Montreal clothier tries AI suggestions to increase online sales


At first glance, Montreal-based Frank + Oak looks like just about any other successful clothier focused on the tastes of Millennials and wannabes. With a dozen stores in hip neighborhoods of Canadian cities and three more in the United States (Boston, Chicago and Washington, D.C.), they expressly focus on the experience of shopping as much as on their products.

Co-founders Ethan Song and Hicham Ratnani created the company in 2012 after experimenting with other retail concepts. In their early days they modeled and took their own pictures for Frank + Oak clothing. Today they’ve grown to 250 employees and more than 3 million customers. With a strong emphasis on personal service, Frank + Oak doesn’t just sell their products to anyone. Online customers need to sign up as “members” to buy, which is intended to create the sense of a unique club.

Their men’s and women’s line has a big emphasis on “lifestyle,” and tries to capture the work clothing tastes of young creative types. Some of their larger stores feature barber shops and coffee bars. They’ve also been known to host events such as whiskey tastings and art shows. The effect is to broaden the definition of a retail experience in the same way that Disney Stores and Bass Pro Shops do for their market segments.

Personalization is a company mantra. Each store has stylists available to assist in clothing selections, and staffers include a handwritten thank-you notes before online orders are shipped (about 50 percent of the company’s sales are through its website).

It’s only natural, then, that when Song was presented with an idea to increase personalization for its online customers, he said yes.

Creating instinct

“From our research we know that customers often may not know what to buy, maybe they don’t shop for clothes that often,” Song says. “But they want to look good, and they appreciate good recommendations.”

Song’s interest dovetailed with the work of Eric Brassard, who formed Propulse Analytics in 2015. Brassard, a former marketing executive with Saks Fifth Avenue, saw the impact of personalized service when he started in retail.

“In the early 1990s I saw that some Saks salespeople were extremely efficient, earning nearly $200,000 per year in commissions,” Brassard says. “I spent some time watching how they worked, and noticed they had an ability to size up clients very quickly.”

As a customer looked at a few items in a department, the salesperson would gather some pieces from various racks and bring them over. Invariably, the customer would almost always choose something the salesperson brought to them.

“It was instinctual. Through experience the salesperson knew what to show the customer by what they appeared to be looking for,” Brassard says. “Over time that’s always stuck with me when it comes to online retail: How do you create that same experience?”

“Everyone talks about AI, but you get the sense in the industry that people are waiting to see how practical it is. In our case, we can say it’s working.”
— Ethan Song, Frank + Oak

Online product recommendations have been around for nearly as long as online retail, usually built on the same principles: taking data points of what people look at on a site and making product suggestions based on what others have bought. The data is based on what typical customers buy, combined with a product’s color, fabric, style and price point.

Brassard saw an opportunity to bring artificial intelligence into the mix.

“Big data looks at previous patterns in a large group of similar customers,” he says.

“Say you’ve bought from my site a handful of times in the past year. That’s still not enough data to tell me where the next dot will plot itself for you. So big data analytics typically averages in similar customer patterns to guess what you might like. You typically see this in the line, ‘Customers who bought this often purchased that …’ But maybe you don’t like what other people bought.”


Using the concepts of AI and machine learning, where algorithms are basically making rudimentary decisions based on data as it flows in, Brassard sought to create a system for Frank + Oak that started not with customer profiles, but with products themselves using image recognition software.

The retailer’s emphasis on personal service matched well with Propulse’s goals.

“For Frank + Oak, we dug into the pictures of all their products and looked at tens of thousands of aspects of each picture to try and determine what might be of interest to a shopper,” Brassard says.

“So it’s not based on a customer’s history. Usually a stylist in real life doesn’t know what the customer has bought before. We’re trying to mimic the stylist’s instincts for what a customer would like.”

The system is invisible and unobtrusive. Click on a woman’s trench coat and Frank + Oak will begin showing similar coats as well as complementary pants and tops. This differs from the company’s previous experiments with product recommendations.


“In the basic algorithms, if you looked at white shirts, you got more white shirts. With machine learning you can go deeper and get more textured recommendations,” says Brassard. “It picks up style and feel — it’s much more organic.”

The system interacts with supply chain software so it knows not to recommend out-of-stock items; when this happens it notifies the retailer that a possibly recommended product wasn’t shown because of an inventory problem. Once it learns the shopper’s size, it also won’t show unavailable sizes.

“Like anything else, it’s only as good as the information you give to make it work,” Song says. “Everyone talks about AI, but you get the sense in the industry that people are waiting to see how practical it is. In our case, we can say it’s working.”

Recommended choices

Like most cloud-based applications, implementation was fairly easy. After product pictures were uploaded and read to “teach” the system, a few lines of code were added into Frank + Oak’s online shopping program. The retailer began working with Propulse last fall and started to see a difference in customer shopping carts.

“The choices our customers make are often based on these recommendations,” Song says. “We’re seeing them select complementary items, because, I believe, there’s a more personal feel to the recommendation. It’s as if a stylist is online with you making these suggestions. Our conversion rates have risen, but more importantly this works well with our branding of creating a custom shopping experience.”

For the future, Brassard looks to AI technology improving to the point where older, big data analytics looks clunky. “Deep learning will continue to be refined to the point where it really will be like talking to a stylist. We’ll see more sophistication in the products chosen, and who knows, maybe AI will begin setting fashion trends for us.”

John Morell is a Los Angeles-based writer who has covered retail and business topics for a number of publications around the world.


Comments are closed.