In retail, it would be a reasonable expectation that the people who spend the most at a company’s stores are also the most loyal customers. That’s certainly what executives at one of the world’s largest oil and gas companies thought when they set out to target new marketing efforts.
But a retail analytics project at the company’s chain of European gas stations uncovered a different result. Working with Beyond Analysis, a business part-owned by Visa Europe, the retailer was able to analyse customers’ total fuel spend paid for by Visa cards. It turned out that the top-spending customers were simply people who bought a lot of fuel—from every retailer in the market.
“The really loyal customers, who bought predominantly from just that one company, were among the Tier 2 customers—people with a lower total expenditure on fuel, but who spent it more or less exclusively at just that one company’s gas stations,” says Paul Alexander, chief executive of Beyond Analysis.
The result: the retailer crafted a retention strategy that targeted a totally different group of customers from that first imagined—Tier 2 customers, not Tier 1.
The best retail analytics projects guide business decisions by providing compelling insights into consumer behavior. In interviews, seasoned practitioners in the field share a common view on best practices for these projects, from establishing clear project goals to securing executive sponsorship, which will be familiar to line-of-business managers and IT executives alike. But because these projects are so strategic—and so challenging—it pays, these experts say, to follow these best practices. The benefits outweigh the costs—and should lead to profitable gains.
1. Link the project to the business’s strategies.
Look at successful retail analytics projects, and you will see a pattern, says Paul Winsor, European retail industry director at Teradata. “Retailers who do analytics well can almost always articulate how projects map onto the business’s strategy,” he explains. “You can’t suddenly wake up one day and decide to do analytics.”
Strategic alignment is essential “to have a clear understanding about how what you are working on will help the business,” says Alexander, whose company analyzes and sells data on every Visa customer transaction in Europe—something like one in four non-cash purchases.
This means mapping every analytics initiative, Alexander adds. “When we spoke to [UK retail giant] Kingfisher recently, we took their eight strategic priorities, and outlined to the board how data analytics could help them in each and every case.”
Max Jolly, global head of digital personalization at retail analytics experts Dunnhumby, whose clients include Kroger and Tesco, stresses that good analytics projects always lead to customers. He asks: “If customers aren’t responding, and you aren’t influencing or changing their business decisions, then what’s the point of the project?”
2. Have clear project objectives.
A common mistake among project managers is to set off on the data analytics equivalent of a fishing expedition, says John Lucker, a principal at Deloitte Consulting. Lucker is global leader of the firm’s advanced analytics and modeling practice, and the co-inventor of three predictive modeling patents and two pending patents.
Even when experimenting with analytics projects, it is necessary to focus a team’s questions, Lucker says. “It’s important to ask: ‘What levers are you trying to get hold of?’ Are you aiming to attract new customers? Upsell? Move customers into a different category? Build loyalty? Drive customers towards higher margin products?”
He adds: “Too often, retailers say: ‘Let’s just sell more to every customer.’ But that is potentially hazardous: it can create noise in the marketplace, as well as creating erroneous offers—offers aimed at products customers already buy. In other words, you have to ask: ‘What’s the mission?’”
Also helpful, Dunnhumby’s Jolly says, is using the experience of past projects, knowing what works and what doesn’t, to influence project objectives. “Our clients often talk to us about ‘upselling’ and moving customers towards more profitable brands,” he says. “We try to dissuade them: in our experience, capturing new customers, and keeping customers loyal, offers a far bigger payoff than trying to move existing customers ‘up-brand’,” he says.
3. Walk in the customer’s shoes.
And start with the consumer’s view of the world, says Deloitte’s Lucker. “Some customers know very clearly what a particular company could do to delight them, while others don’t know—but do know that they will recognize it when they see it.”
That means retailers have to think about “how they are going to figure out what customers are looking for,” Lucker says. That is a different idea than looking at inventory and saying, “This is what I’ve got to sell, so how do I sell it to you?’”
That product perspective represents the traditional view of retailers, says Jolly. Attributes about a product—how big it is, its weight, price and location—are all good data points, “but they don’t actually tell you why a customer is going to buy it.”
Viewing transactions from a customer perspective makes it easier to get to a consumer’s underlying motivation, and thereby craft offers that will make a difference.
“Does a consumer buy a lot of frozen products? That might tell you that they are time-starved. Value products? They’re cash conscious. Big products, or bulk packs? They’re buying for a family. Do that for every product in the shopping basket, and you can build up a very rich picture of the consumer,” Jolly says.
4. Start with the data you have.
With goals and alignment established it’s time to look at the data. This is a key point in a retail analytics project, says Teradata’s Winsor.
Managers “are in a position to articulate very clear business questions which need answering—and from which, you can determine what data is necessary to answer those questions,” he says. “In short, what data do you have—and what data don’t you have?”
This is not the time, though, to go in search of data that fill perceived gaps. Having most of the data will do. Alexander of Beyond Analysis says he’s yet to meet a retailer where truly significant data was actually missing.
Alexander says he sees a lot of clients express interest in creating a customer loyalty program to collect new datasets so they understand their customers better. “It’s a viewpoint that’s understandable, but mistaken. Our response is that you can get 90 percent of the answer just from transaction data, which they already have—so start with that.”
5. Look for quick wins and incremental gains.
Experts say that starting with the data you have, even if it appears that data is missing, can lead to faster progress—and demonstrated business value.
Colin Linsky, predictive analytics worldwide retail sector leader at IBM, says it is not uncommon to encounter strong resistance to starting a project “until all the data issues are sorted out.” But, he adds, “If you wait for the data to show up, what you get is a big project that can be unmanageable. We say: ‘Start small, and then grow.’ Think big—but view the project as a part of that overall bigness.”
“A retail analytics project is like building a house: you have to start with the foundations,” says Diana McHenry, SAS’s director of global retail product marketing who began her analytics career 27 years ago with Procter & Gamble.
“It about defining do-able first steps—easy business wins, to prove the worth of the concept, and then build out from there,” McHenry adds.
For example, she says that optimization projects provide a decent payoff from readily available data: size optimization, space optimization, assortment optimization and mark-down optimization.
After that, go for bigger fish—having learned lessons from catching the smaller fry.
6. Get the team right.
A good analytics team needs technical experts, such as statisticians and data miners, but it is vital to include retail subject matter experts on the team, says IBM’s Linsky. Even better: provide analytics training to retail experts.
“It’s much easier to take someone who’s familiar with retail, and train them in analytics, than to take a statistician or a data miner, and train them to think like a retailer,” he says. “That way they have credibility within the organization, and will take ownership of the results. In other words, they’re using analytics to make better decisions—and not just buying in some analytics.”
Jolly says that the analytics project is not about coding and not only about analysis. “It’s about institutionalizing analysis into tools and methodologies that can be used by people who aren’t analytics people per se. It’s about democratizing data: making it accessible to people who are numerate and interested in the answer, but who aren’t necessarily analysts themselves.”
Colin Haig, program principal at SAP Retail, says the trend for analytics competency centers—an in-house organization devoted to working on analytics projects–helps to formalize the marriage between technical and subject matter expertise.
“They’re perhaps seen as more tactical than strategic, but they’re undeniably linked to success,” Haig says of the competency centers. “You’ve got people with process knowledge regarding doing the job, sitting alongside analytics people. It’s a very, very powerful combination.”
7. Win executive sponsorship.
A failure to have a business champion—emphasis on business—can kill a project, experts say.
“Time and again, we’ve seen retail analytics projects fail because they don’t have business sponsorship,” says IBM’s Linsky. “A retail analytics project isn’t an IT project, or a business intelligence project or an analytics project—it’s a business project.”
McHenry of SAS says that a common—and valuable—way to provide that executive-level business sponsorship is through an executive steering committee. “The best projects are those with an executive steering committee,” she says.
She adds that such a committee might be formed only briefly, but the executive members “provide vital governance, and can provide extra horsepower when required.”
“When you hit an obstacle that can be very difficult to resolve five levels down the organization, or in middle management, then executive-level steering committees can usually resolve it very readily,” she says. “If it’s important, they get it done.”
Freelance writer Malcolm Wheatley remembers analyzing punched card datasets in batch mode, and using SAS and SPSS on 1970s-era mainframes. He lives in Devon, England, and can be reached at firstname.lastname@example.org.