The Promise and the Problem
Artificial intelligence has arrived in haircare. Major brands and startups alike are deploying AI-powered diagnostic tools that promise personalised hair analysis and product recommendations at scale. The technology is real, the investment is significant, and the potential is genuine.
The problem is equally real: the vast majority of these tools have been trained on datasets that underrepresent textured, coily, and kinky hair types. The result is a generation of diagnostic technology that works well for the populations already best served by the industry — and works poorly for those who need precision most.
CROWN has reviewed the major commercially available AI hair diagnostic platforms. Our findings are presented here not as criticism, but as a rigorous assessment of where the field stands and what must be built next.
Commercial Platforms Reviewed
Kerastase Diagnostic Capillaire
Kerastase, L’Oreal’s premium professional haircare brand, offers one of the most sophisticated in-salon diagnostic tools available. The system combines visual assessment with a structured questionnaire and, in some implementations, microscopic imaging.
Strengths: The Kerastase system is well-integrated into the salon workflow and produces actionable product recommendations. The underlying hair science is sound, reflecting L’Oreal’s substantial research investment.
Textured hair coverage: The system’s reference imagery and classification framework are predominantly oriented toward straight and wavy types (broadly, types 1A through 2C on the Walker scale). Curlier types (3A-3C) receive moderate representation. Tightly coiled and kinky types (4A-4C) — precisely the textures most affected by hair discrimination — are underrepresented in both the visual references and the recommendation algorithms.
L’Oreal My Hair Diagnosis
L’Oreal’s consumer-facing diagnostic tool uses a smartphone questionnaire — and in some markets, AI-powered photo analysis — to assess hair type, condition, and needs.
Strengths: Accessibility. The tool brings a structured diagnostic process to consumers who may never visit a professional salon. The interface is well-designed and the recommendations are specific.
Textured hair coverage: The photo analysis component performs best on hair that is worn loose and falls in recognisable wave or curl patterns. For hair worn in protective styles — braids, locs, twists, Bantu knots — the visual AI frequently cannot assess the underlying texture. For very tightly coiled hair, the classification accuracy decreases measurably. Self-reported inputs partially compensate, but the core limitation is in the visual model’s training data.
MyHair.AI
MyHair.AI represents the emerging category of AI-first hair diagnostic startups. The platform uses computer vision to assess curl pattern, density, and apparent condition from user-submitted photos.
Strengths: The platform explicitly acknowledges hair type diversity in its marketing and user interface. The classification system attempts to cover the full spectrum.
Textured hair coverage: Despite the inclusive positioning, the platform’s accuracy — as assessed by CROWN’s review — decreases for tighter curl patterns. This is consistent with training data limitations: publicly available hair image datasets are overwhelmingly composed of straight and wavy hair, and building diverse training sets requires deliberate, resource-intensive data collection.
MDhair
MDhair positions itself at the intersection of dermatology and hair diagnostics, using AI to assess scalp health and hair loss patterns.
Strengths: The dermatological focus adds clinical credibility, and the platform integrates with telehealth consultations.
Textured hair coverage: Scalp imaging on darker skin tones and through densely textured hair presents specific technical challenges that the platform acknowledges but has not fully resolved. This mirrors broader challenges in dermatological AI, where diagnostic accuracy has been shown to decrease for darker skin tones (Adamson and Smith, 2018).
Revieve
Revieve provides white-label AI diagnostic solutions to beauty retailers and brands. Their technology powers personalised recommendation engines across multiple retail platforms.
Strengths: Scale. Revieve’s infrastructure reaches millions of consumers through its retail partners.
Textured hair coverage: As a B2B platform, Revieve’s accuracy depends significantly on the training data provided by each retail partner. Where partners have invested in diverse training data, the results improve. Where they have not, the platform inherits the same biases present in the broader industry.
Becon
Becon (Beauty Connected) develops AI tools for salon professionals, including diagnostic and consultation support.
Strengths: Professional-grade tools that enhance the stylist-client consultation process.
Textured hair coverage: The platform’s European focus means its training data reflects European salon demographics. In markets where textured hair represents a smaller proportion of salon clients, the training data is correspondingly less diverse.
The Structural Pattern
Across all platforms reviewed, a consistent pattern emerges. It is not that these companies are indifferent to textured hair. Several have made genuine efforts to expand coverage. The structural problem is deeper.
Training data scarcity. Machine learning models require large, labelled datasets. Publicly available hair image datasets — the foundation on which most commercial models are initially trained — dramatically overrepresent straight and wavy hair. Building diverse datasets requires deliberate investment in data collection from underrepresented populations, which demands cultural competence, trust-building, and ethical frameworks that go beyond standard data acquisition.
Measurement methodology. Most AI hair diagnostics rely primarily or exclusively on visual analysis — photography, sometimes microscopy. Visual assessment, even augmented by AI, cannot measure fibre diameter, porosity, hydration, protein structure, or chemical treatment history. These properties are precisely what differentiate hair types at a level of precision relevant to both product efficacy and discrimination research.
Validation gaps. Claims about diagnostic accuracy are rarely validated across the full spectrum of hair types. A tool that achieves 90 percent classification accuracy on a test set dominated by types 1-2 may perform significantly worse on types 3-4, but if the test set itself underrepresents these types, the aggregate accuracy metric conceals the disparity.
Commercial incentives. AI diagnostic tools are typically deployed to drive product recommendations and sales. When the most profitable product categories serve straight and wavy hair consumers, the commercial incentive to invest in textured hair diagnostic accuracy is weaker. This is a market failure, not a technical limitation.
What Must Be Built
The limitations documented above are not inevitable. They are the product of specific choices about data collection, training methodology, and investment priorities. CROWN’s approach to AI classification addresses each of these limitations directly.
Multi-modal sensing. CROWN’s diagnostic device does not rely solely on visual analysis. By combining optical micro-imaging, near-infrared spectroscopy, and impedance sensing, the system captures hair properties that visual AI cannot access. This multi-modal approach is fundamental to achieving consistent accuracy across all hair types.
Diverse training data by design. The CROWN Hair Commons is structured from inception to ensure comprehensive representation across all hair types, ethnicities, and textures. Universal coverage is the architecture, not a feature added after initial deployment.
Sensor-verified ground truth. AI classification accuracy depends on the quality of ground truth labels. When ground truth is visual assessment by human raters, inter-rater reliability is low — particularly for tightly textured hair. CROWN’s sensor-verified measurements provide objective ground truth that does not depend on subjective visual classification.
Open validation. CROWN publishes diagnostic accuracy metrics disaggregated by hair type, enabling independent verification that the system performs equitably across the full spectrum. Aggregate accuracy metrics that conceal performance disparities are insufficient.
The Opportunity
The AI hair diagnostics market is growing rapidly. The companies building these tools have the engineering talent, the distribution networks, and the commercial incentive to serve consumers well — including the 150 million Europeans with textured hair who are currently underserved.
What they lack is the data infrastructure. CROWN’s diagnostic technology and open Data Commons are designed to provide precisely this infrastructure — not as a competitor to existing platforms, but as the foundational layer that makes all of them more accurate, more equitable, and more useful for every consumer.