When AI starts "catching" patients after discharge: iFlytek Healthcare's (02506) inflection point lies beyond the medical record.

A 69-year-old coronary heart disease patient living in Tongnan, Chongqing, experienced more frequent episodes of chest tightness on the 106th day after discharge. Family members reported the discomfort through the AI patient management platform; the platform combined collected data on symptoms, medication, and medical history, relied on AI to conduct auxiliary risk screening, triggered a high-risk warning, and pushed it to Dr. Huang Hui at Sichuan Provincial People's Hospital. With coordinated follow-up by medical staff, the patient was readmitted for examination the next day; angiography indicated worsened coronary artery stenosis, and stent intervention surgery was performed promptly, averting the risk of a serious cardiovascular adverse event. This is precisely the pain point repeatedly emphasized by Tao Xiaodong, President of iFlytek Healthcare (02506), at the 1024 "AI Patient Management Platform Innovation Forum": what patients can perceive is only "whether it hurts, whether they feel uncomfortable," while symptoms and medical risk are often two different things. On October 22, at the Beijing National Convention Center, during the 1024 Developer Festival, iFlytek Healthcare turned this technology launch into a narrative about "beyond ten minutes," and completed three things on the same day:. Spark Medical Large Model V3.7 was officially released, topping the MedBench 5.0 agent evaluation with 99.6 points, raising the reasonable rate of end-to-end medical record generation to 85%, and improving imaging-assisted diagnosis capability by more than 15% compared with vertical models; the AI Patient Management Platform 3.0 was released; and, together with the Chinese Medical Association Publishing House, the China Academy of Information and Communications Technology, the China Cardiovascular Health Alliance, Qilu Hospital of Shandong University, the First Affiliated Hospital of USTC, the National Clinical Research Center for Orthopedics and Sports Rehabilitation of the Chinese PLA General Hospital, and other industry-academia-research-application institutions, it launched the industry's first special plan for expert consensus on AI patient management platforms. While most AI healthcare companies are still struggling over in-hospital assisted diagnosis bidding share, iFlytek Healthcare has already moved the battlefield outside the hospital: AI patient management. From "benchmark scores" to "catching patients" The question most easily raised about medical AI is: no matter how high the model score is, will doctors use it? iFlytek Healthcare's answer does not rely on scores, but on a set of real-world weights: Smart Doctor Assistant covers 31 provinces, 828 districts and counties, and 77,000 primary-level institutions, with cumulative assisted diagnoses exceeding 1.2 billion times and assisted medical record writing 490 million times; the imaging cloud has accumulated 260 million images; iFlytek Xiaoyi has 220 million consultations; the doctor side averages 1.71 million calls per day, and the patient side serves 520,000 person-times per day. More critical is the "inconsistent data" recovery mechanism: when an AI judgment conflicts with a doctor's final diagnosis and the doctor does not modify it, the case enters expert reviewcorrect ones are reinforced, and wrong ones become new samples. What makes the flywheel turn is not the volume of data, but hard cases at the boundary of cognition. Therefore, the real meaning of the Spark Medical Large Model V3.7 upgrade release is not another leaderboard climb, but that the model has begun to embed itself into the responsibility nodes of the diagnosis and treatment chain: writing medical records, reading images, managing discharge, and warning of recurrence. It no longer "suggests" outside the screen, but "provides a safety net" within the workflow. Han Hui, Vice President of Qilu Hospital of Shandong University, provided a set of hard data: the "Qilu Spark" full-course management large model covers 20 departments and 70 diseases, has managed 270,000 person-times in total, reduced medical staff follow-up workload by 90%, and achieved patient satisfaction of more than 98%. Tian Yanghua, President of the First Affiliated Hospital of USTC, also introduced that the "Smart Doctor Companion" large model has been launched in more than 30 departments, covers more than 100 diseases, and has patient satisfaction of 97%. In the past, hospitals bought AI as an "innovation budget"; now what is saved is real money in nurse hours and readmission rates. Business model: from "delivering projects" to "operating health" To understand iFlytek Healthcare, one must first understand the business reshuffling in its interim report. In the first half of 2026, the company's revenue was RMB446 million, up 49.4% year on year; gross profit was RMB236 million, up 53.5% year on year; gross margin was 52.9%; loss for the period was RMB48.1 million, narrowed by 41.5% year on year. Three new segments clearly emerged: AI diagnosis and treatment assistant RMB202 million (45.2%), AI health assistant RMB182 million (40.9%, up 74.9% year on year), and AI digital base RMB62 million (13.9%). Compared with the revenue growth rate, the transformation of the revenue structure deserves more attention. As of now, the AI patient management platform has been implemented in more than 300 medical institutions nationwide, providing refined management for chronic diseases and high-risk populations for more than 500,000 person-times in total. Among the key specialty disease samples already launched, medication adherence increased from 49% to 95%, the planned follow-up rate increased from 39% to 67%, and the three-month recurrence rate fell to 0.9%. This progress is pushing iFlytek Healthcare across an important valuation divide: the revenue model is transforming from one-time project delivery revenue to ARR annual recurring revenue. Project-based business is driven by new orders; the core assessment indicators of platform-based business are renewal rate and service coverage scale. What is underestimated is the "right to set standards" The move most overlooked by the secondary market in this launch is that special plan for expert consensus. AI patient management still lacks a unified service process, clear quality control boundaries, and mature evidence-based standards. Whoever first distills the pathway into 160 specialty disease management SOPs and covers more than 30 departments may hold industry discourse power in the next three years. iFlytek Healthcare's joint construction of standards with leading tertiary hospitals and authoritative academic institutions is essentially doing two things: first, "institutionalizing" product experience to raise the compliance cost for latecomers; second, upgrading hospital stickiness from "software procurement" to "co-construction of clinical pathways"if a hospital leaves it, it is not just changing suppliers, but losing an evidence-based system already in operation. Combined with nationwide domestic computing power, a self-developed Harness engineering architecture, a medical knowledge base on the scale of nearly 1 billion, and a dual chain of thought of "basic medicine + clinical medicine," this barrier cannot be caught up with by a single-point model. The moat of medical AI has never been on leaderboards, but in "regulators daring to approve, doctors daring to sign, and medical insurance being willing to pay." As Tao Xiaodong, President of iFlytek Healthcare, interpreted AI healthcare: in a sense, we are building a system for our future selves who may become ill, allowing the judgment of experts at major hospitals to go beyond hospital walls and sink to counties, and to continue to play a role during the stage after patients are discharged when reminders are lacking. From a capital perspective, iFlytek Healthcare's long-term goal is not to sell large models, but to transform "expert-level clinical judgment" into a billable, replicable, cross-institution shared public medical capability, and to continue implementing it in scenarios such as primary healthcare, imaging cloud, and full-cycle health management. While the market is still measuring it with AI concepts and project orders, it has already iterated toward a medical operating system charged by usage, and its broad growth space is worth anticipating.