AI Inbox Management Market Map: AI scrambles to take the load off clinicians
This is part of Elion’s weekly market map series where we break down critical vendor categories and the key players in them. For more, become a member and sign up for our email here.
Last summer, 162 clinicians at Stanford Medicine engaged in a pilot testing Epic’s inbox management AI, which used GPT-4 to draft responses to the patient. Each response was drafted based on the initial patient message, recent clinical notes, and structured data from the EHR, but the clinician decided which messages to actually send.
The study found that while only 20% of AI-drafted messages were actually sent (although specialties like GI and hepatology showed higher rates at 29%), using the tool helped reduce task load on physicians in statistically meaningful ways.
That’s timely, as providers have seen a 157% increase in patient messaging since COVID, essentially because people had to learn how to use these tools to communicate with their doctors. Now, the question is how to use GenAI to reduce the time clinicians spend on paperwork and asynchronous communication.
Differentiating Inbox Vendors
Our category of AI inbox management includes sorting incoming messages, helping prioritize important messages, and automating messages where possible. This includes Epic’s In-Basket ART, Droxi, Mariana AI, and Elaborate, with a focus on understanding and responding to patient messages. Meanwhile, products like Well AI Inbox Admin and Synthpop are focused on triaging and sorting faxes, referrals, and other documents.
Although this is a relatively small category at the moment, helping clinicians manage their inboxes is a high ROI activity for GenAI platforms, and we suspect that many of the AI clinician assistant platforms who can already send automated after-visit summaries to patients and draft referral letters—will start to handle more patient messaging as well.
The Future of AI Inbox Management
While the study seemed to show that the value and potential ROI for physicians might be limited at the moment, part of that stems from the fact that inbox messages that make it all the way to the physician are typically more complex. As a result, these tools are currently seeing more value in automating the inbox for nurses, MAs, and front office staff, according to Keith Morse at Stanford.
It’s pretty clear, however, that these tools are going to add substantial value for both APPs and physicians in the not-too-distant future, not only because vendors are
at prompting and using existing AI models in an integrated manner, but also because the models themselves will continue to move forward. We anticipate that the upcoming generation of models like GPT-5 and Llama 4, expected later this year and next, along with enhanced fine-tuning and thoughtful product design, will significantly increase the proportion of AI-generated drafts that clinicians rely on. This will not only lead to substantial improvements in clinician administrative tasks but also mark a major shift in how clinicians use AI to engage with patients.