Tingmei House Launches AI-Customized Skincare: From One-Time Test to Long-Term Tracking

In the skincare industry, skin assessment is no longer a scarce capability. Most chain stores can generate a skin report in a dozen minutes, but such reports are usually one-off: they only record the skin condition of consumers the exact moment they step into the store, and are later stored in a paper document or data chart. There is no way to tell whether the solution works until consumers "feel" the effect at home, and brands basically lose contact with consumers once they leave the store.

Gracewell House is working to fill this gap. This single-brand cosmetics chain, founded in Guangzhou in 2012 and affiliated to Jiaolian Group, now operates nearly 1,000 stores across 337 cities nationwide, having served more than 1 million registered members in total. It applies AI to a relatively slow-paced task: instead of selling a set of products faster, it builds a continuously updated skin profile for each user that can be iterated for several years, ultimately helping customers find the long-term, proper skincare routines and product regimens that truly suit them.

The industry does not lack detection, what it lacks is the second detection

Personalized attempts in the skincare market have been going on for years, but most of them stop at "one detection for one fixed solution". This model has three unavoidable gaps.

The first is time. A single detection can only capture a snapshot of the current moment, and cannot connect dynamic changes across 7 days, 28 days, and 90 days. The skin itself is in a state of flux: staying up late, seasonal transitions, and menstrual cycles will all cause indicators to fluctuate, so a single measurement can hardly distinguish between "this person's normal baseline" and "this person is in a temporary poor condition recently".

The second is attribution. Does the skin improvement after care come from the skincare solution itself, or short-term fluctuations, seasonal factors, or the change of consumer's daily routine? Without continuous data, neither brands nor users can answer this question. Without clear proof of effectiveness, repeat purchases can only be driven by promotions.

The third is disconnection. After consumers leave the store, brands lose the window to observe their skin conditions, and the skincare solution will stay stagnant. What the industry has long lacked is the second half of the closed loop of "identification - recommendation - observation - re-decision". Almost everyone can complete the first half, but the second half requires time, in-store reach, and data governance capabilities.

Consumers have a more direct intuitive experience of this pain point. One beta tester who has long suffered from sensitive skin said: "I tried more than ten soothing products before, and I never knew if they really suited me. Now I check the skin trend every 7 days and 28 days following the schedule, I can clearly see the fluctuation pattern of redness, and adjust my skincare solutions and regimens according to the actual skin changes, which makes me feel much more reassured with a clear basis." Another offline consumer mentioned that in the past, buying skincare products mostly relied on shopping guide recommendations, and she more or less took a chance on new products. "Now with continuous skin data, I can clearly see whether the solution works and what aspects I need to pay attention to in the future."

Breaking down "one person, one regimen" AI personalized skincare into 5 layers

Gracewell House's approach is to turn the slogan of "personalized customized skincare" into an executable layered structure.

At the top of the project is a digital skin model called SkinTwin, which stores independent skin state parameters for each user: personal baseline, fluctuation threshold, influencing factors, and skincare response. There is one point that is easy to confuse and needs to be clarified: it does not mean "re-training a large model for each individual user". The so-called "one person, one model" refers to retaining a set of exclusive parameters for the individual based on shared population-level rules. The same detection value may lead to completely opposite conclusions when falling on different people's normal baselines: a value that is an alarm for person A may be just a normal daily state for person B.

Supporting this model is a 5-layer collaborative system. The bottom layer completes the capture of basic indicators; the second layer performs cross-device calibration to eliminate the system deviation between store instruments and household devices; the third layer provides a population-level basic model to reduce the cold start cost for new users; the fourth layer is the personal SkinTwin model responsible for long-term memory; the fifth layer outputs decisions through the combination of "rules + large model + manual review", which not only ensures the interpretability of conclusions, but also safeguards the safety boundary.

In this structure, the real difficulty does not lie in the algorithm, but in the second and fifth layers. Cross-device deviation is an inherent trouble for chain business formats: the same face may get different test results on instruments of different models. If the calibration is not done properly, all subsequent trend judgments will be full of noise. Since skincare involves safety, if the AI's recommendation crosses the line, the user's skin will bear the cost. The project team solves this problem by building a calibration layer for devices, setting quality gates for each output, and retaining manual review links for key judgments.

Looking deeper, the technical foundation of this project is not built from scratch. Gracewell House has independently developed and launched the AI Smart Beauty Chain System in 2025, which analyzes skin problems through GENE DERM AI scanning and interacts with the cloud formula library to generate solutions. The group holds more than 30 patents in total, covering 8 major categories and over 200 products, truly realizing the integration of R&D, production and service. What this project is really going to make up for is the weakest link of the previous system - memory and time.

The design of the data closed loop also revolves around "memory": first detection in store, re-test at home, formula adjustment, result write-back. Every skincare action will update the personal profile, so that the model's understanding of the same person will gradually deepen. Once this closed loop is put into operation, the competitive barrier will no longer be the accuracy of a single detection, but the length of the profile.

How to calculate the account of this slow-paced business

From the perspective of business structure, this project does not create a new brand, but is built on Gracewell House's existing chain system. Its revenue comes from three parts: in-store detection services, customized formula products adjusted by cycle, and cross-cycle skin management services. One SkinID connects consumers, stores, headquarters and factories at the same time: stores get service rights and reasons for repeat purchases, headquarters get continuously updatable data assets, and factories get small-batch orders with customized formulas according to consumers' actual skin conditions.

The target users are very clear: first, the group of 18-45 year old core customers who have long been troubled by skin problems and made repeated trial and error, second, high-repeat-purchase members in stores who already have a stable in-store visit frequency. The former contributes high demand intensity, and the latter contributes high fulfillment efficiency - they would have come to the store anyway, but in the past, every visit started from scratch.

As the overall skincare market maintains stable growth, its growth logic is shifting. Data released by the National Bureau of Statistics of China on January 19, 2026 shows that the total retail sales of consumer goods in 2025 reached 50.12 trillion yuan, up 3.7% year on year. Among them, the retail sales of cosmetics of entities above designated size reached 465.3 billion yuan, up 5.1% year on year, outperforming the overall consumer goods retail market and reversing the 1.1% year-on-year decline in 2024. McKinsey predicts that the global beauty market will maintain an average annual growth rate of about 5% from 2025 to 2030, reaching a scale of 590 billion US dollars in 2030, and skincare will still account for about 40% of the market share.

While the growth rate is turning positive, the industry's competition focus has shifted from "launching more new products" to "verifiable efficacy, personal relevance, and sustainable services". This is exactly the window of opportunity for personalized skincare: what it sells is not a single face cream, but a recorded and verified process that truly suits the user.

In terms of progress, the project has completed the layout of in-store skin detection instruments, and entered the first phase of implementation and verification: it prioritizes the launch of SkinID data standards, quality gates and personal baseline models, runs through the three links of in-store first detection, trend interpretation and re-test, and focuses on verifying three indicators: detection pass rate, 28-day cycle completion rate and re-test consistency. The second and third phases will promote the development, pairing and calibration of home detection instruments, complete the learning of skincare and product response, build the fulfillment closed loop, and iteratively update the personal prediction model. Compared with Gracewell House stores that have not accessed the customized skincare system, the stores with the system have achieved higher customer unit price and higher customer repeat purchase rate.

In terms of team, the project is led by Ms. ZHENG Zhihong for overall operation, and YI Da is responsible for omni-channel marketing and online business operation. Relying on Gracewell House's existing offline chain network and supply chain, it also accesses the scientific research resources of Baiyun Meibay International Cosmetics Research Institute Cluster - this cluster was established under the leadership of Baiyun District Government of Guangzhou in 2020, gathering scientific research forces from more than 10 universities, having developed more than 10,000 transformable formulas and served more than 400 beauty enterprises, mainly undertaking raw material innovation and efficacy verification. At the group level, CAI Bingguo, the founder of the brand, takes the lead. The brand has been in operation for 14 years, and has fully experienced the expansion cycle from 1,000 stores to 2,000 stores after the national franchise policy was opened in 2016. This channel experience is exactly what pure online skin detection products lack most: algorithms can be replicated quickly, but more than 1,000 stores that can implement the project cannot.

For the rhythm of this business, YI Da's judgment is "slow is fast". "Many people are chasing the AI trend, trying to build a universal model that can solve all problems at once. But skincare is a very personal long-term matter. We need to dare to slow down, steadily accumulate real data of every user, clarify the basis for every adjustment, and let consumers get real effects." He said that the team does not aim to disrupt the industry, but only wants to fill the gap of "continuous verification".

This statement is logically valid. The real barrier of personalized skincare does not lie in the accuracy of a single detection, but in the number of years of data accumulation: whoever starts continuous recording earlier will get the ability to judge trends earlier. Thousands of stores are its starting advantage, but whether it can convert this batch of offline traffic into continuously updatable data assets, and fully verify and run through the complete private customized in-store and home joint skincare solution still depends on the fission progress of the model.