Chapter 6
Predictive analytics and talent planning
Forecasts, scenarios and uncertainty to support decisions without defining individual destinies.
6.1 Asking useful questions
Talent planning seeks to anticipate needs for skills and resources. The question must be specific enough to guide a decision: which competencies will be lacking, over what period and under what assumptions. Prediction without a planned action can produce costly information of little use.
Aggregate forecasts should be distinguished from individual estimates. A collective training need does not necessarily require scoring every person. Choosing the least intrusive level compatible with the purpose reduces risk and facilitates interpretation.
6.2 Turnover and causality
Turnover models identify associations between historical data and observed departures. They do not automatically establish why someone will leave a job or whether an intervention will prevent departure. A score may change when the market or working conditions change.
Responsible use avoids turning a prediction into a label that limits opportunities. Actions should improve conditions and offer support rather than punish an attributed intention. Hypotheses about causes require designs and evidence different from simple classification.
6.3 Time-based validation
Validation must respect chronological order when forecasting the future. Mixing later information into training data produces a misleading evaluation. Variables that indirectly reveal the outcome to be anticipated must also be avoided.
Comparison with simple baselines, documentation of errors and stability review are necessary. A single accuracy measure may conceal important failures. The cost of a false alarm and of an undetected case depends on the action to be taken and its effects on people.
6.4 Scenarios and capabilities
Scenarios explore plausible futures without presenting them as certainties. They can combine demand, technological change, mobility and skills availability. Their value lies in making assumptions visible and examining decisions that remain robust across several possibilities.
Planning must include internal development, job redesign and hiring where appropriate. A forecast shortage does not mean that replacing people is the only response. It may also reveal a need to improve training, accessibility or conditions for retaining knowledge.
6.5 Monitoring and withdrawal
After deployment, changes in data, errors and the actual use of predictions must be reviewed. A model may continue functioning technically while no longer serving its original purpose. Review must address both performance and unforeseen consequences.
Withdrawal is a legitimate option when value has not been demonstrated or risk cannot be mitigated. Documentation must explain what is being replaced, how continuity will be preserved and how changes will be communicated to those using the results.
Questionnaires
Chapter knowledge & Human Resources dilemmas
Dilemmas receive no moral or employability score and are not used to assess job candidates.