Consumer-facing business have been at the cutting edge of AI adoption, but what about bringing AI to the enterprise? In the last volume of my interview with Kathryn Hume, I ask how AI in the enterprise can become a market reality. What are the barriers to entry, and are enterprises further along in their AI adoption cycles then they themselves even think? Read on to find out.
And if you missed it, here are the first two parts of our AI interview with Kathryn:
- Part 1: How Artificial Intelligence is changing everything
- Part 2: The Ethics of AI and AI in culture
Part III: AI in the Enterprise and the Value of Location-based Data Sets
Let’s move from ethics to revenue. What are some of the obstacles to machine intelligence adoption in the enterprise? I’ve seen a lot about the penetration of data science in (mostly) consumer-facing businesses, but what about AI in the enterprise?
To date, most of the great business applications of AI have existed in consumer-facing companies, like Google, Facebook, Amazon etc. And the enterprise is just now getting excited about AI. So what are the challenges? First of all, a system is only as smart as the data that it’s trained upon. And while everybody thinks they’ve got their data act in order, most of the time enterprises have a long way to go until they have a robust, large and clean enough data set to make headway with these algorithms.
At my company though, we work with the assumption that the data is never going to be perfect. And there can be some paralysis through perfection, where people assume that they can’t even begin with data science until their data is perfect, and they’ve got their enterprise data lake, and everything is ready to go. Companies will postpone experimentations with algorithms until the foundations are in place, and absolutely foundations are really critical. But even the great consumer internet companies have to deal with the fact that data is always a mess, and you’re always doing the best you can with the resources that you have.
In keeping with that, there’s another misunderstanding: people think that AI is like the Matrix, or they just flip in their chip that teaches them karate, and suddenly they are karate experts. That thinking leads to ill-informed project scopes and structures. And that’s why there’s a lot of value to putting what we call a human in the loop. If you don’t have your data all ready yet, or if you don’t have enough to train an algorithm well and get the level of confidence and high-quality outputs that you want, that doesn’t mean that you can’t do anything. There are clever means to design staged processes, where you can gradually extract more and more information from your users, and to one day get to a system that will be completely automatic. The first stage though will be collaborative between the people in the enterprise who know what they’re doing and the machines that can gradually start to encapsulate that knowledge.
There’s a fear though. Everybody is scared that robots are going to take their job, and I think a healthy appreciation of just where this technology stands, and that it can automate some pretty boring and repetitive tasks that most people don’t really want to do anyway. Honestly, there is risk for certain types of work, like truck driving. That’s a big thing with self-driving cars coming down the pike. But in a lot of enterprise work, there will be subtasks that might be amenable to automation, but very infrequently the full spec of jobs.
Finally, in the enterprise, there are essentially two approaches to AI. You can treat it as a sustaining innovation, where it just automates, drives efficiency or does something that you’re already doing a little faster or better. Or, AI can be something that’s a little more disruptive, in the sense that it can open up new product offerings, new possibilities, new ways to package and set pricing around an offering, based on the shift in how something might get done.
Or, as we’re focused on at Integrate.ai, AI can focus on new ways of flipping your perspective to be more customer-centric as opposed to product-centric, so as to really shift around what it means to market, sell and enable customer success as well as enhance customer service to enable your customers to achieve their long-term optimization goals.
Switching tracks, we reveal a lot of information about ourselves through location-tracking apps. But I often think that that kind of information really only replaces demographic information that brands could have formerly obtained through surveys. What do you see as the value that AI can build on top of those location-oriented data sets?
I’m really interested in using people’s behavior as a proxy for some sort of psychographic trait that they have. Location information can tell you things like dwell times, speed, regular paths that you might take; it can basically be an index for your habits.
To date, a lot of marketing segmentation, as an example, has been focused on traditional demographic data coming from the Census, where there’s this assumption that just because you live in a certain zip code, that you’re going to have similarities with the people around you, same income, same gender etc. We create these heuristic-oriented segments that have guided a lot of “personalization” in marketing. But you know, I live an apartment building, and I guarantee you that I live quite differently from my neighbor. Sure, there is going to be some things that are similar about us, but there are going to be many things that are different. Meanwhile, there might be somebody in Norway who happens to like the same literature that I like and has similar habits.
What’s interesting about where you move in the world is that it gives a lot of insights into the real you, as opposed to the “you” you project. There’s interesting value in that. Also, as businesses are doing all sorts of things, like inventory planning and time-of-day optimization, deriving greater knowledge on what, when and where people are actually moving can lead to surprising conclusions.
In the case of iPass, and maybe it was at a convention at some hotel, suddenly at 4:12午後, everybody was clustered in one area. And why were they clustered in one area? They were at the UPS stand, sending back marketing booths or something like that. And it’s like, of course, they’re there doing that. So you see these things in location data that no one would ever think to think about, but that become completely logical in retrospect.
That happens a lot with data and marketing. For example, I was talking to one of our data partners recently who had done a campaign for L’Oreal, I think, or some other makeup company, on when they should send out Instagram promotions. They thought that if they were doing makeup, the promotions should go out Friday night, when the girls are getting ready to go out on the town. That sort of seems to make sense. But no, once they looked at the data, it was like a Wednesday evening or a Tuesday evening, when people are home, and they’re bored. Those are the times that they’re most apt to look at makeup ads.
In retrospect, you’re like, that’s the clearest thing in the world, of course Friday was not a good day, because people are actually getting ready, or they’re out and they’re not scrolling through their phones. So I ultimately see the Sherlock Holmes-value in location-oriented data, where it’ll lead you to the thing that’s the clearest in the world, but that you would never have thought was the truth.