What if, on your next visit to the supermarket, you found the store's selection of beer shelved side by side with boxes of baby diapers? You might think the store manager had gone mad. But would you begin to suspect that the store was putting into practice an advanced, data-driven approach to retail marketing?

Andrew Jordan Group head of data solutions, Dunnhumby

A new marketing strategy is taking hold, changing the business of retail grocery chains, the brands that fill their shelves, and the supply chain that binds it all together—one bag of groceries at a time. It was pioneered at our company, Dunnhumby—a British-born marketing firm with annual revenue of $150 million.

We call our approach "relevance marketing." It tracks and analyzes data on storefronts, products, and consumers to create actionable market intelligence, with an emphasis on proprietary marketing expertise and predictive behavior modeling that sets the strategy apart from traditional business intelligence or data-warehouse operations.

Dunnhumby first established itself in the grocery-retail market, where fractional gains in per-product revenue yield significant impact on the bottom line for retailers as well as for manufacturers of consumer packaged goods (CPG). Its customers include leading U.K. grocery chain Tesco and U.S. supermarket giant Kroger, which relies on Dunnhumby's services to generate intelligence from 2,181 U.S. storefronts and more than 42 million customer households.

For both Tesco and Kroger, the Dunnhumby approach begins with the individual purchase data harvested en masse from the loyalty cards that customers use to earn points or obtain in-store discounts on their purchases, as well as from point-of-sale terminal data. Each purchase is tracked for analysis, giving Dunnhumby a look into every bag of groceries that moves across one of its client's checkstands. In addition to this purchase data, the loyalty cards supply demographic household data. Geographic data on the neighborhoods served by each store is also tracked by Dunnhumby, as well as demographic consumer data for each geography.

Besides customer-centric information, there's data about the storefront itself: The layout of aisles, the arrangement of shelves and products, in-store marketing displays, and outbound coupon promotions all factor into Dunnhumby's intelligence.

At our data centers, this raw data combines with our marketing expertise through proprietary software and algorithms developed from years of constant analysis. Each week, more than 40 terabytes of data is generated per retailer and more than 5 billion pieces of information are processed. The output, marketing intelligence, is served to a Web portal we call "the Shop," which can be accessed by both retail clients and the CPG manufacturers. Additionally, our firm's staff of marketing analysts has real-time access to the data to generate custom reports and data slices as clients demand them.

The information Dunnhumby provides can be simply described as "who buys what, and when." But nothing is simple about the depth and breadth of the information our firm delivers, or its substantial impact on business operations for both retailers and CPG manufacturers. By synthesizing all of the data into actionable, market-driven intelligence, Dunnhumby gives retailers and manufacturers the tools to hone in, at a per-store level, on the inventory, layout, and product mix on each shelf; product-pricing strategies; and promotional opportunities.

Once generated, the information has a significant influence throughout the consumer-goods supply chain. Examples include:

Store inventory. Dunnhumby's predictive modeling uses demographics, buying patterns, and seasonal influences to forecast which products a grocer should stock and in what quantities. Not only is profit enhanced by reducing the amount of unsold products that go to waste, but customer loyalty is improved when the supermarket can offer customers precisely what they're looking for every time they shop.

Store layout. By understanding which products are purchased together and by whom, stores steer the arrangement of products for maximum impact. In fact, as referenced above, we found that in certain geographies, displaying diapers in direct proximity with cases of beer had a multiplying effect on sales of each. Picture the prototypical young father dispatched to pick up diapers at the supermarket after a long day's work and the apparent dichotomy becomes easier to understand.

More Insights

Webcasts

• Getting a Grip on Mobile Malware
• The ROI of Mobile Photo Bill Pay

More >>

White Papers

• Government Analytics: Set Goals, Drive Accountability and Improve Outcomes
• 2012 IBM Chief Information Security Officer Assessment

More >>

Reports

• Strategy: Passwords Don't Work. Now What?
• Strategy: Cybersecurity on the Offense

More >>

Distribution. When we prove to a retailer that a particular segment of its customer base buys a particular brand of soda at significant levels within a certain geography, our findings ripple through the supply chain. Distribution for the preferred brand gets a boost in that region, while less-favored brands are inevitably reduced.

CPG manufacturing. Brands represented on Kroger shelves use our data on a local and global level to define product distribution, adjust manufacturing outputs, project sales trends, plan product launches, and optimize marketing spends.

Retail chains. By knowing as much as possible about individual customers and their buying habits, retailers can ensure that the customer base unique to each store is ideally suited to the store's inventory, marketing mix, distribution network, staffing levels, and layout. This allows the retail chains to maximize customer retention, competitive market share, and profit per product while adding leverage to their ongoing negotiations with distributors and brands.

Marketing. The data enables retailers to engage in one-to-one marketing on a truly massive scale. Tesco has 12 million members in its loyalty-card program managed by Dunnhumby. By using the cards, Tesco customers accrue points toward special offers; through the mail, they receive quarterly statements of their points totals, along with promotional coupons. Dunnhumby tailors each set of coupons to individual households, such that wine drinkers receive discount offers on wine and empty-nesters aren't sent coupons for baby food. In fact, of the 12 million statements sent in the fourth quarter of 2006, 8 million included a mix of coupons unique to that household.

Revitalized infrastructure

Since its founding in 1989, Dunnhumby has brought considerable IT infrastructure to bear on its proprietary, data-driven marketing analyses. However, as we attracted more and more customers, the IT organization found its resources stretched to capacity, with several terabytes of data arriving daily from its network of customers and billions of pieces of information to process weekly. Moreover, sustaining the company's rate of growth is a high priority of management. In October 2005, as the company expanded into North America, it embarked on an ambitious project to migrate its data center to a next-generation framework. Top priorities included a scalable architecture, reliable performance, and cost-efficient capacity for growth.

Prior to migration, Dunnhumby had one data center in the United Kingdom, provisioned with DEC Alpha servers running Tru64 Unix. Scalability was a problem with the rigid Unix boxes, and continued forklift upgrades were hardly a desirable path. With a commitment to restructure its IT infrastructure using best-of-breed building blocks, the Dunnhumby team examined its options and quickly agreed on the combined performance, scalability, and price point afforded by server blades.

With the decision made, Dunnhumby migrated the data center to a foundation of Hewlett-Packard ProLiant BL25p dual-processor server blades running within HP BladeSystem p-Class enclosures. The HP blades, running Red Hat Enterprise Linux Version 4.0, let Dunnhumby add compute power at a granular level for cost-efficient scalability. They also scale out rapidly, without the need for a costly, high-end enterprise server, allowing us the flexibility to adapt to changing work flows quickly and efficiently.

Another important component of the new IT infrastructure was the PolyServe File Serving Utility for Linux, clustered storage software designed to maximize the performance and cost efficiencies of the x86-based servers and industry-standard networked storage. The software united the hardware within one cluster, establishing a robust NAS platform without performance bottlenecks or single points of failure.

With 128 HP blade servers, Dunnhumby's multitude of proprietary data retrieval and analysis processes operate with consistently high I/O performance. Most important for Dunnhumby, the platform permits the data center to seamlessly scale capacity based on workload requirements.

The results achieved from the revamped data center are significant for both Dunnhumby and its retail customers. In terms of pure performance, the data center can deliver complex proprietary analyses 12 to 25 times faster—depending on the process being performed—than the previous infrastructure could manage. And with the timeliness of our insights absolutely essential to grocers, the new IT infrastructure turns around within 90 minutes requests for complex, customized data slices that would have taken five days with the previous technology.

The new data center aligns perfectly with our goals for growth. Presently, the company's IT organization is completing a build-out of two more data centers to the same platform specifications. Furthermore, new growth is being driven not only through the retail grocery channels where Dunnhumby has built its reputation, but through the robustness of its data center infrastructure—which has supported Dunnhumby's entrance into other industries, including financial services and telecommunications, where the value of a data-driven, customer-centric marketing model strongly resonates. — Andrew Jordan, group head of data solutions, Dunnhumby