Salesforce Dictionary - Free Salesforce GlossarySalesforce Dictionary
DictionaryNNext Best Action
AIBeginner

Next Best Action

Next Best Action (NBA) is a Salesforce feature that surfaces recommended actions on a record based on rules, predictive models, or both.

§ 01

Definition

Next Best Action (NBA) is a Salesforce feature that surfaces recommended actions on a record based on rules, predictive models, or both. The user sees a list of recommendations on the page layout (call this contact, offer this discount, escalate this case) and accepts or rejects each one. Behind the recommendation is a strategy in the Strategy Builder canvas: a sequence of load, branch, and prioritize nodes that decides which actions from a defined action set qualify for the current record and in what order.

Next Best Action sits between predictive scoring and automation. A predictive score sitting on a record is dormant until someone consumes it. A flow that fires automatically removes the human from the decision. NBA is the in-between: the score informs a recommendation, the user decides whether to act, and the platform tracks acceptance rates. It is one of the most under-used features in the Salesforce AI portfolio because the setup is unfamiliar (Strategy Builder is its own canvas), but for any team running recommendation workflows it is the right tool.

§ 02

How Next Best Action turns scores into recommended actions

Recommendations as a custom object

Recommendations are records on the Recommendation object. Each recommendation has a name, description, image, accept label, reject label, and a target action (a flow to run when the user accepts). Recommendations are reusable across strategies. The same Offer 10 percent discount recommendation can show up in a strategy for sales reps, a strategy for service agents, and a strategy in a portal. The action link triggers when the user clicks Accept, calling the assigned flow with the recommendation context.

Strategies and the Strategy Builder canvas

A strategy is a graph of nodes that processes recommendations. The Load nodes bring recommendation records into the strategy. The Branch nodes filter (only show this recommendation if the lead score is above 50). The Map nodes set per-recommendation values. The Prioritize node sorts the surviving recommendations and returns the top N. Strategies are versioned and activated independently. Multiple strategies can run on the same page, each scoped to a different surface (sales console, service console, customer community).

Where recommendations show up on the page

The Einstein Next Best Action component is dropped onto a Lightning page in App Builder. The component is bound to a specific strategy and the strategy returns recommendations for the current record. The component renders a card stack: title, description, image, accept and reject buttons. Accept fires the assigned flow. Reject records the rejection and removes the card. The acceptance and rejection history is tracked on the Recommendation Reaction object, which feeds reporting on which recommendations work and which do not.

Predictive inputs through Einstein and Discovery

Strategies can branch on predictive scores. The most common pattern is using a Lead Score, Opportunity Score, or Einstein Prediction Builder output as a branching condition: only surface this recommendation when the score is above the threshold. For more complex predictive logic, CRM Analytics (formerly Einstein Discovery) models can be plugged into strategies as predict nodes. The strategy calls the model, receives the score, and branches accordingly. This is how NBA composes predictive signals with rule-based filtering.

The accept-reject feedback loop

Every acceptance and rejection is a data point. The Recommendation Reaction history shows which recommendations users actually clicked and which they ignored. This feedback is the closest thing to a measurable ROI on the NBA configuration. Strategies that surface ignored recommendations need to be revised. Strategies whose top recommendations are accepted at 60 percent are doing real work. The platform does not auto-tune the strategy based on feedback; the team does, by reading the reaction reports and refining the branch logic.

NBA versus flow versus Agentforce

Three tools can drive in-context action recommendations and people often mix them up. A flow runs automation directly, no user opt-in. NBA shows a recommendation card the user can accept or reject. Agentforce can take action autonomously based on an LLM's reasoning. The right tool depends on stakes and reversibility. Use flow for low-stakes deterministic action (auto-create the follow-up task). Use NBA for medium-stakes recommendations where human judgment matters. Use Agentforce for tasks where the action is bounded by topics and human oversight is configured.

Common pitfalls when teams adopt NBA

Two failure modes appear repeatedly. The first is over-recommendation: the strategy returns ten recommendations on every record, users start ignoring the whole component, and the feature dies of irrelevance. The fix is to set a strict top-N (often three) and prioritize aggressively. The second is recommendation collisions, where two strategies on the same page show overlapping recommendations and confuse the user. The fix is to scope strategies cleanly by audience and surface. NBA does not police strategy overlap; the team must.

§ 03

How to set up a Next Best Action strategy

Setting up NBA is a five-step exercise: define recommendations, build the strategy, drop the component on a page, test, and measure. The order matters. Recommendations are the units the strategy works with, so they have to exist first.

  1. Define the action set as Recommendation records

    Create Recommendation records: name, description, image, accept and reject labels, target flow. Each recommendation is reusable across strategies, so think generic and parameterizable.

  2. Build the strategy in Strategy Builder

    Setup, Next Best Action, New Strategy. Use Load to bring recommendations in, Branch to filter by record fields or predictive scores, Map to set per-recommendation context, and Prioritize to return the top three.

  3. Place the Einstein Next Best Action component

    In App Builder, drop the Einstein Next Best Action component on the target Lightning page. Bind it to the strategy. Configure the layout (cards versus list) and the max recommendations displayed.

  4. Test with a range of record states

    Open records covering high score, low score, missing data, and edge cases. Verify the recommendations match the strategy intent. Click Accept and confirm the flow fires.

  5. Build a Recommendation Reaction report

    Report on the Recommendation Reaction object grouped by recommendation. Track acceptance rate weekly. Use the data to refine branches in the next strategy version.

Key options
Recommendation recordsremember

The reusable action units. Each has a name, description, image, target flow, and labels for accept and reject.

Strategy nodesremember

Load, Branch, Map, Filter, Prioritize, Recommendation Limit. The composition produces the final ordered list.

Predict nodesremember

Calls to predictive models from inside the strategy. Supports Einstein Discovery and CRM Analytics scores as inputs to branching.

Component placementremember

The Einstein Next Best Action component goes on a Lightning page. One per surface; multiple strategies can run on different pages.

Reaction trackingremember

The Recommendation Reaction object captures acceptance and rejection. The basis for measuring strategy effectiveness.

Gotchas
  • Over-recommendation kills adoption. Surfacing ten recommendations per record trains users to ignore the component. Cap at three and prioritize hard.
  • Recommendation collisions between strategies on the same page confuse users. Scope strategies cleanly by audience and surface.
  • Strategy Builder is its own canvas. Teams expecting flow-style logic find it unfamiliar. Plan ramp-up time for the first strategy.
  • Recommendation Reactions are the only feedback signal. Without a recurring report on acceptance rate, the strategy stagnates.
  • Predict nodes call models synchronously. Heavy predictive scoring inside a strategy can slow the page load. Keep predict nodes lean.
§

Trust & references

Sources

Cross-checked against the following references.

Official documentation

Straight from the source - Salesforce's reference material on Next Best Action.

Was this entry helpful?
Help us write better definitions. Quick reactions or detailed edit suggestions.

About the Author

Dipojjal Chakrabarti is a B2C Solution Architect with 29 Salesforce certifications and over 13 years in the Salesforce ecosystem. He runs salesforcedictionary.com to help admins, developers, architects, and cert/interview candidates sharpen their fundamentals. More about Dipojjal.

§

Test your knowledge

Q1. What is Next Best Action?

Q2. What does it combine?

Q3. Where do recommendations appear?

§

Discussion

Loading…

Loading discussion…