Product Bundle Recommendations by Beauty Profile: Building Kits Around One Selfie

Most beauty personalization stops at one SKU: a foundation shade, a shampoo, a serum. But the basket that actually moves revenue is usually a kit — a cleanser plus a moisturizer, a lipstick plus a liner, a shampoo plus a mask. Product bundle recommendations by beauty profile are the next logical step once you already have structured attributes about the shopper: instead of ranking items one by one, you assemble a set that makes sense together, for that person, from your own catalog.

This article is about the mechanics of that: what the profile gives you, how to compose a kit without creating contradictions, and what to watch operationally before you put it on a product page.

Why bundles break where single suggestions work

A single recommendation only has to be defensible on its own. A bundle has to be defensible three ways at once.

First, internal coherence. If the profile says oily skin in the T-zone and the kit pairs a mattifying cleanser with a rich occlusive cream, the shopper notices the contradiction before you do.

Second, shade consistency. A lip kit with a liner one family away from the lipstick reads as a mistake, not as personalization.

Third, non-redundancy. Two cleansers in the same kit is not a bundle, it is a listing error. Most recommendation engines rank items independently and happily return three near-duplicates at the top.

So bundling is less a scoring problem than a composition problem: pick slots first, then fill each slot with the best-ranked eligible item.

What the beauty profile gives you to compose a kit

MaIA Scan returns a beauty profile from a selfie as JSON: skin tone and undertone, skin type and visible condition, hair attributes, and a confidence value per field. These are visible attributes of a photo — not a diagnosis, and not a substitute for a professional opinion. For bundling, each field plays a different role:

• Skin tone and undertone drive colour slots — foundation, concealer, bronzer, blush, lip shades. They are matching fields: the kit either lands in the right family or it does not.
• Skin type and condition drive formula slots — cleanser, moisturizer, treatment textures. They are filtering fields: they remove products rather than rank them.
• Hair attributes drive routine slots — shampoo, conditioner, mask, styling. They tend to define the number of steps in a kit, not just the products.
• Confidence per field decides how assertive the bundle can be. High confidence on undertone lets you commit to a shade. Low confidence means you widen the slot, show two options, or drop the colour item from the kit entirely.

Product bundle recommendations by beauty profile: a composition method

1. Start from a kit template, not from the catalog

Define templates per intent: "complexion starter", "lip look", "wash day", "evening routine". Each template is an ordered list of slots with a role, not a product. This keeps bundles readable and makes the output predictable for merchandising.

2. Choose the anchor

One slot anchors the kit — usually the item the shopper came for, or the one with the strongest profile signal. The anchor is matched first, and everything else is matched relative to it. If the anchor is a foundation, the concealer and powder follow its tone family. If the anchor is a mask, the shampoo follows its hair concern.

3. Fill remaining slots with compatibility rules

For each remaining slot, filter the catalog by the profile fields relevant to that slot, then apply pairwise rules against the anchor: same tone family, compatible finish, no duplicated function, no conflicting claim. Rules should live in your catalog attributes, not hard-coded in the recommendation layer — that is what makes them editable by merchandisers.

4. Apply a price ladder

Offer the same composition at two or three price points — a two-step version and a full version — instead of one expensive kit. The profile stays the same; only the depth changes.

5. Explain each slot in one line

"Shampoo for wavy, dry-feeling hair" beats "recommended for you". The explanation is also your quality control: if you cannot write the line from the profile fields, the slot is probably guesswork.

Where bundles show up

Product page. Anchor is the product being viewed; the kit is the complete-the-look block. The shopper can preview colour items first with MaIA Try-On before adding the set to the cart.

Conversation. On WhatsApp or chat, a kit is naturally presented step by step — one slot per message, with the option to swap an item. This is where MaIA Beauty Advisor works, reading the beauty profile and recommending products from your own catalog rather than generic advice.

CRM and replenishment. A stored profile lets you rebuild the kit later with what is in stock, or send the refill of the fastest-consumed item in the set.

Subscription and gifting. A profile-composed box is still a bundle — the same slot logic, assembled by operations instead of the shopper.

Operational guardrails

• Stock first. Never compose a kit from items that cannot ship together. Filter by availability before ranking.
• Margin floor. Bundles usually carry a discount; set a minimum blended margin per template.
• Fallbacks. If a slot has no eligible SKU, shrink the kit rather than substituting something off-profile.
• Editability. Merchandisers should be able to pin or ban a SKU in a template without a deploy.
• Measurement. Compare bundle attach rate, average order value and return rate against your existing complete-the-look block, on your own traffic. Any benchmark you read elsewhere belongs to someone else's catalog and channel.

Start with one template, one anchor category and one channel. A single well-composed kit teaches you more about your catalog attributes than a full rollout.

Ready to build it? Create an account and get an API key to start from a beauty profile, or talk to the B4A team about a catalog-integrated advisor.