The Search for a Flawless Match
Anyone who has ever bought foundation knows the struggle. Standing under harsh store lighting, swatching shades on your hand, and trying to guess which bottle will actually look natural is a universal pain point. The beauty industry has long sought a better
way, and now, artificial intelligence is being presented as the ultimate solution. Brands are rolling out AI-powered tools on their websites and apps, promising a perfect shade match from the comfort of your home. The process seems simple: you upload a photo, and an algorithm analyzes your skin to recommend the ideal product. This technology aims to eliminate the guesswork, reduce product waste from bad purchases, and make shopping for makeup more accessible and personalised. But for the AI to get it right, it needs to see more than just the colour on the surface.
Tone vs. Undertone: A Crucial Distinction
To understand the challenge for AI, we first need to clarify two terms that are often confused: skin tone and undertone. Skin tone is the surface colour of your complexion—often categorised as fair, light, medium, or dark. This is determined by the amount of melanin in your skin and can change with things like sun exposure. Your undertone, however, is the subtle, permanent hue just beneath the skin's surface. It doesn’t change. Undertones fall into three main categories: cool (pink or blueish hints), warm (yellow or golden hints), and neutral (a balance of both). Two people can have the same medium skin tone but entirely different undertones, which explains why a foundation can match your skin's depth but still look 'off'. One might need a golden-based formula, the other a rosy one.
How AI Tries to See Your Skin
When you use an AI shade-matching tool, its primary job is to analyse the pixels in your photo to determine both your skin tone and your undertone. The technology uses advanced image processing and machine learning models to scan your face. It assesses factors like melanin levels and subtle colour variations to place you on a complex colour map. Sophisticated systems also try to account for lighting conditions, which can dramatically alter how your skin appears on camera. Once the AI has its reading, it compares your unique skin profile to a database of foundation shades, each with its own coded tone and undertone, to find the closest match. In theory, it's a data-driven process that should be far more accurate than the human eye.
The Ghost in the Machine: Algorithmic Bias
The biggest hurdle for AI in beauty is bias. An AI model is only as smart and as fair as the data it was trained on. For years, many AI systems, including those for facial analysis, were trained on datasets that were overwhelmingly composed of lighter skin tones. This creates a massive blind spot. When an algorithm hasn't 'seen' enough examples of deep skin tones or the vast spectrum of undertones present in populations like India's, its accuracy plummets. Studies have shown that facial analysis AI can have error rates over 34% for darker-skinned women, compared to less than 1% for light-skinned men. This isn't because the AI is intentionally prejudiced; it's because it was built from an incomplete and unrepresentative picture of humanity. The result is that the people who have historically been underserved by the beauty industry are often the same people for whom the new 'solutions' fail.
Building a More Inclusive Future
The good news is that the industry is starting to recognise this problem. Forward-thinking brands are actively working to build more inclusive AI. The solution begins with data. Companies are expanding their training datasets to include a much wider range of ethnicities and skin tones, like the Monk Skin Tone Scale, which was designed to be more inclusive for AI applications. In India, there's a growing understanding that products and technology must be developed for diverse local skin types, which have different pigmentary and aging patterns. Researchers are also developing techniques to help AI better correct for tricky lighting and more accurately identify undertones. Ultimately, making AI work for everyone requires a conscious effort to ensure the technology reflects the true diversity of the people it's meant to serve.
