When it comes to the long-promised mass adoption of artificial intelligence (AI), many have been left wondering what exactly is holding up progress. On one hand, AI is expected to have a $13 trillion impact on the global economy by the end of the next decade. On the other, 77.1% of companies report that business adoption of AI initiatives remains a major challenge.
Part of the problem is thought leaders, the media and even the public are talking about AI fervently, not pragmatically. As futurist Martin Ford writes in the introduction to his new book on the subject, Architects of Intelligence, “The result has been a sometimes incomprehensible mixture of careful, evidence-based analysis, together with hype, speculation and what might be characterized as outright fear-mongering.”
Detractors and supporters alike find themselves asking similar questions: How can AI hold the key to long-term success while potentially eliminating thousands of jobs? Why is it touted as a panacea for operational inefficiency when it calls for dozens of tools and constant upkeep? And how far are we from the AI finish line?
At A Crossroads
As we begin a new year, it is once again time to reflect on these questions. There seems to be a consensus among research firms: Enterprises are still clambering onto the artificial intelligence bandwagon, but they haven’t yet made notable progress. According to Forrester, if they are able to “claw their way out of data debt” in 2019, companies will start creating “more potent building blocks” to meet the promise of AI. Last year, the firm warned that the AI honeymoon was over -- choose to naively celebrate the benefits of AI without putting in the work to support it, and face inevitable failure.
Similarly, a recent survey of top-performing companies conducted by Gartner found that 40% of respondents believe AI is a game-changer, up from 7% in 2017. That’s a massive uptick in confidence, “but there is a danger of AI being overhyped in the short term and underappreciated over the long term,” analysts cautioned.
Unsurprisingly, the search for companies that have mastered the science of enterprise AI yields few results. At Oracle OpenWorld in November 2018, I had the rare chance to sit down with the AI point man of one such company: Eric Colson, chief algorithms officer at personal styling subscription service Stitch Fix. Stitch Fix is part of a new breed of companies built entirely on the premise of algorithmic decision making. In fact, Stitch Fix now employs over 100 professionals it dubs “algorithm developers.”
During an OpenWorld fireside chat, I asked Colson, who is also the former Netflix vice president of data science and engineering, to share with me and a small audience of data professionals some of the common misconceptions companies have about AI. While Netflix is often touted as a leader in this space, Colson said, “At Netflix, we used to joke about [how] we’re not very bold about our recommendations because we show about 100 at a time. If we were more bold, we’d show just one.”
Yet Netflix values its recommendation engine at $1 billion annually. That’s because AI isn’t always about bold moves; it’s often about incremental change. Some of the most successful implementations I’ve seen don’t look cutting-edge on paper but result in measurable cost savings or a lift in revenue.
Laying The Groundwork
Colson claims his secret to success is hiring the best people and giving them the leeway they need to experiment. But he also has the privilege of leading a team at a company where the data is so clean it “looks fake,” and data scientists can manage a project from prototype to deployment thanks to the support of a dedicated platform team.
For others, the road to AI is hardly paved. One online auction company I worked with struggled to forecast revenue until a data scientist stepped in. He created a model that predicted the value of 80,000 individual auction items each month, the outputs of which were shared directly with executives.
Two months in, leadership was using the model to predict the total dollar amount of monthly sales with more than 90% accuracy. In the AI world, 90% accuracy isn’t that impressive -- until you consider that this company auctioned everything from end tables to cars to one-of-a-kind artworks, each with a starting bid of $1. Unlike traditional retailers, it simply didn’t have the ability to create standard pricing. This model provided the first glimpse into what being data-driven could really mean.
Laying the groundwork for AI means different things for different companies. It could mean something as simple as automating what products end up on your homepage, or it could mean managing a 40 GB-per-hour stream of data and retraining models daily in order to return results to customers in under one second. Either way, everyone working in this space today is pioneering systems that will inevitably become the future standard.
The Road Ahead
In terms of great strides in enterprise AI, it would appear that 2018 wasn’t a banner year. But on a personal note, I was lucky enough to reach a significant milestone last year: The company I founded, DataScience.com, was acquired by Oracle. Now, as vice president of product management, machine learning and AI at a Fortune 100 tech giant, I get asked every day by people all over the country to reveal the secret to AI success. The answer is there isn’t one.
Take away the buzzwords, and AI is about careful orchestration, experimentation and, most of all, teamwork. The best thing you can do today is to get your big data engineers, data scientists, software developers, IT experts and decision makers together to explore how you can make the most of the data you’re already collecting. If you don’t yet have the right team assembled, all is not lost: There are plenty of out-of-the-box solutions that can help you ease on down the road toward AI competency. Just don’t wait too long to shift into high gear.
