Getting the Data Right

Start with the question

Analytics fails most often not because the statistics are wrong but because nobody established what decision the analysis would inform. Before any extract, write down: what question is being asked, who will act on the answer, what action is available to them, and what result would change that action. If no available action depends on the answer, the analysis is reporting, not analytics.

Data sources

• Core HR system — headcount, roles, grades, contracts, managers, start and leave dates.
• Payroll — pay, allowances, overtime, hours. Usually the most accurate source and often inconsistent with the HR system.
• Time and attendance — hours worked, absence, shift patterns.
• Recruitment system — applications, stages, sources, time to hire.
• Learning system — training completion, competence, expiry.
• Surveys — engagement, exit, pulse.
• Operational systems — productivity, quality, safety, sales, which is where the value usually lies because it connects people data to outcomes.

Definitions matter more than techniques

Most disputes about HR numbers are definitional. Agree and document, in one place:

• Headcount — all contracts, or active only? Are long-term absent, secondees, contractors, agency workers and apprentices included?
• Full-time equivalent — against what standard week, and how are variable-hours workers treated?
• Leaver — does it include end of fixed-term contracts, transfers between entities, deaths and retirements?
• Voluntary versus involuntary — where does a resignation during a capability process sit?
• Absence — calendar days or working days; is authorised unpaid leave included?
• Time to hire — from vacancy approval, advert, or application to offer, acceptance or start date?

A metric without a written definition will be calculated three different ways and produce three different answers, which destroys trust faster than any error.

Data quality

• Check completeness — missing manager links, blank start dates, unassigned cost centres.
• Check consistency across systems — headcount in the HR system versus payroll is the standard reconciliation.
• Check for duplicates, particularly where people have had multiple contracts.
• Test the extremes: negative tenure, ages outside plausible ranges, salaries an order of magnitude out.
• Fix at source. Cleaning the extract each month means the same errors forever.