Online Retail Company Increased Sales with the Perfect Fit

Case Study

Global Fashion and Cosmetic Retailer

Revenue
$3+ Billion

Company Description
As a global retailer, the company handles over 850 brands in 30 different sizes. It serves over 20 million customers worldwide, providing flexible return policies.

“In an online store, providing the right sizing information is critical because a customer receiving the wrong size results in returns, which are costly to our company. By automating our model building efforts with Ople, we were able to very quickly iterate while adding lots more data, resulting in a more accurate model than we were able to develop by hand.”

Key Achievements

  • An iterative process involving more data drove a predictive sizing model with better than 85% accuracy
  • The Ople-generated models were more accurate than the model the company had developed by hand (85% accurate vs. 70% accurate)
  • With 85% accuracy, the company reduced sizing errors, minimized returns, and increased customer satisfaction
  • The Ople Platform was intuitive enough that the models were developed by the customer satisfaction team and a business analyst
  • As the company gathers more data and feedback from customers, they believe they will improve the accuracy of the model even further
manu_bars

When the Shoe Doesn’t Fit

In e-commerce, one of the biggest customer satisfaction issues when purchasing clothing and accessories is sizing. Manufacturers have different ideas about the sizes and fits of shirts, pants, shoes, and other articles. When customers order a certain size, they expect it to fit. And if it doesn’t, they often return the item to the online store. 

 

When returns happen, the retailer loses money. And when customers receive clothing that doesn’t fit, the return process – the return authorization, packing and shipping, and waiting for a credit on their account – often leads to a dissatisfied customer. Dissatisfied customers not only stop shopping at that store, but can also spread the news about bad customer experience. 

 

With the competitive nature of online retailing, it’s essential to long-term success to drive positive customer experiences. One e-commerce retailer looked at the issue of sizing and decided to apply machine learning to the challenge of sizing footwear. 

 

Every shoe manufacturer builds shoes in their own way. Ideally, sizes would be consistent across brands and types, but in actuality, variations exist which can be frustrating to customers. The company decided to tackle the differences in sizes so that they could provide a much more accurate sizing chart for each shoe from every manufacturer they carried. 

 

Using data from the manufacturers, customer feedback on the “trueness” of the sizes (i.e. “this shoe runs small” comments in their feedback forms), and data collected across different clothing forums, the company started working on a predictive model that would assist customers in their purchase decisions.

sizing_v_bar

The data science team started by building a model by hand and testing it against historical data gathered over the past years. Despite the time and effort put into the model building, they were unable to achieve better than 70% accuracy over the course of several months. And based on their projections, they needed the model to be at least 85% accurate when deployed in order to save the company money on returns.

Although the team was frustrated with the results, they knew that more data and more testing and iteration should help improve the accuracy of their models. With a bit of research into automated machine learning (AutoML) platforms, they decided to see if a free trial of the Ople Platform would help their customer satisfaction team improve the accuracy of the sizing model.

manu_quote_pink

With better than 85% accuracy, the sizing model is helping customers find the shoes that fit them, which means happier customers, better reviews on their site, and more recommendations.

Using Automation to Build Accuracy

The first task was to pull together a team to work on the sizing model. Knowing that the Ople Platform incorporated the intelligence of an advanced data scientist, the company assigned an analyst to work with members of the customer satisfaction and data engineering teams. This group was intimately familiar with the business aspects of the sizing challenge and had access to all of the data needed to realize a better model. 

 

To start, they pulled together the data that the data science team had used for their hand-built model and ran it through the Ople Platform. Immediately, they saw a bump in accuracy as Ople was able to use its built-in AI to create a better model, but the accuracy still did not meet the 85% threshold for deployment. The team knew they needed more data to enrich the model. 

 

Data engineering gathered even more data, not only from internal operations like customer engagement but also industry information about sizing based on manufacturers. Each time they added more data, this was fed into the Ople Platform and a new model was quickly built. After multiple iterations using more and more data features, the team finally had a model that exceeded the 85% accuracy mark.

To start, they pulled together the data that the data science team had used for their hand-built model and ran it through the Ople Platform. Immediately, they saw a bump in accuracy as Ople was able to use its built-in AI to create a better model…

sizing_quote

The Results

With the new model in place, the company has been able to reduce the number of returns and frustrated customers when it comes to the issue of finding the right size. With better than 85% accuracy, the sizing model is helping customers find the shoes that fit them, which means happier customers, better reviews on their site, and more recommendations. 


As the company continues to gather sizing data from customer interactions, they can use the Ople Platform to continue to iterate and refine their sizing model. The hope is to provide a model that exceeds 90% accuracy, providing a true competitive advantage for their website.

sizing_middle

Building a Culture of Innovation

As the first project built with the Ople Platform by the company, the predictive sizing model is not only reducing returns and increasing customer satisfaction, but also proving the value of the Ople Platform within the company. Now, the company sees opportunities to embed AI everywhere across the organization. And with the automation and intelligence of Ople, just about anyone in the company can be effective.

sizing_bottom

The Ople.AI Platform

The easiest way to accelerate decision making and reduce risk with predictive analytics

With the Ople.AI Platform, you are only a few clicks away from building predictive models to derive optimal business recommendations with reduced risk.