How to Choose Skin Analysis Technology
Start with the job, not the vendor
Before you shortlist anyone, write down the job in one sentence. Do you want to personalise across every beauty category, get the deepest skin science, ship fast and cheap, or turn scans into sales? The answer points you at a category of tool and saves months.
Then work through five decisions in order. Skip none of them.
| What to evaluate | Oyster | Other tools (what to check) |
|---|---|---|
| Accuracy across skin tones | Built and measured across Fitzpatrick I–VI, accurate on every tone with proven performance on deeper skin | Confirm published accuracy on Fitzpatrick V–VI, not just an overall number |
| Markets & languages | Built in Africa, live in 29 countries across Africa, Europe, the Middle East, Asia and the Americas | Check coverage and language support for your markets |
| Deployment | In-store kiosk, web embed, app widget, plus API and SDK | Check which surfaces are supported and the integration effort |
| Time to launch | Live in under two hours with the widget or API | Ask for a realistic go-live timeline |
| Commerce built in | Scan → recommendation → cart → Oyster Pay in one flow | Check whether recommendations connect to checkout or stop at analysis |
| Data & privacy | ISO 27001 certified; no individual identity exposed in analytics | Confirm security certification and data residency |
| Pricing model | Usage-based, priced for the markets it serves | Compare per-scan or licence cost at your volume |
Decision one, accuracy across your customers
A tool that reads some skin well and other skin poorly is a liability. Define your real customer skin tone mix, then insist on accuracy figures by Fitzpatrick type. If you serve deeper tones, this decision outranks every feature. Oyster is measured across the full Fitzpatrick range and weighted toward deeper skin, and Haut.AI states full Fitzpatrick coverage, so ask both, and everyone, to prove it on your customers.
Decision two, what happens after the scan
A score alone rarely pays for itself. Decide how much of the path from analysis to purchase you want the vendor to own.
- Analysis only. You build matching and commerce yourself.
- Analysis plus recommendation. The tool suggests products, you handle checkout.
- Full commerce. The scan matches to your catalogue and carries the shopper to purchase, then stores the result. Oyster is built for this. See the scan.
Decision three, deployment and channels
The best engine is worthless where your customers are not. Confirm the tool runs on the surfaces you need, whether web, mobile app, in store counter, smart mirror or messaging, and whether one profile follows the customer across them. If messaging first commerce matters in your market, check for WhatsApp and agent support.
Decision four, data and compliance
A face scan is sensitive information. Confirm where images are processed and stored, whether anonymisation is offered, and how the vendor supports regional law such as POPIA and Nigeria data protection rules. Get this in writing. See our guide on data privacy in skin analysis.
Decision five, run a pilot that proves value
Do not buy on a demo. Run a short pilot that produces numbers.
- Pick two vendors and one clear metric set: scan rate, matched add to cart, basket size, returns and repeat visits.
- Use a control group so you can attribute the lift.
- Include a skin tone check on a labelled sample of your own customers.
- Run four to six weeks, then let the data choose.
To include Oyster in your pilot, book a demo and see pricing.
Frequently asked
Start by writing the job in one sentence, then work through five decisions: accuracy across your customer skin tones, what happens after the scan, deployment and channels, data and compliance, and a pilot that proves value. Do not buy on a demo. Run a four to six week pilot with a control group and let scan rate, conversion, returns and repeat visits decide.
Accuracy across your real customer skin tones. A tool that reads some skin well and other skin poorly will misrecommend and erode trust. Define your customer skin tone mix, insist on accuracy figures by Fitzpatrick type, and verify them on a labelled sample of your own customers before anything else.
Yes. Run a short pilot with two vendors and a control group, tracking scan rate, matched add to cart, basket size, returns and repeat visits, plus an accuracy check across skin tones. Four to six weeks on your own catalogue produces the evidence a demo cannot.
Ask for accuracy figures by Fitzpatrick type, how the scan maps to products you sell, which channels it deploys on and whether one profile follows the customer, where data is processed and stored and how they meet regional law, and whether you can run a blind pilot on your own customers first.
See what skin intelligence does for your business.
Oyster reads skin accurately on every tone and turns it into the right recommendation.