What a Virtual Beauty Advisor Needs From Your Product Catalog Before Launch

Most virtual beauty advisor projects do not stall on the AI. They stall on the catalog. The model can read a selfie and describe skin tone, undertone, skin type and hair attributes in seconds, but a virtual beauty advisor product catalog that lacks shade data, consistent naming or stock status will still produce recommendations your team is embarrassed to show a customer. If you are evaluating an AI advisor for WhatsApp, product pages or chat, the preparation work is mostly on your side of the integration — and it is work that pays off well beyond this one project.

This article is a practical checklist, written for the e-commerce lead, CRM manager or product owner who will own the rollout.

Why the catalog is the real dependency

MaIA Beauty Advisor reads the beauty profile produced by the analysis — the same structured output described on the MaIA Scan API page — and then recommends your products. It does not recommend in the abstract. Every suggestion has to resolve to a SKU that exists, is purchasable, and whose attributes justify the match to a customer who asks "why this one?".

That means the advisor inherits every gap in your product data. If two foundation lines use different shade vocabularies, the advisor has to guess. If undertone is missing from half your shades, the match becomes a tone-only match. If a discontinued SKU is still live in the feed, the advisor will happily recommend it.

The good news: you do not need a perfect catalog. You need a well-defined subset, enriched properly, and a plan to extend it.

The attribute checklist by category

Complexion products

For foundation, concealer, powder, BB creams and tinted moisturisers, the advisor needs at minimum:

• A stable shade identifier per SKU, not just a marketing name. "Beige 02" and "02 Beige" should not be two different things.
• Depth level on a consistent scale across all lines you sell. Whether you use light/medium/deep or a numbered ladder matters less than using the same one everywhere.
• Undertone per shade: warm, neutral, cool, olive — whichever vocabulary you standardise on.
• Coverage and finish, because two shades can match equally well and the customer is actually choosing between matte and luminous.
• A hex value or swatch reference per shade, which is also what a try-on palette needs.

Colour cosmetics

Lipstick, blush, eyeshadow and the rest need the shade identifier, the family (nude, red, berry, coral), the finish, and a colour reference. If you also plan to offer try-on, this is the same dataset: palettes in MaIA Try-On are calibrated to a brand's real products inside a project, so enriching shades once serves both the advisor and the photo-based makeup try-on.

Skincare and hair

Here the useful attributes are different: the visible condition or concern the product addresses, the format (serum, cream, oil, mask), the key ingredient or actives in plain language, and the routine step. For hair, add curl pattern suitability, porosity language if you use it, and whether the product is intended for coloured or chemically treated hair. Keep the wording descriptive — the advisor talks about visible attributes and product benefits, never about diagnosing or treating a condition.

Beyond attributes: the operational data

Attributes make the match. Operational data makes the recommendation usable.

Availability and price

The advisor should know what is in stock, in which market, and at what price. A perfect shade match that is out of stock is a worse experience than a second-best match that ships tomorrow. Decide early whether the advisor falls back to the nearest available shade or says the item is unavailable — that is a business rule, not a technical one.

Deep links

Every recommended SKU needs a URL that lands on the product page with the right variant pre-selected, plus whatever UTM or channel parameters your analytics need. On WhatsApp this is the whole conversion path.

Copy the advisor can speak

Product descriptions written for SEO rarely read well in a chat bubble. A short, spoken-style benefit line per SKU — one or two sentences — gives the advisor something to say that sounds like your brand rather than a datasheet. This is usually the most underrated item on the list and the one that most changes how the conversation feels.

Exclusions and guardrails

List the SKUs the advisor must never recommend: clearance, professional-only, gift sets, market-restricted items. Also define your tone rules — what the advisor should say when a customer asks something that belongs with a dermatologist or a physician, and how it should decline.

How to scope a first release

Start with one category and one channel. A complexion advisor on WhatsApp, or a lipstick advisor on product pages, is enough to validate the flow end to end: photo in, beauty profile out, catalog match, deep link, measurable click.

Pick the category where your shade data is already strongest, not the one with the most SKUs. Then measure the things you control — conversation completion, click-through to product pages, add-to-cart from recommended SKUs, return rate on recommended shades versus self-selected ones — and extend from there. Results vary with each brand's catalog, traffic and channel, so your own baseline is the only benchmark worth trusting.

MaIA Beauty Advisor is delivered as a project with the B4A team rather than as a self-serve API, precisely because catalog mapping is where the quality comes from. You can see how the pieces fit on the MaIA Beauty Advisor page, and if you want to test the analysis layer first, the scan and photo-based try-on APIs are self-serve: create an account, buy credits and get an API key in the API documentation.

Bring your catalog export and let's map it together — or start with an API key today and see what a beauty profile looks like against your own shade range.