Automating
Middle Management
Jesse Michael Fagan, PhD
Lecturer of Data Analytics
University of Exeter Business School
https://www.jessefagan.com/automating-middle-management/
Automating Middle Management
Organizations are a form of technology primed for disruption. The fundamental roles of human organizations (e.g. coordinate activity, distribute resources, allocate tasks, etc) are quickly being handed over to algorithmic control piece by piece. In this talk I will address the advancement of AI in the domain of middle management and the coming rise of self-driving organizations. In my work on email networks, social network analysis, and natural language processing in organizations I have discovered the accuracy and usefulness of using machine learning to predict employee outcomes. It’s a short step from this research to the deployment of predictive or prescriptive systems which automate key staffing decisions. These changes could lead to a healthier working environment where systems optimize for human well-being, or they could go a different direction in which human needs are sidestepped in the interest of alternative goals. The goal of this talk is to continue this important discussion.
Source: Marshall Van Alstyne -https://youtu.be/jfd3k4VUYh4
The biggest taxi company owns no cars.
The biggest hotel company owns no hotels.
The biggest media company doesn’t produce content.
Something is different.
Source: Marshall Van Alstyne -https://youtu.be/jfd3k4VUYh4
Organizations are a kind of technology.
A technology on the verge of a major disruption.
The Platform Revolution
&
Machine Learning
Robots are coming quickly.
Source: https://www.theverge.com/2019/3/28/18285923/boston-dynamics-handle-robot-updated-box-stacking
What does it take to succeed following a merger?
Luxury and Standard
Standard, Inc.
(target)
Luxury, Inc.
(acquirer)
Luxury-Standard, Inc.
Luxury
Standard
Standard
Luxury
Time 1
Time 2
Linguistic Signature of Job Attitudes
Linguistic Properties of Emails Authored
Mental State
Linguistic Properties of Emails Received
Personality &
Organizational Situation
Responses to Survey on Job Attitudes
Linguistic Signatures of Job Attitudes
Trained an xgboost model using Bayesian optimization to predict high levels of each scale (6-7 on 7 point scale) from the T2 survey and T2 email.
In the future use changes in linguistic properties to predict changes in attitudes.
Intent to Stay (turnover): 0.77
Job Satisfaction: 0.75
Job Insecurity: 0.82
Org. Identification: 0.85
Procedural Justice: 0.81
Distributive Justice: 0.73
Affective Commitment: 0.77
Continuance Commitment: 0.91
Intent to Stay
Words like:
fuck, damn, shit
Intent to Stay
Words like:
mate, talk, they
Distributive Justice
Words like:
few, many, much
Distributive Justice
Words like:
feel, touch
Affective Commitment
Words like:
cause, know, ought
Affective Commitment
Words like:
ally, win, superior, take
Generalize the results
The intent was to show that we could create a model fitted in one organization that could predict important outcomes
The Problem
There is absolutely no relationship between the quality of the model fit at Luxury-Standard and the predictive quality at Corp-Corp.
Organization 1
Organization 4
Organization 2
Organization 3
Attitude Dashboards
Models like this could give managers accurate, daily updates and the attitudes at their company.
What if I was successful?
Malcom gif - scientists asked if they could, not if they should
Rather than using machines to predict behavior and attitudes..
This study showed everyone that we could use machine learning to change people’s behavior and attitudes.
Rather than using machines to predict behavior and attitudes..
...these studies showed everyone that we could use machine learning to change people’s behavior and attitudes.
Kramer, A. D. I., Guillory, J. E., & Hancock, J. T. 2014. Experimental evidence of massive-scale emotional contagion through social networks. Proceedings of the National Academy of Sciences of the United States of America, 111(24): 8788–8790.
Bond, RM, Fariss, CJ, Jones, JJ, Kramer, ADI, Marlow, C, Settle, JE, Fowler, JH (2012) A 61-million-person experiment in social influence and political mobilization. Nature 489: 295–298.
As social scientists argued over this, many computer scientists and industry practitioners responded with sheer confusion.
The practice of A/B testing is commonplace in and essential to the production of algorithmically produced recommendations, which are the cornerstone of Facebook’s news feed.
It’s not just that it’s standard practice; it’s the very foundation upon which these systems are built.” (Boyd, 2015)
You write an email to Joe. You say “The project sucks”
In transit, the email is changed with your permission to say “The project needs work”
Gmail autocomplete
Source: https://www.slashgear.com/gmail-autocomplete-turn-on-smart-compose-right-now-08530148/
Using gpt-3
https://machinelearningknowledge.ai/openai-gpt-3-demos-to-convince-you-that-ai-threat-is-real-or-is-it/
Automating HR
Employee life cycle
Onboarding
Reward & Performance Management
Recruitment & Selection
Training & Development
Succession Planning
Transition
Deloitte
2019
Descriptive Analytics
After controlling for tenure and performance, the typical leaver makes 20% less compared to stayers
Predictive Analytics
Sarah has a 97.3% chance of quitting next month.
Prescriptive Analytics
Reducing the average working hours to 41.5 from 52 is likely to reduce turnover rates 14.5% and save $140,120 per month.
Causal Analytics
Sarah is more likely to stay if she has her own office.
Automation
I have reassigned Sarah to her new office.
Employee life cycle
Onboarding
Reward & Performance Management
Recruitment & Selection
Training & Development
Succession Planning
Transition
Scenario
Your company has many unfilled positions.
You need to advertise these openings to the right places in the right way.
You need to fill them as quickly as possible.
Some roles are more critical to the mission of the company than others.
Recruitment & Selection
Descriptive
What kind of roles are currently unfilled?
In what departments?
How many applications do we currently have for each position?
What are the characteristics of people applying for each position?
Diagnostic
What kind of characteristics are associated with high performance in each role?
What properties of an application predict a speedy onboarding process?
What kind of job advertisement brought in the best people?
Predictive
How likely is this applicant to accept an offer?
How likely will this applicant succeed in the role we are recruiting them for?
How many quality applicants will we get from advertising on Indeed vs. ZipRecruiter?
Prescriptive
How should we design a job to best suite a high quality application?
Who should be invited for an interview?
Which roles in the organization should we prioritize next?
What should a job advertisement say to get the right person?
Recruitment & Selection
Scenario
Your company has many unfilled positions.
You need to advertise these openings to the right places in the right way.
You need to fill them as quickly as possible.
Some roles are more critical to the mission of the company than others.
Automation?
Organization could use gig-work boards to automatically hire contractors.
Recruitment & Selection
Scenario
How can we persuade managers involved in a talent management program to relocate internationally?
Minbaeva (2018)
Transition
Descriptive
What opportunities are there for international moves?
Which employees have the skills and background to fill these roles?
Which employees have made international relocations, which have declined them?
Diagnostic
What are the characteristics of a job or person or person-job fit where relocations were successful vs. unsuccessful?
What locations are people more likely to relocate to?
What incentives were given to successful relocations relative to the job and location?
Predictive
How likely is Minerva to relocate if we incentivize her with $X bonus?
Prescriptive
What incentives need to be provided for this relocation?
Where is the best place to open a new international office to maximize the likelihood of successful relocations?
Transition
Scenario
How can we persuade managers involved in a talent management program to relocate internationally?
Minbaeva (2018)
Automation?
Using information best places to open office to automatically purchase and lease office space.
Automatically schedule and book international transfers for employees and resources.
Transition
These elements focus on human resource management.
But there are many other aspects of managing organizations that are missing.
Benefits
Ethics of Self-Driving Cars
In the USA there were 37,133 motor vehicle fatalities.
The #1 cause of death for ages 5-24 are motor vehicle accidents, and it’s the #2 cause for ages 25-64.
https://twitter.com/calebwatney/status/747772853102084097
AI for Health?
Bureaucratic Displacement
https://youtu.be/mafghkNmKgo
People Analytics
People Analytics are Here
This technology is already available to employers.
It’s becoming more effective, cheaper, and easier to use.
Vision
Platforms
From Marshall van Alstyne’s slides on the topic
Walmart is estimated to employ 2.2M.
In terms of paychecks Uber employs about 2.25 million people.
But Uber has largely
replaced middle management entirely with software
See talk by Devin Fidler
“Self Driving Organization”
Seek Work
A program seeks out potential clients who would want market research
Customer Defines Needs
The needs of the project generate a list of tasks to fulfill.
Build Team
System uses platforms to find necessary skills.
AI compiles results
A human at his point could curate and determine if the output is satisfactory before sending it to the client.
Finding Additional Resources
Finding the resources needed - computation, project space, labs.
Define Needed Skills
Each task has a required skill set. Can be defined by human or algorithm.
Match Skills to Resource
Tasks are matched to people using a platform, or other AI
Create Task
Director creates list of necessary tasks
Evaluate Outputs
Quality of outputs is rated, or quality is determined by outcomes
Finding Resources
Finding the resources needed - computation, project space, labs.
Risks
Bias
Using the past to predict the future, is a kind of bias.
AI Replacing People in 3D Animation
AI Replacing People in 3D Animation
At Risk Jobs
Safe Jobs
Examples:
Skills matching
Data entry and aggregation
Scheduling
Team matching
Compensation and Reward
Examples:
Conflict management
Negotiation
Strategic Planning
Source: https://youtu.be/HX6M4QunVmA
To Summarize...
Organizations are quickly transforming.
New technologies have the ability to accomplish tasks previously thought beyond automation.
There are great benefits to efficiency and effectiveness.
There are great risks in terms of privacy, bias, and job loss.
Now is the right time to be thinking these things.
Thanks!
end
The big change now is the scale and the efficiency.
The number of connections is magnitudes greater.
The algorithms for matching are much better than wandering from stall to stall.
“Workforce analytics refers to the processes involved with understanding, quantifying, managing, and improving the role of talent in the execution of strategy and the creation of value. It includes not only a focus on metrics (e.g. what do we need to measure about our workforce?), but also analytics (e.g. how do we manage and improve the metrics we deem critical for business success?”
Huselid (2018, p. 680)
Ethical considerations
Privacy concerns
What limits should be placed on employers’ access to data?
Data security
How do you secure your data from breaches?
Data access
How do you determine who should have access and who should not?
Cobots are coming quickly.
Source: https://youtu.be/HX6M4QunVmA
Rethinking Work
We need to rethink what work means and how we as a society reward work.
We derive our identity from our work.
Raising children is work.
Taking care of an elderly parent is work.
Contributing to your community is work.
What are our rights?
The New Deal on Data / GDPR
Sandy Pentland: New Deal on Data
1. You have a right to possess your data.
Companies should adopt the role of a Swiss bank account for your data.You open an account (anonymously, if possible), and you can remove your data whenever you’d like.
2. You, the data owner, must have full control over the use of your data.
If you’re not happy with the way a company uses your data, you can remove it. All of it. Everything must be opt-in, and not only clearly explained in plain language, but with regular reminders that you have the option to opt out.
3. You have a right to dispose or distribute your data.
If you want to destroy it or remove it and redeploy it elsewhere, it is your call.
GDPR adds a requirement that users be notified changes to the analytics, or how the data is shared.
New IEEE Ethical Guidelines
Legal frameworks and enforcement aren’t enough
DATA SHARING
PERSONAL
TARGETING
ANALYTIC SCRUTINY
Democratizing Data
Privacy is important.
… but so is saving lives.
Andrew Beam, a researcher in public health and AI,says the huge issue holding back the benefits of AI in healthcare is the access to data.
We need more evidence, we need more data.
With modern methods of machine learning and causal modeling we can establish useful causal relationships - if we have enough data, diverse data.
Identity Escrow?
Escrow Firm
Identifiable Data
Modeling
Democratizing AI
Democratizing AI
Analytics and AI, if used on the public, must be published and shared with the public.
GDPR currently requires that firms notify users if they apply new analytics to their data, and they have the right to be forgotten if requested.
Creepy or cool?
Deep fakes
Source: https://youtu.be/Nq2xvsVojVo
Creepy or cool?
76% creepy
10% cool
Clothing and wearables include sensors/tracking devices that allow retailers to track you in exchange for a discount.
69% creepy
14% cool
Companies understand your shopping habits so well that they are able to use artificial intelligence to choose and automatically order products on your behalf.
31% creepy
46% cool
You can use fingerprint scanning to pay for items and get automatic home delivery, all from the store floor.
Overall: 41% creepy
Millennials: 27% creepy
Computer programs (such as chatbots) use artificial intelligence to help you answer customer service questions, rather than a real person.
Agenda
Disruptive Technology
Organizations are a technology for distributing resources.
They match tasks to people who can accomplish them, then distribute the gains.
Daft 2008, page 14.
Galbraith 1974
Coase 1937
Zenger 2011
What is workforce planning?
“A core business process to align changing organisation needs with people strategy”
(CIPD, 2018)
Objective of strategic workforce planning:
Achieving sustainable performance
Planning decisions guided by “better information” to make decisions about people in organisations
(Torrington et al., 2017)
How can HR planning support strategy?
Capability Gaps
Workforce Utilisation
Capability Surpluses
Talent Pool Development
Having the right processes in place
Establish systems and workflows that enable efficient data extraction and data organization
Need to ensure data quality
Minbaeva (2018)
From operational to strategic
Organizations struggle to move from operational reporting to using HR analytics for strategic decisions
Not being able to link data to current and future strategy discussions
Minbaeva (2017)
Operational Reporting
Strategic Decisions
Struggle
Platforms and the Environment
Platforms increase efficiencies
Platforms worked in Australia to reduce water use. Those who had low valued crops found they could make a greater return by selling the rights to their water to other farms with higher valued crops.
(from Marshall platform book)
Advanced Chess is a form of chess where each human player uses a computer chess program to explore the possible results of candidate moves. The human players, despite this computer assistance, are still fully in control of what moves their "team" (of one human and one computer) makes.
https://en.wikipedia.org/wiki/Advanced_Chess
Platforms & Machine Learning
From Deven Fidler’s talk. https://youtu.be/moXMXkZ7rXI
Two biggest trends.
Platforms are good at matching people with resources.
Upwork for matching freelancers with work.
Robomanager gets a job request, uses the platforms to find people who can do the work, build teams that should work, and get them the resources.
Disaster relief.
We need to rethink the idea of jobs.
We need to rethink what companies are. Organizations are a form of technology, and a technology that’s ripe for disruption.
In 20 years not only will most of our cars, but most of our organizations are self driving.
https://www.rethinkery.com/insights/building-future-companies
Privacy
Why is privacy important?
We don’t want to hear every thought a person has before they choose the right thing to say in the moment.
Platforms have other weaknesses.
Bias
The need for explanation. A model cannot be a black box. We need to know how it arrived at it’s determination.
If the model explains that a person be denied a loan on a house because of their race, that is a major problem.
Models are trained on a history of past data.
Bias in Machine Learning
Bias in Machine Learning
https://arxiv.org/pdf/1902.11097.pdf
“In this work, we propose the concept of predictive inequity in detecting pedestrians of different skin tones in object detection systems. We give evidence that standard models for the task of object detection, trained on standard datasets, appear to exhibit higher precision on lower Fitzpatrick skin types than higher skin types.”
Google: Machine Learning Fairness
Selection Bias
Using the past to predict the future, is a kind of bias.
This is similar to the idea of modeling creditworthiness based on the experience with past credit customers: those are likely the people whom you had deemed to be creditworthy in the past! However, you want to apply the model to the general population to find good prospects.
This is an example of selection bias—the data were not selected randomly from the population to which you intend to apply the model, but instead were biased in some way
Artificial Intelligence in Organizations
An Integrating Topic Space
Moving People not Ideas
Knowledge is sticky. It is
(Anderson 1999; Szulanksi 2000)
Knowledge is often non-codifiable, and tacit. The best way to transfer the knowledge is to transfer the people who have it.
(Ranucci and Souder, 2015)
Knowledge is moved by moving networks of members, tools, and tasks.
(Argote and Ingram, 2000)
Facebook and Sentiment
In 2014 a paper was published that detailed an experiment on Facebook.
Researchers discovered they could manipulate the emotional state of their users. By filtering news feeds for negative sentiment, users future posts were more negative.
None of the users consented to be in this study, had the option to drop out, or were even aware of it.
Facebook and Sentiment
Rather than using machines to predict behavior and attitudes..
This study showed everyone that we could use machine learning to change people’s behavior and attitudes.
Machine dreaming
https://youtu.be/gvjCu7zszbQ
Deployment
Descriptive
just a description of what is happening and what exists
Diagnostic
why is this thing happening?
Predictive
what will happen?
Prescriptive
how do I make something happen?
Image from: http://www.skmgroup.com/evolving-analytics-from-descriptive-to-prescriptive/
Platforms
Network effects
It’s easier to add new technology to a community, than it is to add a community to technology.
The goal of a platform is “consummate” the match.
Every time you create a match, you create wealth. The better the matching, the better the wealth creation.
Machine learning drives the match.
Merging ideas, but not ties
People moved through the topic space with ease..
..but had less success with forming new ties with people.
Befriending a Roman
or Learning Latin?
AI Replacing People in 3D Modeling Animation
AI Replacing People in 3D Animation
AI Replacing People in 3D Animation
Platform
Standard accounting practices don’t account for the value of communities, but stock markets do.
It is easier to scale network effects outside the firm than inside. The focus of management shifts from inside to outside. The management of human resources shifts from employees to crowds.
The manage of externalities becomes the key leadership skill.
notes
Cool vs. Creepy
https://www.accenture.com/us-en/blogs/blogs-cool-vs-creepy
https://www.sailthru.com/marketing-blog/week-retention-personalizations-cool-vs-creepy-matrix/
Deloitte’s report on the future of organizations and HR
Predicting turnover:
Checking LinkedIn to see how people are networking - Cool
Sentiment analysis of email content to predict turnover - Creepy
Notes
Ethics and stuff
Bias in models
Explainers
The Self Driving Organization: What Happens When Software Can Run the Company
The end of personal privacy?
https://keithlyons.me/wp-content/uploads/2011/01/ed_siii_madan_waber_et_al.pdf
Tonidandel, S., King, E. B., & Cortina, J. M. 2018. Big Data Methods: Leveraging Modern Data Analytic Techniques to Build Organizational Science. Organizational Research Methods, 21(3): 525–547.
The Self-Driving Organization and Edge Analytics in a Smart IoT World
Notes
The Self-Driving Experience
https://www.sisglobal.com/share/4228_The-Self-Driving-Experience-WP160811INT.pdf
Platform technologies, platform companies
Automating management?
https://thefutureorganization.com/self-driving-organization-happens-software-can-run-company/
Flash crash and high-frequency-trading
Customer service
Notes
https://www.rethinkery.com/insights/large-companies-turn-to-bots-to-manage-other-bots
Company offering a platform that automatically generates a website and configures it automatically over time in response to use.
https://www.b12.io/how-it-works