Salesforce Einstein AI Marketing Features: 2024 Changelog

Salesforce pushed more Einstein changes in 2024 than in any prior year — partly driven by the broader Salesforce AI Cloud rebranding, and partly by real pressure to ship generative features before competitors did. For practitioners managing Marketing Cloud or Account Engagement, the pace created a practical problem: release notes were scattered across three separate documentation trees, and several features launched in beta with different availability windows depending on your edition.

This record consolidates the changes that actually matter for how you configure campaigns, build audiences, run scoring, and generate content. Each entry below is dated by effective date, not announcement date — those sometimes differed by weeks.

Q1 2024: Einstein Copilot for Marketing Cloud Launch

Salesforce made Einstein Copilot generally available for Marketing Cloud in the Spring '24 release (effective February 2024 for most orgs). This was the first time a conversational AI assistant was embedded directly into the Marketing Cloud Engagement interface — not as a standalone app, but as a sidebar panel accessible from Journey Builder, Email Studio, and Content Builder.

What changed functionally

• Copilot can generate email subject lines and body copy drafts from a plain-language prompt, using your connected Data Cloud segments as audience context.
• Journey Builder gained a natural-language entry condition builder — you describe the audience in plain English and Copilot translates it to a filter definition, which you then review and confirm.
• Content Builder's "Generate with Einstein" button was expanded from subject lines only to full email sections, with tone controls (professional, friendly, urgent).
• Copilot actions are logged in the Einstein Activity Log, visible to admins — relevant for teams that need an audit trail of AI-generated content decisions.

Practical notes

The natural-language journey condition builder is genuinely useful for teams that struggle with Marketing Cloud's filter UI, but it has a documented failure mode: when the plain-language description is ambiguous (e.g., "customers who haven't bought recently"), Copilot defaults to a 90-day lookback window without flagging that assumption. You need to review the generated filter definition before saving. Teams that skipped this review step reported audience sizes that were significantly off from intent.

Q1 2024: Einstein Send Time Optimization — Model Refresh

Salesforce updated the underlying prediction model for Einstein Send Time Optimization (STO) in February 2024. This is a modification to an existing feature, not a new one, but the model change is worth logging because it altered behavior in measurable ways.

The recency weighting change is the most consequential for active senders. If your list has significant seasonal behavior (e.g., retail holiday patterns), the 18-month window with recency weighting should produce better predictions than the flat 12-month window did. The per-send confidence score in reporting is new and worth monitoring — scores below 0.6 indicate the model had limited data and you should treat the predicted time as a rough estimate.

Q2 2024: Einstein Engagement Scoring — Account Engagement (Pardot)

The Spring '24 release also brought a significant change to how Einstein Engagement Scoring works in Account Engagement (the platform formerly known as Pardot). The scoring model now incorporates Data Cloud behavioral signals when a Data Cloud connector is active — previously, Engagement Scoring only pulled from Account Engagement's own activity data.

What this means in practice

For orgs with Data Cloud connected, Engagement Scoring now has access to web session data, product usage events, and service interactions that Account Engagement couldn't see before. In B2B contexts where a prospect might go weeks without email engagement but be actively using a trial product, this is a meaningful signal improvement.

The catch: the expanded scoring model requires a Data Cloud for Marketing license, which is priced separately. Orgs without Data Cloud will see no change to their Engagement Scoring behavior. Salesforce's release notes described this as "enhanced scoring" without clearly flagging the license dependency — several practitioners reported configuring the feature only to find it wasn't active in their org.

Q2 2024: Einstein Content Selection — Deprecation Notice

Salesforce issued a formal deprecation notice for Einstein Content Selection in April 2024, with end-of-life set for October 2024. Einstein Content Selection was the feature that automatically chose between multiple email content variants at send time based on predicted engagement likelihood.

The replacement path Salesforce pointed to was Einstein Personalization, which operates at the content block level rather than the full-email variant level. The two features are not functionally equivalent — Einstein Content Selection worked with pre-built variants, while Einstein Personalization dynamically assembles content blocks from a defined pool. Teams that built journeys relying on Content Selection needed to rebuild the logic in a different paradigm.

Q3 2024: Einstein Personalization — General Availability

Einstein Personalization moved from pilot to general availability in the Summer '24 release (effective July 2024). This feature had been in limited pilot since late 2022, so the GA release represented a meaningful commitment from Salesforce to support it at scale.

Feature scope at GA

• Supports personalization at the content block level within Email Studio — each block can be assigned a "pool" of variants, and Einstein selects the variant predicted to drive engagement for each recipient.
• Integrates with Data Cloud for real-time audience attributes — product affinity, recency signals, and predicted lifetime value can all be used as inputs to the selection model.
• Includes a holdout group configuration (10–30% of audience) for measuring lift against a non-personalized control group.
• Reporting shows variant-level performance and Einstein selection frequency, but does not expose the model's feature importance — you cannot see which signals drove a particular selection.

The holdout group configuration is the most practically useful addition for teams that need to justify the feature internally. Without a holdout, you have no way to separate Einstein Personalization's contribution from other campaign variables. Salesforce defaults the holdout to off — you have to enable it manually per campaign.

Limitations at GA

Einstein Personalization at GA does not support SMS or push notification channels — email only. It also requires a minimum variant pool of three content blocks to activate; two-variant A/B testing still routes through Marketing Cloud's native A/B testing tool, not Einstein Personalization. The Data Cloud dependency for real-time signals means latency can vary: Salesforce's documentation states signals are refreshed every 15 minutes, but in practice, high-volume sends have reported stale signals at the time of send for segments that changed within the refresh window.

Q3 2024: Einstein GPT Rebranded to Einstein AI

In August 2024, Salesforce retired the "Einstein GPT" product name across all documentation, help articles, and in-product labels. The underlying capabilities were unchanged — this was a branding consolidation, not a functional change. All features previously marketed as Einstein GPT now appear under the "Einstein" umbrella without the GPT suffix.

This matters for practitioners primarily because search results, internal documentation, and third-party guides written before August 2024 will reference "Einstein GPT" for features that are now labeled differently in the UI. If you're troubleshooting a feature and can't find it by name, check whether the documentation predates August 2024.

Q4 2024: Einstein for Flow — Marketing Automation Integration

The Winter '25 release (effective October 2024 for most orgs) added Einstein capabilities directly into Salesforce Flow for marketing automation use cases. Previously, Einstein predictions and recommendations required specific Marketing Cloud or Account Engagement contexts. With Einstein for Flow, you can call Einstein prediction APIs as a step within any Flow, including those triggered by CRM events.

Practical use cases this enables

• Trigger a nurture journey in Account Engagement when a CRM opportunity reaches a specific stage, with Einstein Engagement Score as a branching condition — without needing a custom integration.
• Use Einstein Lead Scoring within a Flow to route inbound form submissions to different sales queues based on predicted conversion likelihood.
• Apply Einstein Next Best Action recommendations as a decision point in post-purchase flows, surfacing different content offers based on product affinity predictions.

Q4 2024: Einstein Trust Layer — Expanded to Marketing Cloud

Salesforce extended the Einstein Trust Layer to Marketing Cloud Engagement in October 2024. The Trust Layer is Salesforce's architecture for ensuring that customer data sent to third-party LLM providers (including OpenAI) is masked, not stored by the provider, and not used for model training.

Before this change, Marketing Cloud's generative features (subject line generation, content drafting) sent prompts to the underlying LLM without the Trust Layer protections that Sales Cloud and Service Cloud already had. The October 2024 extension closes that gap.

For teams in regulated industries — healthcare, financial services, any org with strict data residency requirements — this was a blocking issue for adopting generative features in Marketing Cloud. The Trust Layer extension makes those use cases viable, though you should verify your specific data handling requirements against Salesforce's Trust Layer documentation rather than assuming coverage.

Summary Table: 2024 Einstein Marketing Changes

What to audit if you haven't already

Three of the 2024 changes require active review of existing configurations — they don't break journeys loudly, they just change behavior quietly or stop executing.

1. Check all active journeys for Einstein Content Selection steps. Any journey built before October 2024 that used this feature is now silently skipping that step. The journey will not error — you'll only notice in performance data.
2. Review Einstein Copilot Trust Layer settings in Setup if your org operates in a regulated industry. The Trust Layer extension is not automatically enabled; an admin must configure it.
3. Verify Engagement Scoring data sources in Account Engagement if you built scoring-based automation rules in 2024 assuming Data Cloud signals were active. If Data Cloud is not licensed, those rules are running on Account Engagement activity only.