Why has Oddity's IL MAKIAGE foundation shade matching algorithm seen a 15% decrease in accuracy over the past month?

Introduction

The recent 15% decrease in accuracy of Oddity's IL MAKIAGE foundation shade matching algorithm is a critical issue that demands immediate attention. This decline not only impacts user satisfaction but also threatens the core value proposition of the product. I'll approach this problem systematically, focusing on identifying the root cause, validating hypotheses, and developing both short-term fixes and long-term solutions.

Step 1

Clarifying Questions (3 minutes)

• Looking at the timing, I'm thinking there might have been a recent update to the algorithm. Has there been any change to the shade matching model or its training data in the past month?

Why it matters: Recent changes could directly impact algorithm performance. Expected answer: Yes, there was an update to include more diverse skin tones. Impact on approach: If confirmed, we'd focus on the update's implementation and data quality.

• Considering user demographics, I'm curious about the distribution of the accuracy decrease. Is the 15% drop consistent across all user segments, or are certain groups more affected?

Why it matters: Uneven impact could indicate biases in the algorithm or data. Expected answer: The decrease is more pronounced in users with darker skin tones. Impact on approach: We'd prioritize investigating potential biases in the training data or model architecture.

• Given the nature of foundation matching, I'm wondering about any changes in external factors. Have there been any significant shifts in lighting conditions or camera technology used by customers in the past month?

Why it matters: External factors could affect input quality and thus algorithm performance. Expected answer: No major changes noted, but there's been an increase in users from regions with different lighting conditions. Impact on approach: We'd explore how to make the algorithm more robust to varying lighting conditions.

• Thinking about the measurement process, I'm curious about our accuracy metrics. Has there been any change in how we define or measure accuracy for the shade matching algorithm?

Why it matters: Changes in measurement could explain the perceived decrease without actual performance degradation. Expected answer: No changes to the accuracy measurement process. Impact on approach: We'd focus on the algorithm and its inputs rather than the measurement methodology.