Jul 06, 2014 — Scot Stelter
RFID Journal

At the grocery checkout, there are two printers. One prints your receipt, while the other outputs a strip of coupons just for you. Usually buy Barq’s root beer? Try Mug for free next time. Most likely, Catalina owns the second printer and uses loyalty card data to target promotions, generating substantially higher redemption rates than other methods. The company claims to have the largest shopper history database in existence—some 2,500 terabytes of information.

Data mining for retailers and brand owners is a profitable business. Catalina generated $661 million in revenue during a recent 12-month period with an EBITDA (earnings before interest, taxes, depreciation, and amortization) of about $230 million.

ChainLink Research says that more than 8.5 billion passive RFID tags will be attached to or embedded in merchandise or other items this year, and projects that a cumulative 150 billion will have been deployed by 2020. In addition to basic inventory or asset information, the flow of data from the readers will conceal a rich stream of information valuable to anyone who has the mathematics necessary to extract it. Not everyone has the tools to extract this data, but there are mathematicians and software engineers who do. This article takes a look at this trend, and how it will affect the RFID ecosystem we know today.

For retailers, on-shelf inventory management is just the starting point, according to Stacey Schulman, Zensar Technologies’ senior VP of global retail and former American Apparel CTO. “Any data is valuable if it’s behavior-changing,” she says. Fixed RFID reader infrastructure provides a way of visualizing activity. For example, “we can determine an optimal planogram based on traffic, activity and correlated sales.” It doesn’t have to be labor-intensive: Using reference RFID tags, the system can learn the baseline planogram automatically, after which A/B testing, in which experimental planograms are compared to the baseline, can reveal how to improve on the original.

Combining RFID data with other technologies brings out additional value. For example, Schulman says, “Using Wi-Fi beacons and geofencing, we can correlate a person with an item that has been picked up and infer a relationship. We know what’s moving and what’s selling.”

Fixed infrastructure is a prerequisite for users to realize these benefits, says Caltech researcher Mike McCoy, the founder of Southern California RFID data analytics startup Cofacet Inc. “Data analytics is more restricted with handheld systems. There’s value and information that can be gleaned from [handheld] systems; however, the continuous reads from fixed infrastructure lead to a much richer dataset to analyze.”

Bill Colleran, CEO of UHF RFID technology supplier Impinj Inc., agrees. “Omnichannel demands fixed infrastructure,” he says. “Handhelds have provided a lower hardware investment at the expense of accuracy and time granularity.” Up against Amazon, he adds, “retailers can’t afford to overpromise,” but they also cannot afford to risk overstocks by inflating inventory buffers. “The benefits of [fixed] hardware involve going from periodic to real-time inventory, 24/7. Real-time inventory reduces the necessary inventory buffer size.”

Historically, a fixed infrastructure required larger up-front costs—which, according to Schulman, “were holding the market back.” But the price/performance ratio has been improving steadily since Walmart’s case and pallet mandate in the mid-2000s. In 2006, list prices were about $2,100 for general-purpose four-port readers, and more than $200 for antennas, leading to (uninstalled) hardware costs that exceeded $750 per antenna. A year ago, retail solution provider Senitron Inc. deployed about 15 Impinj readers and 150 antennas to cover the 8,000-square-foot American Apparel store in Melrose, Calif. (see American Apparel Deploys Real-Time, Storewide RFID Inventory-Management Solution). At current list prices, such a deployment would add up to around $300 per antenna—less than half of the 2006 cost.

Lower prices are good, says John Armstrong, Senitron’s CTO, but his company wants to minimize the up-front investment. “Some retailers want to own, some don’t,” Armstrong says, “For those that don’t, we provide a low, up-front cost by building the cost of the hardware into the service fee.” Customers just pay for the data.

Colleran calls this “RFID as a service” (RaaS) and predicts that it will become much more common. “It’s like our photocopy service [at Impinj],” he states. “We’re not interested in the copy machine. We’re interested in being able to make copies when we want to. [RFID] users aren’t interested in newfangled hardware gear. They want the data.”

RaaS will accelerate adoption. On an operational expenditure, rather than capital expenditure, basis RaaS is easier to stomach for most users. Return on investment is easier to predict because RaaS “takes the risk out of it for the retailers,” says Zensar’s Schulman. “If the monthly cost for the service is neutral, why wouldn’t I do this?” Users will exert greater control over what they’re paying for. If they want to upgrade from the basic service, it will increase their monthly expense, but they won’t have to invest in new hardware or software.

The basic service for retailers is real-time inventory visibility, for which there is an established ROI. But for a fee, Armstrong says, Senitron can provide additional value that has the power to drive operational changes, such as telling a retailer “which items get taken to the dressing room more often, or which get picked up.” There is value in knowing if a style is never tried on, or keeps being tried on but is never purchased. He says that there is a lot of opportunity for a retailer with vision.

According to Armstrong, that specialization in the analysis of retail data will create a third-party marketplace for data analysis. “We’ve teamed up with a company that is looking at that market,” he says. “Our value is that we have massive amounts of data. We get value from their labor and skill [in applying specific analysis techniques].”

Cofacet is Senitron’s analytics partner. McCoy, who earned a Ph.D. degree from the California Institute of Technology in computing and mathematical sciences, says: “Cofacet helps retailers understand how people are interacting with physical objects in the real world… such as what areas and what products are seeing activity.” He adds that “activity is a leading indicator” of sales, while sales is a lagging indicator of past decisions. His service will help customers get ahead of sales.

Advanced mathematics is required to interpret the data, McCoy says—it is not simply a matter of counting reads. “The data tends to be very noisy, and fluctuates a lot during the day. How do you know when something has actually been picked up?” He employs statistical methods such as “change-point detection” to know when a change has actually occurred.

Machine learning and automation is part of the solution, McCoy says. For example, when helpful customers re-hang items on the wrong rack, the system can automatically notify employees where to look and what to look for. A learning system doesn’t have to be told where that item belongs. It can infer the planogram from the tag data streaming in. Machine learning is an important element in the larger Internet of Things, of which RFID is a subset. University of Washington professor and MacArthur Fellow Shwetak Patel says that in the sensor systems he studies, “It was previously all a device play. The problem is, simple devices or sensors don’t do much [by themselves].” Machine learning helps sensor systems (including RFID) establish what Patel calls a “baseline of the observables,” which is necessary before changes can be detected.

McCoy’s company is now developing algorithms or systems to take advantage of fixed infrastructure data, just ahead of demand. “The market doesn’t exist yet, because there’s little fixed infrastructure yet,” he explains. “But Cofacet is building systems for what’s coming. It’s less than two years off.”

Diana Hage, CEO of RFID Global Solution (RFIDGS), says other verticals will follow retail, though some will take longer to adopt due to data security concerns. RFIDGS’ customers include government clients. “The majority of them are not comfortable with data going off-premises due to security and firewall issues,” Hage says. RFIDGS has been successful getting its Visi-Trac real-time location system (RTLS) platform adopted broadly in government organizations, such as at Department of Homeland Security (DHS) offices, but regulations like the International Traffic in Arms Regulations (ITAR) restrict access to the data, so “It’s all based on on-premises servers.”

Hage sees similar inertia in some private-sector verticals. There are many regulations that restrict data access in order to protect consumer privacy, including the Health Insurance Portability and Accountability Act (HIPAA) in health care, and the Gramm–Leach–Bliley Act (GLB) governing banking. But she says she is optimistic about the long term: “People will buy actionable information.”

Eventually, Colleran says, people will use third-party analysis “as long as it doesn’t undermine [their] underlying business. They’ll take safeguards.” There are precedents for this, he says. When payroll services were first invented, businesses were suspicious about outsourcing payroll processing. “Now you’d be a fool to do your own payroll,” he says. “It’s a matter of scale and expertise.”

Safeguards enable the U.S. Department of Defense (DOD) to take advantage of RaaS under certain circumstances. Rear Admiral David Baucom, the director of strategy, policy and logistics for the U.S. Transportation Command, says that the DOD often uses “a commercial fee-for-information methodology.” One example is a commercial active RFID network used to monitor cargo passing from Karachi, Pakistan, to the Afghan border. “The DOD pays the commercial vendor for the reads on RFID-tagged shipments passing an interrogator.” The data is secure because readers only return serial numbers, which only secure DOD databases can make sense of.

McCoy, who once co-authored a paper titled “Robust computation of linear models, or How to find a needle in a haystack,” is very optimistic about the prospect for an RFID data analysis industry. “Cofacet helps retailers understand how people are interacting with physical objects in the real world,” he says. All Cofacet needs is data to work on. “When I saw the data that Senitron had, I got excited.”

Scot Stelter has worked in the RFID field for more than 10 years, serving in senior marketing positions at both Alien Technology and Impinj. His company, Stelter Product Strategy LLC, provides guidance to businesses contemplating product strategy decisions in RFID, the Internet of Things and machine-to-machine communications. Stelter is also the VP for RFID and the Internet of Things at ChainLink Research.