Kroger drives data-driven customer experience

US supermarket retailer Kroger has been on a digital mission to use technology to become customer-centric.

Data is often regarded as the fuel of the 21st century. Kroger’s 84.51° subsidiary is its consumer insights business formed through the acquisition of Dunnhumby’s technology assets in 2015. Dunnhumby was the company behind the success of Tesco’s Clubcard in the UK, and was run as a joint venture between Tesco and Kroger.

Scott Crawford, who heads up data science enablement at 84.51°, says the retailer has made a commitment to become a digital business with what he describes as “a truly digitally oriented value chain”.

He says the goal of 84.51° is to conduct longitudinal studies, an analytics technique that follows the same group of people over a long period of time. “We want to understand our customers to drive better customer experiences, make their lives easier or give them an overall better customer experience,” he says.

Crawford leads the Enable the Science team at 84.51°, which is responsible for enabling and empowering data science across the business. He says the team encourages the organisation to develop and deploy machine learning technologies efficiently. The Enable the Science function selects, implements and provides training for machine learning tools, including both “standard” and automated tools. One of these is DataRobot, which provides supervised machine learning.

“My team encourages the rest of the organisation to use machine learning effectively,” says Crawford.”We create toolkits for the rest of the data science teams.”

DataRobot provides the team with an efficient way to deploy data models, he says.“DataRobot enables them to deploy models months faster than if they used traditional tools.”

“The breadth of data science is growing. There are plenty of use cases that add value as a project”

Scott Crawford, Kroger

In 2017, the US grocery giant unveiled its Restock Kroger vision, a digital strategy that combines its food expertise and data analytics to create new customer experiences, delivered both digitally and in stores. The overall goal is to develop customer-centred efforts to create shareholder value.

Crawford believes digital transformation should involve doing something in a new and innovative way, rather than just digitising an existing process. “When we think about operations that are digital in nature, these can be improved by making better decisions through the use of data,” he says. “We have petabytes of Kroger customer history that can be combined with other datasets, and used as the combustion engine of the business to make better decisions.”

Crawford says IT has traditionally delivered capabilities to the business through IT projects, but data science is a capability. “The breadth of data science is growing,” he says. “There are plenty of use cases that add value as a project.”

But there are other use cases that are programmes, he says. “Any data model put in the field has a shelf life. Your very goal is to make your model stale, so there is a need to continually evaluate and evolve.”

In some cases, a feedback loop is built directly into the programme, and can evolve in real time, he says. In Crawford’s experience, there are some situations where such data science initiatives continually evolve  data models on a nightly basis using the most recent data.

For data science to succeed in driving operational change in the business, the analytics team needs a diverse range of skills. Crawford categorises data scientists into three types.

First is the enablement data scientist: “These data scientists have strengths in insights, and are gifted in communications and telling a story,” he says. “They understand the value stream and have an MBA skillset.”

Crawford’s second category is the technology data scientist: “These people are gifted with utilising multiple technologies and tend to be explorers,” he says.

The third category is the statistical learning data scientist, who Crawford says tends to be gifted in understanding the maths and statistics.

Beyond the categorisation of data scientists, he says: “The future of data science in many companies will be determined by their ability to collaborate effectively by bringing together diverse teams that have different skills and perspectives to help them understand how to find the needle in the haystack.” 

But, as Crawford points out, if a data science team is tasked with developing a model for something like forecasting demand, and the model is shown to be more accurate than the existing approach to forecasting, what should the business do?

“How will operations extend data science into the business in a way that evolves a business process to generate a positive ROI [return on investment]?” he says. In other words, for an organisation to become truly data-driven, changes to the business that occur when new data models are deployed need to be part of the iterative data science process.