The use of big data to inform decisions is spreading to more business sectors
Before the infamous breach that compromised millions of personal records, before the disastrous Target Canada campaign, before the layoffs that shook the Twin Cities retail community to the core, there was the Target Pregnancy Debacle. (At least, that’s what we’re calling it.)
As per New York Times Magazine’s excellent 2012 story on predictive retail analytics, the short version goes something like this: Young woman becomes pregnant and starts buying things that pregnant women tend to buy; Target’s super-secret predictive analytics pick up on it; Target sends young woman (who lives at home with Mom and Dad) personalized coupons for things that people with babies tend to buy; Dad, after first demanding an explanation from Target honchos, discovers that father does not, in fact, always know best.
No parent wants to learn that their daughter is pregnant from retail mailings. But, like it or not, retailers have been collecting loads of shopper data for years — as anyone who’s been handed an eerily prescient checkout coupon knows.
Other industries are catching up. Today, data analytics inform virtually every major business process, from marketing and sales to finance and lean production. “Every organization uses data,” says Dee Thibodeau, co-founder and CEO at Plymouth-based Charter Solutions, a management consulting firm that helps decision-makers glean actionable insights from raw data. “Retailers, healthcare companies, banks and financial firms, government agencies, farmers and agribusiness — everyone is asking how to use data-driven insights to serve their customers better.”
They’re getting better at it all the time.
What follows is a mere taste of how Minnesota-based companies are pushing the analytics envelope today — and what could be possible tomorrow.
More bang for your marketing buck
Robert Cooley, CTO of Minneapolis-based OptiMine, is not the nostalgic type. He can't afford to be. His livelihood depends on the brave new advertising world made possible by digital media.
In the past, even the biggest companies might have 10 ads in the field at once: one or two each for print, TV, outdoor, radio and niche media. Today, even a medium-sized company can have 100,000 ads in the field at once, the vast majority of which are digital ads distinguished by minute keyword variations. Those spots are a lot cheaper than national TV ads, of course, but Fortune 100 companies still easily spend at least $100 million on advertising each year.
OptiMine uses cutting-edge analytics to answer, over and over again, what Cooley calls a simple question: “Did I make more money after putting that ad up?” (And that ad, and that ad…)
DATA INNOVATIONS & TRENDS, NOW AND TOMORROW
A type of proto-AI, machine learning systems synthesize vast volumes of data to determine what contributed (and in what parts) to a selected target variable. “Everyone wants to be doing machine learning these days,” says Optum’s Atkins.
True AI is on the horizon – “no longer a pipe dream,” says Atkins. However, early versions are of limited pedagogic value. Atkins: “AI is basically a black box; it’s hard to ascertain how or why the inputs produce the results.”
Data is increasingly dispersed, with multiple copies of the same data and parallel structures that speed access to information. “Think data lakes, not data warehouses,” says Maurices’s Bibelnieks.
Internet-connected sensors are revolutionizing in situ data collection, especially in dangerous and hard-to-reach environments. Among other industries, IoT could revolutionize agriculture: in the future, farm fields could sport digitized “skins” composed of millions of low-cost sensors measuring soil moisture, chemistry and more.
Social Media Analytics
Mastering Facebook ads is one thing; understanding exactly what each like, retweet, pin and share means for your bottom line is quite another. Ever-improving (and ever-closely held) algorithms help marketers and data scientists correlate engagement, evangelism and buying decisions.
“We run software to determine incremental revenue [produced by] each facet of the client’s advertising, using highly automated econometric time series,” says Cooley. If every single ad is a discrete billboard, OptiMine’s software simply counts the cars that drive by, without caring too much about who’s in them or why they’re taking that road. Cooley’s job is much harder than that, of course, because he’s dealing with thousands of ads, millions of drivers, and a “highway” system that makes the Twin Cities’ freeway network look like a donkey path.
This approach makes Cooley something of an iconoclast. Most of his peers relentlessly chase sophisticated tracking techniques that enable personalized (even individualized) marketing: the Target Pregnancy Debacle taken to its logical extreme. Cooley doesn’t deny that tech giants like Google and Facebook have found amazing uses for user tracking, but the value to advertisers has been questionable.
“You can’t track all the people all the time, and not everyone wants to be tracked anyway,” says Cooley. Ad blockers, virtual networks and device switching all stymie tracking efforts, he adds, such that “the leading edge has come to the conclusion that tracking doesn’t work.”
Instead, OptiMine’s value proposition lies in illuminating the “aspects of digital advertising that aren’t particularly trackable,” and blunting counterproductive tendencies like confirmation bias and cognitive dissonance. It’s easy to convince clients that the analysis is right when it confirms something they already believe; it’s almost impossible when the analysis goes against preconceived notions. So OptiMine tries to thread the needle, helping the marketers understand all of their advertising contributions regardless of clicks.
“We play in a narrow window where the analysis provides the data and insight necessary to confirm a client’s suspicion or loosely held belief,” says Cooley.
Who bought X and Y?
Eric Bibelnieks, VP of customer analytics and CRM at Duluth-based clothing and accessory retailer Maurices, is more of a personalization guy.
Like Cooley, he has a vast trove of data at his disposal: “Collecting enough data is not a problem for us,” he says. “Harnessing it to produce actionable insights is more challenging.”
That’s because Maurices, like all modern retailers, has a vast and complex footprint: more than 1,000 retail stores, thousands of affiliate relationships and its own ecommerce portal.
Correlating raw data with final sales only gets Bibelnieks so far. “If a vendor tells us they had X impressions and Y sales, it sounds easy to assign each impression a firm value,” he says. It’s not. Few customers decide to buy after the first impression; as they move toward a buying decision, they touch multiple marketing messages along the way.
“If you just go by impressions, it looks like you made many times more sales than you really did,” says Bibelnieks. “Our job is to figure out which ads are actually making the difference.”
The short answer is, it depends. Comparison shopping drives Maurice’s mobile traffic, which accounts for up to 60% of the retailer’s overall impressions. While browsing the racks at their local Maurices, shoppers might cross-check prices on other clothing retailers’ apps. Or, during free moments, they might simply Google clothing or accessory items and note how Maurices stacks up against the competition. But they rarely buy on mobile — they’re more likely to visit brick-and-mortar outlets (if they’re not there already) or purchase via laptop or desktop. That underscores the importance of serving the right ads, at the right time, on the right devices, and measuring everything that happens next.
Fortunately, the vast amount of data at Bibelnieks’s disposal makes practical “a constant test-and-learn environment” — basically, a real-time retail marketing laboratory in which every Maurices’s customer plays a role, wittingly or not. Bibelnieks’s team synthesizes all that data into digestible reports that drive decision-making internally and at Maurices’s myriad vendor partners. A typical report might plot the number of units moved by a 25% off sale against units moved by a BOGO sale, controlling for extenuating variables such that the comparison is truly apples-to-apples. Or it might contrast loyalty program customers’ responses to a given offer against first-time shoppers’. Every report offers clues into buying behavior, informs subsequent merchandising decisions, and — hopefully — improves outcomes.
“We strive only to present information that’s actionable,” says Bibelnieks. “If it doesn’t drive decision-making, it’s not necessary.”
Up on the North Shore, Bibelnieks lives on the cutting edge of the decades-old retail analytics niche. Back in MSP, Dan Atkins lives in two worlds.
By day, he’s the senior director of patient insights data science at Optum, the fast-growing UnitedHealth Group subsidiary, where (in his words) he “[figures out] how to get your uncle to take his blood pressure medication.” By night, he’s the co-founder and chief analytics evangelist of MinneAnalytics, the (almost as) fast-growing nonprofit dedicated to all things data.
MinneAnalytics makes data fun — almost too fun. The organization puts on three big events per year: Big Data Tech, a geeky confab for data scientists; HALICON, which bridges the technical and business sides of the healthcare industry; and FARCON, which focuses on finance and retail analytics.
In between, smaller, sometimes serendipitous events keep things lively. This January, SportCon drew 700 sports fans and data geeks to the University of St. Thomas, where they spent the day “hacking baseball data.”
In November 2016, 200 students met up for a water quality hackathon that found “crazy correlations between proximity to polluted water and home values,” says Atkins.
And, a couple years back, MinneAnalytics held a “date night” featuring the co-founder and CEO of OKCupid, Christian Rudder. His most titillating discovery: The surest indication that a pair would, ahem, hit it off on the first date was one or both partners’ love for the taste of beer.
Data innovations that save lives
Atkins (left) is just one of the countless life- and cost-saving cogs in his employer’s well-oiled Rube Goldberg machine.
Cutting-edge data innovations drive those cogs’ work too. Dave Anderson, data engineer at Optum Labs, works on The Natural History of Disease, an expansive dataset that incorporates 166 million painstakingly de-identified patient records and claims.
NHD provides unprecedented insights into the contributing factors, progression and cost of “just about every condition known to mankind,” says Anderson. “White papers live and breathe in [NHD’s] data: while the dataset doesn’t quite write white papers for you, it can certainly tell you if you need to write a white paper.”
NHD is basically a time machine that begins at the day of diagnosis, then works backward through the preceding months and years. The system organizes all the risk factors and “future signals” for the diagnosed condition — for diabetes, issues like gout, obesity, sleep apnea — by frequency and cost. An “alluvial flow” view shows the most predictive prior conditions — those most likely to turn up in patients with latent or undiagnosed diabetes.
It’s not that providers and insurers don’t know the risk factors for diabetes. They do. The real problem is that diabetes often goes undiagnosed for years, wreaking havoc on patients’ health in the meantime. In NHD, the line graph of diabetes patients’ healthcare costs spikes dramatically around the day of diagnosis — not because diabetes is expensive to diagnose, but because it’s usually diagnosed incidentally, in the course of inpatient treatment for acute issues like heart attacks and strokes. It doesn’t take a data scientist to see how that’s good for patients, providers, and payers.
NHD is no substitute for clinical research, at least not yet. But it’s a tantalizing glimpse at a possible healthcare future where providers, health systems, and insurers are empowered to leverage indisputable correlations to dispassionately deliver appropriate preventive and acute care at acceptable cost.
Optum Labs is working on forward-looking solutions too. For instance: a predictive model that flags congestive heart failure (CHF) patients for hospitalization risk, empowers providers to step in with preventive actions before patients end up in the hospital — where they’re at far greater risk for infection and other complications. In a published study, the model reduced CHF hospitalizations by 65 percent.
“We’re empowering providers to say ‘Come into the office; let’s see how you’re doing and perhaps make some adjustments,’” says Dr. Paul Bleicher, CEO of OptumLabs.
That conversation is a lot easier — and cheaper — to have over the phone.