Key Takeaways

  • Abridge, the AI company revolutionizing healthcare documentation, attacks doctor 'pajama time' by personalizing AI output at three distinct levels: the individual clinician, the medical specialty, and the health system.
  • Janie Lee, from Abridge, emphasizes the critical difference in workflows, stating, “A cardiologist note or workflow is going to look very very different from a dermatologist workflow.” Their AI adapts to ensure notes are complete, compliant, and billable for each.
  • Embedding specific hospital guidelines and best practices directly into AI clinical decision support, as Abridge does, creates a powerful 'moat.' It makes the product stickier by integrating unique customer data and operational standards.
  • Abridge maintains clinical safety and quality with a rigorous evaluation stack, including internal 'clinician scientists' performing 'LFD' (Look For Data) reviews, LLM judges calibrated with annotated data, and external evaluators.
  • The core insight for ambitious founders is Abridge's Three Levels of Personalization for Clinical AI, a framework for building deep product utility and defensibility in specialized markets.

The Abridge's Three Levels of Personalization for Clinical AI

When you're building AI for a high-stakes, deeply personal domain like healthcare, generic output just won't cut it. Abridge tackles this head-on with a three-pronged personalization strategy, described by Janie Lee as "massive for us." This isn't about cosmetic tweaks; it’s about making AI outputs truly useful, safe, and billable.

Here's how they structure it:

  • Individual Level: Addresses individual clinician preferences on note style (bullets vs. paragraphs, concise vs. comprehensive), preferred phrases, or templates for note structure. Focuses on balancing stylistic preferences with accuracy.
  • Specialty Level: Tailors the product and backend evaluations to the specific workflows and requirements of different medical specialties (e.g., cardiology vs. dermatology). Requires deep understanding of what constitutes a complete, compliant, and billable note for each specialty.
  • Health System Level: Incorporates a health system's own hospital guidelines and best practices into clinical decision support products to inform clinicians. This also deepens the 'moat' by embedding unique customer data into the workflow.

When This Works (and When It Doesn't)

This framework shines for AI products operating in domains where outputs are deeply personal, vary significantly by sub-specialty, or where local institutional practices must be honored for adoption and trust. Think highly regulated environments, specialized professional fields (like law, engineering, or finance), or large enterprises with deeply ingrained internal processes and compliance needs. As Janie Lee noted, “health systems... want to know hey we love your clinical decision support product but how do we embed our own hospital guidelines into them”—that's where this level of personalization becomes non-negotiable.

However, this approach isn't a silver bullet. It demands significant data investment and engineering effort at each tier. If your target market is broad consumers with less specialized needs, or if the 'personalization' adds complexity without a clear, measurable uplift in utility, compliance, or user trust, it can become an over-engineered burden. It works when context is the product, but it breaks down if you're chasing personalization for personalization's sake.

What to Do With This

Don't just build a generic AI and hope users adapt. If you're building any B2B AI product for a specialized professional, take a page from Abridge this week. Map out your own three levels of personalization:

Imagine you're building an AI assistant for project managers, specifically for generating project updates or status reports. Instead of a one-size-fits-all solution, apply Abridge's framework:

1. Individual Level: Allow early users to train the AI on their specific reporting style. Do they prefer concise bullet points or detailed narrative paragraphs? Do they always include a 'Next Steps' section before 'Risks'? Let the AI learn their preferred phrases and tone for client communication.

2. Specialty Level: Think about the different types of project managers. A PM leading agile software development will need different AI capabilities (e.g., generating daily standup summaries or sprint retrospective insights) than a construction PM (e.g., budget variance reports, safety compliance checklists). Build distinct features and evaluate success based on their specific workflow requirements.

3. Organization Level: Approach potential enterprise clients and ask: "What are your internal PMO guidelines? Your standard risk reporting matrix? Your brand-specific client communication templates?" Then, embed those directly into your AI. This makes your product indispensable and incredibly sticky. You’re not just a tool; you're an extension of their institutional knowledge, building a moat around your product that generic competitors can't easily replicate.