Buyer’s guide

AI skin analysis for beauty brands: a buyer’s guide

Thinking of adding AI skin analysis to your store? The questions to ask a vendor about accuracy, safety, privacy, white-labelling and your own catalog.

10 November 2026 · 8 min read · By the Claré AI team

AI skin analysis is easy to demo and hard to get right. Almost every tool can produce a result that looks confident on a good photo. What separates a tool worth putting in front of your customers is what happens when the photo is bad, the lighting is odd or the face is not the one the model expected. This guide gives you the questions to ask before you sign.

Question 1: how does it handle a wrong answer?

Vision models make mistakes, so the real question is whether anything catches them. Ask whether the system describes what it sees before it scores, and whether a separate check removes claims it cannot justify. A tool that never says I cannot tell is a tool that is guessing.

Specific behaviours to test

  • Does it drop findings under flash or glare instead of inventing oiliness?
  • Does it separate raised blemishes from flat dark spots, so one spot is not counted twice?
  • Does it ignore beard stubble and brow shadow instead of scoring them as skin concerns?
  • Is the overall score computed by a fixed, testable function, or taken straight from the model?
  • Does age change the wording only, and is age never guessed from the face?

Ask for a demo with your own team’s faces, including different skin tones and a person with a beard. Anything that only works on a studio photo will disappoint in a shop. Technology like on-device face detection helps with framing and quality checks before anything is sent (MediaPipe is one widely used toolkit for this), but the checks that matter most come after the model answers.

Question 2: where do the recommendations come from?

If the tool recommends products that are not yours, it is a lead generator for someone else. You want a routine built from your own catalog, with allergens treated as a hard wall and budget as a soft preference. Ask how many products it recommends and whether each one has a reason. Claré recommends four, each tied to a concern the scan detected.

Question 3: what happens to the photo?

Your customers will ask you, so you need a clear answer. Find out whether the image is stored, for how long and where. The cleanest answer is that it is never written to disk or a database. Ask whether the report contains the photo, and whether any voice features send audio to a server. See how we handle this on our privacy by design page.

Question 4: can it look like yours?

Customers should feel they are still in your store. Check whether you can set your own colour, welcome message and disclaimer, and whether the report carries your brand name. Ask whether it runs on your website, on a shop tablet or both, and whether you can create a scan key per outlet so you can see where scans happen.

Question 5: how are you charged?

Ask exactly when a credit is spent. The fairest model charges only after a successful analysis, so a blurry photo, a rejected photo or a failed scan costs nothing. Ask what happens when your balance runs out, and what the shopper sees if the service is unavailable.

Question 6: what can it not do yet?

Every honest vendor has a list. Ours includes scoring weights fitted on a small set of scans, left and right localisation that is close to chance, and catalog import that is not built yet. You can read it on the technology page. A vendor with no limits to share has not looked hard enough.

Question 7: does it stay cosmetic?

A skin tool for shoppers should not diagnose anything. Check that the product stays on the cosmetic side of the line in its prompts, its output and its reports. It protects your customers and your brand. We explain our approach in cosmetic, not medical.

A short scorecard

Give each vendor a mark from zero to two on the seven questions above. Anything below ten out of fourteen deserves a second look. More importantly, watch how they respond when you push on the hard questions. Their answers tell you more than the demo does.

Red flags worth walking away from

A few signals should slow you down, however good the demo looks. If a vendor refuses to describe how photos are handled, or answers with a long policy page instead of a plain sentence, that is a flag. If results always look impressive on every face you try, be suspicious: real skin is messy and a tool that is never unsure is hiding its uncertainty.

  • Medical language in results, such as condition names or treatment suggestions.
  • Recommendations that include products you do not sell.
  • No clear answer on what is free when a scan fails.
  • No list of known limits, or a claim that there are none.

When you are ready to see how Claré answers all seven, get early access for your brand or look at the sample report first.

Frequently asked questions

How much does AI skin analysis cost?

It depends on the vendor and plan. With Claré, a credit is spent only after a successful analysis, so failed scans are free. Talk to us about plans.

Does AI skin analysis work for every skin tone?

Test any tool on a range of skin tones and with beards before you buy. We found and fixed one fairness problem, stubble scored as blemishes, and we publish our remaining limits.

Cosmetic analysis only. Not medical advice.

Ready when you are

See Claré on your catalog.

Book a demo and we will walk through a scan, the checks and a routine built from a sample catalog, then talk about how it fits your brand.