Lessons from Building AI Tools for the Beauty Industry

At Lunaray Beauty Factory in Bandung I work on software, and increasingly on AI tools for a cosmetics business. Two of them are the Cantik.AI Wellness Analyzer and an AI Product Concept Simulator, and we introduced Cantik.ai at Cosmobeauty. Here is what I took away from building them.

A conversation beats a long form

A wellness analysis normally starts with a long questionnaire. People abandon those. The analyzer instead asks a few questions and decides what to ask next based on the previous answers, wired together with n8n workflows. The visitor feels they are talking, not filling in boxes, and the result still has enough information behind it. The lesson: make the interface as short as the situation allows, and let the logic do the branching.

The domain sets the rules, not the model

Skin and wellness are areas where a confident wrong answer is worse than no answer. That changes how the tool is written. It should stay inside what the product knows, say plainly when something is outside its scope, and avoid claiming medical results. The expert in the room, a formulator or a brand specialist, sets those boundaries. My job is to turn them into prompts, data and checks.

AI that supports experts, not AI that replaces them

The concept simulator takes a rough product idea and walks it through guided steps toward a structured concept with a price simulation and a report. It does not decide what to make. It saves the first hours of structuring so the specialists can spend their time on judgement. The most useful AI tools I have built are the ones that make a skilled person faster.

Show it to real people early

Seeing visitors use the analyzer at a trade show taught me more than a week of testing at my desk. People type unexpected things, switch language halfway and put the phone down mid-conversation. A tool that survives that is a product.

What I would repeat

  1. Start from the user's shortest path to a useful answer.
  2. Let the domain expert define what the tool must never say.
  3. Use automation (n8n) for the plumbing and keep the prompts small and testable.
  4. Put it in front of real users as soon as it is safe to do so.
← All posts