KPI

KPI stands for Key Performance Indicator. It is a quantified measure with a target, used to judge progress against an objective.

A usable KPI has four parts: what is measured, how it is calculated, the target value, and the period over which it is assessed. Missing any of them produces the familiar situation where two people disagree about whether a KPI was met and discover they were computing it differently.

The qualifier key matters. An organisation can measure hundreds of things. A KPI is one of the small number that actually informs a decision, and treating every available metric as a KPI defeats the purpose of having the category.

A KRA is the area of responsibility, a KPI is the measure that shows performance in it, and a goal is what is to be achieved in the period. The three appear together on Indian appraisal forms and are used loosely enough to cause real confusion.

The practical distinction is that a KRA persists and a goal ends. Payroll accuracy remains a responsibility indefinitely; halving the error rate is a goal that either happens by March or does not.

Where an appraisal form asks for KRAs and then invites free text, what usually gets written is a list of activities. That is neither a KRA nor a KPI and it makes the review unmeasurable, which is how most appraisal conversations end up being about impressions.

Six measures cover most of what an HR function needs to answer about itself: attrition, time to fill, offer acceptance rate, attrition within ninety days, payroll accuracy and absence rate. Each is only meaningful compared against its own history rather than against a published benchmark, because definitions vary too much between organisations.

• Attrition rate, split voluntary and involuntary, and read by function and manager rather than in total.
• Time to fill, measured from requisition approval rather than from when sourcing started, because the earlier date is what the business experiences.
• Offer acceptance rate, which detects compensation and process problems earlier than attrition does.
• Attrition within ninety days, which tests selection and induction rather than retention.
• Payroll accuracy, as errors per thousand payslips, and payroll on-time rate.
• Absence rate together with frequency rate, since one long absence and eight scattered days mean different things.

Two commonly used measures are worth treating with suspicion. Cost per hire is highly sensitive to what is included and is rarely comparable between organisations or across years. Training hours per employee counts input rather than result and tends to survive because it is easy to collect.

The failure modes are consistent and mostly foreseeable.

• Measuring activity instead of outcome. Calls made rather than deals closed, CVs screened rather than positions filled. Activity is easier to count and easier to inflate.
• Too many. Fifteen KPIs on an appraisal form means the employee will work out which two affect their rating and manage those.
• Targets set without a baseline. A target invented rather than derived from the current level is either trivially met or demoralising, and nobody knows which until the period ends.
• Measures the person cannot influence. Holding a recruiter to an offer acceptance rate when compensation bands are set elsewhere measures the bands, not the recruiter.
• No stated calculation. Attrition alone has several common formulas, and two teams using different ones will disagree indefinitely.

The check worth applying to any proposed KPI is the gaming question: how could someone hit this number while making the real outcome worse? A time-to-fill target met by lowering the hiring bar is the standard example, and the answer is usually to pair the measure with a counterweight such as ninety-day attrition rather than to abandon it.

A lagging indicator reports what already happened; a leading indicator moves before the outcome does and gives someone the chance to act. Most KPI sets are built entirely from lagging measures. Attrition, revenue and payroll accuracy are all lagging. They are accurate and they arrive too late to change the thing they measure.

The pairs worth holding together in an HR function look like this.

The reason organisations report lagging measures almost exclusively is that they are easier to define and harder to argue with. A leading indicator invites the question of whether it really predicts anything, and that question is uncomfortable when the honest answer is that nobody has checked.

Checking is worth doing once. Hold last year's leading indicator against what actually happened and see whether the relationship exists in your organisation. Where it does not, stop reporting it.

A target invented rather than derived produces one of two failures: it is met in the first month, or it was never reachable and everyone knew by week three.

The sequence that avoids both starts with the baseline. Measure the current level for at least two periods before setting a target against it, because a single period may be an outlier and a target set from an outlier is wrong in a direction nobody can predict.

• State the calculation alongside the target. Attrition of twelve per cent means nothing until it says whether it is annualised, whether it includes probation exits and whether the denominator is opening or average headcount.
• Set the period explicitly, and make it long enough for the measure to be stable. Monthly attrition in a team of nine is noise.
• Decide in advance what happens if the target is missed, and whether that consequence is proportionate to how much the person actually controls.
• Review the target when the underlying conditions change rather than holding it for the sake of consistency. A hiring target set before a freeze is not a performance measure any more.

Where a KPI is being introduced for the first time, run it for a period without attaching a consequence. The first cycle almost always reveals a definitional problem, and finding it before anyone's rating depends on the number is considerably cheaper.