ComposableStack.AI CDP
Back to agent
Use Caseuse-case.next-best-action

Next Best Action

Stochastic recommendation of the user's likely next action, evaluated across multiple events and historical context. Lives at the Intra-session latency tier.

confidence 85%v1reviewed Apr 26, 2026prediction, nba, streaming, personalization

Next Best Action

The interaction. As the user takes action in-session, the system evaluates the event against recent history (this session and prior days) and recommends the next-best action — surfacing content, offer, or guidance.

Why it lands at Intra-session latency. The decision needs context that a single attribute lookup can't provide; it requires streaming evaluation across multiple events or a stochastic prediction. That puts the latency floor at ~10 seconds and the ceiling at a few minutes (source.real-time-clarity-md).

Why it's expensive. Increasing data and processing power applied to the decision raises cost-to-deliver exponentially with target latency. NBA at Intra-session is order-of-magnitude cheaper than NBA pushed up to Inter-page.

Architectural implications. Requires streaming infrastructure (Kafka/Pulsar/Pub-Sub class), a real-time feature store, and a decisioning service capable of low-millisecond inference. Composable stacks can deliver this; packaged CDPs offering it ship it as a higher-tier SKU.

Sources

Related

This node →

  • applies-to-latency-tierlatency-tier.intra-session— NBA requires streaming evaluation; sits in 10s-5min tier
  • requirestech-dim.dev-team.backend-stream-processing— NBA requires streaming infrastructure in composable stack
  • exemplifiesconcept.signal-to-activation-time— NBA is a canonical signal-to-activation example
  • requirescapability.identity-resolution— NBA prediction across sessions requires a unified profile; anonymous events cannot be enriched with historical context.
  • applies-to-domainorg-dim.industry.ecommerce— On-site product recommendations and browse-to-purchase nudging are the canonical NBA pattern in ecommerce.
  • applies-to-domainorg-dim.industry.retail— In-store digital experience and app-assisted shopping use NBA to guide customers toward adjacent products.
  • applies-to-domainorg-dim.industry.telecommunications— Plan-upgrade prompts and add-on offers during self-service app sessions are the telco NBA canonical form.
  • applies-to-domainorg-dim.industry.financial-services— Product cross-sell to existing customers during self-service app sessions is the canonical NBA pattern in financial services.
  • applies-to-domainorg-dim.industry.travel-hospitality— Ancillary upsell during booking flow (seat upgrades, add-ons, room upgrades) is the canonical NBA pattern in travel and hospitality.
  • applies-to-modalitymodality.web— OC-058. Web is the primary intra-session NBA channel: on-site content recommendation, product suggestion, and offer overlays fire during the active browsing session. The use-case node body defines NBA as in-session evaluation ('As the user takes action in-session') — web is the canonical digital channel for this pattern. Client-side or CDN-edge decisioning against the session's event stream is the standard deployment architecture.
  • applies-to-modalitymodality.mobile-app— OC-058. Mobile app is the co-primary intra-session NBA channel: in-app content recommendations, product discovery modules, and next-step guidance fire during active app sessions. Architecturally symmetric with web NBA — both require a session context store and a real-time decisioning call within the session window (intra-session latency tier, per use-case node). mParticle Cortex AI (mobile-first CDP) and Hightouch's Lightning Sync Engine both specifically address mobile-app intra-session NBA patterns.
  • applies-to-domainorg-dim.marketing-goal.customer-experience— OC-109. NBA drives contextually appropriate offers/messages at the right moment → CX goal.
  • applies-to-domainorg-dim.marketing-goal.customer-lifetime-value— OC-109. Maximizing long-horizon engagement and relevance via NBA → CLV optimization.