Advancing Hyper-Personalized Cosmetics: A Dual Approach Combining Regulatory Methodology and Live A/B Performance Testing

The demand for personalized personal care products has grown significantly, driving the beauty industry to develop highly tailored cosmetic solutions^1,2. Leveraging artificial intelligence (AI) and modular formulation principles, brands can now customize products to address specific hair and skin needs from thousands to millions of possible ingredient combinations.

This technical paper highlights two fundamental capabilities of a hyper-personalized cosmetic brand operating through a made-to-order, direct-to-consumer (DTC) business model:

• Industrial-Scale Regulatory Management: The capability to systematically design, validate, and document compliance for millions of customized formulas.
• Agile Formulation Optimization: The capability to seamlessly de-risk and validate physical ingredient upgrades across bespoke formulas using real-world consumer data.

Historically, achieving these capabilities has been hindered by two major challenges. First, under strict global frameworks such as the European Cosmetic Regulation (EC No. 1223/2009)³, every single formulation variant placed on the market must undergo a full safety and quality assessment. Second, when introducing formula renovations - such as substituting raw materials or upgrading active ingredients across personalized formulas - predicting how consumers will perceive these changes across highly diverse, customized profiles is notoriously difficult.

Traditionally, the cosmetic industry relies on well-established, regulatory-requested standard tests to clear safety, quality, and performance, typically evaluated in controlled environments with small clinical panels.

This paper presents a comprehensive, two-part framework that unites these standard testing requirements with the unique digital advantages of a made-to-order DTC model. First, we outline a risk-based methodology developed to ensure safety and regulatory compliance. Second, we present a live, large-scale A/B testing methodology (online controlled experimentation inspired from the software engineering⁴) designed to validate performance upgrades in personalized products directly in consumers' everyday lives.

Experimental Design

Part I: Regulatory Compliance, Safety and Quality Methodology

To establish regulatory compliance across millions of potential product variations without testing every single iteration individually, a robust, risk-based validation protocol is required. This methodology leverages standardized testing protocols well-established in the cosmetic industry and requested by regulatory bodies to secure the entire customized formulation space. This work presents a comprehensive and scalable approach developed by a multidisciplinary team of toxicologists, regulatory experts, and formulators, illustrated through the case of a personalized conditioner.

The Formulation Architecture Model

A "Formulation Architecture" is established for each product category (e.g., a personalized conditioner) by mapping all eligible ingredients and defining their precise concentration boundaries.

Rather than testing every unique consumer formulation, a multidisciplinary team of toxicologists and formulators identifies three types of formulations for standardized testing:

• Texture Formulations: To establish a general formulation behavior baseline.
• Realistic Formulations: Mirror the most common customization patterns that reflect major consumer needs.
• Critical Formulations: Represent the most challenging combinations identified through expert-driven risk profiling.

Standard Industry Testing Protocol

These designated formulations are subjected to a rigorous, industry-standard testing matrix to generate the necessary data to build robust regulatory files (such as the Product Information File [PIF] in Europe) and the toxicological safety assessment (such as the Cosmetic Product Safety Report [CPSR] in Europe):

• Microbiological Safety: Challenge testing according to the ISO 11930:2019⁵ standard, where formulations are defined as critical based on the risk profile of their ingredients, specifically their potential to support microbial growth or diminish preservative effectiveness (e.g., water-based, protein-, or sugar-rich actives).
• Physical Stability & Compatibility: Evaluated under accelerated conditions (40°C for three months) and real-time conditions (12 months at room temperature, 4°C, and -18°C) in final packaging to monitor physical and chemical integrity and container-content interactions.
• Ocular & Dermal Tolerance: In vitro ocular irritation testing on reconstructed human epithelial tissue models and 48-hour patch tests, as well as in-use safety evaluations under dermatological supervision.
• Efficacy Evaluation: To support core performance or specific product claims, efficacy is assessed through in vitro/ex vivo testing, use tests or clinical studies. As an example for a conditioner, hair breakage (a key performance indicator) is evaluated using an automated grooming simulator. Hair tresses with different textures treated with various formula combinations containing minimal to maximal conditioning actives (n=10 per group) undergo 2,000 combing cycles to measure the average reduction in broken fibers compared to untreated controls, analyzed via a Student’s t-test at a 95% confidence level.

Part II: Live A/B Performance Testing Methodology

While traditional laboratory testing confirms safety and baseline efficacy, a made-to-order DTC brand can deploy live A/B testing to validate actual consumer-perceived performance during formula renovations across bespoke routines. This methodology is illustrated through a performance-focused renovation of personalized conditioners, where the original conditioning silicone was replaced by a cationic silicone crosspolymer (Silicone Quaternium-16/Glycidoxy Dimethicone Crosspolymer, supplied as a 25.5% aqueous dispersion with Undeceth-11, Undeceth-5, Propylene Glycol, Acetic Acid and Benzyl Alcohol) to enhance overall consumer satisfaction and boost targeted conditioning benefits (e.g., softness, detangling and repair benefits for damaged hair).

Live Test Design

The A/B test was integrated directly into the brand's e-commerce platform, a method uniquely enabled by the made-to-order DTC model. Every customer completed an online consultation defining their specific hair type, goals and preferences. When their bespoke conditioner was formulated, consumers were randomly assigned, blinded to their group allocation, to receive one of two customized physical variations:

• Formula A (Control): Personalized conditioners formulated with the original conditioning system.
• Formula B (Test): Personalized conditioners incorporating the new silicone ingredient, tailored to their individual profile.
• Allocation Ratio: 70% control to 30% test.

A total of 82,447 survey responses were captured (n = 60,090 control, n = 22,357 test).

Performance Evaluation

Three weeks after product delivery, consumers received a digital survey prompting them to rate the personalized product's overall performance and specific conditioning benefits on a 1–5 scale. Collected data were analyzed using two complementary statistical approaches: Chi-square tests comparing Top2Box (ratings of 4 or 5) and Bottom2Box (ratings of 1 or 2), with a significance threshold set at p < 0.05, and a Bayesian statistical analysis to strengthen the robustness of conclusions. Error bars in figures represent 95% confidence intervals.

Results

Part I: Compliance Assessment

Standard industry evaluations of the personalized conditioner's critical formulations met all acceptance criteria across:

• Antimicrobial Preservation: All tested critical formulations (including those packed with high concentrations of microbial nutrient sources like proteins and sugars) successfully met Criterion A of the ISO 11930:2019 standard.
• Stability & Packaging: The formulas maintained physical stability (pH, viscosity, and color) under accelerated and real-time conditions with no packaging degradation.
• Tolerance: Standard evaluations confirmed dermatological and ocular tolerance.
• Instrumental Hair Breakage: For a conditioner, the automated grooming simulator showed a statistically significant reduction in broken hair fibers across all texture variations (even those formulated with minimal conditioning actives), fully supporting the product's core efficacy claims.

Part II: Consumer-Perceived Performance in Personalized Formulas

Integrating the new silicone into the personalized conditioner formulas resulted in statistically significant improvements:

• Overall Consumer Satisfaction: The personalized Formula B incorporating the new silicone achieved 74% overall satisfaction (Top2Box) compared to 72.2% for the personalized control group. Dissatisfaction (Bottom2Box) fell from 8.1% for the control to 7.5% for the renovated personalized formula.

Figure 1 - Comparison of overall satisfaction (Top2Box and Bottom2Box scores) between
control and renovated personalized conditioner formulas integrating the new silicone.Courtesy of Prose

• Perceived Conditioning Efficacy: Reports of dry, under-conditioned hair declined from 29.8% (control) to 26.3% (Formula B), while reports of soft and manageable hair rose to 67.7% (compared to 65.2% for the control), consistent with improved conditioning performance delivered by the new ingredient in tailor-made products. Reports of over-conditioned, heavy hair rose slightly, from 5% to 6%, reflecting the intended shift toward greater conditioning.

Figure 2 - Comparison of perceived conditioning benefits between control and renovated personalized conditioner formulas integrating the new silicone.Courtesy of Prose

• Statistical Significance: A Chi-square test of difference confirmed that the renovated formula demonstrated statistically significant improvements. A Bayesian posterior probability analysis further supported these results, showing a 99.7% probability that the test group had a lower proportion of dissatisfied consumers (Bottom2Box) than the control, and a 100% probability that the test group had a higher proportion of satisfied consumers (Top2Box) compared to the control group.

Discussion

The integration of standard testing and large-scale A/B testing highlights how a made-to-order DTC beauty brand can master both regulatory compliance and physical product development across custom formulations in ways that cannot be achieved through conventional retail distribution.

Mastering the Regulatory Frontier at Scale

First, the regulatory framework resolves the impracticality of individually testing millions of custom formulas. By establishing a master Formulation Architecture and identifying the "critical formulation" (the most challenging combination of ingredients), toxicologists can rigorously evaluate safety and stability boundaries using standard, regulatory-requested tests.

Once this critical envelope has been assessed as safe, the toxicologist can confidently approve the entire customized range and validate a comprehensive and robust regulatory file. This demonstrates that compliance can be scaled across the full set of permitted formulation combinations using standard industry protocols.

Leveraging the DTC Loops

Second, the made-to-order DTC channel bridges the gap between controlled laboratory settings and real-world consumer experiences. Conventional industry tests (such as combing simulators or small clinical panels) confirm product efficacy but cannot replicate how a new active or functional ingredient behaves across thousands of different personalized formula combinations under real-world usage conditions.

By testing the renovated personalized conditioners (incorporating a new silicone) in a randomized, consumer-blinded setting directly within the DTC distribution loop, the brand collected over 82,000 real-world evaluations. This volume provides an empirical sample size that conventional consumer testing panels could not match. The data confirmed that incorporating a new silicone into custom formulas significantly enhanced targeted conditioning benefits - reducing dry hair complaints and boosting softness across diverse consumer profiles - demonstrating the second capability: using physical-digital closed-loop data to validate performance upgrades in personalized cosmetics with high statistical confidence.

Methodological Considerations and Future Perspectives

From a methodological standpoint, because consumers were blinded to their allocation, observational bias is minimized, although INCI lists on pack are theoretically and legally accessible. Furthermore, while three-week post-use evaluations capture acute satisfaction and immediate hair performance, future research avenues could explore longitudinal tracking over extended periods (e.g., six to 12 months) to assess long-term cumulative benefits across customized routines.

Conclusion

Hyper-personalization in cosmetics requires robust, scale-ready methodologies that maintain high standards of safety, quality, and performance.

As demonstrated in this paper, a made-to-order DTC model allows this complexity to be managed by harnessing two essential capabilities:

1. Establishing a risk-based regulatory framework that secures the formulation space by testing critical combinations using industry-standard protocols.
2. Deploying a live, e-commerce-integrated A/B testing model to collect real-world consumer feedback at scale and validate performance renovations within personalized product lines.

Together, these capabilities provide a practical framework for continuous product optimization and consumer satisfaction in the expanding beauty-tech sector.

References

1. Di Gesu, R. (2024). Personalised facial skincare is worth $4bn in the US. Mintel Beauty & Personal Care.
2. Lee, H. J. (2022). Patent insights: personalised beauty. Mintel Beauty & Personal Care.
3. Regulation (EC) No 1223/2009 of the European Parliament and of the Council of 30 November 2009 on cosmetic products (recast), OJ L 342, 22.12.2009, pp. 59-209
4. Quin, F., Weyns, D., Galster, M., & Costa Silva, C. (2024). A/B testing: A systematic literature review. The Journal of Systems and Software, 211, 112011
5. International Organization for Standardization. (2019). Cosmetics - Microbiology - Evaluation of the antimicrobial protection of a cosmetic product (ISO 11930:2019)