Complete the look is a way of recommending products that shows a shopper the other pieces that turn one item into a whole outfit. Instead of offering alternatives to the thing they are looking at, it offers what goes with it, such as the trousers, shirt and jacket to wear with a new pair of shoes. A fashion brand can show it on the product page, in the basket or after the order, built around the piece the customer is looking at or has just bought.
What does complete the look mean?
Complete the look means recommending the pieces that go with an item, so that a single product becomes a finished outfit.
In recommendation research, the term names a specific task. In a paper presented at CVPR 2019, Wang-Cheng Kang and colleagues proposed it: given a photo of a real scene, such as a selfie, recommend products that are visually compatible with it. Photos like that are cluttered and hard to work with, the authors note, but they can carry context that a product shot does not, such as the wearer's body type or the season.
For a brand, the starting point is an item in its own catalogue rather than a photo, but the job is the same: answer the question every new piece raises, which is what to wear it with.
What is the difference between shop the look and complete the look?
Shop the look starts from a finished outfit and helps the shopper buy the pieces in it, or pieces like them, while complete the look starts from one piece, or part of an outfit, and adds the pieces that are missing.
In a KDD 2020 paper, Raymond Shiau and colleagues describe shop the look as visual search that lets people find and buy the products within an image. In their account it works in two steps: detect the items in a styled photo, then, when the shopper picks one, retrieve products that look like it.
Kang and colleagues draw the line between the two themselves: shop the look retrieves products that are similar to, or even the same as, something in a photo, while complete the look learns what complements the scene, and the products it should recommend are typically not in the picture.
So one is about similarity and the other about complementarity. They also fit together: once a look is complete, the shopper can shop it whole.
How is complete the look different from "you may also like"?
A "you may also like" rail of similar items offers substitutes, products a shopper could buy instead of the one in front of them, while complete the look offers complements, products bought to go with it.
In a KDD 2015 paper on how products relate to each other, Julian McAuley, Rahul Pandey and Jure Leskovec define substitutes as products that can be bought instead of each other, and complements as products that can be bought in addition to each other. Their example is about timing: while someone is choosing a phone, other phones are sensible suggestions, but once they have bought one, a case or a charger makes more sense.
Clothes follow the same pattern. While a shopper is deciding between tan brogues and black ones, showing more shoes helps them choose. Once the brogues are bought, the useful suggestion changes: the chinos, the shirt and the jacket that make an outfit with them.
Both kinds have their place. Substitutes help a shopper choose, and complements help them wear what they chose.
Where can a fashion brand show complete the look?
A fashion brand can show complete the look in three places: on the product page, in the basket and after the order.
- On the product page, while the shopper is still deciding, it shows the item as part of an outfit, so they can judge it in context rather than on its own.
- In the basket, it can offer the pieces that finish what the shopper has already chosen, before they check out.
- After the order, it answers the question that arrives with the parcel. This is where McAuley's timing point matters most: the piece is already bought, so the useful suggestions are complements, not alternatives.
Atomic Goats' post-purchase outfit campaigns are built for that last moment.
What makes a good complete-the-look recommendation?
A good complete-the-look recommendation is a whole outfit that works together, made only from pieces the shopper can buy, shown on a body, with anything the customer already owns left out of the basket.
The whole outfit has to work. At ACM Multimedia 2017, Xintong Han and colleagues studied outfit compatibility: suggesting an item that matches the pieces already in a set, and predicting how compatible a whole outfit is. They argue that a compatible outfit has two properties. Its pieces should be visually compatible and similar in style, and together they should make a complete set with nothing missing and nothing doubled: an outfit with no shoes fails, and so does one with two pairs. They also model the outfit as a whole rather than one pair of items at a time. The practical lesson is that every piece has to work with every other, because a shirt that suits the shoes can still clash with the trousers.
Every piece can be bought. A look is only useful if the shopper can buy all of it, so build it from what is for sale today.
Shown together, on a body. A row of separate product shots leaves the shopper to imagine the outfit. The same pieces on a person show the look itself, and the more that person resembles the shopper, the more the picture tells them about whether the look suits them. That is a picture of how the outfit looks, not of how it fits.
What they own stays out of the basket. When the look is built around something the customer has already bought, that piece belongs in every picture and nowhere in the basket.
How does Atomic Goats complete the look?
Atomic Goats completes the look for every customer: around the piece each customer bought, its AI stylist builds complete outfits from the brand's own live catalogue, ready to try on and buy in one go.
After an order, the looks are built once for each purchased item and sent through the brand's existing email platform to everyone who bought it. The email opens a personal lookbook, where the customer can try a look on, restyle it with the AI stylist or add the whole outfit to the basket at once. The piece they already own is left out by default, and they choose their sizes in the basket.
Before a first order, the same stylist runs on the brand's storefront. Any visitor can build a complete outfit from the live catalogue, swap pieces until it suits them and try the whole look on, and what they build shapes the first campaign they receive.
The try-on shows the whole look on the model closest to the shopper's build, on their own full-length photo, or on a body built from a selfie. It shows how the outfit looks together, not how each piece fits: Atomic Goats does not give size or fit advice.
The brand stays in charge. The stylist can run fully automatically, or the brand's merchandisers can set the brief and approve each look before any customer sees it. On-model images come from the product shots the brand already has, each one checked automatically, and because the catalogue syncs, every look is built from what is for sale. Colour analysis is optional and off by default. Atomic Goats runs on Shopify storefronts through its own app.
If you would like to see complete looks built on your own range, book a demo.