Overview
AI Capabilities in CRM was the smallest section of the retired AI Associate outline. Its two objectives were to identify CRM AI capabilities and to describe the benefits of AI as they apply to CRM. The focus was recognition, not configuration: you needed to know what each capability does and which team gains from it.
The capabilities fall into a few families. Predictive features score records, such as lead and opportunity scoring, or forecast outcomes. Recommendation features suggest the next best action or the right knowledge article. Conversational features, from Einstein Bots to Agentforce agents, answer customers and employees in natural language. Generative features draft emails, summarize cases and calls, and create knowledge content. Salesforce has since folded many of these into Agentforce and grounds them in Salesforce Data 360, but the underlying ideas match what the outline described.
How it's tested
Expect a role and a pain point: a sales rep who cannot prioritize leads, a service agent buried in long case histories, a marketer guessing at send times. You pick the CRM AI capability that addresses it, or the benefit a stakeholder should expect from it.
-
Feature for a pain point
A team describes a slow or manual task, and you choose scoring, recommendations, a bot or generated content as the matching capability.
-
Benefit statement
You select the realistic benefit, such as productivity, personalization or faster resolution, rather than an overstated claim about removing people.
-
Cloud alignment
The same idea appears in sales, service and marketing variants, and the correct option fits the team in the scenario.
A contact center says agents spend minutes reading long case threads before replying. The pain is time spent absorbing context, not deciding priority. A generative case or conversation summary addresses that directly. Lead scoring is a sales capability, and a bot would deflect new questions rather than help an agent with an existing case. So the summary feature, with the agent reviewing it, is the best fit.
Objectives 2
Official objectives from the Salesforce outline, in outline order.
2.1 Identify CRM AI capabilities.The predictive, recommendation, conversational and generative features in Salesforce.
- What it means
- Salesforce embeds AI across its clouds. Sales examples include lead and opportunity scoring, forecasting and generated sales emails. Service examples include case classification, article recommendations, bots and agents, and case or call summaries. Marketing examples include engagement scoring, send-time optimization and generated content. Custom predictions on any field were possible with Einstein Prediction Builder.
- What to understand
- Group features by what they return. A score or forecast is predictive. A suggested article or next best action is a recommendation. A bot or agent that converses with users is conversational AI. A draft, reply or summary is generative. Know that Salesforce now presents conversational and generative capabilities largely through Agentforce, while many predictive features still carry the Einstein name.
- How it shows up
- Scenarios give a team and a problem; pick the capability family and the feature that fits that team.
- Common confusion
- Thinking a bot and a recommendation engine are the same because both suggest answers.
2.2 Describe the benefits of AI as they apply to CRM.Productivity, personalization, better decisions and faster customer service.
- What it means
- The benefit of CRM AI is measured in outcomes for users and customers. Reps focus on the deals most likely to close, agents resolve cases faster with summaries and suggested articles, marketers personalize at scale, and customers get answers at any hour through self-service. Leaders get more consistent forecasts and insight into trends.
- What to understand
- Frame benefits as augmentation. AI handles repetitive steps and surfaces insight, while people make the judgment calls and own the customer relationship. Understand that benefits depend on good data and user trust; a score nobody believes delivers nothing. Be ready to connect a benefit to a metric such as handle time, conversion rate or customer satisfaction.
- How it shows up
- Choose the benefit that is realistic and tied to a measurable outcome, not the most dramatic claim.
- Common confusion
- Equating AI benefit with reducing headcount rather than improving productivity and experience.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Lead scoring
Ranks leads by how closely they resemble past conversions.
- Remember
- Built from the org's own conversion history.
- Tested as
- The standard example of predictive AI in sales.
- Trap
- Thinking it is a manual point system set by an admin.
-
Opportunity scoring
Estimates how likely an open deal is to close.
- Remember
- Shows the factors driving the score.
- Tested as
- Pairs with forecasting in sales scenarios.
- Trap
- Confusing it with the forecast category a rep selects.
-
Next best action
Recommends an offer or step for a specific customer at that moment.
- Remember
- Combines business rules with predictions.
- Tested as
- Common answer for guided service or sales actions.
- Trap
- Calling it a purely generative feature.
-
Article recommendations
Suggests relevant knowledge articles to service agents.
- Remember
- Learns from which articles resolved similar cases.
- Tested as
- Fits agent-productivity scenarios.
- Trap
- Choosing a bot when the user is the agent, not the customer.
-
Einstein Bots and Agentforce agents
Conversational AI that handles customer or employee requests in natural language.
- Remember
- Bots follow designed dialogs; agents reason over topics and actions.
- Tested as
- Fits self-service and deflection scenarios.
- Trap
- Assuming any bot uses an LLM.
-
Generative summaries
Condensed overviews of cases, calls or records.
- Remember
- Saves reading time but needs human review.
- Tested as
- Fits long case history scenarios.
- Trap
- Treating a summary as an official record without checking it.
-
Generated sales emails
Drafts personalized outreach using CRM data.
- Remember
- The rep edits and sends; AI drafts.
- Tested as
- Typical generative sales example.
- Trap
- Assuming the email sends automatically.
-
Einstein Prediction Builder
Point-and-click tool for custom yes/no or numeric predictions on Salesforce fields.
- Remember
- Needs enough historical records with the outcome populated.
- Tested as
- Answers questions about custom predictions without code.
- Trap
- Thinking it generates text.
-
Salesforce Data 360
Salesforce's data platform, formerly Data Cloud, that unifies customer data for AI and grounding.
- Remember
- Renamed from Data Cloud in October 2025.
- Tested as
- Explains where unified profiles come from.
- Trap
- Assuming it is a separate AI model.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
CRM AI capability families
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Predictive | Scores, classifies or forecasts records. | Returns a likelihood or value. | Expecting it to write content. |
| Recommendation | Suggests an action, offer or article. | Returns a ranked suggestion for a person to accept. | Treating suggestions as automatic actions. |
| Conversational | Interacts in natural language through bots or agents. | Talks with the user directly. | Placing it inside the agent's console only. |
| Generative | Creates drafts, replies and summaries. | Returns new content for review. | Skipping human review because the text reads well. |
Benefits by team
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Sales | Prioritize leads and deals, draft outreach. | Benefit shows up as conversion and rep time. | Choosing service metrics for a sales scenario. |
| Service | Faster resolution through summaries, articles and self-service. | Benefit shows up as handle time and satisfaction. | Assuming deflection is the only benefit. |
| Marketing | Personalized content and better send timing. | Benefit shows up as engagement. | Equating personalization with collecting more personal data. |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Bot for agent productivity
- What candidates pick
- A customer-facing bot when the problem is agents reading long cases.
- Why it looks right
- Bots are the most visible service AI.
- Why it's wrong
- The user who needs help is the agent, so a summary or article recommendation fits better.
Clue to watch for The struggling user is an internal employee.
-
Headcount as the benefit
- What candidates pick
- An option promising to eliminate the service team.
- Why it looks right
- It sounds like the biggest return.
- Why it's wrong
- Salesforce frames CRM AI as augmenting people, with humans keeping judgment and oversight.
Clue to watch for Answer choices include an absolute claim.
-
Manual scoring as AI
- What candidates pick
- A point-based scoring rule built by an admin.
- Why it looks right
- Both produce a lead score.
- Why it's wrong
- Manual rules do not learn from conversion history, so they are not predictive AI.
Clue to watch for The scenario mentions learning from past outcomes.
-
Generative for ranking
- What candidates pick
- Generated emails to decide which leads to call first.
- Why it looks right
- Generative features are new and prominent.
- Why it's wrong
- Deciding priority needs a predictive score; drafting comes after that choice is made.
Clue to watch for The question says prioritize or rank.
-
Data platform as model
- What candidates pick
- Calling Data 360 the AI that makes predictions.
- Why it looks right
- It is central to Salesforce AI messaging.
- Why it's wrong
- Data 360 unifies and supplies data; models and agents consume it to predict or generate.
Clue to watch for Wording about unifying customer profiles.
Answer framework
Use these steps to choose a CRM AI capability and state its benefit.
- Identify who is struggling: a sales rep, a service agent, a marketer, a manager or an end customer, because that narrows the cloud and feature set.
- Name the output that would help: a score, a suggestion, a conversation or a draft, and map it to predictive, recommendation, conversational or generative AI.
- Check that the needed data exists, such as conversion history for scoring or resolved cases for article recommendations.
- State the benefit as a measurable outcome for that role, like shorter handle time, higher conversion or more relevant engagement.
- Reject options that remove human oversight or overstate the result, and prefer the one where people stay in control of the decision.
Practice question
Want more like this? The SF AI Associate practice test has questions for every domain.
Checklist
0 of 10
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Give this section two to three hours; it is small but easy to lose points on through feature mix-ups. Start by making a grid with three rows for sales, service and marketing and four columns for predictive, recommendation, conversational and generative capabilities. Fill each cell with one Salesforce example and the measurable benefit it delivers. Lead scoring, article recommendations, bots and case summaries should all appear somewhere.
Next, read the Salesforce help pages or Trailhead modules for the features you are least sure of, focusing on what data each one needs. A capability with no data behind it delivers no benefit, which links this section to Data for AI. Then practice rewriting dramatic benefit claims into realistic ones, such as turning "replaces support staff" into "lets agents resolve routine cases faster." Because Salesforce has renamed much of this product area since the exam launched, note the current names, including Agentforce and Data 360, so that newer documentation does not confuse you. Finish with a few scenario questions under time.
FAQ
How much of the Salesforce AI Associate exam was AI Capabilities in CRM?
AI Capabilities in CRM was 8% of the exam, roughly three of 40 questions. It was the smallest of the four sections.
Did I need to configure Einstein features for this section?
No. The objectives asked you to identify capabilities and describe benefits. Setup steps and permissions belonged to admin and specialist exams, not this one.
Are the Einstein names still current?
Partly. Many predictive features such as lead and opportunity scoring still use the Einstein name, while Salesforce presents conversational and generative work mainly under Agentforce. Data Cloud is now called Salesforce Data 360.
What is the difference between a bot and a recommendation?
A bot converses directly with a user, often a customer, to answer or route a request. A recommendation suggests an article, offer or action to a person who decides whether to use it.
Which benefits did the exam favor?
Realistic, measurable ones: productivity for reps and agents, personalization for customers, faster resolution and better decisions. Claims about removing people or never making mistakes were usually wrong.
How does this section relate to the Agentforce Specialist exam?
It gives the capability map that Agentforce builds on. The Agentforce Specialist exam goes much deeper into agents, prompts, actions and grounding, so treat this section as the overview.
Other Artificial Intelligence & Machine Learning certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 2 knowledge
Work through practice questions that pair sales, service and marketing problems with the right Salesforce AI capability and its realistic benefit.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the SF AI Associate AI tutor Get any concept explained your way