Start with decision intent

The first question is not which algorithm to use. It is what decision the organisation must make in the next cycle. Descriptive workflows are ideal when the primary task is to understand current performance, identify bottlenecks, or compare regions, teams, or programmes. Predictive workflows are appropriate when a forward looking decision is needed, such as forecasting demand, estimating risk, or allocating resources before a reporting window closes.

If decision makers need clarity on what is happening now, descriptive analytics is usually enough and often more useful. It produces dashboards, trend analysis, and segmentation that explain why outcomes look the way they do. Predictive models should only be introduced when the decision itself depends on a future outcome that can be estimated with reasonable confidence and the organisation is prepared to act on those estimates.

Assess data readiness before choosing predictive

Predictive models are only as good as the data history they are trained on. If you have short, inconsistent, or biased historical data, a predictive model can create false certainty. In many public and private sector settings, descriptive analytics is safer because it highlights data gaps rather than covering them with model assumptions.

A quick readiness scan includes data volume, stability of definitions, frequency of missing values, and whether the data generating process is consistent over time. If any of these are weak, a descriptive approach paired with a data quality improvement plan will produce more reliable outcomes.

• Use descriptive analytics when data history is limited or inconsistent.
• Move to predictive only when data definitions and pipelines are stable.
• Document uncertainty if predictive outputs will shape budget or staffing.

Cost, risk, and implementation readiness

Predictive workflows often require additional engineering, monitoring, and model governance. That cost is justified when the decision impact is high and the organisation can absorb the operational overhead. If the organisation lacks model monitoring, retraining plans, or clear accountability, then descriptive analytics is more sustainable and less risky.

Another practical filter is change management. If users are not ready to trust a model, the output will not be used. In those cases, a descriptive workflow with strong storytelling and clear KPIs can build the confidence needed before a predictive approach is introduced later.

A simple decision framework

I use a short decision matrix to keep this conversation grounded in reality. It can be used in a scoping workshop or internal planning session.

• Is the decision about understanding what happened, or what will happen?
• Is there enough historical data with stable definitions?
• Will stakeholders act on the prediction, or only use it for insight?
• Can the team maintain model monitoring and retraining?
• Is the cost of a wrong prediction acceptable?

If most answers point to uncertainty or low readiness, use descriptive analytics and plan for predictive later. If most answers point to stability and high decision impact, then predictive modelling is justified.

Implementation note

Add a simple decision matrix visual to this page when available. It helps non-technical stakeholders see the tradeoffs and agree on an approach before data work begins.