HR metrics that matter for small teams

Small companies do not need an HR dashboard with forty widgets. They need five numbers, computed the same way every month, looked at by someone who will act on them. Everything below can be done in a spreadsheet in under an hour a month; no BI tool required.

1. Headcount, split three ways

The base number everything else divides by. Track it as of the last day of each month, split by:

• Employment type: permanent, contract, intern. Statutory obligations and costs differ, so lumping them hides real information.
• Department or function, so growth is visible where it is happening.
• Joiners and exits in the month, which turns a static number into a flow you can reason about.

Compute it by counting active employees on the closing date. The only discipline needed is a clean joining and exit date per person; if your employee records are current, this is a filter, not a project.

2. Attrition

The number leadership will ask about most.

• Monthly attrition = exits in the month ÷ average headcount for the month × 100, where average headcount is (opening + closing) ÷ 2.
• Annualised: sum twelve monthly exit counts and divide by average headcount for the year. Do not multiply one bad month by twelve; a single quarter with three exits in a 30-person company will annualise into a terrifying and meaningless number.

Split it two ways that change its meaning entirely: voluntary versus involuntary (resignations versus terminations and non-renewals), and regretted versus non-regretted, an honest internal tag for whether you were sorry to lose the person. Fifteen percent attrition that is entirely non-regretted is a different company from eight percent that is all regretted top performers. Record the split at exit time, when memory is fresh.

3. Time-to-fill

How long hiring actually takes, measured from the day a role is approved to the day the candidate accepts (or, stricter, the day they join; Indian notice periods of 30 to 90 days make the two very different, so pick one definition and keep it).

• Log two dates per role: opened and closed. Time-to-fill is the difference in days; report the median across roles closed in the quarter, since one hard-to-fill architect role will wreck an average.
• Track offer-to-join drop-offs alongside it. In India, candidates serving long notice periods renege at meaningful rates, and if three of ten offers decline after acceptance, your problem is not sourcing.

This metric's main use is planning honesty: if median time-to-fill is 75 days, a backfill request raised the day someone resigns is already late.

4. Leave utilisation

Underrated, and a genuine early-warning signal.

• Utilisation = leave days taken ÷ leave days entitled for the period × 100, computed per employee and then looked at as a distribution, not just an average.
• Very low utilisation, say under 40 percent by September, predicts two problems: a December stampede of leave requests, and a growing encashment or carry-forward liability at year end. It sometimes also flags a manager whose team feels unable to take leave; compare utilisation across teams, and check the leave records for pending requests that were never actioned.
• Very high utilisation early in the year, concentrated in one person, is worth a caring conversation, not a policing one.

Pull days-taken from your leave system; entitlement is policy times months served. If balances are visible to employees all year, the December stampede shrinks on its own.

5. Payroll accuracy

The metric HR rarely tracks about itself, and the one employees feel most.

• Accuracy = payslips issued without post-run correction ÷ total payslips × 100. Count every correction: wrong LOP days, missed reimbursement, incorrect TDS, a bank rejection from a stale account number.
• Also track on-time rate: months in the year where salary credited on or before the committed date.
• Keep a one-line log of each error's cause. After three months the pattern is obvious, and it is almost always the same handoff, usually attendance corrections arriving after cut-off. Fix the handoff, not the people; a clean attendance-to-payroll link removes most of this class of error.

Anything below roughly 98 percent accuracy on a 50-person payroll means one person is wronged every other month, and they tell colleagues.

Keeping it honest

Three rules make these five numbers useful rather than decorative: define each formula once and never quietly change it; report the same numbers every month even when they are embarrassing, because trend beats level; and attach one sentence of interpretation per metric, since a number without a "so what" trains people to ignore the report. When the underlying records live in one system, most of these fall out as filters; more worked examples are in our knowledge base.

Start with these five. Add a sixth only when someone asks for it twice.