Syda Productions - Fotolia
Fashion is a very difficult place to be for the online retailer, as customers want to try on clothes and accessories to see how they look before they make the purchase.
Amazon has reportedly looked into launching its own fashion label, but Jim Downing, chief technology officer of Metail, claims pureplay retailers such as Amazon need to define themselves through more than price and service to push their way into the clothing space.
“You can’t engage fully with clothes online and you never will be able to, but there’s a lot of mileage in being able to get closer than just a photo and some text,” says Downing.
Digital fashion plugin Metail creates computer-generated models of customers to help them try on outfits virtually through retailer websites, without having to visit the store.
This extra touchpoint helps customers to interact with garments in an online space, an area of engagement Downing says Amazon aren’t set up to address yet.
Retailers have been experimenting with technologies to try to encourage customers to continue shopping with their brand regardless of channel, including responsive websites, applications for loyalty points and smart technologies in stores.
Trying to sell fashion can be a pain point for retailers as customers often don’t know their size in a particular store, or pictures of the garments do not properly represent the cut or material.
The adoption of imaging technology
Using imaging technology, Metail creates digital models of how garments look on life-sized mannequins by taking eight photos of the mannequins for different angles.
“That gives us one reference shape of how the garment would be on that one reference shape, then at runtime we model the user’s shape and have a physics simulation that we’ve fitted to the garment and digitised in order to model how the garment’s going to change,” says Downing.
Users provide their height, weight, bra size, waist and hip measurements, which produces a 3D model so the consumer can see what the clothes will look like on their body type.
Hundreds of garments a day are modelled by Metail all around the world, and currently the digitisation process costs less than £10 per outfit.
But Downing says the company is trying to drive the price down to £3 per garment, and in the future the firm will be looking into increasing the use of machine learning and automation for both the modelling process and to make the models more realistic so that customers have a better idea of how the clothing will look when they move, walk or stand.
Fashion in the cloud
Due to the number of garments processed, digitally modelling clothing for retailers produces a lot of data points.
Metail was originally hosting its data at a small internet service provider in Cambridge, using Mac Minis in its datacentres, but when the firm started to expand globally, this solution did not scale.
Metail migrated its data to the Amazon Web Services (AWS) platform to allow global expansion of the business and make it easier to work on research and development. The rest of the data is stored in a hybrid virtual cloud.
Predicting garment models for the plugin is a semi-automatic process that creates a large amount of data after the initial model is made.
In future, if the firm looks to move away from AWS, the migration would be much more complex than the original “lift and shift” approach, due to the large amount of data the firm has, and the fact that the data relates to various global locations.
“The choreography of data around the world to eventually get a picture of a person in clothes is considerable, even at our scale,” says Downing.
Fashion models in the future
Metail has previously used 3D scanners to create custom models for celebrities, but this is not available for everyone.
Currently, users can choose their model’s ethnicity and customise its shape. Metail’s prediction and machine learning model then generates how a garment will look on the custom body shape – but is this enough to encourage customers to buy the outfit?
As part of its research and development, Metail is looking to add more flexibility to its body models so customers can choose different poses and see how the clothes will react to movement or posture.
“The further we go away from that symmetrical pose, the harder it is to keep on modelling the physics and keep it realistic,” says Downing.
The firm is able to process a high number of outfits daily because it limits the number of photos taken of the model to eight per garment.
Deep learning could be used to further enhance the service, predicting the fit, style and pattern of a garment, whether it would be loose or tight in particular places, and using the website image of the garment to predict the type of material used. This, however, can be quite difficult, as retailers currently discard information about a clothing line after it is produced.
“Information has disappeared as soon as it’s not needed in the immediate next step,” says Downing. “The common idea is that we should build our technology from cutting patterns. It’s just that most retailers don’t have the cutting patterns – the cutting patterns disappear from the supply chain as soon as they’re not needed.”
“We’re starting to see that change slightly with the adoption of product life cycle management tools – keeping information alive for longer so we can use it.”
In future, a 3D scanner could be used for each garment, then using standardised engines to build the digital model so Metail would not need to build its own.
Metail has already started looking into this, but Downing admits it’s “a chunky bit of research and development”.
Read more about retail technology
When the firm was founded, it worked with a third party to allow customers to take a selfie and use this as the face of their model, but the functionality was scrapped as the web-based nature of the plugin meant taking and uploading a photo, which drew consumers away from the shopping experience.
Now customers are increasingly using smartphones and the Metail service can be used through the brand’s mobile web version – the firm is looking to redevelop this functionality to enhance the service.
Customer service is one of the most important aspects of the retail space, and Downing emphasises that usability has been a key focus for Metail since day one.
But this focus on usability has slowed the firm’s research and development into rolling out application programming interfaces (APIs) so that other businesses can use the Metail product.
“For the future, we’re looking at publishing APIs for the back end so people can build their own experiences,” says Downing.
For now, consumers can use the service through retailer websites, and Metail will soon be rolling out the plugin responsively so it can be used as part of the retailer’s responsive mobile sites, replacing its desktop version.
The online fashion customer base
The online purchase of clothing and footwear has been on the rise over the years, but consumers are fickle, and if they do not get the service they are expecting, they will shop elsewhere.
Customers who use Metail online when looking for clothes and accessories fall into two categories: those looking for decision support and those who are using the models for fun.
The slightly older generation are most commonly using Metail’s services to help make purchasing decisions, but millennials are more experimental with their retail habits.
According to Downing, millennials are a “sweet spot” that omni-channel retailers have failed to cater to when using technology to bridge the online-offline gap.
Jim Downing, Metail
Retailers still think of mobile as a separate experience to in-store, but millennials are using their smartphones all the time, often when they are in physical stores to compare products and price-points.
“We’ve been concentrating on nailing the online experience before we get into in-store factors,” says Downing
Regardless of the channel used to interact with customers, the user experience should be the main focus for a retailer to make sure their consumers aren’t tempted to shop elsewhere.
“You can look to Amazon’s success to see how far you can go with better service and better product information without anything too flashy,” he says.
The Metail service originally offered video game-style sliders to adjust the models, asking for the consumer’s measurements.
This is different depending on region. Downing uses India as an example, where customers are asked for their overbust measurement rather than bra size, as they are not sure of their European or US bra size.
Once the firm has further developed its machine learning algorithms, it hopes there will be no need for consumers to follow any instructions in the future.
“It’s not that people are bad at following instructions, it’s that they’re bad at reading them at all,” says Downing.