Like most physicians, I’m cautiously optimistic and sometimes skeptical about the AI-powered tools coming to health care.

As a dermatologist, I understand the dire need in our profession to address administrative burdens that can be soul-crushing–the constant barrage of documentation needs, non-clinical paper shuffling, and ever-changing reporting requirements, to name a few. As a software executive, I also understand the deeply complex technical requirements needed to integrate AI into existing systems safely, and that it’s still very early days to unlock AI’s full potential.

Most of us became physicians because we wanted to help people. But the business of medicine is often convoluted with competing priorities that rob us of this privilege. AI offers the potential to simplify practice operations and expedite the delivery of care, getting us back to doing what we love.

To evaluate AI solutions effectively, it’s important to prioritize those designed to deliver the highest value with the least amount of risk. Any AI implementation in the near term should offer safe, immediate, and measurable results.

Let’s start with the patient. Practice staff are dedicated to helping patients but are often distracted by a constant stream of phone calls and messages. Administrative tasks like scheduling, rescheduling, and answering repetitive questions consume valuable time and resources. AI-powered patient collaboration tools can support patient inquiries by facilitating appointments, automating message routing to the appropriate staff member, or providing routine responses for non-clinical inquiries. This self-service approach delivers multiple benefits–reducing administrative burdens on staff, improving practice response times, and enhancing the overall patient experience.

More on practice efficiency. Another killer of practice efficiency is the current workflows for documentation management. Practice staff spend an excessive amount of time submitting and re-submitting claims; directing, redirecting, and filing faxes; and more, taking time away from assisting patients. These different administrative processes require too many steps and too many clicks from often too many people to do one task. Applying AI to documentation management can save practice workers valuable time, such as automating faxes to the right person and significantly reducing the number of steps to get to the outcome. AI-assisted documentation management can be instrumental in helping alleviate staff burnout and turnover.

Let’s move on to clinical. One of the best applications of AI for physicians is AI-powered documentation or ambient listening tools. But here, we must exercise caution. Some AI ambient listening tools are limited in functionality and can only transcribe and categorize notes into pre-set fields–a limitation predetermined by the data used to train the model. If data is of poor quality, unstructured, or not representative of patient-doctor interactions, time-saving benefits could be lost as providers spend more time fixing errors and rewriting notes.

Another limitation of unstructured data is that it shifts the burden of documentation downstream. While the clinical note may be completed, the doctor winds up spending just as much time creating related documentation, such as the prescription, encounter form, education handout, and pathology requisition. A more advanced and well-trained AI can suggest relevant information for all the clinical downstream tasks. This frees up valuable physician time, allowing for more focused patient interactions.

More on data

Not only can poor data sourcing lead to errors that require providers to spend excessive time correcting mistakes, but it can also lead to biased results. There is a risk that AI models can have bias, favoring the data that it’s been trained on over the scenarios it hasn’t seen. Additionally, regular updates with new health care information are crucial for maintaining AI effectiveness, helping to ensure data is representative and that AIs can provide more accurate suggestions.

As AI applications become more prevalent in health care, conversations will increasingly center around data. Questions such as “How much data was used to train the model?” “Is the data source off-the-shelf or in-house?” and “How regularly is the AI model updated with new data?” will become more common. And with good reason. The data drives the algorithms which determine whether the tools are successful. Physicians need to ensure that the tools they implement into practice are trained on a large amount of high-quality, bias-free, and diverse data.

AI has the potential to revolutionize health care, breathing new life into our profession and allowing us to reimagine the entire practice experience. I, for one, am eager for the day when physicians can reclaim more of the joy and fulfillment of caring for people.

Michael Sherling is a dermatologist.