What is explainable optimization?
Explainable optimization is mathematical optimization that returns, with each decision, the reason it was chosen and the reason the alternatives were not.
What optimization does
An optimizer takes an objective, such as margin or service, and a set of constraints, such as capacity, budget and policy. It then finds the best plan that respects every constraint.
This is different from a forecast. A forecast says what is likely. An optimizer says what to do.
What the explanation adds: why this, and why not that
A plan without a reason is hard to approve. "Because the model said so" is not an answer a manager can act on.
An explained decision shows which constraint was binding, what each alternative would have cost, and which inputs the result depends on.
| The manager asks | The explanation shows |
|---|---|
| Why this plan? | The objective it serves and the limits it respects |
| Why not the other option? | What the alternative would have cost, in the same terms |
| What would change it? | The constraint or input the result depends on most |
| Where do the numbers come from? | The source records and assumptions behind each input |
Who explains and who decides
In OptiBrain the optimizer decides, against your objectives and constraints. Opti explains the result in plain language. Opti does not produce the figures. People approve.
This split keeps the explanation honest: it describes a result that the optimizer reached, and it can be checked against the trace.
Why explainability is part of human control
A person can only approve what they understand. Explained decisions are what make it safe to let software run operations autonomously, under human control.
It is one of the ideas in the Trinity of Intelligence: intelligence, automation and autonomy, bound by explainability.
Where this lives in the journey
Questions about this term
- Why does explainability matter?
- Because "the model said so" is not an answer a manager can act on. Every recommendation should show why it was chosen and why the alternatives were not.
- Is the explanation written by the optimizer?
- No. The optimizer produces the decision and the facts behind it, such as the binding constraints. Opti puts those facts in plain language.
- Can an explanation be wrong?
- It can be checked. Each figure in an explanation opens its trace: what it is, the equation, the rules applied and the source records.
- Does explainable optimization need the earlier stages?
- No, but they help. Reporting gives trusted data and OptiPlan gives the drivers, so a decision can be priced back into the P&L.
Start with a scan.
You get a prioritized AI roadmap. Scope, timing and terms are agreed before any work starts.