Industries / Consumer goods / Cosmetics and personal care
Matching shades and products to people
This is the work of fitting a product to a buyer: choosing which foundation, powder or lipstick shades to make, and helping each customer find the one that matches their skin, by eye, by machine or by phone camera.
6
Milestones achieved, from 1918 to 2022.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1918
Shades matched to coloring
Max Factor's company history says he made Flexible Greasepaint in 1914 as the first makeup for film, and in 1918 introduced "Colour Harmony", in which makeup shades were chosen to suit a woman's complexion, eyes and hair. A cosmetics historian records that his company patented a makeup base for Technicolor film in 1936, used it in the 1937 film "Vogues of 1938" and sold it to the public as Pan-Cake from 1938; the company's own timeline dates Pan-Cake to 1935.
Credit
Max Factor
1975
A sunburn scale becomes standard
Thomas Fitzpatrick, a Harvard dermatologist, sorted skin by how easily it burns and tans in the sun, to set the first dose of ultraviolet light for treating psoriasis in White patients; his 1988 paper describes six types, I to VI. The scale became the usual way to describe skin in dermatology and was later used to label skin tone in image data. A 2025 study in npj Digital Medicine that compared it with a colorimeter, an instrument that measures color, concluded that Fitzpatrick type is not a proxy for skin tone.
LessonA scale built to predict sunburn is a poor ruler for skin color.
Credit
Thomas B. Fitzpatrick
2015
Foundation blended at the counter
Lancôme, a L'Oréal brand, began making foundation to order in November 2015 at two Nordstrom stores in the US. An adviser scans the customer's face to measure skin tone, a machine mixes a matching foundation, and a shade number is printed on the bottle for refills. By 2017, according to the trade journal Cosmetics Business, the machine could mix 72,000 shades and the service ran at 11 US counters and was opening at Harrods in London.
Credit
Lancôme (L'Oréal)
2017
Forty shades at launch
Fenty Beauty, the singer Rihanna's brand made with its parent company Kendo, launched in September 2017 with 40 foundation shades. Time reported that the deepest shades soon sold out at Sephora and that other brands followed with campaigns aimed at women of color. In June 2018 the data journalism site The Pudding measured the lightness of each shade as shown on brand websites: Fenty's 40 shades were spread evenly from light to dark, while 31 of Make Up For Ever's 40 shades sat in one band of light and medium tones.
LessonThe number of shades says little; what serves buyers is how the shades spread from light to dark.
Credit
Rihanna and Kendo; The Pudding (Jason Li, Amber Thomas)
2018
L'Oréal buys virtual try-on
On 16 March 2018 L'Oréal bought ModiFace, a Toronto company founded eleven years earlier by Parham Aarabi, whose software draws makeup in 3D on a camera image of the shopper's face and offers color and skin diagnosis. L'Oréal said the technology was already used by almost every leading beauty company, and that ModiFace, with nearly 70 engineers, researchers and scientists, would serve its 34 international brands.
Credit
ModiFace (Parham Aarabi); L'Oréal
2022
An open skin tone scale
Google and the Harvard sociologist Ellis Monk released the Monk Skin Tone Scale, 10 shades meant to cover human skin tones more evenly than the tech industry's usual scale, free for anyone to use. Google used it to let people filter makeup image searches by skin tone. A 2025 study in npj Digital Medicine found that Monk ratings clustered more tightly in color space and repeated well in person and from photos, although ratings from photos varied with lighting.
Credit
Ellis Monk; Google
Watch for
Phone scan picks foundation shade
Clarins says its AI Shade Finder, built with the startup IlluminateAI, reads skin tone and undertone by taking a rapid series of phone photos while changing the light from the screen, and agrees with a seasoned makeup artist's choice of shade 96% of the time. Clarins also told the trade site Glossy that a 2025 trial in 20 stores in France and the UK raised conversion, the share of served shoppers who bought, to 70% and doubled the average basket. The tool runs at more than 100 boutiques and counters in seven countries. No outside test was found.
Achieved whenan outside lab publishes the tool's picks against instrument-measured colors, across skin tones and shop lighting.
Claimed by
Clarins and IlluminateAI, February 2026
Open
01
A measured library of shades
A free, open table of foundation shades as they look once spread out. Each shade is drawn down at a fixed thickness on standard black-and-white opacity cards and on a synthetic skin substrate, and measured with a spectrophotometer wet, after one hour and after eight hours. Each row gives brand, product, shade name, the swatch color on the brand's website, the measured CIELAB values (the standard color coordinates for lightness, red-green and yellow-blue) and the full reflectance curve, plus the nearest Monk Skin Tone shade.
The briefCredit · unclaimed02
Shade finders on painted heads
A bench test of shade-finding tools that uses mannequin heads painted to known, measured skin colors instead of people. Paint ten heads to span the Monk scale and measure each with a spectrophotometer. Run each public shade finder, such as brand apps and websites with camera try-on, on every head under three lights: daylight, warm shop lighting and a phone's own light. Record the shade each tool picks, look up that shade's measured color in the library above, and report how far off each pick is and whether a head gets the same shade under every light.
The briefCredit · unclaimed
Open 01 · the brief
A measured library of shades
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free, open table of foundation shades as they look once spread out. Each shade is drawn down at a fixed thickness on standard black-and-white opacity cards and on a synthetic skin substrate, and measured with a spectrophotometer wet, after one hour and after eight hours. Each row gives brand, product, shade name, the swatch color on the brand's website, the measured CIELAB values (the standard color coordinates for lightness, red-green and yellow-blue) and the full reflectance curve, plus the nearest Monk Skin Tone shade.
Why it is still open
The largest public shade data, The Pudding's 2018 and 2021 sets (the 2021 one covers the names of 6,816 complexion products), record the swatch colors shown on brand and retailer websites; The Pudding called its measure "a rough proxy" because a foundation changes once applied. A 2025 Scientific Reports study did measure 32 shades from four brands on opacity charts with a spectrophotometer, but to study fading, not range. Arbelle, a seller of virtual makeup software, mapped seven brands' ranges onto the Monk scale in December 2025 and published summary findings only. Searches of Hugging Face and Zenodo found no measured shade library.
Built with what exists
A portable spectrophotometer such as Konica Minolta's CM-700d or X-Rite's Ci64; Leneta opacity charts and a film applicator for even layers; IMS's Vitro-Skin, a test substrate that mimics the surface of skin; the open-source colour-science Python library for CIELAB values and color differences (ΔE00, the standard measure of how different two colors look); Zenodo to host the data.
First step
Buy the full shade range of one widely sold foundation, draw each shade down on opacity cards, and measure them to fix the method and the repeat error.
Done looks like
At least 1,000 shades from at least 20 brands sold in at least three regions, such as Europe, North America and East Asia; each shade measured three times, with repeat readings within 1.0 ΔE00; for each shade, the color difference between its website swatch and its measured drawdown; all data and code under an open license.
Open 02 · the brief
Shade finders on painted heads
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A bench test of shade-finding tools that uses mannequin heads painted to known, measured skin colors instead of people. Paint ten heads to span the Monk scale and measure each with a spectrophotometer. Run each public shade finder, such as brand apps and websites with camera try-on, on every head under three lights: daylight, warm shop lighting and a phone's own light. Record the shade each tool picks, look up that shade's measured color in the library above, and report how far off each pick is and whether a head gets the same shade under every light.
Why it is still open
The accuracy figures for shade finders come from the companies that sell them, such as Clarins' 96% agreement with makeup artists. Published checks of phone color readings against a spectrophotometer measure people's skin for medical uses: in one, two phone apps reading 50 people's foreheads were off by about 7.6 ΔE00 on average. The 2025 npj Digital Medicine study found that skin tone ratings made from photos varied with lighting. Beauty writers' tests of shade finders judge the picks by eye on a few people; one, in October 2025, used three models and gave accuracy ranges without saying how they were measured. No test of commercial shade finders against measured targets was found.
Built with what exists
Mannequin heads; acrylic paints mixed to target colors; the spectrophotometer and shade library from the item above; a light booth with daylight-balanced (D65) and warm 3,000 K lamps; two or three current phones; the colour-science library.
First step
Paint one head to a medium Monk tone, measure it, and find out which shade finders accept it as a face.
Done looks like
At least five shade finders tested on ten heads under three lights, three times each (450 runs); for each tool, the median ΔE00 between the picked shade and the head, and the share of heads that get the same shade under all three lights; all photos, readings and picks published.