Salesforce has been embedding artificial intelligence into its CRM platform since 2016 under the Einstein brand. What began as a predictive scoring feature has expanded into a comprehensive AI layer that now spans every Salesforce Cloud – from lead and opportunity scoring in Sales Cloud, to case classification in Service Cloud, to personalisation in Marketing Cloud, to the fully autonomous Agentforce AI agents introduced in 2024. This guide explains how Salesforce uses AI across its platform, which Einstein features are included with standard licences versus add-ons, and where Agentforce fits into the overall AI architecture.
The best guide is the one that makes AI feel useful in context.
A useful explanation should help the reader see how AI fits into the platform.
That means the guide should focus on practical assistance rather than abstract AI language.
For many organisations, the value is in making the system more responsive without making it less controlled.
It should also show where AI helps most and where humans still need to make the final call.
A good guide should explain what the AI is meant to do and how it supports the CRM workflow.
That makes the topic important for teams that want AI to improve real work rather than create hype.
How Salesforce uses AI in CRM is useful to study because AI features are increasingly built into the everyday work of sales and service teams. Salesforce Einstein and Agentforce are part of that shift, helping users get support from the platform itself.
The best guide is the one that makes AI feel useful in context.
A useful explanation should help the reader see how AI fits into the platform.
That means the guide should focus on practical assistance rather than abstract AI language.
For many organisations, the value is in making the system more responsive without making it less controlled.
It should also show where AI helps most and where humans still need to make the final call.
A good guide should explain what the AI is meant to do and how it supports the CRM workflow.
That makes the topic important for teams that want AI to improve real work rather than create hype.
How Salesforce uses AI in CRM is useful to study because AI features are increasingly built into the everyday work of sales and service teams. Salesforce Einstein and Agentforce are part of that shift, helping users get support from the platform itself.
The Einstein AI Brand: What It Covers
Salesforce uses “Einstein” as the umbrella brand for AI across the platform. Einstein encompasses:
- Einstein Prediction Service: the underlying machine learning engine that trains models on org-specific Salesforce data
- Einstein Generative AI (Einstein GPT / Einstein 1): large language model capabilities for content generation and natural language interaction
- Agentforce: the autonomous AI agent platform where AI takes actions within Salesforce – not just predicting or generating, but doing
- Feature-specific Einstein tools: Einstein Lead Scoring, Einstein Opportunity Scoring, Einstein Activity Capture, Einstein Next Best Action, Einstein Bots – each a distinct product within the Einstein family
Einstein AI Features in Sales Cloud
Einstein Lead Scoring
Einstein Lead Scoring analyses historical Lead data to identify the patterns that correlate with Lead conversion. The model trains on converted and not-converted leads in the org’s history – field values, demographic data, behavioural signals – and assigns each new Lead an Einstein score (0-100) and a tier (High, Medium, Low). Sales reps can sort their Lead view by Einstein Score to focus on the most likely converters. Unlike Zoho Zia’s scoring (which works with as few as 75 converted leads), Einstein Lead Scoring requires a minimum of 1,000 converted leads and 1,000 non-converted leads in the past 6 months to generate a model – making it best suited to established sales teams with significant historical data.
Einstein Opportunity Scoring
Einstein Opportunity Scoring assigns each open Opportunity a score based on: how similar the opportunity’s current state is to historical won deals, activity levels (are emails and calls happening?), and deal characteristics (amount, close date, stage). The score updates automatically as the opportunity progresses. Einstein Opportunity Scoring is included with Sales Cloud Enterprise and above – it does not require Sales Cloud Einstein as a separate add-on for the standard version.
Einstein Activity Capture
Einstein Activity Capture automatically syncs emails and calendar events from Gmail or Outlook/Microsoft 365 to the related Salesforce records – eliminating the need for reps to manually log activity. When a rep sends an email to a Contact or meets with an Account, Einstein Activity Capture detects the Salesforce record match and creates an activity log. It also keeps Salesforce Contact and Account data in sync with the connected email directory. Einstein Activity Capture is included with Sales Cloud Enterprise and above.
Einstein Conversation Insights
Einstein Conversation Insights transcribes and analyses sales call recordings (integrated with Zoom, Salesforce Meetings, Dialpad, and other call platforms) to surface: competitor mentions, pricing discussions, next steps committed on the call, and sentiment trends. Managers use Conversation Insights to coach reps at scale – reviewing call summaries without listening to full recordings. This is Salesforce’s native equivalent to tools like Gong and Chorus.
Einstein Deal Insights and Next Best Action
Einstein Next Best Action uses predictive models and rule-based strategies to surface contextual recommendations on Salesforce records: “Offer a discount – this deal is at risk,” “Send a proposal – this contact has visited the pricing page,” “Request a reference call – this opportunity is in the late stage.” Recommendations are configured by admins combining Einstein predictions with business rules and are displayed on record pages as action cards.
Einstein AI Features in Service Cloud
Einstein Case Classification
When a new Case is created (via email, web form, or customer portal), Einstein Case Classification analyses the Case subject and description and predicts the appropriate field values: Case Type, Case Reason, Product, Priority. The predicted values are applied automatically or surfaced as suggestions to the agent, reducing the manual triage effort. Einstein Case Classification requires 1,000 closed cases with consistent field data to train an effective model.
Einstein Article Recommendations
Einstein Article Recommendations analyses the content of an open Case and recommends Knowledge articles that are likely to resolve the issue – surfaced in the Service Console for the agent while they are working the case. The model learns from which articles were actually attached to resolved cases, improving recommendations over time.
Einstein Reply Recommendations
Einstein Reply Recommendations surfaces suggested response text for agents in live chat conversations – based on how similar conversations were resolved historically. Agents can select a suggested reply and personalise it rather than typing responses from scratch, improving handle time and consistency.
Einstein Bots
Einstein Bots are Salesforce-native chatbots deployed on Experience Cloud portals, Salesforce’s messaging channels, or embedded in websites. Bots handle common service queries without agent involvement: password resets, order status lookups, appointment scheduling, case deflection. Einstein Bots integrate natively with the Service Cloud case management system – when a bot cannot resolve an issue, it creates a case and routes it to a human agent with full conversation context preserved.
Einstein AI in Marketing Cloud
Einstein Send Time Optimisation
Einstein Send Time Optimisation analyses each individual subscriber’s email open history and predicts the time of day and day of week when they are most likely to open an email – delivering the email at that individual-optimal time rather than at a scheduled batch time. According to Salesforce’s published data, send time optimisation typically improves open rates by 2-5% on large subscriber volumes.
Einstein Engagement Scoring
Einstein Engagement Scoring in Marketing Cloud scores each subscriber on their likelihood to: open an email, click a link, convert, or become a disengaged subscriber. Marketers use engagement scores to suppress disengaged subscribers from campaigns (protecting sender reputation) and to create re-engagement journeys for medium-engagement subscribers before they disengage completely.
Einstein Content Selection
Einstein Content Selection dynamically selects the most relevant content block or product recommendation to display in an email or web personalisation zone – from a defined content library – based on the individual recipient’s predicted preferences. Each subscriber sees a different version of the email with the content most relevant to them, without requiring manual segmentation or A/B test setup.
Agentforce: The Next Generation of Salesforce AI
Agentforce, announced in 2024 and generally available from late 2024, represents Salesforce’s shift from AI-as-insight to AI-as-action. Agentforce AI agents can autonomously take actions within Salesforce (and connected systems) without human initiation – operating as always-on digital workers.
How Agentforce Works
Agentforce agents are built on the Einstein 1 Platform using three components:
- Atlas Reasoning Engine: the underlying LLM-based reasoning layer that processes context, determines the appropriate action, and verifies the output before acting
- Data Cloud: provides agents with unified customer data context – so the agent knows the full customer relationship before taking any action
- Salesforce Actions: pre-defined actions the agent can take – create a record, send an email, update a field, run a Flow, query external data – defined by admins using Agent Builder (a low-code configuration interface)
Agentforce Use Cases
- Sales Development Agent: autonomously researches inbound leads (using Data Cloud and web search), generates personalised outreach, schedules meetings, and updates Salesforce records – without SDR involvement
- Service Agent: handles customer service queries end-to-end across chat and email – resolving common issues, creating cases, and escalating to humans only when needed
- Customer Success Agent: monitors customer health scores, proactively reaches out to at-risk accounts, and schedules QBRs (Quarterly Business Reviews)
- Marketing Agent: creates campaign briefs, generates email copy, builds Journey segments, and monitors campaign performance – triggering optimisations autonomously
Agentforce Pricing
Agentforce is priced on a consumption basis – approximately $2 per conversation (an agent-customer interaction handled end-to-end). Volume pricing is available for high-volume deployments. This consumption model means Agentforce cost scales with usage, making it appropriate for organisations that can demonstrate clear ROI per agent interaction (e.g., deflecting a service case that would otherwise cost $15-50 in agent time).
What’s Included vs Add-On
Understanding what is included with standard Salesforce licences versus requiring additional purchase:
- Included with Sales Cloud Enterprise: Einstein Opportunity Scoring, Einstein Activity Capture, basic Einstein Insights
- Requires Sales Cloud Einstein add-on: Einstein Lead Scoring, Einstein Conversation Insights, full Einstein Forecasting with AI predictions
- Requires Service Cloud Einstein add-on: Einstein Case Classification, Einstein Article Recommendations, Einstein Reply Recommendations, Einstein Bots (beyond basic)
- Included with Marketing Cloud: Einstein Send Time Optimisation, Einstein Engagement Scoring for most editions; Einstein Content Selection in higher editions
- Separate Agentforce licence: Agentforce agents (consumption-based pricing, approximately $2/conversation)
- Einstein Copilot (now Agentforce Copilot): the conversational AI assistant embedded in the Salesforce UI – included with Salesforce licences but with usage caps
Deploying Salesforce Einstein and Agentforce Successfully
What is the difference between Einstein AI and Agentforce?
Einstein AI covers predictive and generative AI features (lead scoring, email generation). Agentforce is the autonomous agent platform where agents take actions, not just predictions.
Is Einstein AI included in standard Salesforce licenses?
Some features like Einstein Activity Capture are included. Others require add-on purchases or Einstein 1 edition.
What data does Einstein use for predictions?
Einstein uses your org historical CRM data – past deals, lead outcomes, email activity – to train models specific to your business. It does not use data from other customers.
How long until Einstein starts producing predictions?
Einstein needs 3-6 months of historical data and typically 2-4 weeks after activation to build initial models.
Can small businesses use Salesforce Einstein?
Yes, but small businesses often lack the data volume Einstein requires. Start with Einstein Activity Capture (free in Sales Cloud) and add predictive features as data grows.
Problem: Einstein Lead Scoring Producing Inaccurate Predictions
Einstein Lead Scoring needs at least 1,000 converted and 1,000 rejected leads in the past 6 months. Fix: Cleanse historical lead data before enabling. Remove duplicates, fill missing fields, ensure conversion outcomes are recorded. For low-volume orgs, use manual scoring criteria while accumulating data.
Problem: Agentforce AI Agents Giving Hallucinated Responses
Gaps in your knowledge base cause AI agents to fabricate answers. Fix: Audit Salesforce Knowledge for completeness before deploying agents. Enable Trusted Answers to restrict agents to verified articles only. Set human escalation triggers for queries with confidence scores below 70%.
Problem: Einstein Analytics Dashboards Showing Stale Data
Tableau CRM datasets refresh on schedules, not in real time. Fix: Use Live connections for dashboards where real-time accuracy is critical. For historical analysis, keep scheduled refreshes. Communicate refresh cadences to users in dashboard descriptions.
The best AI setup is the one that improves actual CRM work. If the data or process is weak, the AI layer will not help much.
The best AI setup is the one that improves actual CRM work. If the data or process is weak, the AI layer will not help much.