If there is a single product that embodies the power of data analytics gone awry, it is this: a cell phone cover displaying an adult diaper worn by an elderly man with a crutch.
This is the result of a mischievous bot on Amazon that collected frequently searched images, applied them to iPhone covers, and then put them up for sale. Apparently the bot’s algorithm went rogue, and its chosen imagery took a curious, and sometimes sordid, turn.
Amazon should be forgiven for not catching the bot, called My Handy Design, sooner. A massive wave of consumer data —2.5 exabytes, or 2.5 billion gigabytes — is generated every day, and a significant portion flows into retail.
That wave is expanding as quickly as the online shopping crowd on Prime Day, thanks to mobile devices and the data-gathering technologies they enable. From artificial intelligence to digital wallets, much of the sources creating data today are inhuman. So how do retailers make relevant human connections, not to mention desired experiences, through them?
“Retailers these days mainly face massive machine-generated data. It is projected to occupy almost half of all data in the next several years,” said Sansom Lee, chief scientist at Zero Gravity Labs, a Toronto-based innovation and experimentation group operated by LoyaltyOne. “The challenges are size and relevancy.”
Asked how retailers can clean out the “gunk” in all this data and get to the good stuff (meaning no phone cases featuring adult diapers or collections of toilet paper tubes), the team of data experts at Zero Gravity Labs offered these four steps. Combined, they help locate the best insights.
- Let the machines deal with the machines. It’s pretty straightforward: The larger the group of machines working in unison, the more data they can manage collaboratively. When the data surge reaches fire hose scale, retailers can turn to distribution systems that portion out and store the data in multiple locations and process it similarly across various nodes. On this front, the cloud comes to the rescue.
- Let the math deal with the complexity. A challenge of vetting such massive sets of data is that the rules designed to detect anomalies (called a rule-based system) can’t keep up with the task. The solution: statistical methods that filter and smooth out the signals.
For example, if the shopper data varies for some reason — let’s say seasonality — the rule-based systems likely would not be able to cope with the changes. Statistical methods can, however, by selecting random windows of data through which to detect the anomalies and derive conclusions. Statistical methods could detect whether a Best Buy store that posts high sales volume during the holiday season is a high-performer in general.
- Let deep learning deal with discovery. Next, retailers can seek out behavioral patterns in larger data sets by using deep-learning networks. Deep learning is the product of neural networks — advanced systems of hardware and/or software patterned to operate like neurons in the human brain. Self-driving cars use this technology to gauge the environment and steer themselves.
Neural networks can decipher various forms of data, including detecting sentiment in human dialogue. The technology is not a far reach and, importantly, the learning curve is not steep. An online course or two would be enough to understand the basics of deep learning and possibly to build some models to act on a company’s data.
- Let artificial intelligence deal with humans —to a point. From digital house-remodeling to virtual fashion shows, artificial intelligence uses these perceptions enabled by deep learning to emulate human intelligence. When used to meet specific shopper needs, AI can help retailers interact with customers in ways that demonstrate they are evolving their understanding and driving interactions due to new reasoning and insights. The key is offering something useful, not just flashy (or, as in the adult-diaper phone case, cringeworthy).
The luxury department store chain Neiman Marcus uses AI technology to enable what it calls Memory Mirrors. The mirrors enable shoppers to digitally compare themselves in different outfits by capturing full views that shoppers can share with friends by email and social media. Again, the shared data is used to inform more relevant interactions.
And squeezing relevance out of data is the point of the entire exercise. Data well used — including responsibly used — would produce for retailers and consumers a pattern in which both parties mutually benefit from sharing information. True, it might result in the occasional odd product or rogue algorithm, but risk and mistakes will be part of the process.
Just ask Amazon.