MARKETING OPS TEARDOWN · OCTOBER 2026

Three ways to put AI in an MCAE program

This is a map I drew from Salesforce documentation, not a field report. Every Salesforce fact on this page comes from Salesforce Help, Salesforce Developers or Trailhead, checked October 2026.

Download the one-page PDF

One-page teardown comparing three ways to put AI in an MCAE program: Salesforce-native Einstein and Agentforce, External Actions (developer required), and working outside the platform. The full text is below.

1. Salesforce-native Einstein plus Agentforce, if your editions line up

  1. Einstein analyzes Account Engagement and Salesforce data for Behavior Scoring, Lead Scoring, Send Time Optimization, Engagement Frequency, Key Accounts Identification, Campaign Insights and Attribution.
  2. Six Agentforce actions view, create and update prospects, read lists and list memberships, sign prospects up for lists, and look up Engagement Studio programs.
  3. Deploy them from Account Engagement Settings, then add them to an Agentforce topic to build a Lead Management Agent.
  4. Before deploying: the latest Account Engagement AppExchange package, Agentforce turned on, an External Client App and a Named Credential.

Watch out: Einstein needs Advanced or Premium with Salesforce Enterprise, Performance or Unlimited. Send Time Optimization also lists Professional, and wants 90 days of email engagement per prospect. The Agentforce actions work on every Account Engagement edition, but only with Salesforce Enterprise, Performance or Unlimited. Every agent use of an action consumes credits, and a deployed action can't be deleted.

2. External Actions developer required · calling an LLM here is my own design idea, not a documented pattern

  1. A developer builds an invocable action: Apex with the InvocableMethod annotation, a Flow, or an External Service.
  2. Salesforce recommends an auth provider and a named credential for the outside system, and invocable actions built with TLS.
  3. It gets registered as an action type (MarketingAppExtAction) on a Marketing App Extension, with Active in Automations checked.
  4. HML merge fields like {{Recipient.FirstName}} pass prospect data in, and the action runs as a step in Engagement Studio.

Watch out: Plus, Advanced and Premium editions only. External Actions don't retry when the outside service is slow, so Salesforce says to build your own retry logic and target a 2-second response. Apex actions need custom exceptions or they can fail silently. Editing an action changes every program that uses it. None of the Salesforce sources on this page describe calling an LLM from this path.

3. Outside the platform export, analyze, human review, import

Salesforce's own API use cases include exporting visitor activity to your own analytics tools and doing segmentation outside Account Engagement, then importing the result as a list for engagement programs. The rule for this path: a person reviews every output before it goes back in, and any demo runs on fake or de-identified data.

Watch out: Daily API requests are capped by edition: Growth 25,000 · Plus 50,000 · Advanced 100,000 · Premium 100,000. Go over and requests fail with error code 122 until the daily reset; more than five at once returns error code 66. As with path 2, none of the Salesforce sources on this page describe the LLM step.

The tell

Salesforce's AI path puts anything it sends to an LLM outside its trust boundary under a zero data retention contract, and turns LLM data masking off for agents. The External Actions guide's data advice is TLS, an auth provider and a named credential. Same prospect record, different paperwork. Pick the path, then find the sentence above that describes your data.

Sources

Verified October 2026.

Molly Purcell-Weatherwalks
mollypw.com