AI Field Guide: What to Know Before Contract Renewal

So, you’re ready to explore the AI health technology landscape. We’ve got some checklists you can throw in your backpack, some myth-busters to help you wade through the hype, and a few success stories for telling around the virtual campfire.

Before you set out, understand that although AI tools are being promoted for many parts of your practice, some applications are much further along than others. The most advanced are AI scribes, which are proving hugely popular with physicians. At Permanente Medical Group in Northern California, an AI scribe saved the physicians using it an hour a day, on average, in a 10-week pilot. Physicians install the app on a secure smartphone, and it transcribes patient encounters and produces a note documenting the visit. (Yes, it filters out small talk.) In contrast, AI apps for front office, medical billing, charting, clinical decision support, and medical documentation hold huge promise but are still in early stages.

It’s also imperative to understand that some AI being touted in the press and promoted by vendors is traditional AI, and some is generative AI. “Smart” technology like speech recognition and scheduling that “remembers” a physician’s preferences is traditional AI that has proven its value and stability. Generative AI (as in ChatGPT) is newer and more glamorous, but it has a troubling tendency to “hallucinate,” which means it provides false information every now and again.

With that in mind, let’s lace up our boots and walk through the details.

What to Know About AI Tools

Front Office

The main thrust of AI’s role in front office is eliminating repetitive tasks, letting your staff concentrate on more complex patient issues and in-office interactions. It can assist with:

• Scheduling
• Prescription refills
• Check-ins
• Appointment reminders
• Report on call volumes by call type
• Report on types of patient inquiries for marketing
• Highlight areas for improvement during difficult calls
• Biometric check-ins to reduce paperwork and improve data security

Revenue Cycle Management

Automation and enhancement are both possible with AI RCM tools. AI can assist with:

• Data entry
• Eligibility verification to reduce claim denials
• Patient registration
• Updating patient information
• Providing real-time updates for patients
• Payment posting

It can also:

• Identify denials trends
• Find claims inconsistencies (missing codes, for example)
• Predict patient payment behavior
• Use advanced edit engines to facilitate more accurate charge capture
• Smooth out cash flow with up-to-date patient statements

Clinical

Physicians and practice managers will be pleased to learn that a 2024 study by the British Royal College of Physicians found that AI-produced documentation shortened consultations by 26% on average, without impacting patient interaction time. AI has the potential to:

• Make using an EHR easier and faster by speaking rather than clicking
• “Learn” medical language and make helpful suggestions about similar cases that could be helpful in creating a clinical note
• Provide a list of recommendations based on symptoms and other information
• Create a note intended to be read prior to an encounter with recent medication changes, medical history, and labs
• Serve as an automated medical scribe
• “Read” a medical note and suggest the right code(s)
• Search multiple approved medical databases and drop relevant research into a clinical note

What You May Not Know About AI: The Risks

There are several risks associated with AI, some that are standard with any new technology, and some that are unique to AI.

• The real price. Many AI products are free to practices who want to use them on a trial basis. Although this is a neat way to get a glimpse of the future, remember that free isn’t actually free in this case. Consider the amount of work your IT department must do to understand the new system, install it, and integrate it with your existing systems — not to mention the time your physicians and staff will spend learning the new system.
• HIPAA. HIPAA and AI have a complicated relationship, and unless your vendor fully understands this and commits to making its products compliant now and as changes occur, you run the risk of non-compliance.
• Review needed. Generative AI is new and known to hallucinate. Any output, especially on the clinical side, should be reviewed by a human before becoming part of the patient record.
• Cybersecurity. As with any new system, AI apps must include data encryption, intrusion detection, and access controls, as well as notification of data breaches involving PHI.
• Lack of evidence. Simply because it’s new, there is little research on the usefulness of AI in real-world medical settings. Efficiencies no doubt exist (or will), but the best places for AI and best practices for using it have yet to be fully explored.

Questions to Ask Before You Sign a New AI Contract

First, here are some general questions that practice managers should ask themselves and their experts when considering an AI solution.

• Can this problem be solved by AI?
• Will this solution integrate with our current electronic systems?
• Is the product ready for use (tested and vetted)?
• Does the product integrate with our existing systems? 
• Do we have sufficient computer power and data storage to use the product?
• Do we need new/different security to use the product without increasing the risk of a breach?
• Is the product HIPAA compliant? What level of encryption is used? Where is user data stored?
• Is the product non-discriminatory?
• Is our staff amenable to using the product?
• What is the “sticker” price? What additional costs might the practice incur? IT support? Additional hardware?
• How will we measure ROI?

Next, here are some questions to ask when considering an ambient AI scribe, which (as stated) are popular and in wider use than other AI healthcare products.

• How much time does it take, on average, for a clinician to learn how to use the solution?
• Can the AI-generated note be edited? How?
• Can certain portions of the generated note be deleted?
• Can a clinician start a note in the EHR and continue it in the AI solution?
• How do completed notes get put into the EHR?
• Can the solution accommodate templated content that clinicians have created?
• Can the solution also be used for dictation?
• Can the solution help with tasks beyond documentation, such as coding or finding patient information (medications, allergies, etc.) in the EHR?
• Can the solution help with answering inbox messages or writing referral letters?
• What are the options for creating a note if a clinician doesn’t want to use ambient?
• How does the solution measure accuracy? Will the solution get more accurate with use?
• How does the solution handle hallucinations? Will it learn to prevent further hallucinations in the future?

Finally, whether you’re just beginning your AI journey or well on your way, remember that applications like ambient AI scribes use generative AI technology, so their outputs (encounter summaries, clinical notes, etc.) should always be checked by a human for accuracy.