One face may have an oily T-zone and cheeks that behave differently. Combination skin is among the most commonly self-reported types, but it is a weak basis for product matching. The label records a real observation, yet misses the degree of difference between zones and whether the T-zone or cheeks should guide the recommendation.
Product labels cannot sort this out. A combination-skin moisturizer is usually a compromise: light enough for an oily forehead, but not rich enough for dehydrated cheeks. For many people, one product made for combination skin serves neither zone particularly well.
We read a face photo as a map of zones, not one continuous surface. Sebum across the forehead, nose, and chin can look unlike the barrier texture around the cheekbones and jawline. For many people who call their skin combination, these zones genuinely need different things. Recommendations based on zone signals, rather than one broad label, therefore produce a different set of products.
What "Combination" Actually Describes
The combination category came from clinical observations of patients with oilier T-area zones and relatively drier cheeks and jawline. That observation was useful. The difficulty is that "combination" compresses a distribution problem into one label: how far apart are the zones, under what conditions, and which creates the most immediate concern?
If the T-zone is only slightly more active than the cheeks, a lighter general moisturizer will probably work well. If the forehead and nose produce significant sebum while the cheeks show barrier disruption, one product is being asked to perform two incompatible jobs. The label includes both people, but the resulting recommendation cannot serve both equally well.
Pixel-level readings across facial zones show more than a binary of oily here and dry there. They show gradients, zone-by-zone sebum variation, cheek texture that may signal dehydration instead of inherent dryness, and forehead pore geometry indicating different activity levels. The label misses this resolution. The image contains it.
Zone-Level Reading vs. Global Category
Early in building the skin reading model, we decided against producing one global skin type score. It would have been easier to output "combination" or "oily-combination" and recommend from there. We rejected that format because it would repeat the resolution loss that makes single-product recommendations for combination skin unreliable.
We instead extract features from the forehead, nose and immediate perinasal area, cheeks, chin, and perioral area. These zones can correlate, but their signals may still diverge. That divergence is often the key matching signal. A formula addressing an oily T-zone and a dehydrated cheek makes a compromise for the shopper. Reading each zone lets the recommendation prioritize the dominant concern while noting the secondary one.
This method has failure modes. Consumer photos taken under uncontrolled lighting add noise to zone readings. A shadow over one cheek changes the texture signals, while camera characteristics affect how pores appear at close distances. We account for these variables in the pipeline, but low-quality inputs remain less accurate than we would like. A single phone photo is not the resolution of a clinical imaging system.
What This Changes for Product Matching
For someone who has bought combination-skin products for years, zone-level reading may produce a different recommendation, especially for an evening routine. The T-zone may call for a lighter texture with sebum-regulating actives, while the cheeks may call for ceramide-supportive or humectant-rich formats. Some formulas address both. The matching question becomes specific: does the product address what this zone is doing, rather than simply belong to the combination category?
For brands, the value lies in seeing what their product is matched to. A moisturizer for combination skin may suit T-zone regulation while being neutral or suboptimal for cheek dehydration. Which zone drives the match, and which is merely tolerated, can inform product development and targeting. A skin type label on a return form does not provide that information.
The Limits of What a Label Can Carry
This does not make combination skin a wrong label. Printing zone-level behavior on a moisturizer jar is impractical, and the category is useful shorthand for many people whose zones differ only mildly. It suits its communication task: helping a shelf shopper decide within perhaps thirty seconds whether a product is for them.
A recommendation engine should not share the limits of a label. A shopper using Cutib is not browsing a shelf. They have taken a photo, and the model has read it. That context permits greater specificity, especially for people at the more extreme end of the combination spectrum. It can separate a product that works from one that helps for two weeks before the cheeks feel tight again.
We are still working out how to present zone differences to the shopper. Too much detail is harder to use than a clear recommendation, while too little undermines a more precise reading. This is a communication design issue as well as a model issue. The proper sequence is to read the skin more accurately, then decide how much of that resolution belongs in the recommendation. We are not reversing that order.