E19 MIT, AI, & the Power of People Analytics
In this interview, Ben Waber - an innovator in the human capital space and co-founder of Humanyze, discusses the power and potential of people analytics. Ben explains the innovative uses of AI-powered behavior analytics. He describes numerous creative ways the technology can be used to make impactful decisions for organizations, the importance of ensuring employee privacy in data collection, and the need for emotional intelligence when leadership interprets the data. Ben also shares his thoughts on the future of artificial intelligence and people analytics.
Ben Waber is President and Co-Founder of Humanyze, a behavior analytics company. For over 15 years, Ben has studied, researched, and implemented new technology to better understand human behavior.
Ben Waber:
Companies fundamentally, for the vast majority of roles, I don't care about what you as an individual produce, I care about what the team, what the organization produces. And so, the challenge is I need to figure out how your behaviors and your work contributes to that whole. That whole is now on longer and longer time scales, we’re making more and more complex stuff. It gets even harder for me to see how your behavior today impacts that. But I think that what is going to start to happen is much more focus on that behavior.
Don MacPherson:
For years, companies have tried to unleash the potential of their employees. Different management tactics have been used, companies have tinkered with the physical environment, and some have tried to design a superior employee experience. But what if the secret to unlocking employee potential is in the mountains of data your organization already has? Our guest today is Ben Waber. Ben literally wrote the book on people analytics. He is President and Co-Founder of Humanyze, a behavior analytics firm. Ben has a PhD from MIT and is a visiting scientist to the MIT Media Lab. He has dedicated his career to using analytics to improve the workplace for both individuals and the companies where they work.
Ben, welcome to 12 Geniuses.
Ben Waber:
Thanks so much for having me.
Don MacPherson:
Let's start talking about your background. Can you talk about where you grew up and where you went to school and what you studied?
Ben Waber:
Well, I grew up in Philadelphia, and I guess I was there until I was 18, and then came over here to Boston where I've been for now over 17 years. Really, what I first started studying was computer science, also Japanese, but that's a separate topic. I started to do research in computer vision, trying to use images and video to, not just understand what people were doing, but really use that to make interfaces for people with disabilities. So, that's what my master's thesis was on. At the time, I started reading papers, obviously in the field. And it was interesting because I was starting to read some of these more recent papers, which weren't about computer vision at all. They were about social networks, about the relationships between people and how that predicted things like depression, performance, all these things.
I started to read about that and got more and more interested in that. And then when I got time to do my PhD, applied to be in the group and was fortunate enough to get into the Media Lab at MIT, which is where I, again, I did my degree there. When I first entered the lab, what we were doing was using sensors in the lab to try to understand the outcomes of specific conversations. For instance, you and I having a salary negotiation, trying to figure out who's going to win, not based on what a human would think of the conversation, but just by automatically analyzing the way we talk to each other. Not even words, just the way we talk. And you could do that really accurately. You could get over 85% accuracy on who's going to win, how much we actually make in a salary from just the way… How much I interrupt you, how do I talk? All those things.
Around that time we had a professor in the business school from Sloan and MIT come over and say, “It looks like really interesting what you're doing in the lab, but I'm wondering what would happen if you collected that data over much longer periods of time in something like a workplace. Now I've got this bank in Germany where I'm looking at all their email data, surveys every day. Again, not content, really just the patterns of communication. But I really think that face-to-face communication is a really important component of all that. Do you think you could use these sort of sensors to measure that?” We'd never done anything like that before. We'd, at most, gotten those badges to work for two hours at a time. So, when people are saying, “Hey, could we get those things to work for a month unsupervised?”
We were a little bit scared of that, but we thought that sounded pretty cool. So, we went over to Germany, we actually rolled them out. We got the sensors to a point where we could actually do that. But at the time, we collected all this data on all these people, and because this was really the first time we collected anything at that scale, I mean, there's very little pre-processing we could do on the data, which meant we had about four gigabytes per person per day of data from across this bank. It was an awful lot of data. We ended up writing a bunch of papers on this. The first paper we wrote, though, we sent to the executive team of the bank because they had let us do this experiment. Again, all general team at higher-level results. They saw this and they said, “Wow, this is amazing.”
And they reorganized the entire company — this is a multi-billion-euro institution — based on the paper that a couple of grad students wrote. That was really amazing to us, right? That this gigantic organization would make a pretty big change based on a paper a couple of grad students wrote. So, we started to do more of that. And towards the end of our PhDs, we were actively changing how companies were being managed based on these analyses. And it was from that that then it made a lot of sense to start the company, which is what I'm still doing today.
Don MacPherson:
So, every person in the organization is wearing a piece of hardware that's tracking this information or tracking movement, tracking conversations. Is that basically how you're collecting the data?
Ben Waber:
Well, so I do want to distinguish what we're doing at MIT. And this is eight years ago what we're doing today.
Don MacPherson:
Correct.
Ben Waber:
At MIT, what we were trying to do was really get as much data as we could about human interactions, that we could understand it as deeper level as possible. What that enabled us to do is first figure out the things like content don't matter, right? They give you very little predictive power and there's lots of privacy implication with that. at the same time, what we do today is use data the companies already have about what goes on internally and use that to do similar sorts of analyses, but we're also spreading software. It's not like a one-off thing. Again, so universally, that's things like email, chat, meeting data. On the sensor data side, it's really interesting. If I go back eight years, when we started the company, I mean, there's very little sensor data that companies had internally.
They had some things like badge in, badge out from buildings. What's been interesting is that, even in the last year, we've just seen a pretty dramatic uptick in the amount of in-office sensor data that really exists today.
Don MacPherson:
For example.
Ben Waber:
Yeah, so for example, I have a couple IDs here, people have to imagine, I've got twos here. I've got one to get into the building, but I've got another one that is just a little bit thicker than…It looks like a regular ID. But a couple million people in the U.S. have those today. Globally, it's tens of millions from a bunch of different manufacturers, Panasonic, HID, Kronos. They don't have things like volume or audio, but they do know, they're using Bluetooth within an office, like roughly where I am. And by combining this different data together, right? For example, imagine you had one right now, we'd see that we're fairly close to each other. We're in this room. By itself that I can estimate some probability that we're interacting, but if you combine that with things like calendar date of course, or how often do we chat? Then you're able to estimate that better.
And what enables us to do is not again, at an individual level, but at a team in higher level, estimate really good probabilities and just the amount of interaction between different teams, different parts of the organization, things like where people spend time. Then using that, we're able to, not really predict large outcomes for organizations, but really what that enables you to do is, particularly when it comes to the physical workplace, is really understand, how can I change that to change dynamics of how people actually collaborate, and look at that, not just on the digital side, but on the face-to-face side as well?
Don MacPherson:
I want to get back to MIT and your time there. You were there for five years doing your PhD, is that right?
Ben Waber:
Exactly. I would say I'm still visiting scientists there, so I can legally hang out there. I actually teach the MIT executive people in all their course still. I still go over there occasionally.
Don MacPherson:
You're a busy guy.
Ben Waber:
Yeah.
Don MacPherson:
How would you describe the approach to creativity and innovation at MIT?
Ben Waber:
The Media Lab is sort of a weird child of MIT, but I think lots of very interesting things came out of the Media Lab. But I'll tell you a story of one of the most interesting things that came out of the Media Lab and sort of this really emblematic at the way that things work at the lab. I mentioned before, the lab is funded mostly by corporate sponsors, but they don't sponsor any specific project. That's not the way it works. You can only put your money into a general pool, and then each of the groups is just given a cut of that and they can do whatever they want with it. The thinking behind that was that if a company already knew what they wanted to research, it's probably not that interesting. In one case, so one of the groups at the Media Lab, one of the things they do is make new kinds of musical equipment, new instruments.
And one of the things they made was this thing called the hyper cello. I don't know if anyone plays the cello or any string instrument, you know that you can actually, with different physical movements, play the cello and make the same sound. If you can see the musician, you can tell they're playing it in a different way. But if you're just listening to the music, it could actually sound the same. And they wanted a way to encapsulate those differences. What they did is then added some sensors to essentially the bow, and then it modulates the sound that actually comes out of the cello. It's pretty cool. Again, it's a fun project. It was interesting because at the time, a researcher from NEC, a big Japanese electronics company, was in residence at the lab and saw that and said, “Hey, do you think you could put that same sort of sensing technology in a chair?”
Because what I'm trying to do is figure out in cars, if there's a kid in the front seat of a car, such that we don't deploy the airbags. And the researchers are like, “Yeah, probably. You could probably do that.” One thing that they did is they just, I thought it was interesting, they made a chair. They've sort of converted the hyper cello to like a chair. So, you don't even need the bow. You could just sit in the chair and play it in a different way. But then NEC took that and built it into their car products. And now today, if you were in a car, the sensor technology that detects, like you're in the car and you put your grocery bag in the front, maybe it thinks there's a kid there, but pretty good. That's because this group of the Media Lab made this next generation cello.
And that's really the ethos behind the lab is that the really interesting things, they're going to come from weird places, but because you're really valued on not just writing papers, but also a couple times a year, all of the sponsors, all the members in the lab, all these companies come by and you have to show them your research and talk about what you do, then they’ll, of course, tell you, “Well, hey, if you did things in this way, it’d be more relevant for us.” And of course, as a researcher, you want your work to be relevant to more people. Now maybe they're going to do or suggest something that is not really what you're doing or really interesting, but a surprising percentage of the time, it actually does help change your research in certain ways. But these are things that, within a company, just would've never been created, because really what you're doing is you are creating solutions that you don't know the problems for, and really work with people with many different backgrounds. So, you just get a very interesting mix of things that happen there, which is probably why the lab keeps making up pretty cool stuff.
Don MacPherson:
So, if I'm to summarize this, you have very smart people from diverse backgrounds. You give them a lot of space, and then they present it to people who imagine how it might be applied commercially for their organization. And then maybe there's a refinement period in there?
Ben Waber:
I mean, the things that come out of the Media Lab are almost never anywhere close to a finished product, but they are going to be the seeds of ideas that then some of which could be refined. And whether that's people spinning off companies, which about a third of Media Lab graduates do, or that's going into industry, which again, another third do, or some or the rest typically go into academia. But that really provides a foundation for getting to a point where, yeah, this is something that could be used more broadly in the real world, right? Because it's been validated at some level, or at least created at some level, and has been demonstrated to show, “Hey, this actually could work.”
Don MacPherson:
Now you are President and Co-Founder of Humanyze. What is the creative process here?
Ben Waber:
Here we have both the near-term things that we're working on, which is very much on a biweekly basis, updating both our technology as well as the products that we provide to our customers, and as well as we have on the R&D team, especially much longer-term efforts. And some of that is directed towards, “Hey, this is going to go into a product in the next year or so.” There's other things that we work on that we're pretty sure in the next couple of years will eventually be some sort of product, but we're not sure. And at the very least, in the near term, there are things that at least our customers are interested in. So, we, at this point, because we're deployed across some of the largest companies in the world, across like every single employee, at over a hundred thousand employees at some of these companies, we have all this data about work from across organizations.
That means there's all this interesting stuff we can learn about how people work, some of which could be fed back in a product. But other things where it's unclear in the near-term how we could do that, but what it enables us to do is not just for ourselves, but also for our customers. And then, even at a broader level with the public, try to share some of the things that we're starting to see, and also help them start to see, well, here's where it looks like, not just work is going, but here's… I should probably think differently about work and about management. Because we're able to pretty authoritatively say certain things about what makes effective work.
Don MacPherson:
Is it accurate to say you're taking the data that exists within an organization and helping them synthesize it so it makes more sense and associate the data with key organizational outcomes?
Ben Waber:
There's an interesting process for companies trying to use this technology, okay? Because again, I can go to any company in the world and asks pretty basic questions about what goes on internally today. They can't answer. But people have never seen these kind of metrics before. I could ask you how much management talks to sales and you can't answer, can't even answer how many hours people work if you have information workers, which is crazy. And so, then if I ask you, “Okay, well, what percent of the time should the engineering team talk to the sales team?” You might be able to ballpark it, but you can't actually give me a hard answer. No one can say, “Ah, 32% or 64%.” Like, you're totally guessing. Those are the sorts of things that we measure and that we can't provide. But typically, when we start working with companies, like what we're providing with them, the software we're providing is quite simple.
The vast majority of the time, the way companies start using our technology is really around workplace decisions. And why is that? First of all, those are gigantic investments. You're building a new headquarters, you are restacking existing location. Or a Fortune 500 company, you're probably spending on the order of a billion dollars on that. So, it's a big investment. Fundamentally, you are making that investment to change how people work. Of course, if you're head of corporate real estate, if you're in HR and you're helping with that process, you have an existential issue. You're paid more if you cut your budget, which, of course, this means everyone eventually works from home. And for some organizations, that could be right, but if you're investing all this money in space, you're doing it because you believe that's going to change how people work. Of course, what that means is increasingly, CFOs, CEO are saying, “Okay, prove it.”
And you can't without these kind of metrics. And so, at first, just being able to say, “Okay, well, here's who works with each other. That's from all the data that already exists. We're just able to analyze it and understand it. And so, here's who it looks like should sit together. So, here's how big that place needs to be. Here's roughly who needs to sit together.” And then, yeah, if you actually change a couple different configurations, so rather than, maybe I have 1,000 people here, or maybe I have 2,000 people here, here's how that would change all these different metrics. That's a useful thing to do that dramatically speeds up the process. And that's the firs, this is a fraction of what we can technologically do. The next step tries to get more interesting. That's where you say, that's where these sort of metrics become more and more normal.
And we've only got a couple of our customers there so far. Where they get to is essentially using this technology to really AB test how they manage the business. You have enough people, you don't even need gigantic numbers. You got a couple hundred people, you can start to test things. The issue has always been the speed with which you can test it. But if I can say, I am implementing this change to get this behavioral change, right? Of course, eventually, you wanna relate that to some quantitative outcome. Maybe it's turnover, maybe it's performance. But if you know, hey, this behavior predicts those things in long-term, then I'm reasonably sure that it's going to have an impact. That means you can much, much more quickly run through these things. You name it.
Don MacPherson:
And how long do these tests typically last?
Ben Waber:
Rough rule of thumb is if you're making a change to a certain… You want to change a certain behavior, I want to give it at least a month, but it really depends on the cycle of the business. For some of these teams, for example, they're working on projects, using project management software that could take years. And you don't want to necessarily wait years, but you could at least, over the course of one cycle, that project might be a couple months. I want to see one of those cycles, at least.
Don MacPherson:
I have a question about performance management, but can you talk about how you're helping to make performance management more objective?
Ben Waber:
Performance is a difficult thing to define for information workers. It's something where a lot of times we feel like we're being productive if writing emails, we're banging out reports, but eating lunch with a coworker or having coffee with them, that's not productive. When I could argue, it's almost the exact opposite. Increasingly, the value that we create, it's not on our individual work. It's how we enable everyone we work with. And, of course, at the end of the day, some individual work needs to get done. But it is interesting, so if I look across our data, over the course of a year, you pretty much max out at about 40% of your time in focus work, so it’s 15 minutes or more of uninterrupted work.
There are, of course, periods of time where you can go beyond that, but again, I can look over literally tens of millions of people. You're not going to get much more than that. Given that there's just 60% of other time, it's really about coordinating and interacting with other people. And companies fundamentally, for the vast majority of roles, I don't care about what you as an individual produce, I care about what the team, what the organization produces. And so, the challenge is I need to figure out how your behaviors and your work contributes to that whole. But I think that what is going to start to happen is much more focus on that behavior, and that here's the way that we'll likely need to collaborate to be effective for making this thing. Are we matching that or not? And then, of course, what we need to constantly check as an organization is, are we getting that right? But that means then, looking at those behaviors, of course, at a team in more macro level, but then also filtering that down to individuals.
I think the challenge there is, does that mean you’re exposing individual behavioral data to companies? Which is something that then gets very sensitive. Is it something where-
Don MacPherson:
From a privacy perspective.
Ben Waber:
From a privacy perspective. Or is it something where you're not exposing the actual data itself, but a high-level metrics that help both people and the companies figure out here's how cohesive your network is. That here's to everyone you talk to. Here's how cohesive your network is, here's the breakdown of your time between meetings, focused work. Again, so that, how does that fall compared to what the group as a whole wants? While at the same time acknowledging, and this is the challenge, that there really needs to be very high variance there, that each role is unique, even if I have the same job title as someone else. And so, ensuring that people are paying attention to those behavioral metrics, but at the same time that you're contextualizing that for each person. Again, it's sort of unsatisfying because what I think a lot of people want, I want some number that’s going to tell me the truth. And that's just not how it works.
Don MacPherson:
Today's guest is Humanyze President and Co-Founder, Ben Waber. When we come back from this short break, Ben will talk about the future of people analytics.
Hi everybody, this is your podcast host, Don MacPherson. At 12 Geniuses, we write, report, and speak about the trends shaping the way we live and work. As we look toward entering a new decade, technologies like 3D printing, artificial intelligence, gene editing, and more and more sophisticated robots will continue to disrupt and change our society. If these trends are important to you, we invite you to follow us on social media. And to book me to speak at your next event, contact us at future@12geniuses.com.
We are back with Humanyze President and Co-Founder, Ben Waber. In this segment, we discuss practical applications for data collected in the workplace, employee privacy, and how AI can be used to augment leader decisions and employee behavior.
And how can analytics be used to minimize or reduce or eliminate bias? Or is that impossible?
Ben (21:25):
Okay. How would I define bias? I would define bias is if behaviors are the same and outcomes are different. That's measurable. It's the sort of thing where I can look across our customers and I can say, all right, let's say I break people down by gender, by different demographic groups. Do they behave in similar ways? Do they have different outcomes? And we can see that. Absolutely. It's interesting if, again, generality, and we obviously don't have a representative sample of the entire U.S. or global working population. What we tend to see is that for large portions of many companies, actually, there's very low levels of bias, but there are pockets that are pretty severe. And even the low levels, again, the issue here is that a lot of the differences you can show with analytics and bias, they're not something that I can just flick a switch and change.
If, for example, I show women are invited less to meetings than men, it's one of the things we can show. You could fix that. I mean, I could mechanically just fix that. Would that fix the differences in pay? Nah. It would've some effect, marginal effect, right? But the fact that exists is not because people are actively choosing to, for example, exclude women from meetings. It’s their subconscious biases that exist that then again, with data you can show. Now, what you can see is then, so I wouldn't recommend just mechanically changing some of that. There are some things that you should change. People have the same job title. You can show in with metrics, and Salesforce does this, for example. When they look internally, they see, do people have the same job title? Are they paid differently? Yes. Then at the end of the year, I make all those salaries the same. Easy, good, you eliminate that thing, but you don't actually fix what created that disparity in the first place. That you would keep measuring essentially with those analytics, but that requires societal change
Don MacPherson:
And awareness.
Ben Waber:
And awareness, yes.
Don MacPherson:
You're creating the awareness.
Ben Waber:
Yeah.
Don MacPherson:
At least you're identifying…
Ben Waber:
You can create awareness and you can show, in a default state, are we making progress against that? But actually changing it, right?
Don MacPherson:
That’s something entirely different. Yes.
Ben Waber:
Is a multi-generational change, which is really unfortunate, but I think that's what it's going to require. But at least now we do have some metrics that are going to be useful to see how much progress are we making on that front.
Don MacPherson:
How do you think performance management is measured, let's say five years from now?
Ben Waber:
I do think that, again, much more an emphasis on the team and people's contributions to the team. And again, I think a more rigorous emphasis on actual behaviors rather than subjective assessment. I think that's been possible in some roles, right? You look at sales, you look at customer service, you could do that.
Don MacPherson:
All sorts of metrics there.
Ben Waber:
Exactly. And I think it's not going to be exactly the same, but I think you're going to see some similarities. Again, for example, if you're a software developer, there's a question of like, for what percent of code that you write did you talk to the people who also worked on the same code? That is an extremely useful metric actually.
Don MacPherson:
Is that easy to measure?
Ben Waber:
Oh yeah. Then there's one role like, hey, we actually wrote some papers on that as well. For example, I mean, all the code you commit, I mean all your code is obviously, by definition, digital record of it. You can see who else worked on the same code. And through any communication platform, you can say that you communicate. And I mean, there's decades of research on if those things match and really good things happen, and if they don't match, really bad things happen. And so seeing that, whether there's that mismatch or not, how much you contribute to that or not is a really big deal, as one example. But again, I do think it's something where it's hard for… I think a lot of people want, like, I want a general productivity metric in terms of behavior, and that does not exist.
It is going to have to be customized very, very much, down to specific roles and work you need to do. So, I think there is going to be this increasing sophistication around how we think about those things and not… Unfortunately, it's not as easy as me just putting a question on a survey, like, how productive is a person? It's going to require actually a lot of work. But that when you do that, then it's going to make it a lot easier for people to figure out, well, here's what I need to change, and here's as a manager what I can change about the work environment, about the way we manage to improve those behaviors as well. That also, of course, becomes interesting because, not just for my current job and for performance management, my role, but then eventually I could say, “Hey, do I look like I'd be a good fit in another company or another role in my own company?” That makes a lot of sense in that I could say, “Hey, I want to be promoted to be a manager here. What would I have to do? What do the best managers do? How do I work differently than them?”
Or then I could say, “Hey, maybe I want to work at Google or LinkedIn or something. Do I look like I work like those people?” And me being able to use my own data in a certain way to actually index into that. So, it's a little very interesting.
Don MacPherson:
How would that work?
Ben Waber:
I think that from a privacy perspective, it has to work on the search. I can't submit my data. I think there’s a lot of issues there. But if I could say, “Here's how I work.” For people who are rated highly at company X, do I work like them? Because if you do, that's actually a pretty good signal…
Don MacPherson:
That's really interesting.
Ben Waber:
… that you'd probably be really good there. Right now you would need data with enough coverage across enough companies, people to do that. But I think something like that will certainly happen at some point. I think there's a question of time horizon there. Yeah, I think there's a real application.
Don MacPherson:
Where will we be with wearables that work five years from now?
Ben Waber:
Wearables are sort of a weird thing in that there's just more and more data about what we do in the real world, right? And some of that is from devices that we have. Again, there are now these smart ID badges that can do certain location information within the office. There'll be some more of that. People are really good at creating new ideas, at sharing complex information. Computers are terrible at that stuff. Similar way, computer's very good at quantitatively analyzing large patterns, making inferences from those patterns, and measuring things at large scales. And humans are terrible at those things. The combination of those two is very interesting. On the wearable side, the ability to measure more stuff about the real world, and some of that is going to be through, again, whether it’s IDs or cell phones.
Some of it might be through environmental sensors that can look at air quality or temperature, other things. And using that in the background to not just understand the organizational context, but then also essentially use that as a management tool to be able to say, because there's a question of, I could, based on my individual preferences, which I could sense, like I know I am in this room, so I could say, “Well, maybe I like rooms to be brighter, making it up, but maybe when it gets brighter, maybe that makes me a little bit more physically animated. And maybe for the work of the team, actually, me being a little bit less animated would tend to be better. I’m making that up. But the idea that there could be this mediation effect there, where there's individual preferences, which there are some systems which do that today with things like temperature.
But the idea that it's not just about using individual, it's about what the whole group does. I think the ability, not just to collect data, but then also influence the environment really just automatically, both through your presence, but then also this more organizational data about what's going on. Again, this is obviously a longer-term thing. I think one of the interesting things is this is technologically possible today, but no one's going to buy this today. It's not something that I could make as a product, but that I could argue that the role of management is continuously moving towards a place where we're trying to just naturally shape behaviors to be in the right way. And it's not really about micromanagement. And that that's a lot of what this kind of technology and wearables can enable.
Don MacPherson:
Practical applications like safety and movement, lack of movement. I remember wearing a Fitbit for, a similar product for a while, and just getting information about how sedentary I was, and getting a reminder to move every 20 minutes or 30 minutes or things like that.
Ben Waber:
I personally am less bullish on the individual feedback mechanisms in a lot of this. I think that the environmental shaping things are interesting and much will be used more, mostly because again, so I have a step tracker thing. I, of course, look at my individual dashboards every day, right? But I do this stuff. This is like my job. I'm not a good model for the average consumer. And for most people, your job is not to look at metrics on how you work. That's not your job, or how you behave in general, right? Because you see your behavior, you're like, “Okay, made a change, I'm done.” But if you say, “Well, here's the goals that I have.” I don't actually care about the metric. I mean, I don't care about how many steps I take.
I care about what is my overall heart health? What's my longevity? If you say, “Well, here's what I care about and I'd like that outcome to be more likely. Could you make that happen more?” That's it. And maybe that's through change in temperature. Maybe that's through change in light. Maybe it's through, if I don't have an assigned seat, and I get in the office and I look on my cell phone and it says, “You should probably sit here today.” And maybe I don't want to sit there, and so I choose something else, but I'll probably go with the default. If you do that, I know the right stuff will happen more, but that seems more like people will probably do that stuff. I think that those devices and their ability to generate data about what we do and using that to sort of passively shape the environment, and then occasionally, intentionally, us to use that too to make bigger changes. But I think that's very infrequent. I think that's every couple of months or something like that.
Don MacPherson:
Based on the data you collect from clients, how do you see artificial intelligence being used to augment leader decisions or augment individual employee performance?
Ben Waber:
What at least we're seeing is our sort of analytics become an input into these large management decisions. But what I tell everyone of our customers is that if you blindly follow those predictions, you're going to make a lot of dumb decisions. Because no matter how smart the algorithm is, they just not know the full context of your business. And so, smart leaders will use that and say, “Well, okay, I know that, for example, it looks like these groups should sit together and they'll be more productive. But actually for branding purposes, I need to make the office for this group very different because we have media come in there, and that's why it's got to be like that.” Again, no algorithm’s going to know that. But I think the power of the AI in this space is the ability to narrow that universe of possible solutions, right?
Is to say, here are the things where it looks like it should happen. Probably not all of them should, but rather than look at thousands of potential, for example, interfaces between groups, here's who should potentially sit with each other, for example. You say, “Oh no, I know I need to talk to like these three groups, and that's it.” That really helps. And that dramatically speeds this process. Again, eventually you'll be able to automate more and more of that, but there's still gonna be a very strong place for humans in that. I think really getting people to a point where they trust these algorithms enough to use them as inputs for the decision, but not so much that they blindly follow those.
I think we're still very much before that, we're still very much before. People are saying, “Well, I've always managed in this way, and so I don't need any data to tell me how we do things.” Well, so that's where most people are today. And so, I don't think people are just going to blindly follow these things in the near-term. I think that's what, of course, we do need to look for as people get on this path towards using these algorithms to make better decisions.
Don MacPherson:
Privacy is really important to your clients, to you as a business. It's all over your website. Let's talk about first, how do you get employees confident that their data will not be used for purposes that really are negative to them?
Ben Waber:
I mean, part of it is we contractually guarantee it with each of our… Again, if we collect new data the company doesn't have, again, that's just an opt-in basis, but we sign the consent forms with the individuals, right? And we even have it on our site, so you can look at that. Even online, we're just, here's how we deal with data, here's the metric feedback to companies. We're not going to share, even if the company owns the data, we're not sharing your location, we're not doing content. I don't even have access to names and email addresses. Like, we even built our technology such that we don't even have access to that stuff. And it was very intentional. On our customer server is where all the actual data is collected, but it's effectively names, and all identifying information there is removed. It’s processed on our server and then sent back, right? So that actually, even if someone hacked into our servers, it is essentially impossible to figure out who's who and stuff like that.
Don MacPherson:
I spent about 25 years working in the employee survey business. What do you see the future of employee surveys being based on the way that we can really understand performance much better with these massive amounts of data?
Ben Waber:
Well, but it's different, right? Because surveys can tell you about people's subjective experiences, about qualitative things that again, this kind of data does not get to. Yeah, again, a lot of people think, oh, we should just be able to throw out surveys and things like that. And actually, I don't see that's the case. I believe it's a combination of those things that's very interesting. So, imagine people take an engagement survey. I don't like working here. Okay, well, what behaviors tend to correlate with that? I can do that, right? Right now it’s, I don't know if it was just a rainy day or something like that, that cause that. But if I can actually isolate, here, it looks like are the behaviors that it looks like are relating to those outcomes. And, of course, I wanna follow that up with some interviews to see, is that the case?
But then I can actually try things out. And the benefit is that, how often can you really do some of those, especially larger surveys? You don't do that very often. But if I can figure out here are the behaviors that relate to those responses, and then I can see change in those very, very quickly in a couple weeks. Then before the next time we do that survey, I'd have a much better idea if I'm going to make progress towards that or not. The subjective experience still matters, and maybe we'll be able to measure more and more of that, but we're never going to… Never. Never is a long time. In the foreseeable future, we will not get to a point where we just are so predictive of that, that we would never need to do that. I think we're very, very far from that.
Don MacPherson:
I've mentioned this book by Kai-Fu Lee, which is AI Superpowers: China and Silicon Valley. And one of the things that was remarkable that I remember from the book was that there was a predictor in terms of people who would pay back credit in China. And that predictor was when people paid bills on Tuesdays or Wednesdays, they were far more likely to pay their loans back, which is more predictive than a credit score is in the United States or something like that. Have you ever seen a wild variable like that that is predictive of employee behavior, like turnover or engagement or performance or something like that?
Ben Waber:
At one large technology company, the biggest predictor of performance was the size of your lunch network, in that if you ate with a lot of people at lunch, on average, you were a lot more productive. We're only speaking… We saw something, there was really weird artifacts in the data, where you saw there were really almost two types of people. There are people who ate lunch with 11 other people in groups of 12. There are people who ate lunch in groups of four, so three other people. The people in these bigger groups were just much more productive. They would hit their milestones more quickly, things like that. And you could say, “Well, if I ate lunch with those people, then I was much more likely to talk to them later in the week.” Which if you're a programmer, you've got thousands of dependencies, I'll be more likely to know someone in a random group than I would if I only ate in a smaller group.
Okay, sort of makes sense. But it was weird because through these really defined peaks, you almost always ate lunch with 11 other people, and occasionally it was 10, and very rarely it was nine, but it was almost always 11, right? And again, with a smaller group, similar thing, you almost always ate lunch in groups with other people. And sometimes it's two, very rarely. You go to company cafeteria and it becomes immediately clear what happened. There are two doors to this cafeteria. And you can easily go to either door. It's a big cafeteria, but it appears that when people first start working at this company, they just get a normal path to the cafeteria, and that's the path they always take. And but one set of doors, all the tables are small, and they have four seats. But the other set of doors, the big tables, they have 12 seats.
So, it wasn't that you get these… It's a psychological difference that these people just, I'm more outgoing, blah, blah, blah. No, no, no. It's that just the environment made it much more likely for that stuff to happen. And, of course, there's still individual variants there. It's not that it was the only predictive performance. It was incredibly powerful and sort of crazy that you could talk about 10 plus percent differences in predictive power by changing the kind of lunch tables people sit at. I think that's a really interesting thing, right? Is that you see all these behaviors around the way that we collaborate where… How many CEOs are thinking about how big the lunch table is? Who cares? You spend all the time thinking about, of course, org charts, and things like that. yeah, and you should, those things matter. But what we consistently see there are these environmental decisions we make typically without really considering the impact of them that have really outsized impacts on outcomes and how people work.
Don MacPherson:
I want to I ask you about artificial intelligence, and as you look into the future, do you see AI as being this great jobs eliminator? Do you see it as a jobs creator? Do you see it as an equity disruptor, meaning, those who harness it will be mega powerful and those who don't will be left behind? What’s your prediction.
Ben Waber:
I don't see, frankly, at this point at least, how it's any different than other technological changes that have come in the past. The steam engine was a disruptor, Computers was a disruptor, the internet was a disruptor. They all do eliminate some class of jobs. They create some other jobs, and they also change the way that work happens. The nature of most AI algorithms at this point is incredibly brittle. They're very, very good at automating rote tasks, especially in more general areas, they're just not very good at it. So, I would say that this current wave of hype that we're seeing is not without reason, and there are going to be changes, but I don't think it means that in 10 or 20 years we're going to see unemployment at 10%.
Don MacPherson:
Where can people learn more about you and about Humanyze?
Ben Waber:
They can learn more about both at our website, www.humanyze.com, and that's Humanyze with a YZ.
Don MacPherson:
And I'll put that in the show page notes. Ben, thank you for being a genius.
Thank you for listening to 12 Geniuses. Thanks also to the amazing team that makes this show possible: Devon McGrath is our production assistant; Brian Bierbaum is our research and historical consultant; Toby, Tony, Jay, and the rest of the team at GL Productions in London make sure the sound and editing are top-notch. To learn how 12 Geniuses can prepare leaders for a rapidly changing business world influenced by shifting demographics, new technologies, and innovative business models, please go to 12geniuses.com.