In this segment, we're going to talk about some of the unique challenges associated with managing people and human resource issues broadly, as may be particularly relevant for folks who are coming to this topic for the first time. Let's talk a little bit about what we know about managing people issues, particularly as they intersect this topic of data science and the underlying paradigm of optimization. If we turn back the clock, 100 years or so ago to the beginning of the modern factory era, one of the things that you would have seen, and a lot of this actually happened in Philadelphia. Before the Civil War, 1930s or so, Philadelphia was the second largest English speaking city in the world and a great manufacturing center in the United States. If you went into a factory around Philadelphia around 1900 or so, one of the things that would have surprised you if you stopped and asked people, is that a lot of the people working in that building were not employees, they were independent contractors. By some measures, by about 1910 or so, as many as 40 percent of the workers in U.S. factories were actually independent contractors, not employees as well. One of the things you also would have seen is that the organization was really chaotic. It was just a mess inside the building. You could see why just from that earlier description. You have a building, you have an owner who owns it, but he's also got all these contractors in there who are working by themselves, and sometimes they actually got a piece of the action, that is a share of the profits, that came out of the business as part of the work that they did there. The other thing that we would see is if the people came and left and new workers came tomorrow and existing ones left, the turnover in these factories, be 300 percent or so was not uncommon, and the management practices were really casual. At this point, the world's largest manufacturing operation was near where City Hall is now in Philadelphia, was the Baldwin locomotive works. They made trains. Here's how the foremen would operate at the Baldwin locomotive works. They would go in at the beginning of the day and they'd see how many workers from the day before showed up and if they were missing two, they didn't have time off and policies for sick leave. If you were missing two that day, they would go out to the gates of the factory and there'd be a crowd of people looking for work, and they would throw a couple of apples in the crowd. If I needed two people I threw two apples. If you caught an apple, you came to work for me. You followed me into the plant. Somebody else on the other side of the plant does show three people, they might throw three pears into the crowd. If you caught a pear, you work for him, if you caught an apple, you work for me. Pretty simple, pretty straightforward, pretty fast, but also really chaotic. One of the things that companies began to discover slowly first, at the Ford Motor Company, is that this system of really high turnover and really casual work didn't work very well. Once people stop working simply with their hands, hammers and saws and carrying stuff around. On the assembly line, for example, Ford Motor Company realized when they first started it up that the cars were terrible coming off the assembly line. They didn't start, they wouldn't work, they're missing parts. It was just a mess. One of the engineers had the idea that maybe this had something to do with the incredibly high rates of turnover. Ford ran a big experiment. They called it Henry Ford's $5 a day plan. If you knew economic history, you may have heard this description. Ford famously says, "I want to pay my employees enough so that they can afford to buy my cars." That isn't what he did at all. He did it to cut turnover and he did a bunch of other things that he invented as well at the time, like exit interviews, asking people why they quit. What they said is much the same thing they say today, "I hate my boss." He fires the supervisors, brings in new ones, creates a department of industrial sociology, he calls it. He did double the wages. Turnover falls like a stone from 300 percent to about 50 percent per year, almost within a week or so. They discovered that the quality of the cars jumps enormously. We think about assembly lines as being really low-skilled kind of jobs. But if you make a mistake on an assembly line, it is way worse than if you're making a mistake working with a hammer by yourself, because you stop the assembly line and you might break a very expensive piece of equipment. Ford begins to discover, as other employers do as well, that stabilizing your workforce really matters, talent management, getting those right people in the right jobs really matters as well. The other big development starts at the Western Electric plant, maybe about 10 years later or so in the mid to 1920s or so. Where they began playing around with work design. It first started as part of a government study to look at lighting, and the idea was, maybe better lighting would improve performance, and you may have heard of this if you've ever taken a psychology course. This study was done at the Hawthorne plant of Western Electric where they made telephones. What they did is they took out the lights and they put in new fluorescent lights, and they discovered productivity went up, which is great. That was obviously lighting, great, we learned that. Being good researchers, though, they put the old light bulbs back and they discovered that productivity went up again. What they learned from this eventually, was that it wasn't the lights at all, it was talking to the employees, asking them how things were going, asking what the new lighting was like, paying attention to them seemed to matter. Western Electric, although the study of lighting went away, Western Electric decided to keep playing around with it. They had this small part of the phone assembly operations where they just played around with different ways of organizing work, paying more attention to workers, paying more attention to their needs and interests. Into this came a guy named Elton Mayo, who began his career at the Wharton School. I should have said earlier that Frederick Taylor, who created this model in the first place of scientific management that is industrial engineering applied to work, also began his career here around Philadelphia; his great patron was Joseph Wharton, who ran Bethlehem Steel, and what Wharton had done is he hired Frederick Taylor to come in and see if he could make his workers more productive. Taylor's insight was that you could take jobs, break them down into individual tasks, and treat them as if they were a machine. We'll take a job a machinist is doing, we'll identify each separate task the machinist has, and we will see whether we can break those up, give the simpler tasks to unskilled workers, and keep the machinist doing just the things that are unique to machining. We break those down into parts, we can time them, we can see how long each one should take, and then we can start telling the machinist exactly how much to do, how long it should take to do each task, hold them accountable and move them along. Back to Elton Mayo, he's the opposite of Taylor. He started at the Wharton School, he moved to Harvard Business School, and they went to this Western Electric study and began to check it out and follow it. They thought there was something really interesting going on here, because in these small groups, they could see productivity differences going up by orders of magnitude five times greater when the employees were being engaged in this more. Mayo creates a literature and a set of theories called the human relations movement. Basically boils down to workers have psychological needs, they're not simply rational machines that you can tell what to do. If you pay attention to those needs, productivity and performance would go up dramatically. After Mayo, in 1930s or so, some people grabbed this, so Edwards Deming was the quality guru. In World War II he had helped the US improve quality in manufacturing during the war time, was really important then. But then after the war, the engineers came back in and we more or less had Frederick Taylor's scientific management approach; break jobs down, time them, set standards, pay people for achieving them, etc. By the 1970s, though, we discovered that manufacturing wasn't working very well, we had a lot of problems with quality, etc. We started to see during the oil crisis Japanese cars coming to the US. Fuel efficient small cars, they started to get a market share, but then the other thing we discovered was that the quality of these cars was really better. We discovered that the Japanese had learned to adopt the Deming's quality approaches, particularly quality circles, which mean you're empowering the employees to make decisions over quality. They're checking, not somebody at the end of the line. They're responsible if the quality isn't good, they're going to try to fix the quality. The last step in this in terms of lessons from Japan, was to learn about the Toyota manufacturing system, which they called lean production, or we call lean production, and this was a model that went a step further from the quality circles. Here, workers and their teams are empowered to fix quality problems, find them and fix them, but also to figure out how to make their part of the operation more productive, how to move the cars through faster. Maybe with less labor, maybe sometimes even with fewer parts or cheaper parts. We moved from industrial engineering, Frederick Taylor, assembly line, quality circles, empowering the workers on quality, lean production, empowering them on quality and productivity to this last step, in the mid 2010s coming out of information technology, which was the rise of agile project management and agile project management was the discovery, starting in software first that the best way to manage software and get it built, was to empower these teams, leave them alone basically, support them, give them what they need when they need it, and let them decide not just quality, not just productivity but even design issues to figure out the best way to meet the customer needs at the end. A lot of face to face interaction, a lot of communication, a lot of transparency. What Douglas MacGregor described in our introduction would refer to as theory Y. We went from recognizing it Western Electric, that this really mattered to pay attention to worker needs to the quality circles, where we start to empower them on quality. To lean production where we empower them on productivity, to agile where we empower them on virtually everything. What's the theme about theory Y? It is empower, empower, empower, empower. Let employees do more and more. Now in order to make this work, you have to surround the employees with a system that supports all of this. The reason this is important is because when we talk about data science now, we're talking about introducing it into some systems, which are already that far along in empowering employees. It means we're going to be talking about, how data science can make decisions. But it's going to be in this context where a lot of organizations have already had their employees making a lot of these decisions themselves. We talk about managing people as well. In addition to these questions of how we organize work, there's of things we have to think about that affect how employees are motivated? Which means how much are they going to get done? Things like how do we pay them? How do we measure their performance, and decide what's good and what's bad? What is their relative pay compared to other people? They care a great deal about that and what is my perceived contribution versus others relative to my pay? This is a fairness question. Quirky thing I learned from my MBAs here over 30 years or so we've had the same midterm exam for incoming MBA students, asking them to write about their last job. The main reason they quit, their last job is because of these fairness issues. That is mainly, they perceived that they were getting more done than other people and it wasn't being recognized. There's all issues that matter to employees that are going to affect their performance, which is going to affect the organization's performance and data science in this context have to be very careful about not messing that up. Let's talk a little bit about some of the policies, and procedures, which are in place around managing people that go outside of simply performance. Employers, also care about things, like these days the demographic balance of the workforce across racial groups, across gender and sex based groups as well. Diversity inclusion an important topic today. We're interested not just in having people hired who we think are the best performers, but we're interested in having an overall workforce that represents the demographic balance in society. We are interested in the public perception of our workforce practices. We want to be seen as a good employer and a desirable place to work. A few years ago, there was a movie called The Intern that some of you may have seen. It's been a few years now. Google is the site of that movie and as I understand it, Google supported that movie financially. You may wonder why would they do that? Well, because Google's brand as an employer is being broadcast out and Google benefits from that. It reminds us that, in addition to just having effective work practices, it matters whether people believe and see, that what you are doing in managing people is something that is desirable. Of course, the other thing which looms large in the US, in other countries, it's equally important, just executed differently is compliance. Compliance, when you think of human resources in the US a lot of people begin to make a frowny face, because they're thinking about the legal issues. About being sued for discrimination, or actions that affect people with disabilities, or wage and hour violations. There are things that constrain what otherwise might be optimal decisions with respect to performance and productivity. But they have to play within the lines of these legal issues. There's all parameters here to worry about. Into this jumps data science. Data science begins with the assumption of optimization, we can optimize on one outcome at a time. If for example, we want to say, we want to hire the people who are going to be the best performers on this job, that sounds reasonable. How do you measure performance? Well, performance appraisal scores, it's pretty good, but does that include whether they're going to quit? Well, probably not. Does that matter? Yeah, it probably does. How about their interpersonal skills and ability to get along with other people or their possibilities for advancement, do those matter? Yeah, they probably do as well. Well, pick which one of these you want to optimize on because we can't optimize on two at the same time. At the very beginning that begins to throw a little wrinkle on us here. It's one thing to predict when a ball bearing is going to shatter, it's another to try to predict who is going to be a good performer when we can't completely define an easy way, what does it mean to be a good performer. The second problem we're going to bump into and you've heard this from our colleague Sonny Tambe and Will Thuet talking about tools like machine learning is they require a lot of data. A lot of data means if you're a reasonably small employer, let's say 500 employees or something, you probably don't have enough employee transitions, enough quits to build a machine learning model, even if you extend back historically over time. That becomes a problem because as we'll see a little later, questions about privacy start to come up. How long should we keep employment data on somebody who left 10 years ago, for example? Raises interesting questions about privacy. The other thing about data science, once we roll out a technique data science like hiring, and we're using it across the whole organization, if we make mistakes, the mistakes are easy to spot because they're happening now at scale. Something else to think about, the risk tolerance of making a one size fits all approach. It might be much better than what we were doing before. But if you make a mistake, easy to spot, and if it breaks the law, easy for plaintiff lawyers to spot as well. One of the big issues for us to think about is that when we introduce machine learning models, algorithms, data science models, that tell us how to make decisions, we're taking those decisions away from somebody else and typically we're taking them away from supervisors. What happens when we make that transition? Here's a quick example. Let's say right now we have to set a schedule for our employees. Who comes in on what days and at what times, let's say. We get the employees together and we talk this through. We maybe set up a system, and the supervisors there a little. We do a bunch of negotiation, I can't work on Saturday, well, I'll take that if you'll take Sunday, a kind of messy thing works this out. But then we decide; let's optimize this and go to scheduling software so we make sure we get exactly the right number of people at the right time. Then we spread the work out completely evenly. What happens now? Well, let's say I'm an employee and I'm scheduled to work this Saturday, which I don't particularly like, and next Saturday the schedule comes out, and I'm scheduled to work next Saturday as well, and I'm really angry about this. The current system, I go to my supervisor and complain to him about it and say, this is terrible. Can you do something about it? She might say back to me, "I understand this is really a problem. I'll take care of it later on. We won't do this to you again." We replace that with the scheduling software and I get scheduled on Saturday. Not happy about it. Next Saturday it's scheduled again. I go to my supervisor and say, hey, what's going on here? My supervisor says, "here's the software. It did it." I say, this is terrible. My supervisor says, "here's the phone number of the programmer who made the software. Why don't you talk to them?" We've got a problem now of fairness and equity and we don't have a good way to solve it. What are the big decisions we have to make about data science is where can we introduce it in ways where it will make things better and where if we introduce it, might make things worse. This is the important question that we're going to be dealing with throughout this program, and that is the question of judgment. We understand what data science can do. We understand the complexities of the workforce. Where could this be helpful? Where might it actually create more problems? That's judgment that we're hoping to get out of the end of this program.
