A try-on tool is judged by its worst match, not its average one. If it fails on deep or very fair skin, the customers it fails are the ones most likely to have given up on buying shade online.
Virtual try-on for beauty uses a phone or laptop camera to overlay make-up on a live face, and AI shade matching estimates skin tone and undertone to recommend a foundation or concealer shade. The overlay is a solved problem in most respects. The match is not, because a camera does not measure colour: it reports it after white balance, exposure and screen settings have had their say.
So we design around calibration. The system checks lighting before it estimates, asks for a retake when the light is wrong rather than guessing, and recommends a shade with an honest neighbour (the next shade lighter or deeper) when the estimate falls between two. It is tested against the full depth of your range, with results reported separately for each tone band so a weak band cannot hide inside a good average.
The face is personal data under GDPR, and the EU AI Act prohibits using biometric data to infer characteristics such as race or ethnicity. We match on measured tone and undertone only, process images for the session, and do not store them unless the customer explicitly asks.
Validation uses images of real people across the full range of skin depth and undertone, in the lighting your customers actually shop in, with the correct shade agreed by trained artists. Accuracy is reported per tone band, and launch is gated on the weakest band, not the overall figure. After launch, returns tagged as wrong shade are fed back into the evaluation set.
It depends mostly on lighting and on how the system was tested. A good shade matcher works well in reasonable light and knows when the light is too poor to answer. The common failure is weaker performance on deeper skin tones, so we report accuracy per tone band and gate launch on the weakest band rather than an average.
A face image is personal data. It becomes biometric data in the special-category sense when processed to identify a person, which shade matching does not need to do. We still treat it carefully: processing for the session only, no retention by default, and no inference of sensitive characteristics, which the EU AI Act prohibits. Your data protection assessment is yours to make.
It can reduce returns caused by a wrong shade, if the matching is accurate across your customers’ skin tones. It does less for returns caused by texture, finish or reaction. We measure it by tagging returns by reason before and after launch, rather than quoting a reduction in advance.
What we built, what stage it is at, and a link to the live thing wherever there is one.
yourOS is an operating system with an AI at its heart. You talk to it like a person, out loud or typed. It makes your calls, plans your days and works on your projects with a crew of agents, on your own private server. Our own product, built from the first commit; the waitlist is open.
Website, backend and app assets for Sonora, a free AI sound-therapy app on iOS and Android published by Cognitive AI.
Brand, website, prototype, apps on five platforms, the AI backend and the algorithm-intelligence pipeline for Podswap, a social growth engine that runs without a human in the loop.
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
Live at coherence.com: brand identity generated with AI agents we directed, the motion system and website for the group behind Sonora, AboutAI, WakeUp and MudShadows. Agent work remains in build.
Live at domainbooth.com for DomainBooth, a premium domain brokerage: a new animated mark, every page redesigned and rebuilt with designed mobile layouts, and a custom CMS for its sales record, featured names, press and enquiries.
We start by testing your current shade finder, or a prototype, across every tone band you sell to.
30-minute technical call · no deck