projects are slow and the taxonomy delivers value earlier.

Should we use an AI tool to score candidates?

Use AI to draft, extract, structure and summarise; keep scoring and rejection with humans. A model can suggest which skills appear in a resume and turn six scorecards into one comparison table. It should not rank candidates out of your funnel or decide a no. Disclose where you use it, get consent before recording, offer an alternative to candidates who decline, and audit pass-rates on any stage where a tool is involved.

How do we stop the taxonomy from going stale?

Give every skill group a named owner in the business — not in HR — with a light quarterly review of fast-moving groups and one full annual pass. Show a last-reviewed date wherever skills appear. And keep it inside the systems people already use daily; a taxonomy living in a shared drive will be out of date within a year.

Where to Start

If you take one thing from this, take the sequencing. Pick one function, harvest skills from the job descriptions you already have, cap the vocabulary, write anchors only for the skills you will actually assess, calibrate them on three people you already know well, and run three live requisitions with a proper scorecard. That is a quarter's work, not a transformation programme.

Everything else — internal mobility, capability mapping, pay linked to demonstrated skill, succession that runs on evidence — is built on the same vocabulary. Which is why it is worth getting the foundation right and keeping it small enough to maintain.

If your spreadsheets are starting to strain, CozyHR handles the structural side: skill fields on requisitions and employee records, scorecards captured in the hiring workflow, evidence stored against the person, internal job posting, and reporting sitting on the same data as payroll and performance. Worth a look when your pilot is ready to become a system.