How Predictive Analytics Shape Strategy

CPG retailers and their supplier partners have long used predictive analytics for tactical applications such as optimizing promotions and pricing, but there may be opportunities to use the modeling of data more strategically.

Experts say retailers could make more use of the vast amount of consumer and other data available to them to assist in such tasks as evaluating acquisition and new-store opportunities, customizing planograms, and making decisions around their supply chains, their marketing strategies and various other aspects of their operations.

Predictive analytics is defined as using existing data to forecast future outcomes, behaviors and trends. It generally involves sophisticated modeling using large amounts of data to assign probabilities to various potential outcomes.

"Predictive analytics is a basket of sophisticated tools that allow us to use what we know today to try to anticipate what is likely to happen," says David Meer, principal at PwC Strategy&.

Achim Schneider, global head, SAP Retail Industry Business Unit, says retailers who invest in predictive analytics solutions with a "digital core" as the platform for their businesses "will be more nimble, flexible, innovative and able to capitalize on new opportunities."

"Traditional methods of data mining and analysis provide a view of past performance, but digital consumers adapt quickly to new trends and preferences," he says. "Retailers who want to thrive in this digital environment must be able to learn from past performance, but also be able to predict future needs based on a broader set of data and influencing factors."

MARKETING STRATEGY

One of the key areas of focus for predictive analytics today is determining how to allocate marketing spending among digital and other platforms.

"Everyone who is spending money on marketing is trying to answer the questions of how much money they should spend on digital, what's the ROI on digital, and can they afford to do less television and other media," says Meer, who founded the Mindshare media-buying operations of advertising agency WPP in 2000. "This is one of the most interesting areas of analytics right now."

Online shopping lends itself to analytics because people's actions can be more easily attributed through the specific digital trails they leave, but once consumers start mixing their online and offline behaviors, attribution and analysis become much more difficult.

"As soon as someone researches online and then buys at brick-and-mortar [store], or sees a TV commercial and then buys online, you have broken the chain of evidence, and then you need time-series regression models to piece it together to view what should get credit for the sale," Meer explains.

Older models for optimizing marketing allocation "are becoming more and more obsolete as the focus turns to digital," he said. "I would say this is an area where a lot of vendors are trying to create solutions, and I think there will a lot of innovation and development on this front."

Schneider of SAP notes that as new digital advertising channels emerge, "retailers must also become more sophisticated in their marketing effectiveness and ROI measurements." 

"Effective advertising across channels will require retailers to trace customer search patterns and sales execution in order to evaluate their full picture of financial performance by channel," he says.

EVALUATING A LOYALTY PROGRAM

Another way predictive analytics can be used is to evaluate the viability of establishing a retailer loyalty program, Meer says, explaining that he recently worked with a retailer on such a project using sophisticated analytical tools.

The research involved modeling different levels of discounts customers could receive, how much they had to spend to earn a reward, and other components of a loyalty program. That data was married with a financial model that considered data around traffic, transactions and other information.

Models were evaluated to determine first whether or not a loyalty program would be viable, and second to make choices about how it should be structured.

"Part of it was very strategic in terms of determining which segment we wanted to attract," Meer says. "Do we want to attract new customers, or do we want to reward existing customers, or do we want to get people who come to us sometimes to come more often and give us more of their business? That was answering some very fundamental strategic questions about what the goal of the loyalty program should be."

One area of opportunity for retailers to make better use of predictive analysis could lie in evaluating store performance, says Kurt Jetta, CEO and founder of TABS Analytics.

"There's a need to crack the code on what are the attributes that go into a great store location," he says.

Retailers already know the attributes of their best performing stores and their weaker stores, but discerning what the exact variables are that drive the differences between them can be challenging.

"What modern predictive analytics can do is, I can throw a whole lot more variables at it," Jetta says. "We can apply machine learning, and run through several iterations and apply all of these different models, and come up with a better way to predict what location and what model has a higher likelihood of success."

Such an analysis might also be used to improve weaker-performing stores, he notes.

SEGMENTING LOCATIONS

Meer of PwC Strategy& notes that retailers can use predictive analytics to help them segment their product assortments by location to achieve the optimum product mix based on local demographics.

"We worked with a retailer who wanted to go beyond a one-size-fits-all approach to merchandising and assortment, even down to what brands to carry in various outlets," he explained.

The retailer sought to identify segments based on the socioeconomic and ethnic compositions of the areas they served, the degree of competitive intensity within those locations, and other factors. The PwC analysis allowed the retailer to group its locations into about 10 to 15 different buckets, so it could optimize product mix based on the likely interests of consumers in that specific geography.

By looking at actual store performance, that kind of analysis could be extended to evaluate store acquisition opportunities, he explains.

"You can figure out where your offer is outperforming the average and underperforming the average, and then you can use that as the basis to acquire stores in underperforming areas where you think someone might have a better business model and might do a better job of serving those geographic locations," he says. "Or you could decide that you have a better business model, but you just don't have enough brick-and-mortar in the places we need to be."

In that way, a retailer could model geographies that are similar to the areas where it currently performs well as the basis for determining where it might look to either open new stores or acquire locations from others.

The same kinds of analytic matrices can then be used to determine which post-merger store locations might not be viable, which stores could be improved through remerchandising or reformatting, and which might need to be closed.

"It's not rocket science, but it is an analytical framework that retailers can use to evaluate who to acquire and what to do with the stores once you acquire them," Meer says. "We've worked with companies that have done this, including large grocery chains that have wanted to expand to other parts of the country where they were not operating."

Schneider of SAP says the ability of predictive analytics to allow retailers to evaluate customer demand and preferences across a market or sales area is important when evaluating merger opportunities.

"It provides the company with insights that can drive more localized pricing and assortments, as well as aligning selling and fulfillment locations with customer demand," he says.

"These predictive insights can help retailers determine if another brand or store location is worth the investment based on predicted sales volumes," Schneider says. "Additionally, retailers can use this information to rebrand existing locations and attract consumers to new locations through personalized marketing and promotions."

LOGISTICS AND OTHER USES

Meer notes that as retailers grapple with online ordering, delivery and store pickup, predictive analytics can help them make decisions about how to adjust their supply chains and their fulfillment models.

"We have used analytics to help retailers think about their supply chain strategy by using customers' behavior and anticipated future behavior to understand how they might have to change their supply chain," he says. "How important is same-day delivery or next-day delivery or two-day delivery, or three to five days? Which segments of customers are going to gravitate toward that, how much are they willing to pay, and is it worth it to try to revamp your supply chain to try to serve that need?"

Other less common uses for predictive analytics include assessment of risk from such issues as loss prevention and food safety, Meer explains. Historical store data can be used to determine the variables that affect outcomes in those areas, and then models created to pinpoint stores that might share those attributes.

POTENTIAL PITFALLS

One of the potential pitfalls in using predictive analytics include the timeliness of the data used in generating predictive models.

"A lot of predictive analysis relies on empirical data—you take what's happened in the past, apply some statistical models, and use that to predict the future," Jetta says. "The problem with that approach is that the shelf life on empirical data wears out very quickly."