Next best action (NBA)

Next best action (NBA) is a decisioning approach that picks the most appropriate action for each customer from a set of eligible options, balancing customer needs and business goals.

Part of the Learn hub and the customer data glossary.

What it means in practice

NBA usually runs as a pipeline. First, eligibility and constraints remove actions that must not be offered: products the customer already holds, people who have opted out, those in collections or flagged as vulnerable, and anything the contact policy caps. Second, suitability or propensity models score what remains. Third, arbitration ranks the candidates by value, priority and fatigue and picks one, which may be "do nothing". Finally the action is delivered and the outcome is logged to feed back into the models.

The constraint layer matters most in regulated sectors. It is where policy, consent and regulation become rules, and it should run before any model so that a clever score cannot override a prohibition. Each selected action should carry a reason code recording the winning rule or top factors, which explainability later depends on.

NBA is a single decision; journey orchestration coordinates a sequence over time. They are often combined: orchestration owns the plan and NBA picks the step.

Under the EU AI Act, marketing personalisation is not high-risk in itself, but a system whose output determines access to credit or prices life or health insurance falls within Annex III, and Article 5 prohibits certain manipulative AI practices.

Example (retail banking): a customer showing signs of overdraft pressure is shown budgeting guidance instead of a credit-card offer, because a policy rule removes credit offers for that segment before any scoring.

Related terms

• Real-time decisioning: the machinery that runs it.
• Journey orchestration: the sequence it plugs into.
• Explainability: reason codes for each choice.
• Consent state: a hard constraint.
• Activation: delivers the chosen action.

Where it matters

• Real-time next-best-action in banking: without sending data to a SaaS.
• Real-time decisioning: latency budgets and where rules beat ML.