Decision support and decision intelligence
A decision support system prepares work in advance so that a person decides faster and more consistently. It does not take the decision. It assembles the evidence, applies the standing checks, states what is missing and presents the position in a settled form. Decision intelligence is the wider practice of designing decisions that way. The distinction from automation is simple: automation removes the person from the step, decision support keeps them there.
What is a decision support system?
It is the assist outcome expressed as software. The work that precedes a decision, gathering the record, reconciling sources, applying the standing checks and surfacing the exceptions, is done before the person arrives. What reaches them is a prepared position with its evidence attached.
The measure of such a system is not how often it is right. It is whether the decisions taken with it are faster and more consistent than the decisions taken without it.
How does decision support differ from automation?
Automation runs the step without anyone present. Decision support keeps the person in the step and changes what they see.
The choice between them is not a matter of ambition. A step is automated when it can be specified precisely enough that its behaviour is determined. Where it cannot, and where a person's judgement is what the step is for, the correct answer is assist. Choosing automation for a step that has not been specified does not remove the judgement. It hides it.
What does assist look like in practice?
A reconciliation arriving already matched, with only the exceptions presented and each one carrying the reason it failed. A file arriving with the relevant clauses extracted, sourced and dated, and the missing items named. A periodic report drafted from the record, with the figures traceable to their origin and the assumptions listed.
In each case the person still decides. What has changed is that the preparation is complete, uniform and inspectable, and no longer depends on who happened to do it.
When must a step stay with a person?
When it cannot yet be specified precisely enough to be anything else. That is a statement about the current state of the evidence rather than about the dignity of the work, and it is reported plainly.
Steps also stay with a person by design, where the specification names what the system must be gated from supporting. Regulated determinations, matters of professional judgement and decisions with an accountable signatory frequently sit there.
What goes wrong with decision support?
Three things, reliably.
It presents a conclusion without the evidence, so the person either accepts it without checking or rebuilds the work themselves. Neither is faster.
It presents everything, so the exceptions are not distinguishable from the routine and the reviewer's attention is spent on the wrong items. A prepared position that has not decided what matters has not prepared anything.
It is built without a record of divergence, so nobody can say later whether the person's judgement and the system's proposal ever differed. Without that record the step cannot be narrowed, widened, promoted to automation or retired, because there is no evidence on which to argue any of them. All three are specification failures rather than modelling failures, and all three are settled before anything is built.
What does a decision support system need in order to be trusted?
The same properties as anything else that is built: a stated quality threshold, the evidence it relies on and a schedule for refreshing it, a record of which sources were used, a record of where human review was applied, and a cost to serve justified against the commercial value it creates.
Without those, a prepared position is an opinion with better formatting. With them, it is an artefact the person can interrogate before signing.
Common questions
-
What is a decision support system?
A system that prepares the work preceding a decision so that a person decides faster and more consistently. It assembles the evidence, applies the standing checks, surfaces the exceptions and states what is missing. It does not take the decision.
-
What are the four types of decision support system?
The usual classification is by what the system is organised around: data-driven, model-driven, knowledge-driven and document-driven, with communication-driven sometimes added as a fifth. The classification describes the input. It says nothing about whether the decision itself has been specified, which is the part that determines whether the system gets used.
-
What is an example of a decision support system?
An exception queue for a periodic reconciliation: matched items settled without anyone present, unmatched items presented with the reason each one failed and the supporting records attached, and a note kept of what the reviewer concluded. The person still decides. The preparation no longer depends on who happened to do it.
-
What is decision intelligence?
The practice of treating a decision as a designed object: its inputs, the evidence it requires, the threshold it must meet, who takes it, and how it is recorded. A decision support system is one implementation of that.
-
Is a dashboard a decision support system?
Rarely. A dashboard reports state. Decision support prepares a specific decision, with the checks already applied and the exceptions already surfaced.
-
How do you measure whether it works?
By the speed and consistency of the decisions taken with it, and by the divergence between what the system prepared and what the person concluded. Both are recorded from the start, or they cannot be measured later.
-
Can a decision support step become an automated one?
Sometimes, once the divergence record shows that the step is determined in practice. That is an evidence-led promotion rather than a planned one.
-
Does this need a large model?
Not necessarily. Much of the value in assist lies in assembling and reconciling the evidence, which is often deterministic work. A model is used where language or judgement support is genuinely required.