The truth is you can’t go through a day without hearing another twist on the wonders of artificial intelligence and machine learning in the world of e-commerce.
AI is B-I-G — whether it’s self-driving cars, playing Go, diagnosing disease or writing news stories.
In fact, the machine learning and AI buzz around e-commerce has been so pervasive for so long that it sometimes feels like we’ve been talking about this forever. But it’s helpful at times to slow down and consider not only how far we’ve come, but how far we have to go when it comes to automating digital commerce.
Let’s start by separating the hype from reality with the help of an AI expert who’s seen this happen before (in the digital advertising field).
Long-Ji Lin is Signifyd’s chief scientist and a guy who thinks about machine learning a lot. He holds a doctorate in computer science and artificial intelligence and has been working in the field for years. He’s a big believer in the power of machine learning to augment human intelligence and, naturally, he’s especially bullish on the technology’s ability to improve fraud prevention.
Think of the problem: many millions of online orders flooding into merchants daily. Fraudsters lurking, constantly upping their game in an attempt to beat the barriers to fraud that merchants erect. And so how do you comb through the orders and determine which are legitimate and which are fraudulent?
“There are so many people out there trying to cheat,” Lin says of the fraudsters who prey on e-commerce merchants. “It’s not feasible to hire a lot of humans to review orders and make decisions. It’s just too much work.”
And so humans devise the models that can spot trouble. They teach the machines to do the big, repetitive tasks and then work alongside them. The machine issues a thumbs up or thumbs down on the overwhelming majority of orders. The trickier cases are referred to a fraud analyst who can turn to human instinct, experience, reasoning and research to make a decision.
Machine learning is our friend, really
Of course Lin is familiar with the school of thought that the rise of the machines will one day be detrimental to the human race. Elon Musk is a well-known machine-learning worrier and MIT physicist Max Tegmark is out with a compelling book, “Life 3.0,” which envisions a world in which machines do not play nice.
“Think drones the size of bumblebees that could be programmed to kill certain people, or certain categories of people, by grabbing their skulls with tiny metal talons and drilling into their heads,” is the way a Wall Street Journal review of the book describes Tegmark’s machine-crazy future.
Of course, machine learning and artificial intelligence haven’t achieved anything close to the kind of sophistication to create such a nightmare scenario. We’ll take that as good news.
Even a less dystopian view — a time when machines in a sophisticated way teach machines to get better at what they do — is a ways off, Lin says.
“We still have challenges teaching machines, let alone teaching machines to teach machines,” he says.
Machine learning today isn’t about the scary rise of the machines that tech luminaries like Musk worry about. Machine learning today is a way of scaling a business. It’s a way for digital retailers to increase revenue while providing a better shopping experience for their customers.
Increase revenue and better the shopping experience how, you ask?
Consider fraud protection, just one slice of the machine-learning revolution as it relates to e-commerce and one that Signifyd is particularly interested in. How can a human-powered or a rules-based system keep up with fraudsters who are constantly changing and improving their tactics and techniques? How can merchants using traditional systems keep up with rapid growth in orders — the kind of growth that’s good for business, as long as the orders are legitimate.
A well-crafted, machine-learning-based fraud prevention system allows a business to scale. It dramatically reduces the number of orders that need to be manually reviewed, while also dramatically reducing the number of orders that are incorrectly declined for fear of fraud.
Fewer declined orders and real-time decisions on whether to ship mean fewer disappointed and frustrated customers wondering why their stuff never came or took so long to arrive.
And yet, the proliferation of machine learning technologies in a broad swath of industries is relatively new. As always, change requires new thinking. But it’s fair to say that among the leading themes in retail today is the need to embrace innovation, to behave like a startup, to be willing to try, fail and iterate.
Fraud prevention has turned to machine learning
Even big consultancies are paying increasing attention to machine learning’s role in fraud prevention. Forrester in its June 2017 “The Total Economic Impact of Guaranteed Fraud Protection,” a commissioned study conducted by Forrester Consulting on behalf of Signifyd, looked at how machine learning, paired with a financial guarantee for approved orders, can bulk up a retailer’s bottom line.
Others have suggested that such systems can run in parallel with legacy rules-based systems to help merchants handle seasonal spikes, such as the holiday shopping season. And they’ve noted that running the machine-learning models in parallel provides an opportunity to test the efficacy of the cutting-edge tools.
Consultants have noted the huge buzz around machine learning and warned retailers to carefully evaluate the alternatives, making sure vendors can explain exactly what they mean by machine learning. And then, of course, comes the work of determining whether their technology is worthy of the phrase.
“All these companies are saying they have AI,” Lin says. “But some have real AI. Some have fake AI.”
His advice? Remember that the quality of a machine learning technology comes down to the quality of the algorithms and the quality of the data. Bigger companies have more data. More data means more revenue. More revenue means better talent and better algorithms.
And of course, the companies selling AI have a track record and a list of customers. Learn what you can from them.
And after all that, if you’re still uncertain, just ask the nearest smart machine.
Mike Cassidy is Signifyd’s lead storyteller. Contact him at email@example.com; follow him on Twitter at @mikecassidy.