Decision and evidence brief
A focused question, accountable decision owner, agreed definitions and the evidence needed to answer it.
07HR Service and Analytics
HR Analytics Framework
Turn a clear people question into trustworthy analysis, an informed decision and a review of what happened next.
A clear question before analysis
Findings with visible limitations
Action followed by a learning review
HRAFQuestion-to-Decision Chain
Start with your situation
When this framework helps.
- Dashboards show patterns but leaders disagree about their meaning.
- Definitions differ between HR, Finance or operational reports.
- A people initiative needs an evidence-based review of its contribution.
The decisions it supports
- What decision will this analysis inform?
- What can the available data support, and what remains uncertain?
- What action is proportionate to the findings?
The framework at a glance
Question-to-Decision Chain
Turn people and service data into useful insights for decisions and improvement.
Explore the complete model
The approach
Begin with a decision worth improving
Start with a useful question, establish trustworthy evidence and connect analysis to a decision, an accountable action and a test of what changed.
An expanding dashboard does not automatically improve a people decision. Leaders may see turnover, recruitment and absence figures without knowing which question matters, whether definitions are consistent or what action the evidence supports. More complex analysis can amplify those weaknesses if the foundations remain unclear.
The HR Analytics Framework starts with the decision a person needs to make. It helps HR teams agree the question, check the data, compare explanations and recommend a proportionate action. Workforce planning, service performance, employee experience and programme reviews can all use this approach; the question determines the analysis, not the dashboard already available.
Useful starting questions include:
- Monitoring: where is recruitment taking longer, using consistent stage definitions?
- Diagnosis: which evidence could explain a delay, and what else might account for it?
- Evaluation: after a process change, did the intended experience or outcome improve?
Each requires a different evidence plan. People analytics can support business problem-solving with relatively simple descriptive work; predictive methods are not a prerequisite. This question-led position is consistent with the CIPD's people analytics guidance. CIPD people analytics factsheet.
What you can take forward
Useful work. Clear next steps.
Interpretation with limitations
Findings that distinguish observed patterns, alternative explanations and conclusions the data cannot support.
Action and learning log
A record of the decision, expected change, review evidence and reasons to revise the response.
HRAF · In practice
Five links in the Question-to-Decision Chain
Each link has an explicit handover test. If that test cannot be met, the next step is to resolve the gap, narrow the question or explain why a decision must be made under uncertainty.
Question
State the decision, who owns it, when it is needed and what would change as a result. Define the population, timeframe and outcome of interest. Ask what evidence might challenge the initial explanation. Separate a monitoring request from a diagnostic question or an evaluation of whether an intervention worked.
Handover test: A decision owner can explain how the answer will be used.
Evidence
Identify the minimum necessary sources and agree definitions, permissions and data responsibilities. Examine completeness, accuracy, timeliness, consistency, coverage and the origin and handling of each source. Reconcile headcount, dates, organisation boundaries and the population used to calculate each rate before combining records. Qualitative evidence and professional judgement can explain context that administrative data does not capture.
Handover test: The evidence is sufficiently relevant, usable and permitted for the stated purpose, with remaining limitations documented.
Analysis
Choose a method proportionate to the question and available evidence. Begin with distributions, rates, trends and carefully selected comparisons. Examine missing records, unusual values, competing explanations and how findings change when assumptions change. Use advanced models only when they add decision value and can be tested responsibly.
Handover test: Another competent reviewer can understand the method, check the result and identify its limitations.
Decision
Explain what the evidence supports, what remains uncertain and which options are available. Consider operational feasibility, employee impact, cost and risk. Separate observed findings from interpretations and recommendations. A named human decision owner weighs the analysis alongside relevant context and records the rationale.
Handover test: An authorised owner selects an action, further investigation or a reasoned decision to make no change.
Evaluation
Agree how the action will be reviewed before implementation. Set a baseline, intended outcome, review period and signals of unintended harm. Record what was actually delivered. Compare results with the expected change and revise the original explanation when the evidence requires it.
Handover test: The organisation can explain what it learned and whether to continue, adapt or stop.
HRAF · In practice
Establish data quality before claiming insight
A compact data agreement records the business definition of each measure, source system, steward, refresh frequency, population and calculation. Changes to definitions or system coverage are visible in the reporting history. Missing data is shown as missing, rather than silently converted into zero or a favourable outcome.
For example, turnover analysis needs an explicit departure definition and a stated workforce population for the rate. Comparing functions can be misleading if one includes temporary contracts, another excludes recent starters and a third has reorganised during the period. The first useful output may therefore be a reconciled measure and a qualified baseline.
The diagnostic reviews a sample of current reports and follows selected figures back to source. It tests access, calculations, explanations and whether anyone uses the output. Findings are recorded as Fit for this decision, Usable with conditions, Further evidence needed or Do not use. These are decision-specific judgements, with reasons and review dates, rather than permanent labels for a dataset.
HRAF · In practice
Distinguish patterns from causes
A relationship between workload and resignations may justify investigation, but it does not establish that reducing workload will produce a particular retention benefit. Leadership changes, labour-market conditions, role mix or measurement choices may also explain the pattern. A before-and-after improvement can coincide with an intervention without being caused by it.
Where the question concerns impact, plan an appropriate evaluation with a credible comparison where feasible. Consider timing, selection effects and alternative explanations. Use suitably qualified analysts for experimental or quasi-experimental designs, including comparisons intended to estimate impact, and explain the assumptions. Where causal evidence is unavailable, describe the result as an association or observed change.
Report uncertainty, the size of any observed difference and its practical importance together when the method allows. Statistical significance alone does not establish importance or fairness. Small samples, selective responses and repeated exploratory comparisons require particular caution. A predictive model's accuracy also does not establish that its inputs are fair, necessary or causally relevant.
From insight to action
Four records that make the work usable
The decision brief captures the question, owner, options and timing. The data agreement documents sources, definitions, access and quality. The analysis record preserves methods, assumptions, checks, versions and limitations. The action-and-learning log connects the decision to delivery and subsequent review.
Depending on scope, these can be accompanied by an executive insight brief, a small set of decision-focused visualisations and an analytics improvement roadmap. Findings include the period, population, source and comparator. Charts use readable labels and accessible alternatives; colour is supporting information rather than the only signal.
A reusable dashboard is appropriate only when there is a recurring decision and an owner for its maintenance. A one-off question may be better answered through a short, well-evidenced brief.
HRAF · In practice
A consulting process built around a live question
We begin by agreeing one or more priority decisions with HR and business leaders. We review the available evidence, clarify definitions and identify privacy or quality constraints. A focused analytical pilot then demonstrates the method and exposes practical capability gaps.
Interpretation sessions bring together analysts, HR specialists and the people responsible for acting. The final stage establishes review routines, documentation and knowledge transfer. Wider capability development follows the gaps found in the pilot rather than assuming that every organisation needs a large analytics team or new platform.
HRAF · In practice
Protect confidentiality and retain human judgement
Use purpose-limited access, minimum necessary data and agreed retention. Aggregate reporting by default. Set disclosure controls before analysis, including small-group suppression, complementary suppression where totals could reveal hidden values, and restrictions on repeated filtering. Removing names alone does not make records anonymous.
Consider how sensitive data, linked sources and third-party tools alter risk. Involve privacy, information-security and relevant legal specialists where appropriate, including for cross-border processing. Do not infer protected or sensitive characteristics from names, images or other proxies. Avoid individual surveillance and unsupported employee risk labels.
HRAF supports organisational and service decisions. It does not provide automatic individual employment verdicts, psychometric validation or a guarantee of legal compliance. Any high-impact application requires a separate, explicit review of purpose, evidence, fairness and decision safeguards.
Your questions, answered
Frequently asked questions
Do we need a sophisticated HR system?
No. A focused question and a small, well-understood dataset can provide useful evidence. System investment should follow the requirements and constraints identified.
Can analytics explain why people leave?
It can reveal patterns and plausible explanations. Causal conclusions require stronger evidence than a dashboard or a single exit-survey result.
Can we benchmark against other organisations?
Only where the source, population and definitions are comparable. Internal trends may be more useful when external comparisons are weak.
Is this a validated assessment?
It is a consultancy framework for analytical practice. Any survey or model used within it needs its own evidence and interpretation limits.
An illustrative application
Turn a disputed dashboard pattern into a useful decision
Illustrative scenario.
The situation
A recruitment dashboard reports a longer time to hire. Leaders propose more recruiter capacity, but functions use different start dates and some include paused vacancies in the calculation.
Explore the evidence
Use Question to establish the decision owner and the capacity question. Evidence reconciles vacancy definitions, stage dates, paused roles and the relevant population. Analysis examines where delay occurs and how role mix or approval dependencies could explain the pattern. Missing or inconsistent dates remain visible as limitations.
Agree the next step
Present findings, plausible explanations and uncertainty separately. If delay is concentrated in approval, the owner may test that workflow before increasing recruiter capacity. Record why the chosen action is justified and what further evidence could change it; the example illustrates a decision route, not a predetermined diagnosis.
Review what changes
Evaluation checks implementation, stage timing and the candidate experience against the agreed baseline. Consider changes in vacancy mix and other conditions before claiming impact. Use the action-and-learning log to decide whether to continue, revise or investigate further.
Put the framework to work
Make the next people decision better informed
HR Delivery can help you select a worthwhile question, establish reliable evidence and build an analytical approach that leaders can understand, use and review.