HR Analytics Dashboard: Metrics That Change Decisions
A metric-by-metric blueprint for building an HR analytics dashboard that people actually use, with exact formulas, definition traps, dashboard layouts and a 90-day rollout plan.
Somewhere around 80 people, the spreadsheet stops working. Up to that point, your founder or HR lead can hold the whole company in their head — who joined last month, who is unhappy, which team is short-staffed, what the salary bill looks like. Past that point, memory fails and the questions get harder: is attrition getting worse or is it just noisy? Which manager is bleeding people? Are we paying the Bengaluru backend team below what we promised in the band? Why did payroll cost jump 6% when we only added two people? An HR analytics dashboard is how you answer those questions in five minutes instead of five days of Excel archaeology.
But here is the uncomfortable truth: most HR dashboards built by Indian SMBs and startups get looked at twice — once when they are launched and once when the founder asks for something that isn't on them. They fail not because the charts are ugly but because nobody ever decided what decision each number was supposed to change, and because the underlying data was never clean enough to trust. You can't run a monthly people review on numbers that three people in the room quietly believe are wrong.
This guide is the practical version. It covers the data foundation you need before you build anything, a metric library with exact formulas and the definition traps that make the same metric produce three different answers, two fully worked calculations, dashboard layouts for three different audiences, how to run a monthly people review that actually changes decisions, the anti-patterns that kill credibility, and a 90-day roadmap you can start on Monday. It is written for companies between roughly 50 and 1,000 employees in India, where you have real complexity but not a dedicated people analytics team.
Why Most HR Dashboards Get Ignored
Walk into a 300-person company and ask to see their HR reports. You will usually get one of three things: a monthly deck of 22 slides that took HR ops four days to build and that nobody reads past slide 3; a BI tool with 40 tiles where half show "no data"; or a WhatsApp message from the HR head with headcount and last month's exits.
The failure modes are consistent.
No metric owns a decision. The dashboard shows "Average tenure: 2.3 years." Fine. Now what? Who does something differently because of it? If you cannot name the person and the action, delete the tile. Every metric on a useful dashboard should be traceable to a decision someone in the company is empowered to make — approve a backfill, escalate a retention risk, change a hiring source, fix a pay outlier, adjust a shift roster.
No owner. A number with no owner drifts. Time to fill sits at 61 days for four months and nobody is accountable for moving it because "recruitment is a shared responsibility." Assign every metric to one named person. Not a team. One person.
Definitions are ambiguous. Three people compute attrition three ways and all three are "right." When leadership notices the inconsistency, they stop trusting the whole dashboard, including the parts that were fine. Definitional discipline is the single highest-leverage investment in people analytics for SMBs.
The data is dirty and everyone knows it. Exit dates entered a month late. Two employee IDs for the same person after a rehire. Managers not updated after a reorg. Location field with "Bangalore", "Bengaluru", "BLR", "blr" and one blank. Once managers spot one wrong number in their own team's tile, they discount everything.
It reports the past with no lead time. A dashboard that tells you attrition was 22% last quarter is a coroner's report. A dashboard that tells you seven people in one team have crossed 30 months tenure without a role change, and that the team's manager has had four skip-levels cancelled, gives you something to act on.
What a Useful HR KPI Dashboard Does Differently
A good HR KPI dashboard has four properties.
- Every tile has a decision, an owner and a cadence. Written down, visible, ideally in a tooltip or a linked definitions page.
- It is small. Twelve tiles on the executive view, not forty. Depth lives in drill-downs and in the HR ops view.
- It mixes lagging outcomes with leading signals. Attrition is lagging. Tenure concentration, internal mobility rate, offer acceptance rate, manager span, and unplanned absence trend are leading.
- It comes with a narrative. Numbers without interpretation get interpreted by whoever speaks loudest. The HR lead writes 300–500 words of explanation each month: what changed, why, what we are doing.
If you build nothing else, build the discipline of point 1. A five-tile dashboard where each tile changes a decision beats a fifty-tile one where none do.
The Data Foundation: You Cannot Analyse What You Have Not Modelled
Analytics is 20% charts and 80% plumbing. Before you pick a visualisation tool, get your core HR data model straight. In most Indian SMBs, this data is scattered across an HRMS, a payroll sheet, a biometric or attendance app, an ATS, a Google Form for leave, and an offer-letter folder. Every one of those is a source of truth for something and a source of contradictions for everything else.
The Core HR Data Model
Think in nine domains. Each has a natural grain (the level at which one row means one thing) and a set of fields you genuinely need. Do not model what you will not use.
| Domain | Grain (one row =) | Fields you actually need |
|---|---|---|
| Employee master | One employee | Employee ID (immutable), full name, date of birth, gender, date of joining, employment type (permanent / fixed-term / intern / contractor / consultant), status (active / on notice / exited / on long leave), confirmation date, exit date, exit type, exit reason, rehire flag, personal email |
| Org / reporting structure | One employee-manager relationship, effective dated | Employee ID, manager employee ID, effective from, effective to, dotted-line manager (optional) |
| Position / job architecture | One position or job code | Job code, job title (standardised), job family, sub-family, level / grade, IC-vs-manager flag, department, business unit, location, work mode |
| Compensation | One compensation record, effective dated | Employee ID, effective from, fixed CTC, basic, variable target, one-time components, band / grade, band minimum, band midpoint, band maximum, currency, change reason (hire / annual / promotion / correction / market adjustment) |
| Attendance | One employee-day (or one punch pair) | Employee ID, date, shift, first-in, last-out, hours worked, status (present / absent / weekly off / holiday / leave / half day), regularisation flag, overtime hours |
| Leave | One leave transaction and one balance snapshot | Employee ID, leave type, from date, to date, days, applied date, approved date, approver, status; and per-period opening balance, accrued, availed, encashed, lapsed, closing balance |
| Payroll | One employee-month | Employee ID, month, gross, all earning heads, all deduction heads, employer PF, employer ESI, gratuity provision, bonus provision, net pay, cost to company for the month, cost centre |
| Recruitment | One application; plus one requisition | Requisition ID, role, level, location, hiring manager, opened date, approved date, closed date, source, candidate ID, stage, stage entry date, stage exit date, outcome, offer date, offer acceptance date, joining date, offer status |
| Performance | One employee-cycle | Employee ID, cycle, rating, calibrated rating, potential (if used), promotion flag, promotion effective date, PIP flag, review completion date |
Two things people skip and later regret. First, an immutable employee ID that survives rehires, name changes, marriage, department transfers and system migrations. If your ID encodes department or joining year, it will break. Use a meaningless sequential ID and store everything else as attributes. Second, cost centre and location on the payroll record, not just on the employee master — otherwise you cannot reconcile people cost to finance's ledger, and finance will win that argument.
Effective Dating and Why Point-in-Time Snapshots Matter
This is the concept that separates an HR analytics dashboard you can defend from one you cannot.
Most HRMS tables store the current state: current manager, current salary, current department. If you report from current state, then a reorg in July silently rewrites your January numbers. Your board deck from Q1 will not reconcile with the same report re-run in Q3, and you will look careless.
Two fixes, and you want both.
Effective-dated records. Any attribute that changes over time — manager, department, location, grade, salary, employment type, status — is stored as a row with effective_from and effective_to. The current value is the row where effective_to is null or a far-future date. To ask "who reported to Priya on 31 March," you filter rows where effective_from <= 31-Mar and effective_to >= 31-Mar. This is how you get historically correct answers.
Monthly snapshots. On the last calendar day of every month, write a full copy of the active employee population with all their attributes as of that date into a snapshot table with a snapshot_date column. It is cheap — a 500-person company generates 6,000 rows a year — and it makes everything downstream trivial. Average headcount, headcount bridges, attrition denominators, cohort tenure curves, cost per employee trends: all become simple queries against the snapshot table. If you take one technical action from this article, take this one. Start snapshotting tonight, even if the rest of your data is messy. You cannot retroactively create history.
A third habit worth adopting: freeze and archive. When you present numbers to the board or to investors, save the exact extract that produced them as a dated CSV. When someone asks six months later why the number moved, you can show what you knew at the time.
Common Workforce Data Quality Defects (and How to Fix Them)
Every company I have seen at this stage has some subset of these. Fixing them takes two to four weeks of unglamorous work and it is the highest-return activity in the whole project.
| Defect | How it shows up | Fix |
|---|---|---|
| Duplicate employee records | Same person with two IDs after rehire, transfer between entities, or a migration | Dedupe on PAN / Aadhaar-linked identity or personal email; keep one master ID, link the old one as an alias, add a previous_employee_id field |
| Free-text location and department | "Bangalore" / "Bengaluru" / "BLR" appearing as separate groups | Convert to a controlled dropdown backed by a master list; map historical values once, block free text going forward |
| Missing or late exit dates | Attrition understated this month, overstated next month | Make exit date mandatory to close the exit workflow; run a weekly check for employees with status = exited and null exit date |
| Manager field not updated after reorg | Span-of-control and manager-wise attrition are wrong | Quarterly org data review where every manager confirms their direct reports; block confirmation until it matches |
| Job titles as free text | 300 people, 240 distinct titles, no meaningful grouping | Build a job architecture: job family, sub-family, level. Map every person once. Keep the vanity title as a separate display field |
| Salary changes without a reason code | You cannot separate promotion increases from market corrections | Mandatory change-reason dropdown on every comp record |
| Attendance gaps for field or hybrid staff | Absenteeism looks inflated for teams who never punch in | Define expected-attendance rules per employment type and work mode; exclude non-applicable populations rather than counting them absent |
| Contractors mixed into headcount | Headcount does not tie to payroll or to finance's numbers | Separate employment_type; report own-payroll headcount as primary, total workforce as secondary |
| Leave balances not carried correctly | Leave liability wrong, employees dispute it | Reconcile opening + accrued − availed − encashed − lapsed = closing every month, per employee; investigate any non-zero difference |
| Test or dummy records in production | "Test User" in the headcount | Flag and exclude system accounts; check the list quarterly |
A Data Hygiene Checklist You Can Run Monthly
Make this a 45-minute checklist that HR ops runs on the first working day of every month, before any report is published. It is boring and it is the reason your numbers get believed.
- Every active employee has a non-null manager, and no employee is their own manager.
- No circular reporting loops (A reports to B reports to A).
- Every employee has department, location, grade, employment type and work mode populated.
- Every exit in the last month has an exit date, exit type (voluntary / involuntary) and a reason code.
- Every joiner in the last month appears in both the employee master and the payroll run — or has a documented reason if not (mid-month joiner processed next cycle).
- Headcount from the HRMS reconciles to the payroll count, with a written explanation for every difference.
- People cost from payroll reconciles to the finance ledger for the month within a defined tolerance.
- No duplicate PAN or personal email across active records.
- Location and department values all exist in the master list; zero free-text entries.
- Leave balance equation reconciles for every employee.
- Any employee marked "on notice" has a last working day recorded.
- Snapshot for the month-end has been written and row count is within expected range of last month.
Publish the checklist result alongside the dashboard — a small "data quality: 12/12 checks passed" badge does more for trust than a beautiful chart.
The Metric Library: Exact Formulas and Definition Traps
Here is the core of it. For each theme: what to measure, the precise formula, and the traps that make the number wrong or argued about. Resist the urge to implement all of them. Pick 10–15 for year one.
Headcount Metrics
Headcount seems trivial until you try to reconcile three reports and get three answers.
| Metric | Formula | Definition traps |
|---|---|---|
| Active headcount | Count of employees where status = active as at a specific date | Do you include people serving notice? Include interns? Contractors? Employees on long unpaid leave or maternity leave? Pick one primary definition and label the tile with it |
| Headcount on notice | Count where status = on notice as at date | Often reported inside active headcount, which overstates effective capacity. Show it as a separate tile |
| FTE | Sum of FTE fractions of active employees (full-time = 1.0, half-time = 0.5, and so on) | Most Indian SMBs have almost no part-timers, so FTE ≈ headcount. If you have interns at 0.5 or consultants, FTE is the more honest capacity number |
| Total workforce | Own-payroll headcount + contractors + agency staff + consultants | Never mix this with headcount in the same trend line. Two separate series |
| Average headcount (period) | (Opening headcount + Closing headcount) ÷ 2, or better, the average of month-end snapshots across the period | The simple opening/closing version distorts badly when hiring is lumpy. Use the snapshot average if you have snapshots |
| Headcount growth rate | (Closing headcount − Opening headcount) ÷ Opening headcount × 100 | Meaningless without noting whether M&A, entity consolidation or an intern batch is inside it |
The headcount bridge is more useful than headcount alone, and it is the tile executives actually engage with:
Opening headcount + Joiners + Rehires + Transfers in − Voluntary exits − Involuntary exits − Transfers out = Closing headcount
If this equation does not balance to zero, you have a data defect. Make it balance before you publish anything else. Many teams discover their first real data problem the day they build this bridge.
Joiners and Leavers
| Metric | Formula | Notes |
|---|---|---|
| Joiners (period) | Count of employees with date of joining inside the period | Rehires: count separately, and flag them. A rehire is a signal worth tracking |
| Leavers (period) | Count with exit date (last working day) inside the period | Use last working day, not resignation date. Resignation date is useful separately as a leading indicator |
| Resignations submitted | Count with resignation date inside the period | This leads exits by 30–90 days depending on notice period. It is your earliest warning signal — put it on the dashboard |
| Net change | Joiners − Leavers | Positive net change with high gross churn is a treadmill: you are hiring hard just to stand still |
| Replacement ratio | Backfill hires ÷ Total hires × 100 | High ratio means growth budget is being consumed by churn |
Attrition: Where Most Dashboards Go Wrong
Attrition rate calculation is where definitional sloppiness does the most damage, because attrition is the number the board asks about.
The base formula:
Attrition rate (period) = Leavers in period ÷ Average headcount in period × 100
Annualised attrition from a monthly figure:
Annualised attrition = Monthly leavers ÷ Average headcount for the month × 12 × 100
Rolling 12-month attrition (the one to lead with):
Rolling 12M attrition = Leavers in the last 12 months ÷ Average headcount over the last 12 months × 100
Where average headcount over 12 months = sum of the 12 month-end headcounts ÷ 12 (or the average of 13 month-end points including the opening).
Why the rolling 12-month version should be your headline: single-month annualised attrition in a 200-person company is violently noisy. Three exits in a month of 200 people annualises to 18%. Four exits annualises to 24%. That six-point swing is one person. Reporting it as a trend invites bad decisions. Show rolling 12-month as the headline, monthly annualised as a secondary sparkline with an explicit noise warning.
Attrition cuts you should maintain:
| Cut | Formula | Why it matters |
|---|---|---|
| Voluntary attrition | Voluntary leavers ÷ Average headcount × 100 | The number you can actually influence through management action |
| Involuntary attrition | Terminations + performance exits + role eliminations ÷ Average headcount × 100 | High involuntary attrition points at hiring quality or unclear expectations, not at retention |
| Regretted attrition | Regretted voluntary leavers ÷ Average headcount × 100 | Regretted = you would rehire this person. Decide the flag at exit, with the manager, using a fixed rule (for example, last rating at or above "meets expectations" and manager confirms rehire eligibility). Decide it before you see the monthly number, or it becomes a way to make the number look better |
| Early attrition (first 90 days) | Leavers with tenure < 90 days ÷ Joiners in the corresponding period × 100 | Measures hiring accuracy, role clarity and onboarding, not retention. Use joiners as the denominator, not headcount |
| First-year attrition | Leavers with tenure < 12 months ÷ Joiners in the comparable 12-month window × 100 | Same logic, longer window |
| Attrition by manager | Leavers reporting to manager ÷ Average team size × 100 | Only report where average team size is above your minimum threshold — see the privacy section |
| Attrition by tenure band | Leavers in band ÷ Average headcount in band × 100 | Bands: 0–6m, 6–12m, 1–2y, 2–3y, 3–5y, 5y+. Shows where your leak is |
| Attrition by location, department, grade, gender | Same structure, different grouping | Watch small denominators |
Cohort attrition is the most decision-useful version and almost nobody at SMB scale does it. Take everyone who joined in a given month or quarter, and track what percentage is still with you at 3, 6, 12, 18 and 24 months. Plot the curves. If the Q1 cohort retains at 88% at 12 months and the Q3 cohort at 61%, something changed in how you hired or onboarded in Q3 — different sources, a rushed hiring push, a manager change, a compensation shift. Cohort curves point at causes; aggregate attrition points at nothing.
The denominator problem, stated plainly. These all describe "attrition" and give different answers on the same data:
- Leavers ÷ opening headcount
- Leavers ÷ closing headcount
- Leavers ÷ average of opening and closing
- Leavers ÷ average of monthly snapshots
- Leavers ÷ (opening headcount + joiners)
In a fast-growing company, opening headcount as the denominator inflates attrition and closing headcount deflates it — sometimes by several percentage points. Pick the average-of-snapshots method, write it into the definitions page, and never change it silently. If you must change it, restate history and say so.
One more trap: exclude interns, fixed-term contract completions and contractors from headline attrition, or you will report scary numbers every time an internship batch ends. Report them separately if they matter.
Tenure and Stability
| Metric | Formula | Notes |
|---|---|---|
| Average tenure | Sum of (as-at date − date of joining) for active employees ÷ active headcount, in years | Heavily skewed by fast growth. A company that doubled headcount last year will show falling average tenure even with zero attrition |
| Median tenure | Middle value of the tenure distribution | More honest than the mean. Use this as your headline |
| Tenure distribution | Count of active employees in each tenure band | The shape matters more than any single number. A big bulge at 2–3 years with nothing above is a promotion and progression problem |
| Stability index | Employees with ≥ 12 months service at period end ÷ Headcount 12 months ago × 100 | Tells you how much of last year's team you still have. Complements attrition nicely |
| Manager tenure in role | As-at date − date they took the current role | New managers correlate with team disruption; useful as context, not as a judgement |
Internal Mobility and Promotion
| Metric | Formula | Notes |
|---|---|---|
| Internal mobility rate | Internal moves in period ÷ Average headcount × 100 | Define a "move": lateral transfer, promotion, or change of function. Exclude title changes with no scope change |
| Promotion rate | Promotions in period ÷ Average headcount × 100 | Count only grade or level changes, not title inflation. Enforce via the job architecture |
| Internal fill rate | Roles filled internally ÷ Total roles filled × 100 | The cleanest measure of whether your career paths are real |
| Time since last move | As-at date − date of last promotion or role change | Excellent leading indicator. People at 30+ months with no move are a flight risk cluster. This is a tile, not a report |
| Promotion-to-exit lag | Median months between promotion and voluntary exit | If people leave 4–6 months after promotion, you are promoting into unsupported roles or promoting without pay correction |
Span of Control and Organisational Layers
| Metric | Formula | Notes |
|---|---|---|
| Average span of control | Total employees with a manager ÷ Number of people managers | A single average hides everything. Look at the distribution |
| Managers with span of 1–2 | Count of managers with ≤ 2 direct reports | Usually accidental structure or a title given instead of a raise. Each one is a conversation |
| Managers with span above 10 | Count of managers with > 10 direct reports | Often a real overload signal, especially with junior teams. Pair with that team's attrition and skip-level feedback |
| Organisational layers | Maximum depth from CEO to the deepest individual contributor | Above 6 layers at 300 people means you have built a mid-size company's bureaucracy without its scale |
| Manager ratio | People managers ÷ Total headcount × 100 | Watch it creep up during growth; it rarely comes back down on its own |
Span data is only as good as your reporting structure data, which is why the quarterly org data review in the hygiene checklist matters.
Diversity Representation and Progression
Keep this simple, general and privacy-aware.
- Representation: count and percentage of each group at total company level, by level or grade band, and by function. Trend it over time.
- Hiring representation: share of each group among joiners in the period, compared with representation among applicants and among people at the offer stage. This tells you where in the funnel the composition changes.
- Progression: promotion rate by group, computed as promotions in group ÷ average headcount in group. Compare across groups rather than reading any one in isolation.
- Exit representation: attrition rate by group, subject to minimum group size.
Three cautions. Collect only what you have a lawful and stated purpose for, and tell employees what you collect and why. Self-identification should be voluntary with a "prefer not to say" option, which you then report as its own category rather than dropping. And apply your minimum-group-size threshold ruthlessly here — reporting attrition for a group of four people is both statistically meaningless and a privacy breach in a small company where everyone knows who is in that group.
Recruitment Funnel Metrics
Recruiting metrics get muddled constantly. Be precise about clock start and stop.
| Metric | Formula | Definition traps |
|---|---|---|
| Time to fill | Offer accepted date − Requisition approved date (in days) | Measures the process end to end including approval delays. Use requisition approved, not requisition raised, or you measure your own budget process |
| Time to hire | Offer accepted date − Candidate's first application or sourcing date | Measures candidate experience and speed once someone is in the pipeline. Different metric, different owner |
| Time to productivity | Date of independent delivery − Date of joining | Rarely instrumented, but if your managers can mark it, it beats every other recruiting metric for real quality |
| Offer acceptance rate | Offers accepted ÷ Offers extended × 100 | Decide how to treat offers that were accepted and then reneged before joining. Recommendation: count as accepted here, and track drop-offs separately |
| Offer drop-off / joining ratio | Candidates who joined ÷ Offers accepted × 100 | Critical in India where candidates hold multiple offers through a 60–90 day notice period. Often the single biggest leak in the funnel |
| Stage conversion rate | Candidates entering stage N+1 ÷ Candidates entering stage N × 100 | Compute for each stage. The stage with the worst conversion is where to focus |
| Source effectiveness | Hires from source ÷ Applications from source × 100; plus quality: 12-month retention and first-rating of hires by source | Volume-only source metrics reward whichever job board sends the most unqualified applicants |
| Cost per hire | (External recruiting costs + internal recruiting costs) ÷ Hires in period | Include agency fees, job board subscriptions, referral bonuses paid, and a fair share of recruiter salary. Excluding internal cost makes agencies look worse than they are and in-house look free |
| Requisition ageing | Days since approval for all open requisitions, bucketed | Shows where capacity is stuck right now, unlike time to fill, which only counts closed roles |
| Interview load | Interview hours per hire, by function | Protects engineering capacity. Useful when hiring volume spikes |
One structural warning: time to fill only counts filled roles. If your hardest roles never close, they never enter the metric, and your time to fill looks great while the org starves. Always show requisition ageing beside it.
Attendance, Absenteeism and Overtime
| Metric | Formula | Notes |
|---|---|---|
| Absenteeism rate | Unplanned absence days ÷ Total scheduled working days × 100 | Unplanned only — approved planned leave is not absenteeism. Mixing them makes a well-run leave policy look like a problem |
| Attendance regularisation rate | Regularisation requests ÷ Employee-days × 100 | High rates usually mean a broken capture process, not dishonest employees |
| Late arrivals / short hours | Count of days below the defined threshold ÷ Employee-days × 100 | Only meaningful for shift-based or client-hours-based roles. Do not apply to salaried knowledge workers; it reads as surveillance and buys you nothing |
| Overtime hours | Sum of approved overtime hours per period | Track per team. Sustained overtime in one team is a staffing gap, not a work-ethic story |
| Overtime cost ratio | Overtime cost ÷ Total payroll cost × 100 | If it is climbing, compare the cost of the overtime against the cost of an additional headcount. Often the hire is cheaper |
| Shift coverage | Filled shift slots ÷ Required shift slots × 100 | For operations, support, retail, manufacturing, healthcare |
Leave Balance and Liability
Leave liability is the metric your CFO cares about and HR usually cannot produce on demand.
- Leave liability = Σ (encashable leave balance in days × per-day cost) across all active employees. Per-day cost is usually basic + DA ÷ working days per month, but follow your own policy and applicable state rules. Get the definition confirmed with finance once, in writing.
- Average leave balance per employee = Total encashable balance ÷ Active headcount.
- Leave utilisation rate = Leave days availed ÷ Leave days accrued in the period × 100. Persistently low utilisation is a burnout warning and a growing balance-sheet liability at the same time.
- Balance concentration: count of employees whose balance exceeds 80% of the carry-forward cap. These are the people who will either lose leave or trigger a cash outflow. Both are worth a conversation before year end.
- Zero-leave employees: people who have taken no leave in six months. Flag for their manager. This is genuinely useful and genuinely well received.
Payroll and Cost Metrics
| Metric | Formula | Notes |
|---|---|---|
| Total people cost | Gross salaries + employer PF + employer ESI + gratuity provision + bonus provision + insurance + other benefits | Agree the inclusion list with finance once and freeze it |
| Cost per employee (monthly) | Total people cost for the month ÷ Average headcount for the month | Rises when you hire senior, falls when you hire junior. Always read with the level mix |
| Compensation cost ratio | Total people cost ÷ Total revenue × 100 | The number the board tracks. In services businesses it is the primary lever |
| Revenue per employee | Revenue in period ÷ Average headcount | Only meaningful with a stable revenue definition and consistent contractor treatment |
| Payroll cost variance | (Actual people cost − Budgeted people cost) ÷ Budgeted people cost × 100 | Explain variance by driver: headcount variance, mix variance, increment timing, one-time payouts, overtime |
| Month-on-month payroll movement | Current month cost − Prior month cost, decomposed by driver | Build a bridge: prior month + joiners + increments + arrears + variable payouts − exits ± other = current month. Finance will love you |
| Compa-ratio | Employee's fixed pay ÷ Midpoint of their salary band × 100 | Only works if you have bands. If you don't, build them before attempting this |
| Salary range penetration | (Pay − Band minimum) ÷ (Band maximum − Band minimum) × 100 | An alternative to compa-ratio; better when bands are wide or asymmetric |
| Out-of-band population | Count and percentage of employees below band minimum or above band maximum | Below-minimum cases are retention risks and often compliance risks. Above-maximum cases limit your future increment flexibility |
| Increment spend | Total incremental annual cost of the cycle ÷ Pre-cycle annual fixed cost × 100 | Compare planned versus actual, by department |
| Cost of attrition (indicative) | (Recruiting cost + onboarding and training cost + estimated productivity ramp cost) × Number of regretted exits | Label your assumptions clearly. This is an estimate for prioritisation, not an accounting figure — do not present it as precise |
Performance and Engagement
Performance distribution = count and percentage of employees in each rating category, shown by department, level and manager. Watch for two patterns: heavy central clustering, which means the process is not differentiating; and wildly different distributions between managers of similar teams, which means calibration is not working. Also track review completion rate (reviews submitted ÷ reviews due × 100) by manager and the median days late — process compliance is easier to fix than rating quality and often unblocks it.
Cross-cut worth building: performance rating against compa-ratio. Your high performers sitting below band midpoint are the retention conversation you should be having this quarter. This single scatter plot has probably saved more people than any engagement survey.
Engagement and eNPS. Handle with care.
- eNPS = % promoters (score 9–10) − % detractors (score 0–6), on the "how likely are you to recommend this company as a place to work" question. Report the response rate alongside it, always. An eNPS from a 34% response rate tells you about the 34%.
- Never report eNPS or engagement scores for groups below your minimum threshold. Confidentiality promises broken once are never repaired.
- Do not chase the score. Report the two or three themes from free-text comments and what is being done about them. A rising score with no action is theatre, and employees know it.
- Survey no more often than quarterly, and only if you will publish what changed as a result. Survey fatigue is real and self-inflicted.
Two Worked Examples
Formulas are abstract until you run them. Here are two calculations end to end, with the judgement calls made explicit.
Worked Example 1: Annualised Attrition
The company. A services startup, offices in Pune and Hyderabad. Monthly headcount snapshots at each month-end for the last 12 months (April through March):
412, 419, 428, 431, 440, 447, 451, 448, 455, 462, 470, 478
Opening headcount on 1 April was 405.
Exits in the 12 months: 74 total. Breakdown: 61 voluntary resignations, 9 performance-related exits, 4 fixed-term contract completions.
Step 1 — Decide what counts as a leaver. The 4 fixed-term completions are excluded from headline attrition; those roles ended as planned. That leaves 70 exits for the headline figure, of which 61 are voluntary and 9 involuntary.
Step 2 — Compute average headcount. Using the average of the 13 points (opening plus 12 month-ends):
Sum = 405 + 412 + 419 + 428 + 431 + 440 + 447 + 451 + 448 + 455 + 462 + 470 + 478 = 5,746
Average headcount = 5,746 ÷ 13 = 442
Step 3 — Overall attrition.
70 ÷ 442 × 100 = 15.8%
Step 4 — Split it.
- Voluntary: 61 ÷ 442 × 100 = 13.8%
- Involuntary: 9 ÷ 442 × 100 = 2.0%
Step 5 — Regretted attrition. Of the 61 voluntary exits, the manager and HR flagged 38 as regretted using the pre-agreed rule (last rating at or above "meets expectations" and rehire-eligible).
38 ÷ 442 × 100 = 8.6% regretted attrition
This is the number that should drive action. The gap between 15.8% headline and 8.6% regretted is the story: roughly 45% of the churn is either performance-related, contract-related or genuinely fine.
Step 6 — Show what the denominator choice would have done. Same 70 exits:
| Denominator | Value | Attrition |
|---|---|---|
| Opening headcount (405) | 405 | 17.3% |
| Closing headcount (478) | 478 | 14.6% |
| Simple average of opening and closing | 441.5 | 15.9% |
| Average of 13 snapshot points | 442 | 15.8% |
A spread of 2.7 points on identical data. In a company growing 18% a year the effect is moderate; in one that doubled, it would be far larger. This is exactly why you write the definition down once and stop debating it.
Step 7 — Early attrition, separately. Joiners in the same 12 months: 143. Leavers with tenure under 90 days: 17.
17 ÷ 143 × 100 = 11.9% early attrition
Note the different denominator — joiners, not headcount — because this measures hiring and onboarding accuracy. Nearly one in eight new hires left inside three months. That is a hiring-process problem, and it will not be fixed by a retention initiative aimed at the whole company.
Step 8 — Cut by cohort. Split the 17 early leavers by joining source: 11 came through one agency that supplied 34 of the 143 hires. That agency's early attrition is 32%, versus 5% for everyone else. You now have a decision to make on Monday morning, which is the entire point of the exercise.
Worked Example 2: Compa-Ratio and Range Penetration
Setup. Software Engineer II band in your Hyderabad office:
- Band minimum: ₹12,00,000
- Band midpoint: ₹16,00,000
- Band maximum: ₹20,00,000
Six people in the band:
| Employee | Fixed CTC | Last rating | Tenure |
|---|---|---|---|
| A | ₹13,20,000 | Exceeds | 2y 4m |
| B | ₹15,60,000 | Meets | 1y 8m |
| C | ₹17,80,000 | Meets | 3y 1m |
| D | ₹11,40,000 | Exceeds | 2y 11m |
| E | ₹16,00,000 | Meets | 0y 9m |
| F | ₹20,80,000 | Below | 5y 6m |
Step 1 — Compa-ratio for each (fixed pay ÷ midpoint × 100):
| Employee | Calculation | Compa-ratio |
|---|---|---|
| A | 13,20,000 ÷ 16,00,000 | 82.5% |
| B | 15,60,000 ÷ 16,00,000 | 97.5% |
| C | 17,80,000 ÷ 16,00,000 | 111.3% |
| D | 11,40,000 ÷ 16,00,000 | 71.3% |
| E | 16,00,000 ÷ 16,00,000 | 100.0% |
| F | 20,80,000 ÷ 16,00,000 | 130.0% |
Step 2 — Group compa-ratio. Sum of fixed pay = ₹94,80,000. Sum of midpoints (6 × ₹16,00,000) = ₹96,00,000.
94,80,000 ÷ 96,00,000 × 100 = 98.8%
The group average looks perfectly healthy. This is the trap: aggregate compa-ratio conceals the distribution almost completely. Always show the scatter, never just the average.
Step 3 — Range penetration ((pay − min) ÷ (max − min) × 100), range width = ₹8,00,000:
| Employee | Calculation | Penetration |
|---|---|---|
| A | (13.20 − 12.00) ÷ 8.00 | 15.0% |
| B | (15.60 − 12.00) ÷ 8.00 | 45.0% |
| C | (17.80 − 12.00) ÷ 8.00 | 72.5% |
| D | (11.40 − 12.00) ÷ 8.00 | −7.5% |
| E | (16.00 − 12.00) ÷ 8.00 | 50.0% |
| F | (20.80 − 12.00) ÷ 8.00 | 110.0% |
Step 4 — Read it against performance and tenure.
- D is the emergency. Below the band minimum, an "exceeds" rating, nearly three years of tenure. Negative penetration means you are paying below your own stated floor. Fix this off-cycle, this month, before the market fixes it for you.
- A is the second call. "Exceeds", 82.5% compa-ratio, over two years in. Underpaid relative to contribution. Plan a correction in the next cycle at minimum, and tell A it is coming.
- F is a different problem. Above band maximum with a "below expectations" rating and long tenure. You cannot cut pay, so the answer is either a performance conversation with a clear plan, a role that justifies the cost, or freezing fixed pay and moving future reward into variable. Whatever you decide, decide it — drifting is the expensive option.
- B, C, E are fine. No action.
Step 5 — Compute the cost of the fixes. Bringing D to ₹13,80,000 (86.3% compa-ratio) and A to ₹14,80,000 (92.5%) costs ₹2,40,000 + ₹1,60,000 = ₹4,00,000 a year. Against the recruiting cost, ramp time and delivery disruption of replacing two capable engineers, that is not a close call. Putting the number in the room is what makes the decision easy.
This is what an HR analytics dashboard is for: not the chart, the Monday-morning decision.
Designing the Actual Dashboard
Build three views for three audiences. Same data model, different questions, different permissions. Do not try to serve all three with one screen.
View 1: The One-Page Executive View
For founders, the leadership team and the board. Monthly. Rolling 13 months of trend on every chart so seasonality is visible. Ten to twelve tiles, one page, no scrolling.
| Tile | Metric | Cut / display | Cadence | Owner | Decision it drives |
|---|---|---|---|---|---|
| Headcount | Active headcount, on-notice shown separately | Total, with department split | Monthly | HR Head | Hiring pace, budget re-forecast |
| Headcount bridge | Opening + joiners − exits = closing | Waterfall | Monthly | HR Ops | Validates all other numbers; explains growth |
| Attrition | Rolling 12M total, voluntary, regretted | Three lines, 13 months | Monthly | HR Head | Whether retention becomes a leadership priority |
| Attrition hotspots | Rolling 12M attrition by department and location | Bar, above minimum group size only | Monthly | HR Head | Where to run a targeted intervention |
| Early attrition | Under-90-day exits ÷ joiners | Single number + trend | Monthly | Talent Acquisition Lead | Whether to change sourcing or onboarding |
| Open roles | Open requisitions, ageing buckets | Stacked bar by function | Monthly | TA Lead | Where to add recruiter capacity or reset the spec |
| Offer health | Offer acceptance rate and offer-to-join ratio | Two numbers + trend | Monthly | TA Lead | Compensation competitiveness, closing process |
| People cost | Total people cost, cost per employee | Trend + variance to budget | Monthly | HR Head + Finance | Budget action, hiring freeze or release |
| Compa-ratio distribution | Compa-ratio by level | Box plot or scatter | Quarterly | Comp Lead | Off-cycle corrections, next increment budget |
| Manager span | Span distribution, layers count | Histogram | Quarterly | HR Head | Org design changes |
| Internal mobility | Internal fill rate, promotion rate | Two numbers, rolling 12M | Quarterly | HR Head | Whether career paths are real |
| Engagement | eNPS with response rate | Trend, company level only | Half-yearly | HR Head | Prioritising the next people initiative |
Rules for this view: no tile without an owner. No tile below your minimum group size. Every tile shows a comparison — prior month, prior year, or plan. A number with no reference point cannot be interpreted.
View 2: The HR Operations View
For the HR team. Weekly or daily. This is a working screen, not a presentation. It is allowed to be dense.
| Tile | Metric | Cadence | Owner | Decision it drives |
|---|---|---|---|---|
| Data quality checks | Pass/fail on the 12 hygiene checks | Daily | HR Ops | Fix before anything is published |
| Pending exits | Employees on notice, last working day, clearance status | Weekly | HR Ops | Full-and-final planning, backfill trigger |
| Joiners this week | Names, documents pending, asset and access status | Weekly | HR Ops | Onboarding follow-ups |
| Probation due | Confirmations due in the next 30 days | Weekly | HR Business Partner | Chase confirmation decisions before they lapse |
| Attendance exceptions | Missing punches, pending regularisations, unapproved absence | Daily | HR Ops | Payroll input accuracy |
| Leave anomalies | Negative balances, balances above cap, zero-leave employees | Monthly | HR Ops | Policy enforcement, burnout flags |
| Leave liability | Total encashable liability, movement | Monthly | HR Ops + Finance | Provisioning, year-end planning |
| Payroll variance | Month-on-month movement by driver | Monthly | Payroll Lead | Catch errors before the run is locked |
| Requisition tracker | Open roles, stage, days open, recruiter | Weekly | TA Lead | Reallocate recruiter effort |
| Funnel conversion | Stage-to-stage conversion by role family | Monthly | TA Lead | Fix the weakest stage |
| Review completion | Percentage complete by manager, days overdue | During cycle | HRBP | Chase specific managers |
| Compliance calendar | Statutory filings and deadlines, status | Monthly | HR Ops + Finance | Avoid penalties |
View 3: The Manager View
For people managers, showing only their own org — direct and indirect reports. This is the view most companies skip and it is where behaviour actually changes.
| Tile | Metric | Cadence | Owner | Decision it drives |
|---|---|---|---|---|
| My team | Headcount, on-notice, open roles | Live | Manager | Capacity planning |
| Team tenure | Tenure distribution, months since last role change per person | Monthly | Manager | Career conversations before people leave |
| My team's attrition | Rolling 12M, versus company and versus function | Monthly | Manager | Whether to ask for help |
| Leave and absence | Team leave taken, upcoming leave, zero-leave flags | Weekly | Manager | Workload balance, burnout check |
| Pending actions | Approvals pending, reviews due, confirmations due | Live | Manager | Unblock HR processes |
| Compa-ratio (if permitted) | Team compa-ratio distribution, out-of-band flags | Quarterly | Manager | Increment recommendations |
| 1:1 and skip-level coverage | Percentage of team with a 1:1 in the last 14 days | Monthly | Manager | Restart the ones that lapsed |
Two design rules for the manager view. Compare fairly — a manager of six people needs to see their attrition against similar teams, not against the company average, which is dominated by larger teams. And never show tiles the manager cannot influence. If they cannot change it, it is not a management tool, it is a stick.
Leading Versus Lagging Indicators
Most HR dashboards are entirely lagging. Attrition, cost, headcount and performance ratings all tell you what already happened. You need leading indicators to intervene while intervention is still possible.
| Lagging (outcome) | Corresponding leading indicators |
|---|---|
| Voluntary attrition | Resignations submitted this month; months since last role change; compa-ratio below 90% among high performers; manager 1:1 frequency; internal application rate; leave balance spikes (people banking leave before exit); sudden drop in leave usage |
| Early attrition | Offer-to-join drop-off; onboarding completion rate; 30-day new-hire check-in scores; first-90-day manager 1:1 coverage |
| Time to fill | Requisition ageing; stage conversion rates; interview scheduling turnaround; candidate drop-off rate at each stage |
| Payroll cost overrun | Approved but unfilled requisitions with their budgeted cost; overtime hours trend; offer-versus-band variance on recent offers |
| Manager churn risk | Span above threshold; number of new managers with no manager training; team review completion rate |
The practical rule: for every lagging metric on the executive view, put at least one corresponding leading indicator on the HR ops or manager view. You cannot change last quarter's attrition; you can change what happens to the eleven people who have crossed 30 months without a move.
Segmentation, Small Samples and Privacy Thresholds
Segmentation is where the insight is — company-level attrition of 16% means almost nothing, while 34% attrition in one team of thirty means everything. But segmentation is also where you break confidentiality and generate garbage conclusions.
Set a minimum group size and enforce it in the query, not by policy memo. For a company between 50 and 1,000 people, a threshold of 5 for operational metrics like attrition and headcount, and 8 to 10 for anything sensitive — engagement scores, diversity cuts, performance distribution, compensation — is a reasonable starting point. Below the threshold, the tile shows "group too small to report" rather than a number.
Watch out for differencing attacks. If you report a department of 12 and also its two sub-teams of 7 and 5, and you suppress the team of 5, someone can still derive it by subtraction. Suppress complementary cells too. This is a real risk in small companies where everyone knows the org chart.
Small numbers are volatile, not just private. One exit from a team of six is 17% attrition. Do not let anyone build a narrative on it. Where sample sizes are small, report counts rather than percentages — "2 exits from a team of 9" is honest; "22% attrition" implies a precision you do not have.
Restrict manager views to their own org, enforced in the tool with row-level security, not by asking people not to look. And be deliberate about compensation visibility: many companies give managers compa-ratio for their team but not absolute salaries of skip-level reports. Decide once, document it, and configure it.
The Monthly People Review That Actually Changes Decisions
The dashboard is the input. The monthly people review is where it turns into decisions. Without this meeting, you have built a very expensive wallpaper.
Who Attends
Keep it to six to nine people. Founder or CEO, HR head, finance lead, TA lead, and two or three functional leaders on rotation. Ninety minutes, monthly, same slot, never cancelled. The HR head owns the meeting and the pre-read.
Do not invite everyone. A people review with twenty attendees becomes a status update because nobody wants to raise a hard issue in front of that many people.
The Pre-Read Rule
The pack goes out 48 hours in advance: the executive dashboard plus a written narrative of 300–500 words. Nobody presents slides in the room. The first ten minutes are silent reading if people did not do the pre-read — awkward once, and then everyone reads it.
The narrative is the discipline. It forces the HR head to have a point of view rather than reciting numbers. Structure it in four parts:
- What changed. "Rolling 12-month attrition moved from 14.9% to 15.8%. The move is entirely in the Hyderabad delivery team, which contributed 9 of the 12 exits this quarter."
- Why we think it changed. "Six of the nine cited compensation in exit interviews. We checked: 40% of that team sits below 90% compa-ratio, against 18% company-wide. Two of the nine cited the manager change in October."
- What we are doing. "Off-cycle correction for 7 people, ₹9.4 lakh annualised, effective this month. Manager coaching for the new team lead starting next week. Skip-levels with the remaining team in the next two weeks."
- What we need from this room. "Approval for the ₹9.4 lakh off-cycle spend. A decision on whether to backfill all nine or hold three positions until Q3."
That last section is what makes the meeting worth attending. If nothing is being asked of the room, cancel the meeting.
The Agenda
Minutes 0–5 — Data confidence. HR ops confirms the hygiene checks passed and flags anything restated since last month. Thirty seconds if all is well. This ritual is why nobody spends the meeting arguing about whether the numbers are right.
Minutes 5–15 — Headcount and cost. The bridge, plan versus actual, cost per employee, and the forward view: what are we committed to hiring, what does that cost, are we still inside plan.
Minutes 15–35 — Attrition and retention. Rolling 12-month with the voluntary and regretted split. Then straight to hotspots — which team, which cohort, which manager, which tenure band. Named risks: who are the ten people we would be most damaged to lose, and what has been done for each in the last 30 days. This is the highest-value twenty minutes in the meeting.
Minutes 35–50 — Talent acquisition. Open requisitions by ageing, offer acceptance and offer-to-join, funnel bottlenecks, and any role that has been open more than 90 days — for each of those, a decision: change the spec, change the pay, change the source, or close the role.
Minutes 50–65 — Organisation and capability. Rotating focus by month: span and layers one month, internal mobility and promotions the next, performance and calibration in cycle months, compensation review in the quarter before increments.
Minutes 65–80 — One deep dive. One topic, pre-selected, with real analysis. Cohort retention by hiring source. Compa-ratio versus performance for a specific function. Absenteeism in a shift-based team. Rotate it so that over a year you have covered everything properly at least once.
Minutes 80–90 — Decisions and owners. Read back every decision made, with an owner and a date. Circulate within 24 hours. Open the next meeting by reviewing the previous month's decisions. This closing loop is what separates a people review from a people discussion.
Writing the Narrative, Not Just the Numbers
Three habits that make the narrative land.
Lead with the exception, not the summary. Not "attrition is 15.8%." Instead: "Attrition is stable everywhere except Hyderabad delivery, which is at 31% and accounts for most of the company movement."
Quantify the "so what" in money or capacity. "Nine exits at an average replacement cost of roughly ₹3.5 lakh plus a three-month ramp means about ₹32 lakh and eighteen person-months of delivery capacity." State assumptions plainly.
Say what you do not know. "We do not have a clear read on why the Q3 cohort is retaining worse. We are running exit-interview text analysis and five stay interviews, and I will have a view by next month." Admitting uncertainty buys enormous credibility and stops the room from inventing explanations.
Anti-Patterns to Avoid
Vanity metrics. Total applications received. Total training hours delivered. LinkedIn followers. Number of engagement events. These go up when you do more of something, not when something improves. If a metric can only rise, it is not a metric, it is a scoreboard for effort.
Metric proliferation. A dashboard with 60 metrics is a dashboard with zero priorities. Every metric costs data quality effort, definitional debate, and someone's attention every month. Ten metrics you trust beat sixty you don't. When you add a metric, retire one.
Borrowed benchmarks. Someone shares an industry attrition figure, and suddenly your target is to be below it. Benchmark numbers vary enormously by industry, city, company stage, function mix, seniority mix and how the source defined the metric — and you almost never know their definition. A published figure computed on closing headcount, including contract completions, in a different sector, tells you nothing about your business. Build your own baseline instead: run your metrics for three or four quarters, understand your normal range and its seasonality, and set targets against your own trend. Your history is the only benchmark whose definition you control. Use external figures for rough orientation at most, and never as a target.
Surveillance-style metrics. Keystroke counts, screen time, minute-level login tracking, idle-time reports for salaried knowledge workers. These reliably damage trust, produce gaming rather than performance, and in a tight talent market they show up in exit interviews. Attendance metrics are legitimate where shifts, client hours, statutory records or safety require them. They are not legitimate as a general productivity proxy.
Punishing managers with numbers they cannot control. A manager whose team's attrition spiked because the company froze increments while a competitor opened an office next door did not fail. If you put a number in front of a manager, ask whether they have levers — pay, promotion, hiring, workload, role design. If they have none, giving them the number without the levers just teaches them to hide problems and argue about data instead of fixing anything.
Precision theatre. Reporting attrition as 15.83% implies a precision your data does not support. One late exit-date entry moves it. Round to one decimal at most, and show counts alongside percentages for small groups.
Changing definitions quietly. If you must change how a metric is computed, restate the whole history on the new basis, announce the change, and show both series for one period. Silent redefinition is the fastest way to lose the room permanently.
Reporting without a comparison. "Cost per employee: ₹94,000." Against what? Last month, last year, plan, or a different segment. A standalone number cannot be acted on.
The annual report trap. People analytics done once a year for a board deck is archaeology. Monthly cadence with a stable definition beats an annual deep dive every time.
A 90-Day Implementation Roadmap
You do not need a data team or a six-figure BI budget. You need a clean HRMS, month-end snapshots, a definitions document and a recurring meeting. Here is a sequence that works for a 50–1,000 person company.
| Phase | Weeks | What you do | Output | Owner |
|---|---|---|---|---|
| Scope and decisions | 1–2 | List the 10–15 decisions leadership actually makes about people each quarter. Work backwards to the metrics that inform them. Kill everything else. Agree audiences for the three views | One-page metric shortlist with a decision and an owner per metric, signed off by the founder | HR Head |
| Definitions | 2–3 | Write the exact formula, inclusions, exclusions, denominator and data source for each shortlisted metric. Settle the hard ones: notice-period inclusion, contractor treatment, regretted-attrition rule, people-cost inclusion list | Metric definitions document, agreed with finance | HR Head + Finance |
| Data audit | 3–5 | Run the hygiene checklist against live data. Count the defects. Reconcile HRMS headcount to payroll count, and people cost to the finance ledger | Defect list with counts, owners and target dates | HR Ops |
| Data cleanup | 5–8 | Fix duplicates, standardise location and department masters, backfill missing exit dates and reasons, run the manager-confirmation exercise for reporting lines, build the job architecture and map everyone | Clean employee master, validated org structure, controlled value lists | HR Ops + all managers |
| Snapshots and history | 6–7 | Turn on month-end snapshotting. Reconstruct the last 12 months of month-end headcount from joining and exit dates so you have trend on day one | Snapshot table with 12+ months of history | HR Ops / whoever owns the HRMS |
| Build v1 | 8–10 | Build the executive view only. Twelve tiles maximum. Use whatever you have — the HRMS's own reporting, a spreadsheet, or a BI tool. Do not let tool selection delay this | Executive dashboard, refreshed monthly | HR Head |
| Validate | 10–11 | Have three people compute five key metrics independently and reconcile every difference. Review with finance. Fix the definitions document where reality disagreed with it | Signed-off numbers; a version-1 definitions doc that matches the build | HR Head + Finance |
| First people review | 11–12 | Run the meeting. Pre-read 48 hours ahead with a written narrative. Capture decisions and owners. Ask everyone what was missing | Meeting notes, decision log, v2 metric backlog | HR Head |
| Manager view | 12–13 | Build the manager view with row-level security. Pilot with three or four managers before rolling out. Run a 30-minute session on how to read it | Manager dashboard live for pilot group | HR Head + HR Ops |
| Iterate and harden | Ongoing | Retire unused tiles. Add leading indicators. Automate the hygiene checks. Publish a data quality badge alongside the dashboard | Stable monthly rhythm | HR Ops |
Deliberately not in the first 90 days: predictive attrition models, sentiment analysis, engagement platforms, anything with "AI" in the pitch. They are not useless, but they are useless on dirty data and without a decision-making forum. Earn them in year two.
The single most common way this goes wrong is spending eight weeks choosing a tool and two weeks on definitions. Invert it. A correct, agreed metric in a plain spreadsheet beats a wrong metric in an expensive dashboard, every time.
Frequently Asked Questions
How many metrics should a company of 200 people actually track?
Ten to fifteen on the executive view, twenty to twenty-five across the HR operations view, and six to eight on the manager view — with heavy overlap between them. The constraint is not what you can compute, it is what you can keep clean and act on every month. If a metric has not informed a single decision in two quarters, retire it. A short dashboard that people trust is worth far more than a comprehensive one they have learned to ignore.
Should I include employees serving notice in headcount?
Report both, and label clearly. Your primary headcount tile should show active employees including those on notice, with on-notice broken out beside it, because for capacity and budget planning you need to know that 8 of your 478 people are leaving within 60 days. What matters far more than the choice is consistency: pick one for the headline, write it in the definitions document, and never switch without restating history.
What is the right attrition denominator for a fast-growing startup?
Average headcount computed from month-end snapshots across the measurement period. In a company growing 40% or more a year, using opening headcount inflates attrition substantially and using closing headcount deflates it — on the same data, with the same exits. The snapshot average tracks the population that was actually exposed to the risk of leaving. If you do not yet have snapshots, use the simple average of opening and closing as an interim, start snapshotting immediately, and note the method on the tile.
How do I handle attrition metrics for very small teams without breaching confidentiality?
Set a minimum group size — five is reasonable for operational metrics, eight to ten for sensitive ones — and enforce it in the query so undersized groups render as "group too small to report." Suppress complementary cells too, or people will derive the hidden number by subtraction. For small teams, report absolute counts rather than percentages, because one exit from a team of six is not "17% attrition" in any meaningful sense. Roll small teams up to the next organisational level for reporting.
Can I benchmark our attrition against industry figures?
Treat external figures as rough orientation only, never as a target. Published benchmarks vary widely by industry, city, company stage, function mix and seniority mix, and — most importantly — by how the source defined the metric, which you usually cannot see. A figure computed on closing headcount, including fixed-term contract completions, in a different sector and a different city, is not comparable to yours. Build your own baseline from three or four quarters of your own consistently-defined data, understand your normal range and its seasonal pattern, and set targets against that. Your own history is the only benchmark whose definition you fully control.
Do I need a BI tool, or can I do this in spreadsheets?
Spreadsheets are entirely adequate up to roughly 300 people if two conditions hold: your HRMS can export clean, consistent data, and you have automated month-end snapshots so history is not being reconstructed by hand each time. The point at which you should move to a proper tool is when manual report preparation exceeds two days a month, when you need role-based access so managers see only their own org, or when you need to join data across attendance, payroll, recruitment and performance regularly. Start in a spreadsheet, prove the metrics matter, and let the pain tell you when to upgrade.
Who should own the HR analytics dashboard?
One named person — usually the HR head at this company size, with HR operations owning data quality and production. Individual metrics get individual owners: talent acquisition owns funnel metrics, payroll owns cost metrics, HR business partners own the metrics for their client groups. Finance should co-own anything involving money so you never present a people-cost number that contradicts the ledger. What does not work is "HR owns it" as a collective, which reliably means nobody chases the numbers that are drifting.
How long before an HR analytics dashboard actually pays for itself?
Usually within the first two or three monthly people reviews, and typically through one specific find rather than gradual improvement — a hiring source with 30% early attrition that you stop using, a cluster of high performers sitting below band minimum that you correct before they resign, an overtime pattern that turns out to be cheaper to solve with one additional hire, or a leave liability nobody had provisioned for. The first cycle rarely produces insight because you spend it fixing data. Budget three months to trustworthy numbers, and expect the value to come from decisions, not from the dashboard existing.
What if our managers push back on being measured?
They are usually right to, if the numbers are wrong or if they lack the levers. Deal with both. Show managers their own data privately before it appears anywhere else, and give them a straightforward way to flag errors — every correction they submit improves your data and buys you credibility. Then check honestly whether they can influence what you are showing them. If a manager cannot affect pay, promotion, hiring or workload, showing them attrition is not management, it is blame. Give the levers along with the number, compare them against similar-sized teams rather than a company average dominated by bigger ones, and frame the conversation as "what support do you need" rather than "explain this figure."
How do we keep the dashboard from going stale after six months?
Two mechanisms. First, the standing decision log: open every people review by reviewing last month's decisions, which forces the numbers to stay tied to actions. Second, a twice-yearly tile audit where you ask, for every tile, "which decision did this change in the last six months?" and delete the ones with no answer. Dashboards die from accumulation, not neglect. The discipline is subtraction.
Bringing It Together
An HR analytics dashboard is not a reporting project. It is a decision-making habit that happens to need some charts. The companies that get value from people analytics at 50 to 1,000 employees are not the ones with the best tooling — they are the ones that wrote down their definitions, cleaned their data, snapshot their history every month, kept the metric list short, tied every number to a named owner and a real decision, and held the same 90-minute meeting every month without cancelling it.
Start smaller than feels satisfying. Twelve tiles. One narrative. One meeting. One decision log. Get those working, get people trusting the numbers, and then earn the right to add cohort curves, leading indicators and the deeper cuts. The failure mode is almost never "we did not build enough" — it is "we built a lot and nobody used it."
And do the boring part first. Month-end snapshots starting tonight, the data hygiene checklist on the first working day of every month, and a definitions document that settles the arguments before they start. Those three things will carry more weight in your second year than any visualisation you build.
If your data currently lives across a biometric device, three spreadsheets, a payroll vendor's portal and a WhatsApp group, the dashboard is not your first problem — the plumbing is. CozyHR keeps attendance, leave, payroll and employee records in one system, with effective-dated records and clean exports, so headcount, attrition, leave liability and payroll cost all come out of a single source of truth instead of a monthly reconciliation exercise. If you are about to start building your people reporting, it is worth seeing what it looks like when the underlying data is already tied together — take CozyHR for a spin and start your next people review with numbers nobody has to argue about.