Why I Built Clad—and Where the Product Stops

The path into personal color analysis is not especially clear. A professional consultation requires time and money. Online guides tend to stay generic. Even after receiving a season label, a person may still need to work out which clothing or makeup colors fit the result.

Clad started in the gap between those options. It is not an attempt to deliver an unquestionable diagnosis from one photo. It is a lower-friction first analysis for someone who is curious about personal color and wants to translate the result into a practical palette.

The job the product needs to do

Users need more than a category name. They want to know what the product observed in their photo, why certain colors are suggested, and what they can carry into the next shopping or makeup decision.

That creates three product responsibilities:

1. Help the user prepare a photo suitable for analysis.
2. Explain the result, the reasoning, and any meaningful uncertainty.
3. Connect the analysis consistently to palettes and beauty recommendations.

If any one of these steps is missing, a technically plausible prediction can still be a poor product result.

The boundary matters as much as the capability

Clad cannot reproduce every part of an in-person consultation. A photo does not perfectly represent skin tone, and it is affected by lighting and camera processing. Personal preference, context, and style also influence whether a color feels right.

The service should not disguise those limitations. It should reject unsuitable inputs, explain lower-confidence results, and present recommendations as a useful starting point rather than a fixed rule. A product earns more trust by being precise about its scope than by suggesting it has replaced expert judgment.

Why use AI at all

AI makes it possible to analyze visual features and assemble a personalized explanation and palette within one on-demand flow. It can reduce the cost of producing an individual result and meet the user at the moment of curiosity.

But the presence of AI is not the value proposition. If the result varies without explanation or the recommendations feel generic, the service can be less trustworthy than a conventional guide. Model and prompt work have to be developed alongside input validation, result structure, recommendation rules, and feedback collection.

The current hypothesis

My current hypothesis is that users prefer a concise explanation they can understand, a palette they can compare, and recommendations they can reference immediately. I also expect honest treatment of uncertainty to make the result easier—not harder—to use.

A launch does not prove that hypothesis. I need to observe whether people read the explanation, continue into the palette and recommendations, and understand when they should try another photo. Those behaviors will show whether Clad is doing a useful job rather than simply producing an interesting answer.