Skip to main content

MIC AI

MIC AI is a collection of AI add-ons for Medical Information Cloud, delivered through the AIMI package as part of Mavens Labs. It brings Agentforce-powered automation into the Medical Information workflow: triaging email intake and summarization; detecting signals for potential adverse events and product quality complaints; request categorization, product search, and finding relevant content. The AIMI Tier 1 Orchestration Agent calls different subagents to recommend the complete interaction and fulfillment. AIMI Recommends provides a human-in-the-loop review panel to insert checks on agentic action at key decision points. System and business guardrails determine how much is automated.

Definitions​

  • Agentic fulfillment is the concept of introducing agentic action into the fulfillment workflow. Rather than a specialist handling every step of an Interaction by hand, AI agents read the inbound request and recommend (or, where configured, carry out) the work to fulfill it: summarizing the Interaction, logging the Request(s) with suggested content, and flagging potential adverse events or product quality complaints. Recommendations are tracked in response logs that document what agentic action was suggested and what was taken (if any), and allow direct reporting between the agentic action and the action taken by the Medical Information specialist. Over time, additional agents and gates (such as the Interaction QC Judge) will join the flow.
  • AIMI Tier 1 Agent is an orchestration agent that coordinates a series of MIC subagents, instructed with system and business guardrails, to create recommendations and take action regarding the Interaction, Request, Adverse Event, and Product Quality Complaint objects. Because those subagents, prompts, and flows are reusable building blocks, they can be assembled into a fully autonomous fulfillment workflow, though the accelerator does not ship that configuration; out of the box it pairs the agent with AIMI Recommends review.
  • AIMI Recommends is the human-in-the-loop layer that allows business rules to define any required review of agentic recommendations at key decision points before agentic action is taken. This gates agentic fulfillment so that inaccuracies are caught before records are created.

A foundation for AI across MIC​

MIC AI is a foundation rather than a single feature you call. It puts AI at the center of Medical Information by shipping configurable building blocks that teams assemble to fit each program.

Out of the box, the AIMI Tier 1 Agent runs as an Agentforce Service Agent on email-origin Interactions, and the inbound email is the only source it draws on today, but that is only how the building blocks ship, not a limit on where they apply. Its subagents, the Interaction Management Agent, Request Management Agent, Adverse Event Management Agent, and Product Quality Complaint Management Agent, are published in the Agentforce Builder Asset Library, so they can be reused beyond the default agent:

  • added to custom agents designed for specific countries or regions, and
  • added to Employee Agents, so internal users can chat with a subagent and have it take action.

Mavens believes Medical Information is uniquely positioned to benefit from AI: the work is definition-driven and evidence-based, which is exactly where grounded, well-guarded agents do their best work. MIC AI is built so that AI is part of the foundational core of Medical Information Cloud.

Our AI principles​

MIC AI is designed around a consistent set of principles:

  • Low-code templates, each agent runs from a structured prompt template rather than bespoke code, so its behavior is predictable and reviewable.
  • Configurable building blocks, reusable AI assets, subagents and actions, that can be assembled to fit each program.
  • Strict guardrails and secure data retrieval, prompts ground their answers in the Interaction's own data and the relevant definitions, retrieve it securely, and are told how to handle missing or ambiguous information instead of guessing.
  • A complete audit trail, every recommendation and the action taken on it is recorded in the AIMI Response Log, related to core MIC records, and reportable. The result is no black boxes.
  • Configurable oversight, the shipped AIMI Recommends experience keeps a human in the loop, requiring specialist review before records are created. Because the subagents, prompts, and flows are reusable building blocks, the same assets can also be assembled into more autonomous workflows where a program allows it. See AIMI Recommends.

These templates are written to a shared standard and refined against real interactions over time, not set once and forgotten.

In this section​

  • AIMI Recommends, the review panel where a human specialist accepts or rejects recommendations.
  • Ask AIMI, run individual AIMI prompts on an Interaction, on demand.
  • Agents, the AIMI Tier 1 Agent orchestration agent and its four subagents, plus the planned Interaction QC Judge.
  • Set up MIC AI, prerequisites, permission sets, agent connection, and prompt configuration.
  • Recommendation types, reference for each recommendation MIC AI produces and what it creates on accept.

Requirements​

MIC AI is an extension of Medical Information Cloud that puts AI at the center of Medical Information inquiry and content management workflows. It extends both the inquiry and content management modules. An additional Salesforce license for Agentforce is required; Data 360 is strongly recommended. See Agentforce pricing and licensing for Salesforce's licensing details.

  • Agentic actions create records in the new AIMI Response Log object and relate them to standard MIC objects like Interactions, Requests, and Adverse Events.
  • Document recommendations use the MIC Content Search engine (MED_CMSOrchestrator) and the Mavens Content Data Cloud search handler. For more relevant results, enable vector search through Data 360.
note

Document recommendations only surface content published in Mavens Content (MCM). A document that is not published in MCM cannot be recommended.

See Set up MIC AI for prerequisites, permission sets, and the Agentforce connection.