‘Data trumps opinion’: How Oh Polly uses AI-powered merchandising

UK-based online fashion brand Oh Polly, known for its dresses and occasion wear, is using AI-powered personalisation to impressive effect, with category page merchandising in particular referred to as “the biggest game changer for our team” by product owner Lauren Muir, who said this tactic drove a 270% increase in clicks and 80% increase in conversion rate.

Muir spoke at Ecommerce Expo last week about her work over the past three years with Nosto – a digital experience platform focused on commerce – and she shared the ways in which the tech has challenged “conventional beliefs” about merchandising across the business, in order to drive sales and time-savings.

AI-powered or personalised merchandising, once called dynamic merchandising, is the sort of ‘set and forget’ technology that allows ecommerce marketers to show each shopper the content that is deemed most relevant to them.

In the case of category merchandising, Nosto defines this as controlling product sorting on category pages using product attributes and performance metrics, and then personalising this approach to segments and individuals based on their preferences.

This gives shoppers “some really relevant and intentional experiences when they land on our website,” said Muir.

“We sell, primarily, dresses, but dresses come in many different shapes, forms, sizes – and people have got their own preferences in terms of what they like.”

A ‘two-pronged approach’ to segmentation: Product affinity and price sensitivity

The social-first brand takes a “two-pronged approach” to serving people what they might want to see on site, said Muir: segmenting by product affinity and price sensitivity.

“In terms of product, someone who likes long-sleeve maxi dresses is not going to want to see a strapless mini dress on their homepage and at the top of their categories. …Our chance of making them convert is so much higher if we just show them what they want, or what we think that they want.”

Our chance of making them convert is so much higher if we just show them what they want, or what we think that they want.

On price, Muir describes a varied customer base, and said that, for those “customers who consistently shop full price and show us that they have got a higher affinity for full-priced products and not shopping discounts… there’s no point us showing them 70%-off sales, because we’re just reducing our profit margin by doing so.”

“So we segment quite significantly our… higher AOV customers… [And] when we do launch sales, they’ll potentially see products which are only 10%, 20%, 30% off, because we know they’re more likely to convert on that.”

Those customers that, “want to shop discounts, want to use discount codes ,and shop with a lower AOV, we will show them products at 50%, 60%, 70% off,” Muir said. “That protects not only our margin but also our brand,” she added.

“When you land on our website as a new customer, you’re not being hit with loads of sale messaging… and people [don’t] think that we’re a sale brand all the time, because we’re really not.”

Category merchandising – “We had to challenge conventional beliefs in our business”

Muir described a cautious approach to making changes across their ecommerce site, with as much A/B testing as possible.

“As a brand, we’re not [like] Boohoo or Asos, with a continual flow of products. We do launches, our conversion rate is quite cyclical and it can change. So, I’m quite scared to make changes and then not really know what’s driven them.”

With this in mind, Muir’s team launched category merchandising on selected category pages to begin with, with a month of A/B testing revealing a 70% increase in clicks using the new sorting technology.

“When we rolled that out to all categories… our clicks actually increased by 270%, our conversion rate increase by 80%,” said Muir.

“There’s products which in a normal conventional merchandising strategy might sit on page 5 of a category, but for specific customers, they should be on page 1 because it’s what that customer really wants to see.”

There’s products which in a normal conventional merchandising strategy might sit on page 5 of a category, but for specific customers, they should be on page 1…

Muir admitted that this departure from manually merchandising some of the most important pages on an ecommerce site can feel like a big change in an industry like fashion.

“We had to challenge conventional beliefs in our business that things had to be merchandised really prettily. We put a lot of money and effort into our visual content and really nice shoots… and our team wanted everything merchandised nicely side by side, or by colour. And we had to challenge that belief and come at them with data. Because at the end of the day, data does Trump opinion.”

Muir also shared detail on how Oh Polly uses Nosto for product recommendations across the online store, with different approaches to different stages of the shopper journey, such as on product pages, on-site search, or cross-selling at the checkout.

The brand has also implemented a solution for stockouts, where visually similar products can be recommended to shoppers.

“We do get a lot of products, potentially go viral and fully sell out, and they’re getting loads of traffic to them, and we can’t convert on those pages. So… rather than just redirecting them somewhere else, let them land on the page, let them sign up for back-in-stock notifications, but also show them other recommendations and other alternatives. And we do the same for pre-order, if you’ve got a product on pre-order, you don’t want to wait six to eight weeks to get it, shop something else that’s really similar.”

Nosto’s Phil Gregory, regional head of customer experience for UK, Ireland and the Nordics, described the next step with Oh Polly to factor in returns rates to merchandising decision making. “We’re working… on how we can bring in the attribute to analyze what the returns rate is on certain products, to make sure that we’ve got a strategy within category merchandising and recommendations that’s taking into account all of that data… that’s coming in.”