In this video, John Sumser discusses new management approaches for HR in AI enabled enterprises.

We’d love to hear your thoughts about this video. Post your comments on or John’s YouTube Channel »





Important: Our transcripts at HRExaminer are AI-powered but there are still instances where the robots get confused and make errors. Please expect some inaccuracies as you read through the text of this conversation. Thank you for your understanding.



JS – John Sumser, Principal Analyst, HRExaminer

00:00:00:01 – 00:00:22:03
I’m John Sumser. And let me introduce myself very quickly I am the principal analyst at the HR Examiner. If you haven’t had a chance to get a good look at the HR Examiner it’s it’s worth seeing and I’ll tell you about that in a second. I also write a monthly column for H.R. Executive Magazine on intelligent tools.

00:00:22:03 – 00:00:33:27
I do selection and implementation consulting and help people with product development and envisioning the integration of intelligent tools in their organizations.

00:00:33:27 – 00:01:00:00
My home is a it operation called The HR Examiner and at The HR Examiner we explore the edges of HR and HR technology. It comes out as a weekly newsletter. We do a couple of podcasts a week and we do intensive research every year into the emergence of intelligent technology in HR.

00:01:00:00 – 00:01:13:02
We publish an annual report in the fall that looks at trends and growth and success and failure with this new form of technology.

00:01:13:09 – 00:01:35:12
So what I’m going to talk to you about today is is the changes that happen as we move into into a time where all software has some intelligence function to it in one way or another it’s a it’s a big deal change in the old model.

00:01:35:21 – 00:02:11:11
We had software you put data into the software and that same data came out and it was reformatted in some way in the emerging model of digital management. You put data in and then the machine does something to it. And what comes out of it is different than what you put it to it. And that difference is what they’re calling intelligence. You’ll notice that I work pretty hard not to use the term artificial intelligence because I don’t think that’s what we’re seeing yet.

00:02:11:26 – 00:02:35:06
Artificial intelligence is a larger more conversational kind of thing and what we’re seeing today are the extraordinary things that you can do with math and statistics. Once you have a good clean set of data so just to orient it very quickly there’s there’s a lot of material here and I’m going to scoot through some of it.

00:02:35:09 – 00:02:52:16
You’ll be able to get a copy of the slide deck once once the podcast is died once the webinar is done so so I will I will pick and choose my battles in this slide deck as we go along.

00:02:52:17 – 00:03:15:24
First thing is the terrain that we’re talking about each our tech is a an enormous sprawling bag of cats and dogs. One of the hardest things to get one’s arms around with H.R. tech is that it is so various and is so different.

00:03:16:08 – 00:04:49:19
Almost all companies have a unique configuration of their H.R. text stack and so the fundamental elements of it are talent acquisition which much of H.R. thinks are the cowboys Talent Management and Development the organization of all the data or the human capital management function hygiene which are the things that H.R. does that have to be executed perfectly that don’t motivate. But if you get the wrong way demotivated and then the off boarding problem now that looks like in practice this enormous pile of discrete technology silos there is software available at each one of these 75 blocks. And the truth about this particular picture is there are more like 140 discrete H.R. technology silos but it’s impossible to portray that at a single slide. And so this map is worth having if you’re thinking about how the pieces relate as you build the strategy for intelligence tools at your system. But if you look at it let’s let’s take the example of talent acquisition this shows maybe 15 subsets of talent acquisition and there are really 40 different areas inside of talent acquisition for which intelligence tools are already available.

00:04:51:18 – 00:04:58:23
The thing that’s going on in the business in general is this startling curve.

00:04:58:24 – 00:07:02:17
This is the growth of vendors of intelligent tools in H.R. technology over the last three years. When we started looking at it and doing research it looked like an anomalous thing there were 40 vendors in 2016. By the time we did our second round of research in 2000 17 there were 200 at the end of 2018 what we’re going to look at for the next report. There were six hundred and forty five. We expect to see fourteen or fifteen hundred at the end of 2019. And the truth is that you won’t be able to find a piece of software that doesn’t have intelligent components to it within the next couple of years. This is this is the new way of doing digital delivery inside of organizations so. So the slide says Smart Data is the new software and I want to go back to what I said at the beginning of the conversation that that what happens with this new stuff is the data you put in is not the same as the data that comes out. The machine does things to the data that you give it to produce a predictive result or a recommendation resolved or some other value added insight merged it into the data that you gave the system in the first place. And that is where value is going to be created in all digital offerings. The current view and I’ll talk about this cyberwar over the course of the presentation of the current view is that that this is a sort of an add on to your existing processes at what you do is take these bits and pieces of intelligence and add them into the existing business practice.

00:07:03:19 – 00:07:24:01
What’s starting to happen is the intelligence embedded in these offerings is showing new ways of doing business. And so rather than workflow that’s defined by the organization.

00:07:24:02 – 00:12:17:03
And in the somewhat political process we’re going to start to see workflow that’s defined by the way things actually happen right the object here is to get the friction out of the workflow and the results into the workflow by having smarter recommendations and more compressed processes that focus on results over process so just to back up for a second. I said in the beginning that I don’t like to use the term A.I. and and you may I spent a long time with a book called The Book of y by today a Perl it’s been a hovering sort of at the low end of the New York Times bestseller list for a couple of months now and today a pearl is somebody who is thought heavily about statistics and intelligence and is a proponent of the idea that you can generate ideas of causality from statistics and and that’s that’s a big change. It’s been the case I think almost everybody listening to this will will recognize the idea that statistics can only tell you about correlation and there’s an emerging school that’s part of this intelligent technology that suggests that you can actually get to what causes X in with with the use of statistics. So so Judea Pearl says there are three levels of intelligence. The first level of intelligence is learning from association and you know that’s this thing is like that thing and everything that’s in the marketplace today is a learning from association kind of product. If you think about the non H R implementations of intelligence they’re things like image recognition where you collect a bunch of data and then you can see the next thing that comes along as like the last thing that came along. So the algorithm learns by imitating and matching and copying it’s more a good copy and paste exercise taken to extremes than it is actual intelligence and everything that’s available to us in H.R. tech right now works like this even the fantastic accomplishments of the systems that play chess and beat go and do extraordinary feats of learning are all about this learning by imitation thing. The second level of intelligence which is where we’re headed in the sort of medium term 10 20 years is learning from planning and reflection. Currently it’s not possible for a machine to understand where it is and importantly what it doesn’t know. As we get to the next level of intelligence which is learning from planning and reflection it’s possible for the machine to start a path of sizing about intervening in a process. What happens if you do this. What are the consequences of this. But today we don’t have anything like that. The machines have zero understanding of the consequences of the recommendation and then the world that I would call artificial intelligence is a world in which machines will be able to imagine worlds that don’t exist. And using that ability to imagine worlds that don’t exist to offer reasons for things that are happening. So this is like this is like the way you plan for certain things you say we’re going to go on a summer vacation. Why do we need to pack. We need to pack these things. Well what if it rains all the time if it rains all the time. We should pack an umbrella. What if there are a lot of bugs or is there a lot of bugs we should pack bug spray. And so so human intelligence does that human intelligence does that naturally we imagine things that don’t exist and we use that as a way to plan and navigate inside of the world and machines are completely incapable of doing that. But this learning from counterfactuals in Pearl’s view and I share it is is that you can’t really think of it as artificial intelligence until it’s capable of doing this sort of thing.

00:12:21:24 – 00:13:37:16
So it’s not software that we’re talking about intelligent tools today use statistical methods to stimulate to simulate and predict the world and this this introduces the thing that that I think if there are if there is a single takeaway from this webinar it should be the idea that machines now have opinions. It used to be that machines reflected the data that they were given. But today they’re being constructed so that they have an opinion based on history and what machines are great at doing is predicting a world that will be yesterday plus whatever the factors are. You know this is this is the fundamental approach to sales planning an objective setting in in sort of primitive environments as you take last year and you add 10 percent to it. And so just so machines can do that sort of thing. They can’t predict unseen things are logical consequences of things they can only predict more of the same.

00:13:37:16 – 00:14:20:09
And so what you get out of a machine is some kind of opinion of some kind of quality that predicts that you’re going to have more what you had yesterday which is great unless stuff changes. So so for example on machines opinion about what makes a great call center employee based on what has always made a great call center employee is just fine until you change the technology or change the product or change the way you organize the call center. And so so this this intelligent capacity is perfect as long as nothing changes in the future from the past.

00:14:20:09 – 00:14:29:17
It’s rooted in probabilities rather than certainties. And so you get this output of the machine. That’s an opinion.

00:14:30:08 – 00:15:12:29
And what you do with an opinion is what one always does with an opinion. If it’s a if it’s a direct opinion from your boss you generally take it. If it’s a direct opinion from a particularly angry spouse you often take it otherwise with most opinions what you look for is another opinion before you make a decision. So so machines are going to fairly rapidly start to assume roles where they’re left to make decisions but they’re not equipped to do that just yet. They’re they’re equipped to offer your opinion and an opinion is a good thing but it’s an input to a decision rather than the decision itself.

00:15:13:00 – 00:15:46:27
And this is where the requirement for new forms of management actually emerges so the five kinds of product being offered in the in the world right now that have some form of intelligent capabilities that there are and you should expect that this will have massive impact in H.R. technology over the next two to three years.

00:15:46:28 – 00:16:30:29
There’s A.I. as a platform service Amazon Google IBM Microsoft Facebook all offer cloud based environments for executing various kinds of intelligent work and they offer the tools to it because each of those companies is fundamentally in the business of selling storage and processing time in their cloud systems so what they want to do is help you get better at using their tools so that you’ll be a higher volume customer. It’s in their interest to deliver A.I. as kind of a commodity that you take and apply into your business.

00:16:31:09 – 00:16:57:18
One of the central problems with with using intelligent tools is your data has to be in good enough shape to use them. You have to have enough data and you have to be able to process the data. There’s generally a painful data governance process that precede the active use of intelligence tools.

00:16:57:20 – 00:17:32:08
Companies like one model swoop talent and vizier all come at this data workbench and cleaning process from a different angle vizier gives you a sort of a template view of analytics so that you can start to make intelligent decisions from a dashboard swoop. Talent is the most innovative of these offerings and they they give you a way to continue to use your data while their system processes and cleans it out.

00:17:32:12 – 00:18:08:29
And then one model is kind of stuck between these two things and they offer the ability to do dashboards and analytics plus some predictive capabilities. Most of what you’ll see in the in the environment is this microsite this category that I called micro services. There are about 450 current vendors offering little tiny solutions to little tiny problems.

00:18:09:00 – 00:18:36:24
There are examples all over the place I’ll give you a list of a couple of slides of all of the various kinds of micro services that are out there. But but each one solves a little bit of a bigger problem. And this is the new venture capital model for product development is that the narrower the focus the more likely you are to have a success from the minds of the of the venture investors.

00:18:37:02 – 00:18:45:16
And so we’re seeing an explosion of these little tiny solutions.

00:18:45:19 – 00:19:15:24
The next category is pretty interesting if you’ve followed technology over the last couple of decades the operating theory has always been that companies like these micro services providers have a distinct advantage over legacy systems because they’re more able to be rapidly adaptive and not burdened with all of the history in this case.

00:19:15:25 – 00:21:29:10
There’s there’s an argument that says the more data you have the higher the likelihood that you are able to deliver a quality solution. And so workday Oracle Ultimate Software civilian Kronos paychecks salary the hardware store smart recruiters of ADP are these legacy operators who have huge slugs of data and because they have huge slugs of data they have a serious operating advantage in the delivery of these tools and so. So there’s there’s this emerging choice that looks like the old point solutions sweet dichotomy where micro services are starving for data and legacy systems are drowning in data. And the the challenge for the legacy services is how to get out of their own way. And the challenge for the micro services is how to get enough data to actually make their use case and then we’re starting to see that there’s going to be a lot more of this. The the tools that are built with A.I. in mind from the very beginning the first suites and there are a couple out there now ascend to Phi as a San Francisco based talent management and talent acquisition tool that ranges from an 80 Yes. All the way out through learning and development. It’s built around competencies and is entirely focused on the Fortune 100 bridge buy in structure is a learning system that that has that same idea as a center FI but they start from the learning piece and our branching out into talent management also starts as recruiting and branches out of the talent management. And Google is doing really interesting things in the evolution of their talent management products that makes them and a first suite as well as being a platform service provider.

00:21:29:10 – 00:21:56:03
So these are when you when you go and you look out in the market and you see what’s there. These are the five large categories of tools that are out there the other thing that you want to understand that’s happening right now is that all of the legacy enterprise providers are building marketplaces or ecosystems.

00:21:56:04 – 00:22:13:02
And you know there’s there’s on the one hand there’s this explosion of new providers of micro services and on the other hand there’s these suite providers and the the place where those two things intersect is in the ecosystem. And

00:22:13:02 – 00:29:17:26
And what I see happening is that as the legacy providers build out these fantastic ecosystems they do it based on their point of view about how the world works. And so what you’re buying into what you buy tools from a legacy enterprise provider is more of an editorial point of view than it is raw functionality. While there are some observable differences in functionality generally speaking the enterprise providers all provide the same functionality and differentiation occurs out in this ecosystem and the ecosystem is is always going to be built around the values of the core legacy provider. And and so what you’re buying into. So let’s let’s take some examples. If you buy Ultimate Software and you start to build out your solutions inside of their ecosystem you are buying into the idea that the most important thing about the world is the way employees feel about the company and the way employees interact with the company and that’s the heart of ultimate software’s value structure. If you buy workday the heart of work days value structure is that planning and data consistency or the heart of the enterprise. And so you can see just from those two things that that one’s going to prioritize one set of functionality and the other’s going to prioritize the other. And so the choice that you make any more with a let with an enterprise provider is about their personality and their core values rather than their functionality. The functionality gets solved. Any any decent nuance and functionality gets solved inside of the ecosystem and they’re going to pick who’s inside of their ecosystem based on those companies alignment with the core values of the larger enterprise. I started to talk about this in the very beginning we’ve we’ve had a long run of of best practices and benchmarking where where what what people have said is that H.R. is the same everywhere and that what you need to do is align yourself with the way that other people are doing and what’s becoming clear as we start to have enough data to actually make sense about this is something like 60 percent of what all companies do in H.R. is very similar. And there are best practices inside of that 60 percent but what makes the difference between one company and another is the 40 percent of difference. And so the best way to think about those 75 silos that I showed you at the very beginning is that H.R. is a spice cabinet and depending on what you’re trying to do with your company you pick and choose between the elements of H.R. in order to accomplish your larger mission and you can’t benchmark that you can’t find best practices for that because the unique be all that you’re making in your H.R. department with your H.R. technology is dependent on the company not all of H.R. H.R. H.R. exists to serve the larger company rather than it being the other way around. And so the organization and execution of H.R. technology has to be more creative than the set that sparks. And here are the spaces currently in the cabinet. These are the things that if you wanted to tomorrow morning you could go out and buy these are all going to be subscription services. You could buy a subscription to any one of these 45 things ahead and so the question for you the really big question is how do you want to put these things together. No single legacy enterprise company offers all of these capabilities there’s some of these some things on this list like text augmentation that probably are never going to really be part of a legacy system but they are powerful tools for accomplishing very specific things. There is nowhere near enough time in this conversation to walk you through all of the functionality that’s available. We have an annual report that gets several layers deeper into what each of these things are and who the vendors are. But it’s a rapidly growing and expanding universe. So the number of micro services I imagine will continue to explode until there is a micro said intelligent micro service available for every place that you could possibly stick one in your organization. And so your question for managing this is how to understand the entire array of possibilities and then how to understand how to envision where you’re going and then prioritize the the acquisition and deployment of this functionality. And so it’s infinitely more complex than a traditional enterprise software implementation problem because it requires a level of clarity that you don’t have to have to get in bed with the devil price provider. There are all of these pieces and underneath each one of these categories there are multiple ways to skin the cat and. And so the overwhelming problem in acquisition and execution is understanding what these things being what the ethics associated with the more what the unintended consequences of using them might be. And it’s enough to give you an anxiety attack if you’re not careful. It’s a rich environment with an exploding amount of possibility and an increasing likelihood that if you get this right you’re going to be able to catch the Holy Grail which is proving that your work in nature has actual direct consequence on business outcomes. That’s where we’re headed with this stuff.

00:29:20:14 – 00:29:31:15
When you think about managing this this is like my favorite slide in the whole world. When you think about managing these intelligent tools.

00:29:32:10 – 00:29:52:21
I like to use this analogy of a fruit bowl. This this this is my favorite thing to eat. It’s a bowl of fruit with a little bit of whipped cream and it’s got all sorts of things all sorts of things in it. When a machine looks at it this is what it sees.

00:29:52:21 – 00:30:35:07
So if you go back it’s a polka dot bowl with a spoon and a bunch of fruit in it. When the machine sees it it sees a bowl a spoon different quantities of fruit and then that line of polka dots. Because how does a machine know the difference between fruit and a polka dot. It doesn’t it doesn’t it won’t. It can’t. It counts things. And if you look at this picture and you think I don’t know if you like if you love fruit salad the way that I love fruit salad. But what’s missing from this picture is the best part of the fruit salad and that’s the juice at the bottom of the bowl right.

00:30:35:07 – 00:35:16:17
So this is a quantitative look at a fruit salad and and everything that you can get with the machine today is a quantitative look at stuff and quantitative looks are fantastic but they never ever ever capture the essence. This is this is a way of saying that the thing that that you should always understand about every output you get from an intelligence system. Is that all models are fundamentally wrong. All models are simplifications and they’re great. You can do a lot of things with the simplification the euro so. So if you’re smart enough you could come and look at look at the picture on the right and go oh if I want to make half as much fruit salad I take half as much stuff. But if you’re a machine that might mean a half a bowl or half a spirit and a half a pile of polka dots those are irrelevant to the heart of this question is not about all of the elements it’s about a subset of the elements. And it’s not always the case that the data model that you’re looking at is robust enough to be compartmentalized that way so you have to be extremely conscious of the fact that what the machine offers you is an opinion about how this works not some ultimate reality about how it works and you take it like the opinion that you take from anybody who’s got a simplistic view of what you’re doing. That may be something important there. It may be very useful but you should never ever rely on a single opinion any more than you rely on a single opinion anywhere else in your life. The ethics questions are huge. I’m sure everybody who’s who’s who’s on this call has spent some time thinking about bias. I’m here to tell you that you can’t remove bias from systems you can manage it you can mitigate it but the essence of being human and the essence of being a machine is having a bias when people talk about bias in. In this world what they’re really talking about is discriminatory biases that are not allowed by regulation or law and and those are extremely important things to pay attention to for a whole host of reasons the least of which is their legality. But it’s not possible to eliminate these things. And so if you look at the example of Amazon and their attempt to remove bias the problem that they had was their historical data was biased and they couldn’t remove the bias of the data but culture is sort of a hologram and it permeates everything. And you can’t really scrape it away and if you do sort of neuter the bias in and in a setting what you effectively do is eliminate the culture from from the construct. And that’s as dangerous as having bias. We don’t know all the bits and pieces about privacy yet. What’s clear is that privacy is rapidly evolving. If you are not aware of the California privacy legislation that goes into effect in 2020 it effectively makes the United States even more hard to navigate than GDP are. And so privacy is going to be an ongoing thing there’s a bigger question which is which is what does data tell you about people and how do you use it. That’s really the core of the privacy ethical question. Total cost. I don’t think very many people include total cost in a conversation about ethics but I want to be sure to tell you that as you look out at the vendors who are out there when they talk about costs they talk about the cost of acquiring their service the cost of managing the service is liable to be significantly higher than the cost of acquiring the service.

00:35:16:17 – 00:37:19:23
There is this sort of implication that if you buy an intelligent tool you could save money by firing people and that doesn’t appear to be true. There is a a a complicated process by which you have to keep the people who you’d want to replace in place while you train the new system so that you can make sure that the new system knows everything that those people know and they have as you might guess some reluctance to train the system. And so so you end up with a lot of duplication in the early days which means that the idea of a return on investment may fundamentally be an ethical problem. You may you may not be able to get the returns that you think even though the reason that you have to do this stuff is for survival down the road data models wear out. And so the cost of maintenance associated with with a system is always understated by the vendors and the evangelists because what you have to do the way the data models work is they accelerate learning up to the point that it’s learned everything that they can and then the learning goes flat. And so so you have a sort of an S curve of returns from the new tool. And when you get to the plateau at the top of that line you have to replace the data model it whereas out at that point it’s very difficult to rationalize some of these systems because they learn it means that the results are not reproducible. The system that you use tomorrow is not the system that you used yesterday. And if it’s doing what it’s supposed to do it makes different recommendations based on what it’s learned. And so you’re not going to get an even stream of repeatable stuff and there’s some question about how this affects the bias conversation.

00:37:19:23 – 00:38:42:11
This is uncharted territory. Latency is the idea that a data model knows about history up to some certain point and then there’s a gap between the end of that history and where you are today and the latency problem means that recommendations coming out of any intelligence system are always suspect. And you have to start by asking what’s wrong with this. The last bit of ethics is liability for a long time. I thought that liability was going to be borne by the vendors of these tools. But it turns out I’ve talked to I don’t know maybe 20 employment lawyers about this it turns out that whether or not the software company is liable for the errors introduced by their data models the person holding the bag is the company and so. So when you deploy a intelligence systems you’re liable for the unintended consequences. It’s your career it’s your company’s body. And it bears a conservative viewpoint. So with that in mind I’m going to give you these basic principles for a I enabled enterprises first always know which problem you’re solving and why.

00:38:42:28 – 00:38:51:08
And this is the first principle and it is the hardest principle you can be I’m sure.

00:38:51:08 – 00:47:06:22
I’m sure your desk looks like by and I am flooded with offers from a million vendors who appear to think they’re the only ones who ever send out a piece of email about selling an intelligence system offering to solve this problem or that problem or the other problem. But but what you have to remember is that H.R. is the spice cabinet and that that you can’t really control this stuff until you have a clear picture of what recipe you’re making tonight. You can’t just throw things willy nilly. You have to have some larger picture. And that’s a hard process that requires building a vision amongst the leadership of the H.R. function in order to sustain the process which will have failure in it of moving from the old way of doing business to the new way of doing business in order to get anything done. Your data has to be clean. And the last decade of legacy software has left us with a lot of very tailored workflows for very specific problems and very tailored workflows for very specific problems don’t aggregate into enough data to learn from. And so so what happens as you do data governance and unless you use swoop talent is that you have to compress the workflows till they start to look more and more like each other knowing the limits of your helpers means that currently intelligent tools are like toddlers. They have the attention span and the ability to learn of a four year old. And if you ask them to do more than that they may not know how to say no. And so so it’s on you. The liability is on you to understand the precise limits of your helpers and as you come to understand the precise limits of your helpers. That’s what enables you to start to be able to see the things that might be unintended consequences of your actions. As I’ve been preaching literally preaching you never rely on a single model or algorithm to do everything. They’re always somewhat wrong. And so you need multiple things that leads you to. It’s my view that that five years from now most of the people on this call are going to have 15 or 20 data models for each employee for each work group for each project for each vendor relationship for each governance process for each division. So. So if you’ve got 5000 people in your company you’re liable to have a million data models under management and they are not going to agree with each other they’re absolutely not going to agree with each other. So the the competency based training model of development of employee X is not going to mesh smoothly with the aggregated social media profile that you’ve created for that person and that’s not going to mess weirdly with the performance management recommendations you can have a stew of opinions you can it’s it’s like machines as multi raters you’re going to have a stew of opinion. That’s exactly what you need to make good decisions. But this this describes a world in which there is much more input in every possible decision and so decisions are going to get harder and take longer you can never take a machine recommendation at face value. You always have to think about what are the unintended consequences liable to be if I take this recommendation and you shouldn’t ever act on a recommendation until you’ve done the process of considering what might go wrong. All of this stuff is based on statistics and you don’t need to understand how everything works all the time but you do need to know enough to call bullshit when it’s time to call bullshit. And so that means reading and learning about numbers and about statistics. The great the great leaders in the industry are companies like Allstate who admittedly have an edge because they’re in a in a business that requires being numerate. But they they have two days a year of training for everybody in the H.R. department in analytics and statistics. Everybody in the huge Allstate H.R. department goes to this process and it’s worth copying. They train their people to be literate in statistics and numbers the next piece is machines have opinions. Machines are wrong. But it’s going to take some courage to argue with the machine when it makes a recommendation because the machine has the data and you don’t and so so the workforce is going to need to learn how to argue with the machine and the way that you argue with a machine. It’s an interesting it’s an interesting thing. People who play video games understand how to hack and arguing. Another term for arguing with the machine is hacking. You have to be able to understand what makes the system work how the system arrives at its conclusion and what it’s liable to be missing. So if you think back to the fruit salad example you know that the that the recommendation of the machine is not going to take the juice at the bottom of the bowl into account. And sometimes the juice at the bottom of the bowl is all that matters. And so your people have to be able to understand the output of the machine and take that opinion and do something with the opinion which often means it being able to explain why and how you think the machine is wrong in this process. One of that one of the most interesting things about the the movement of data to center stage which is what we’re talking about is that data often shows that the existing way of doing business is mistaken that the core assumptions about the business are mistaken and that means that leadership has to learn the skill of being wrong in public on a routine basis and that’s not our conventional sort of male view of what leadership is. But but increasingly the primary characteristic of leadership is going to be imposter syndrome because it’s not possible to understand all of these things. And so so there’s going to be well justified feeling that you’re in over your depth that everybody has to figure out how to deal with and the only effective way to deal with being in over your head is to be able to say man I don’t get that man. I don’t know what you’re talking about help me here. And so this this idea of being willing to embrace being wrong on a routine basis is is part of the structure of the new style of management. You know I talk a little bit about the the degree to which maintenance and replenishment are important.

00:47:06:22 – 00:48:39:11
But imagine you’ve got 20 data models for every person and every function in the organization and they age and they stop working because they age and so you have to be able to when they stop working replace them you know it’s like tires on a car when they when they were out you get a new one and you have to be able to imagine setting new goals for these things. And so. So you’ve got a million models under management and you have to have some idea about how the replacement process works you probably have to deal with experiment going on to replace the stuff that’s currently in use. And you have to spend a lot of time thinking about that because the future of your enterprise depends on the maintenance and replenishment of your data models. It’s easy to get caught in unintended consequences. The hardest thing to see for most of us is the problems with our assumptions and how they create the world that we existed. That gets amplified when you have intelligent tools all over the place. And so having a way to step back and look at what’s going on so that you can see if there is a skewed towards something that you don’t want in the process is going to be a the stuff of staff meetings rather than Are we on track to the goal that the question might start to become. How are we skewing away from our intention and what do we do about it.

00:48:41:12 – 00:48:59:16
Systems thinking is an M.I.T. engineering discipline about how bits and pieces of systems go together. And it’s the it’s the troubleshooting tool for all of this technology.

00:48:59:16 – 00:49:24:13
You have to be able to understand how engineering style feedback loops work and change the behavior of the system that you’re in and this applies to the HR Tech system and the HR tech system has a deep influence on the larger human social and political system that it’s operating inside of and you need to be able to have an ongoing monitoring process.

00:49:24:13 – 00:50:27:15
for looking at the relationship between those two things and then the final thing and number 13 for a reason is they’re going to be a lot of bugs. You’re going to fix a lot of bugs. And generally speaking every bug fix creates a new bug. And so this is these are the management principles for these things takeaways. Machines have opinions bias can’t be eliminated get started. It’s a big project and it’s coming fast and it’s the new way of doing business. It’s a world in which always being better is the name of the game. When you talk about using intelligent systems with your leadership and your your purchasing functions resist the temptation to talk about there being an immediate return on investment there may not be. You’ve got to get started now and the longer you wait too to wade into these new styles of management the harder it’s going to be.


Watch more of John’s talks on his YouTube Channel »


Tagged with:  
Read previous post:
HRExaminer Radio Executive Conversations Badge Podcast Logo
HRExaminer Radio – Executive Conversations: Episode #346: Drew D’Agostino, CEO and Greg Skloot, President of Crystal

John Sumser speaks with Drew D’Agostino and Greg Skloot, the CEO and President respectively of Crystal. In early 2015, Crystal...