Where to begin with AI implementation
At the recently concluded 2026 World Artificial Intelligence Conference, enterprise-level AI emerged as a highly competitive field. Compared to previous years, which focused on model parameters and generative capabilities, this year's exhibition featured more intelligent agents, industry solutions, and enterprise deployment tools. AI companies are now extending their products to connect enterprise data, systems, and business processes.
However, when AI truly enters the physical world, things do not go as smoothly as product demonstrations suggest.
For example, in a clinic, which room a patient walks into, what procedures a doctor performs, where equipment is moved, or why a medication is retrieved and then returned—these events do not automatically become data.
Recently, So Young founder Jin Xing engaged in a nearly two-hour discussion with media outlets, including Wall Street News and Knews24.
So Young is currently implementing AI transformations across its 50-plus light medical aesthetics clinics, aiming to apply AI to areas such as face-to-face consultations, treatment quality inspections, patient triage, and clinic operations.
What stood out in this discussion was the extensive groundwork So Young has done to make AI feasible in its clinics.
For instance, to track a consumer's waiting time at different stages within a clinic, So Young tried various methods, including tablet-based check-ins, wristbands, Bluetooth, and Wi-Fi. To enable headquarters to remotely monitor whether doctors follow standard treatment procedures, the company installed cameras in treatment rooms. It later discovered that the network bandwidth across all clinics nationwide needed to be individually upgraded.
So Young began building this digital system in 2023, deploying over a hundred engineers. Three years later, Jin Xing believes the company is still primarily in the "first half" of AI implementation—data governance.
"Only with this data accumulation can AI learn from your business," Jin Xing said. "We are still in a massive phase of digitalization."
So Young's practices also reveal a hidden threshold for AI deployment in chain enterprises: before AI can be implemented, companies must first convert the ever-changing physical world into real, continuous, and machine-understandable data.
The hidden costs of data collection
For chain enterprises, one of the most appealing values of AI is the ability to replicate the capabilities of top-performing clinics and employees.
A single clinic can operate smoothly with a skilled manager, doctor, or experienced staff member. But when the number of clinics grows from ten to dozens or even hundreds, companies must ensure every clinic is working according to the same standards.
So Young aims to become a standardized chain of light medical aesthetics clinics. According to Jin Xing's vision, if the same consumer with the same request visits different clinics and is examined by different doctors and consultants, the treatment plan they receive should be highly similar.
However, So Young is not yet able to achieve this. The final treatment plan a consumer receives still largely depends on the individual knowledge and experience of the doctor and consultant.
To narrow this gap with AI, the first step is to know exactly what happens inside the clinic. A complete medical aesthetics treatment includes at least: consumer presenting a request, skin assessment, doctor's diagnosis, treatment plan design, preparation of medications and supplies, treatment execution, and post-treatment follow-up testing and feedback.
But this information exists in various forms, scattered across different locations. The consumer's request is a conversation during consultation; skin condition comes from testing equipment; the medications and dosages used are recorded in the supply chain system; the treatment process is a video; and the final outcome is judged through pre- and post-treatment photos, follow-up reports, and consumer reviews.
For different data types, So Young has adopted different collection methods. Consultation conversations are recorded via speech-to-text; skin testing equipment is connected to the backend to extract structured indicators from entire reports; medications and supplies are tracked through the ERP system; and the doctor's treatment process is recorded on video.
According to Jin Xing, So Young has accumulated over 3 million pre- and post-treatment comparison photos from approximately 1.75 million treatments, along with nearly 530,000 follow-up reports.
But data governance is not just about storing as much information as possible. The first challenge is ensuring data authenticity. In practice, So Young once wanted to track how long a consumer spent at each stage—consultation, skin testing, pre-treatment preparation, and treatment itself.
The company initially placed tablets at different stations for staff to manually check in. However, during busy times, staff often forgot to check in. When headquarters began monitoring check-in rates, some staff started checking in ahead of time or retroactively.
The system records became complete, but the time data was no longer real. "This data is dirty data for us, and dirty data is useless," Jin Xing said.
So Young then tried using wristbands with identification codes, allowing consumers to scan and record their presence at different areas. But wristbands could interfere with the consumer experience. Why would a consumer need to wear a wristband? Would different colors be interpreted as a classification of membership tiers? To make the wristband seem valuable, the team even discussed adding features like opening lockers.
Bluetooth, Wi-Fi, and other methods did not solve the problem alone. Manual collection increases the workload, while passive collection requires a trade-off between accuracy, cost, and customer experience.
Ultimately, So Young had to combine multiple methods to best recreate the real location of consumers, staff, and equipment within the clinic. This reveals a fundamental contradiction in data governance for offline chain clinics: poorly designed collection mechanisms can increase staff burden and even alter staff behavior, creating a system that appears complete but is actually distorted.
AI enters treatment inspection
Although So Young is still in the midst of data governance, expected to be completed by the end of the year, AI is already gradually entering the clinics. Video-based treatment inspection has become one of the scenarios for AI implementation.
For So Young, the core problem that video inspection aims to solve is monitoring whether different clinics are truly implementing uniform treatment standards. To this end, So Young has developed standard operating procedures (SOPs) for different treatment projects. For example, for a BBL project with 10 steps, the SOP specifies the required operation and standard duration for each step.
Additionally, treatment rooms are equipped with dual-screen displays. One screen shows the operational flow of the current treatment, allowing doctors to follow the procedure during treatment, and consumers can also see the progress.
To ensure each step is accurately executed, So Young monitors the treatment process through manual video inspections. A dedicated inspection team at headquarters remotely views treatment sessions from different clinics and rooms to supervise whether doctors are following the SOP.
However, as the number of clinics and treatments increases, the volume of daily videos grows, and manual inspection becomes inefficient. So Young is now trying to change this process with AI. The company told Knews24 that a new intelligent inspection system is scheduled to go live at the end of July.
The new system will introduce AI vision capabilities, performing frame extraction, object detection, and behavior analysis on treatment videos. It will identify who appears in the video, which step the doctor is performing, what equipment is being used, whether the actions meet standards, and whether any anomalies are suspected.
According to So Young's plan, this system will gradually achieve 24/7 intelligent inspection, with continuous human confirmation and correction of recognition results to improve model accuracy.
But the primary challenge for covering all clinics nationwide is not model recognition, but network bandwidth. Medical aesthetics clinics originally did not need to continuously upload large amounts of video. The existing network must support customer service and other business systems. After the intelligent inspection system goes live, treatment room footage must be continuously transmitted to headquarters, creating a bottleneck with the original bandwidth.
After actual operation, So Young found that each clinic's bandwidth needed to be expanded to at least 200 Mbps. Otherwise, video uploads could affect the operation of other systems. The company had to contact network providers in different regions and shopping malls to upgrade the network for each clinic individually.
From manual sampling to AI inspection, a function that seems like a simple AI vision task ultimately involves cameras, network bandwidth, video storage, SOP definition, and human annotation. This also means that integrating AI into clinics is not an upgrade that can quickly generate profits. To enable the model to see and understand the real treatment process, companies must first invest significant engineering resources to transform existing systems, hardware, and business processes.
Jin Xing does not believe that AI will make a single clinic immediately more profitable in the short term. However, for chain enterprises, the true potential return from AI is not reflected in the profit statement of a single clinic, but rather in expanding the organization's management reach. It transforms management capabilities that were previously dependent on individual experience into system-level capabilities that can be deployed, monitored, and replicated by headquarters.
But before that, companies must first see the real world. Converting an offline clinic into a world that machines can understand still requires a great deal of slow, meticulous work.
Comments