Skin Analysis API for Skincare: From Selfie to Care Routine

Skincare brands sell routines, not single items. The problem is that the customer arrives at the online store without knowing where to start: which cleanser, which moisturiser, which texture. A skin analysis API for skincare solves the first step of that reasoning — it reads a selfie and returns, as JSON, visible skin attributes that your recommendation rule can use to assemble a sequence of products from your own catalog, instead of pushing the best-seller to everyone.

What the analysis returns (and what it does not do)

MaIA Scan works as an API: your system sends the photo and receives a structured beauty profile. The fields include skin tone and undertone, skin type and condition, hair attributes and a confidence level for each field. The model was trained on Brazilian skin, which matters for a customer base with the variety of tones and undertones that Brazil has.

It is essential to be precise about what this means. The analysis describes visible attributes in a photo. It is not a clinical assessment, it does not treat anything, it does not replace a dermatologist and it should not be presented to the customer as a clinical evaluation. The language of your interface needs to reflect that: "what we see in your photo", not "your skin verdict". That copy choice protects the brand and, in practice, creates less friction with the customer, who understands the reading as a starting point rather than a ruling.

The confidence field is part of the rule, not decoration

Every field comes with an associated confidence. Use it. If skin type came back with low confidence — dark photo, heavy makeup, bad angle — your rule can ask for a new capture, fall back to a broader recommendation or ask the customer directly. The classic mistake is treating the JSON as uniform truth and building the whole routine on a fragile field.

Skin analysis API for skincare: from JSON to routine

The API delivers attributes. What turns an attribute into a routine is your business rule, and it depends on an organised catalog before the first line of code.

Map the catalog by step and by attribute

Build a spreadsheet with your skincare SKUs and classify each one along two dimensions:

• Routine step: cleansing, toner/essence, moisturising, sun protection, targeted treatment, eye-area care.
• Skin attribute addressed: what the product addresses according to your own marketing claim and product specification — a lighter texture for oily skin, a richer formulation for dry skin, a tinted sunscreen compatible with certain tones and undertones.

Without that map, no API will solve it. With it, the logic becomes simple: the profile returns condition and type, the rule filters the eligible SKUs for each step, and the routine comes out assembled with three to five items numbered in order of use.

Decide the depth of the routine

A ten-step routine has a high abandonment rate. For most brands, the best result comes from a short, complete routine — cleanse, moisturise, protect — with an optional treatment upsell. You can always offer "see full routine" for those who want more.

Write the justification alongside the product

Each recommended item needs a sentence connecting the attribute read to the choice: "gel texture, suited to skin with visible shine in the T-zone". Without a justification, the recommendation looks like a raffle. With one, it looks like consultation — and supports the higher ticket of a kit.

Where to place it in the journey

Site entry point or routine-finder landing page

A dedicated page, with a clear explanation of the capture and of how the data is used. It works well as a destination for media campaigns, because it gives the visitor something before asking for the purchase.

Product page

Here the question changes: it is not "which routine" but "is this product right for me?". The same call to the selfie skin and hair analysis returns the profile and your page responds with a compatibility badge and, if there is none, an alternative from the same catalog.

Post-purchase and conversation

The profile stored in the CRM feeds the next communication: a refill at the right moment, the complement for the missing step, a change of texture when the season changes. For brands that want this in conversational form — WhatsApp, chat, product page —, the path is the virtual beauty advisor integrated with the catalog, which reads the profile and recommends the brand's own products. That product is delivered as a project with the B4A team, not as a self-serve API.

What to measure

Define the metrics before go-live, because afterwards nobody goes back to instrument:

• Analysis completion rate: how many start the capture and reach the profile. A drop here is a photo-instruction problem, not a model problem.
• Add-to-cart for the full routine versus a single item.
• Average order value with and without analysis, broken down by channel.
• Repurchase at 60 and 90 days among those who received a recommended routine.
• Distribution of the attributes read: it shows whether the catalog covers the base well. If a condition appears frequently and you have a single SKU for it, that is portfolio information, not technology information.

Do not import third-party numbers to promise internal results. Every brand has a different catalog, traffic and channel, and the gain only shows up in your own measurement.

How to start without a long project

MaIA Scan is self-serve: you create an account, buy credits and receive the API key. A developer can make the first request the same day and validate the fields with real photos from your base before designing the interface. Start with the catalog mapping, test the rule with a small group and only then take it to the home page.

Create your account and generate your API key in the MaIA Scan documentation — or talk to our team if you want to design the recommended routine together with us.