John Sumser, aka HR Examiner, has been focusing on "intelligent technology" including AI and machine learning as it morphs the HCM function. Trends in that function are a forerunner of what we will see across the enterprise. I recently caught with him on his latest report and what he learned at the recent HR Tech event in Las Vegas
John, what are you excited about when you look at "intelligent technology"?
In the HCM world, it's pretty clear to me that data is becoming infrastructure rather than the stuff that passes through the software. Everybody I have talked to said, in one way or another, that first you have the data and then you have the software, so the software sits on top of the data . Getting the data right is the key to allowing the software to do its thing.
The second thing that was pretty uniform, although I had to prompt it sometimes, is that what now comes out of a machine is an opinion. It used to be what came out of a machine was the facts that you put into it. But, it is decreasingly the case that what the machine gives you is the facts that you gave it. It is almost always, at some level, providing an improvement on those facts, some benchmarking, forecast, context or some insight.
Rather than software being a re-packager of the data that you give it, it's now an interpreter of the data that you give it. That's a big change. That's a really big change. And so, people are starting to talk about that. People are starting to try to figure out how you deal with that.
I ran a session a called "How to Evaluate an AI Vendor," The gist of it was how do you understand data quality and what does it mean? Does the vendor understand their data models at scale? If you've got 5,000 people in your company, you're going to have a million data models.
How do you monitor their health? And, how do you make sense out of the variable quality of the overall data farm when you're making decisions?
When you say it's more opinion and less fact, is that surprising to the audiences? How did they react to that?
It was a lightbulb. It was actually a surprise. Nobody -- I've worked a long time to get that idea that simple, and the notion that machines are now not as reliable from a pure fact perspective causes lots of heads to bounce up and down. "Oh, that's what intelligence means."
One of the things that I've been probing a lot of the bigger enterprise vendors is, you don't have enough enterprise class data. Most of your data is locked up in customer, on-premise systems. You don't have permissions to access that. A lot of what they're talking about is stuff that consumer tech is doing, the image recognition, the voice recognition, and so on. There's very little enterprise data that is really being used for AI use cases.
You're right. The notion that enterprise technology is going to resemble consumer technology is one of the most damaging ideas that we've entertained in enterprise technology. I was looking the other day. I made a reservation on OpenTable for some restaurant in Las Vegas. OpenTable has handled 820 million reservations. With 820 million reservations in your database, you could do things like predict what the menu ought to be and where you ought to locate the restaurant if you want to have a successful restaurant in Western Sonoma County, California. It's a huge database of consumer interest and preference.
If you've got 10,000 people in your organization, you're never going to have that. You can't get to a trillion data points like OpenTable can in a 10,000-person organization.
What happens in those smaller settings is latency becomes a primary problem. Fidelity becomes a primary problem. You don't have those when you have a ton of data points.
We're about to see some expectations being reset in the enterprise space. That doesn't mean that there isn't cool stuff being done. It just means that you will not see what Cornerstone is busy suggesting that they're going to have a recommendation engine for learning videos. You can't do that with little tiny clients.
If you took all the clients in the world, you wouldn't have enough data to make sense out of it. If you had all of the employee data, employees don't handle their interactions with the company in the same way that they do with a consumer product. They don't have daily interactions in their learning system.
I don't have a clear picture of how it shakes out. But, the difference between consumer technology and enterprise technology is going to become increasingly apparent.
Well, tell me where it is promising in HCM AI
First of all, if you wanted to find where a technology rooted in statistics was going to take hold, you'd look at insurance and banking, maybe pharma
So, Allstate Insurance is a really good example. At Allstate, they have built by integrating - it is a patchwork quilt of vendors integrated into a single system that can go from identifying names of people who might be good employees all the way out through the entire acquisition and development process to identifying the kinds of skills that you'll need in the future and the sorts of places where you're going to need to open offices. And, if you get to the place where you're opening offices up, it can help you navigate all the way through the staff problem. That's pretty interesting.
What's even more interesting there is that team is run by librarians. The team that built that structure are the librarians they hired. There are five librarians, five Masters of Library Science on the team.
The reason that's interesting is because, in order to use the next technology, you have to be better at asking questions than having answers. This is what it means when you say that the machine gives you an opinion. But, if you have a boss or a very upset spouse, these are the times when you only need one opinion to make a decision. Every other time, you need multiple opinions when you make a decision.
The worker of the future is going to be better at gathering opinions in order to make decisions. Places where it's easier to have ambiguity be the center of the equation are the places where this stuff is going to tick.
There is a vendor called Bridge that has an integrated LMS and performance management system. The idea of performance management stops being, "Here's how you screwed up and missed your objectives last year," and starts being, "Here's how we're developing you." The manager gets coached through the available material in order to bring the employee along as a developing expert in something.
There's a great one that's for interview scheduling - Ascendify. Their interviewing system identifies the interviewing team by expertise, gives that interviewing team a set of questions to ask individually, and evaluates the results of the interview to see if the information that was asked for was obtained and then revises the question set for the people down the road. It does this while doing all of the coordination necessary to keep hiring interviews scheduled.
Ascendify only works in the Fortune 100. If you're a big company, that process is cumbersome and expensive and often has ten interviews as a part of the overall process. It takes all the administrative load out of that. That's significant. The estimate for Microsoft's interview coordination budget, so just scheduling and rescheduling the interviews, is $20 million. They can zero that out and give you a faster process.
The first wave of things look like those sorts of incremental improvements to existing processes. Over time, what's going to happen is the silos, which look to me like things that are held in place by existing enterprise technology, will start to collapse. It doesn't take very long, if you stand far enough away, to see that employment branding, recruitment marketing, onboarding and orientation training- are all the same task. Because they are understood to take a different point in a workflow or done by a different expert, you have the same task performed multiple times by five different people for different reasons. You can consolidate that under a goal that looks like, how do we shorten the time to productivity for a new employee?
If that's the goal of the entire talent acquisition process, including the training and admin functions that wrap around it, then what you do in the initial moments of contact with somebody who might come to work for you are different than what people are doing today. That becomes easier and more intelligible when you've got data as infrastructure underneath it all
I heard you say some customers are doing interesting stuff. Some startups are doing some cool stuff. Where do the mainstream vendors fit in?
The emerging battle is for the replacement generation for enterprise software. The people who are in that battle are IBM, Google, Amazon, Microsoft and -- maybe Facebook.
In that group, the name of the game is selling commodity processing and storage. And so, they all -- IBM and Google being the first into the market, they all have an idea of an overlay of technology that expresses as functionality. They all already have applicant tracking systems. They all, for the most part, believe that the world is better if you have a taxonomy as the fundamental structure for HCM.
IBM and Google are very quietly duking it out over which is the better taxonomy, but it's pretty clear that, in this particular game, taxonomy is going to be the key because you can organize everything around the taxonomy. If you go taxonomy plus competencies in the places where people want to use competencies, you start to have a magic formula. That's the initial data structure.
If you start with the initial data structure, then you can lay workflow on top of that pretty easily. The next generation of large-scale enterprise providers looks like it's coming out of that area, and they all have recognizable teams.
If you look at what they're doing, it's really interesting. The Microsoft people analytics team, which is both in HR and product development, has 140 people in it. People analytics, the HR analytics team, 140 people at Microsoft. At Facebook, it's 60, and that's 60 people doing HR analytics for 30,000 employees. These are environments where budget doesn't mean the same thing as it does in the customers you and I understand, but partly because both companies are using their internal processes as a way to accelerate product development. There's real interesting stuff up there in the sort of serious cloud providers area.
The next tier down, if you look at Workday, even Kronos -- Workday, Kronos, or Ultimate Software, the ones that pop up on my radar as the best of the bunch, it looks to me like one of those situations where you lift the house up and install a new foundation underneath it. At Workday, it's a planning and data management system. And, at Kronos, it's a layer of intelligence that uses 40 years' worth of time clock data to predict some interesting stuff.
I don't expect to see anything resembling innovation out of Oracle or SAP. They won't go out of business because of it. They have a long, long, long tail on their cash cow, but people are going to have to innovate around them in order to get real intelligence in their enterprises.
Then the rest of it is small point solutions and data manicuring plays.
That's another fairly prophetic statement that it is the infrastructure-oriented Microsoft, Amazon, Google, IBM that are leading the way on big machine learning. Are HR executives willing to accept them as their HCM providers?
Well, I think it's interesting. If you look at Google and IBM's strategy, and they're really out in the market right now, they are largely going for a sort of a Trojan horse play. IBM is inside of Workday and inside of Cornerstone. Google gets their foot in the door by being the frontend of employment websites for 3,000 companies.
I don't think that the way that it happens is these companies come out of the gate on day one selling a comprehensive enterprise solution. They come out of the gate on day one solving actual, significant problems with a meaty difference so that they have the relationships in place that they can build on. They will ankle bite their way into taking accounts. It's not going to be an RFP level competition for a long time but, by the time it becomes an RFP level competition, the game will be over.