AI Routing the Patient to the Right Branch
From which department should a patient with a stomachache book an appointment: internal medicine, gastroenterology, or should they go to the emergency room? Most people do not know this and book an appointment with the wrong branch. As a result, the patient loses time, and the hospital bears an unnecessary appointment burden. The Hospital Department Routing System is an AI-supported decision support infrastructure we developed precisely to eliminate this confusion. By analyzing the patient's complaint in natural language, it predicts the correct branch with high accuracy and operates with a 98.7% routing success rate on real case data.
Featured Results
The system understands the patient's complaint, routes it to the correct medical branch, and does this with 98.7% accuracy. Behind this accuracy lies a question-and-answer mechanism that narrows down uncertain cases step by step; when the system is unsure, it asks the patient additional questions to clarify the probability. It generates an automatic emergency alarm for critical symptoms and directs the patient to the emergency room. On doctor screens, it works as an auxiliary layer that accelerates the diagnosis process. All of this is positioned on top of the Phexum Analytic Platform, easily integrating with the hospital's existing digital structure.
Cross-Cutting Concepts and Architecture
The most defining idea running through this entire product was narrowing down uncertainty step by step. A patient's complaint is often unclear and can point to multiple branches; the system's job is to gradually turn this uncertainty into certainty. The first leg of this involves understanding the complaint in natural language. The technical equivalent for this is natural language processing (NLP): transforming free text, meaning the patient's own sentences, into structured meaning that the machine can process. The system doesn't stop there; instead of settling for a single symptom match, it runs a dynamic decision tree that progresses based on the patient's answers and narrows the probability in fuzzy cases by asking the patient additional questions. The 98.7% accuracy comes from this: it is the result of this adaptive Q&A mechanism, not a one-time prediction.
Alongside this, the idea of decision support is equally defining. The system does not make the decision for the patient or the doctor; it accelerates and enriches it. On the patient side, it recommends the correct branch and directs to the emergency room in critical combinations. On the doctor side, the physician sees the patient's complaint along with pre-findings analyzed by the system. This distinction is critical: the goal is not to change the physician's decision, but to offer them an additional data layer to make their decision faster and more solid. Thus, the system works as an auxiliary intelligence supporting both the patient and the physician over the same structure.
The World the System Lives In
Misrouting in healthcare is a silent but expensive problem. On the patient's side, this means reaching the right treatment late; on the hospital's side, it means the operational burden and cost created by appointments falling into the wrong branch. Moreover, this is a two-way problem: on one side, there is the patient who doesn't know which department to go to; on the other, polyclinics filled with incorrect appointments. The classic solution is making the patient fill out a lengthy form on the website. However, forms are static; they cannot change questions based on the patient's answer or clarify a fuzzy picture. The gap this system fills is exactly here: instead of a static form, establishing an interactive clinical pre-evaluation layer that progresses by conversing with the patient.
How the Story Was Built
We designed the system with the idea of routing patients to the right branch in the fastest and most accurate way, and we built it on top of the Phexum Analytic Platform rather than as a separate infrastructure from scratch. This decision was decisive from the start: thanks to the platform's flexible and expandable structure, the system can easily integrate with different data sources, adapts to the existing digital structure of each hospital, and data processing and decision support processes work as a single unified whole. We didn't leave the routing as merely a prediction; we designed it as a flow that operates from the first moment of contact with the patient, asking questions when necessary, and generating alarms when needed.
Not a one-time prediction, but an adaptive Q&A that narrows uncertainty step by step.
Building Blocks of the Solution
The smart branch routing engine analyzes the complaint in natural language and predicts the correct branch with a dynamic decision tree; narrowing the probability with additional questions in fuzzy cases. The emergency detection layer generates an automatic alarm and directs to the emergency room when it catches critical symptoms and risky combinations; this structure was explicitly designed to increase the probability of early intervention in time-critical cases. The system also works as an analytical layer, analyzing patient traffic and symptom distribution; reporting insights at the management level such as which branch is busy and which complaints are increasing. A customized model is created for each hospital, taking into account its existing branch structure, the clinics serving, and operational workflows.
The Impact We Created
The system simultaneously solves a two-way problem: it resolves the patient's confusion about which department to go to, and it reduces the hospital's costs arising from incorrect appointments and operational burden. As the misrouting rate decreases, the appointment system is used more efficiently, the chance for early intervention in critical cases increases, and the physician begins the diagnosis with richer pre-findings. The 98.7% accuracy is merely this value reduced to a single number.