If Luming Wang knew about the news that was set to rock Uber just as he delivered a speech in Toronto, he certainly didn’t show it.
Wang is the head of deep learning at Uber Technologies Inc. He took the podium at Big Data Toronto around the same time the New York Times broke the story that Travis Kalanick had resigned as Uber’s CEO.
When Wang stepped up to the stage at about 8:30 a.m. (or 5:30 a.m. back at Uber’s Silicon Valley headquarters), did he know his boss had just stepped down? Was the impact of that news already weighing on his mind?
If so, Wang didn’t let on. But he did share his thoughts on other pressing issues, namely how enterprises can deploy artificial intelligence (AI) to fit their business model. Here are Wang’s top AI tips, plus other highlights from Big Data Toronto.
While AI is all the rage right now, it’s evolved beyond some homogenous catch-all technology. With a nod to that, Wang pointed out some key distinctions between deep learning (DL) and machine learning (ML) applications in the enterprise.
For example, he said ML is usually a good fit if you’re looking for churn prediction, product recommendation, budget optimization, sales prediction, pricing models and ETA. (As you can imagine, predicting estimated time of arrival is kind of a big deal for a ride-hailing service like Uber.)
DL, however, may be a better choice if your business needs to process large volumes of image or sound data.
That said, Wang listed some potential limitations to be aware of before diving into DL solutions: DL hardware still hasn’t “caught up” with DL software; DL training takes a long time compared with ML; DL modeling can be 10 to 100 times slower than ML modeling; as with any emerging technology, it can be tricky to call in extra DL help when you’re stuck because “there’s not a lot of people who have experience tuning a deep learning model.”
One of the coolest things about AI is that it allows you to use vastly different data sets in ways that either weren’t previously possible or never occurred to you before, said Dalia Asterbadi, CEO and chief data scientist at Verve.ai, a Toronto firm that specializes in AI-based customer data analytics.
Taking part in a C-suite panel at the conference, she said companies should enhance their analytics by thinking of creative ways to harness diverse data sets they haven’t crunched together in the past.
For example, Asterbadi said a new restaurant might not have a high credit score simply because it hasn’t been in business very long. If a bank included social media reviews and Yelp ratings in its analytics, however, it could use that data “to supplement their traditional credit score” when deciding whether to give the restaurant a loan, she suggested.
Time it right
Don’t relegate data insights to an afterthought; by that point, it’s already too late.
“A lot of times, data science is treated as a post-activity function at the tail end of your campaign or whatever to say ‘we found out why we lost this customer,’” said Asterbadi. “But the moments that make an impact are when you focus on your customer’s life events, not your company’s.”
There are, of course, limitations to how ‘real time’ any analytics program can actually get. The ultimate goal is to someday have widely available software that delivers insights pretty much as they happen, said fellow panelist Bijan Vaez, co-founder and chief technology officer at EventMobi, a Toronto firm that makes mobile apps and platforms for event planners.
For example, if meeting planners can analyze real-time data on attendance, foot traffic or lineup waiting times, “will that tell me how to change my event in real time?” Vaez wondered aloud. Event organizers could perhaps shift to a smaller or larger meeting room within their venue, he said, or attendees could make sure to network with a certain convention delegate when notified they’ve arrived.
Your AI solution may be able to spit out a dazzling array of charts, dashboards and pie graphs. It may even attain 100 per cent predictive accuracy. To which Wang retorts: “So what?”
“The customer doesn’t care about that,” he reminded the audience. “Customers won’t ask for machine learning specifically. They just ask for better products and better service.”
Remember: data has little value if it doesn’t help your business. “A common mistake is using a non-business-related (AI) goal,” said Wang.
Focus on people, not just numbers
A common goal of many businesses, of course, is to cut costs wherever possible. Yet Wang cautioned the crowd not to view AI as a tool to slash the bottom line or reduce staff headcount.
“Try to help people, not just replace people, in the beginning,” he said, with the word ‘people’ meaning employees. “Try to minimize cost while still reaching the (business) goal as closely as possible.”
See how Wang qualified his comment by adding the phrase ‘in the beginning’? It’s his subtle acknowledgement that AI and automation will eventually replace many people’s jobs — more than five million of them by 2020, according to the World Economic Forum’s estimate.
Ignore all that, said Wang, and look at how you can use AI to help your employees (the ones you still have right now) do their jobs better.
And then Wang exited the stage, while half a continent away, his boss was exiting Uber, the search to replace him with another CEO already underway.