Decision intelligence: From data to decisions

Three different questions

Descriptive business intelligence shows what happened and how things stand: stock levels, backlogs and utilisation. A forecasting model estimates what is likely to follow under particular assumptions. Decision intelligence adds the question of which available action makes sense under the actual conditions. These tasks build on one another, but require different evidence.

Consider an illustrative case: a planner expects a shortage of components. An express shipment might protect the delivery date while consuming the transport budget. Changing the production sequence might buy time while delaying another order. A more accurate arrival forecast cannot resolve this trade-off. Alternatives, consequences and business priorities must become explicit. The team can then explain why it accepts additional cost or negotiates a revised date.

Context connects data to action

Palantir describes its Ontology through data, logic, actions and security. This architectural idea illustrates why operational decisions need more context than a collection of separately considered metrics.

In our example, stock belongs to a location, a quality status and existing order commitments. A transport has capacity, an arrival window and a responsible planner. These relationships establish whether an apparently available quantity can actually be reassigned. Data age and provenance matter too. When a warehouse entry conflicts with a receipt notification, the system should expose the disagreement. A shared definition of “available” is more useful here than another generated summary of inconsistent figures.

Sources: Palantir: Why create an Ontology?

Define what a good outcome means

Optimisation tools such as Google OR-Tools address allocation, scheduling and transport problems using mathematical constraints. The modelling of objectives and limits determines which solution the system favours.

In the example, approved materials and binding capacity limits may be hard constraints. Transport effort, schedule stability and delivery reliability require trade-offs. The business must decide which deviations are negotiable and who may authorise them. Comparing a few understandable options with their assumptions can make this discussion practical. Teams can also examine whether the preferred option remains workable if arrival slips further. If every option violates a hard constraint, a reasoned escalation is the appropriate result. Making priorities visible also exposes changes in weighting that could unexpectedly favour different orders.

Sources: Google: About OR-Tools

Use feedback to examine decisions

The NIST AI Risk Management Framework connects governance, context, measurement and risk management throughout the lifecycle. Our implication for decision systems is that responsibility and review must continue during operation.

After an approved stock transfer, expected and observed consequences are compared. Did the material arrive? Did the shortage move elsewhere? A favourable outcome alone does not prove that the recommendation caused it. A pilot therefore needs a documented comparison with the existing process and evaluation criteria established in advance. NyxAI approaches sovereign decision intelligence through this connection between understandable options, controlled optimisation and operational responsibility. The components required to support it must be established for the particular process.

Sources: NIST: AI RMF Core