Features Hub

eCommerce Expo 2019: How Simon Calvert helped Bonmarché balance machine learning with human intelligence

Wed 21 Aug 2019

As of this summer, the fashion retailer is using machine learning to ensure it has the right stock, in the right place, at the right time. Techerati talked to eCommerce Expo speaker Simon Calvert about his role in helping the retailer intelligently pre-empt markdown

It has been a disruptive start to the century for traditional retailers as they grapple with the shift towards online shopping together with economic ups and downs that have made consumer habits unpredictable. Retailers have adapted by spreading business across multiple channels – incorporating offline brick-and-mortar stores and mobile apps and e-commerce sites – and of course, embracing new tech trends.

During this digital transition, data has been an invaluable tool to help retailers make better decisions and learn more about customers, enabling them to keep up with fast-moving currents with agility and speed. Although retailers have been buoyed by technology, the commercial focus remains the same, says Simon Calvert, former trading director at Bonmarché.

“The channel you’re selling on might have changed and continues to do so, and the possibilities may have become bigger in terms of what you can do and how you can learn in order to satisfy the customer. But at the end of the day, it’s still a customer-driven business,” Simon said.

“I got a phrase I borrowed off a knowledgeable, well-connected friend in retail: The traditions of retail may be dead but traditional retailing is alive and well,” he said.

Machine learning

Across the board, machine learning has breathed new life into long-standing industries. Retailers are no exception. Numerous household names are using algorithms to make accurate predictions about the future, enabling actionable business insights and innovations in marketing strategy, customer relationships and operations.

While other retailers were distracted by how the technology could be used to understand customers, Calvert saw the potential for machine learning to assist in allocation decisions.

“Using data to understand what customers want is great and machine learning can be used to make those decisions better,” he said. “But since putting the right stock, in the right place, at the right time is still as important to traditional retailing as it ever was. I thought maybe we should be using that technology in a more effective way to make those decisions.”

For middle to lower middle-market retailers, markdown is the single biggest cost to business – making effective stock replenishment and allocation vital. When stores are overstocked, not only are retailers forced to discount, but it means there are stores elsewhere that are understocked – transporting stock to another branch in order to sell it more effectively is costly and complex.

The traditional approach to merchandise allocation is “push” orientated: The retailer decides how much stock they want to sell, breaks it down by product type, timeframe and location, buys it and then pushes it out. A far more effective method is to flip the process and let the same variables dictate how much stock retailers should purchase and distribute.

Nowadays, data pertaining to these variables is far more accessible and abundant. It is possible to gather this data and make informed allocation decisions without machine learning, but there is enough data out there to support a machine learning system capable of far more granularity than humans.

“Don’t think you know too much, because at the end of the day, from a granular perspective, you probably don’t”

Bonmarché’s allocation and replenishment system, launched in summer of 2019, takes a product and uses this data, along with previous demand history, to calculate demand for similar products in a particular store or on a particular day in the future, enabling the retailer to allocate accordingly. The results of those decisions are fed back into the system to fine-tune it, and so on.

After noticing immediate results, the system is now allocating stock across 99 percent of the company’s business.

“Because we’re in effect sending less out, we’re able to keep that availability in the warehouse for longer, so we are flattening out the rate of sale decline,” Calvert said, referring to a product’s sales performance over time.

Assisting not replacing

Like we have seen in other industries, the system is not replacing humans but complementing them, as it lacks emotional and aesthetic intelligence, and awareness of each store’s particularities. In this context, human decision making becomes even more important, Calvert said, as it is only humans who are able to make the ‘big picture’ decisions from the granular detail provided to them.

For instance, if Bonmarché followed the system’s instruction without cross-referencing against store profiles, they would end up with overstocked stores, because branches are smaller than their turnover denotes they should be, Calvert said.

In the absence of emotional or aesthetic intelligence, the system also lacks the capacity to predict the ebb and flow of fashion trends.

“There’s nothing to say that red dresses are going to be exactly the same demand this spring/summer as they were last spring/summer. So you do need that emotional intelligence that machine learning algorithms don’t have,” Calvert said.

Nonetheless, Calvert’s advice to other retailers considering developing their own machine solution is to largely respect the recommendations it provides.

“Human instinct is ‘better safe than sorry’. But when you’re dealing with data-driven decisions there is no point overruling [the system] too often. It’s always going to be 80 percent data and 20 percent intuition.”

When it comes to delegating important tasks to autonomous systems or other emerging technologies, retailers can learn a lot from Bonmarché about achieving that balance; open-mindedness about technology’s benefits and a critical distance that allows them to challenge them from a theoretical perspective without interfering too much.

“Don’t think you know too much, because at the end of the day, from a granular perspective, you probably don’t,” Calvert said.

Join Simon eCommerce Expo 2019, 25 - 26 September 2019, Olympia

26 Sep 2019
Delivery & Fulfilment Theatre


machine learning
Send us a correction Send us a news tip