Ask the expert: Using AI to support compliance

Q: How should quality and accreditation leaders partner with IT and clinical leadership so that artificial intelligence (AI) supports compliance instead of creating new blind spots?

Wael Khouli, MD, MBA, co-founder and chief medical officer of Authsnap, Inc.: First, recognize that quality must lead AI governance, not IT.

AI implementation isn't an IT project with quality input. It's a quality initiative that requires IT infrastructure. The quality department defines what accreditation compliance means, sets the standards for AI performance, and owns the validation process. IT ensures the technology works, and clinical leadership ensures it supports care delivery, but quality drives the strategy.

Next, establish clear standard work and guidelines for AI models. Before any AI tool goes live, quality leaders should define the following:

• What constitutes compliant documentation for each process the AI will touch. This includes medication reconciliation, fall risk assessment, surgical timeouts, etc.
• Acceptable error rates. What's the threshold for false positives and false negatives before the AI is deemed unreliable?
• Validation protocols. How often will AI outputs be audited against surveyor standards? Who conducts the audits?
• Feedback and correction processes. When the AI makes a mistake, how is it flagged, reviewed, and used to retrain the model?
• Override requirements. Under what circumstances can clinicians override the AI, and how is that documented?

These aren't IT decisions. They're accreditation decisions. Quality owns them.

Then, implement staged validation before rollout. Run side-by-side validation: AI flags compliance, human reviewers audit the output. Track false positives (AI says compliant when it's not) and false negatives (AI misses actual gaps). If the AI's error rate exceeds your defined threshold, it doesn't deploy. Period.

Next, create clinical feedback loops to improve the AI. Vendors need real-world feedback from clinicians and quality teams, not just IT performance logs. When the AI makes a mistake, clinicians flag it, explain why it's wrong, and that feedback must loop back to retrain the model. If the AI keeps making the same errors month after month, your feedback mechanism is broken, and quality needs to hold the vendor accountable.

Lastly, audit AI outputs against surveyor standards quarterly. Run mock tracers where you compare AI-flagged "compliant" charts to what a surveyor would actually accept. If the AI says a medication reconciliation is complete, but a surveyor would cite it as deficient, your AI is creating liability, not reducing it.

The bottom line: Quality-led AI governance means quality sets the rules, validates the results, and holds vendors accountable. If IT is driving AI strategy without quality leadership defining what compliance actually looks like, you're automating deficiencies, not eliminating them.

Editor’s note: This Q&A was excerpted from our Inside Accreditation & Quality newsletter.