AI Applications in People Management

AI Application in HR

Matthew Bidwell, Associate Professor of Management

AI Applications in HR

• Figure out what kind of data we are going to use

• What data can we use to make these predictions?
• How would we actually use these algorithms?
• Engagement
• Attrition
• Internal career paths
AI Applications in People Management
AI and Engagement

Matthew Bidwell, Associate Professor of Management

Sentiment Analysis

Basic principle is looking at emotion in text

• Starts with a pre-defined dictionary of words
• Long list of all of the words we associate with positive emotions and another
lost list of all of the words we associate with negative emotions
• Count the number of words with positive emotions and words with negative
emotions
• Compare the frequency differences between them
Sentiment Analysis

Obvious problems
• What if somebody says that they are "not happy"
• Contorted example — “Inadequately excited”
• Differences in how people express themselves
• People might be strategic, particularly if they know we are looking at what
they say
Sentiment Analysis

Accuracy considerations
• Decent correlations with human raters
• The accuracy of any one piece, particularly a short piece of text, could be
questionable
• When you look at large bodies of text, or text written by a variety of different
people, generally the accuracy is pretty good
Where Do We Use Sentiment Analysis?

• Tradeoff between comprehensiveness and invasiveness

• Emails and instant messages
• Legal and ethical issues
• What people post about the company on social media
• If it's on social media, it's public
• Can give you a sense of overall morale, and a warning when things are
changing, but not to the granularity of emails
• Open-ended pulse survey questions— just asking people for a couple of
sentences on how they feel, on a more regular basis
AI Applications in People Management
Topic Modeling

Matthew Bidwell, Associate Professor of Management

Topic Modelling

A technique for extracting meaning from text data through themes

• Trying to figure out the key themes being talked about in large bodies of text
• Being able to code each of the different pieces of text to tell you which
theme is present
Topic Modelling Example

“Very strong culture, people

genuinely loved each other and “A young and inclusive
working there. There was very much team-based environment
a "work hard, play hard" mentality.” that cherished flexibility and
hard work.”

“Our culture was very collegial and impact-oriented.

Working on big societal challenges attracted people
who were very driven, purposeful, and passionate
about their work. There was also a strong spirit of
collaboration and respect across programs and
hierarchies.”
Assumptions of a Topic Model

Words Topics Documents

Wibble Topic A
Text 1
F’tang
Aark
Topic B
Ribbet
Text 2
Wiing
Aggle Topic C
Blang
Riggon Text 3
Topic D
Krogg
Topic Modelling

Important limitations
• We have to tell it how many topics (themes) to look for
• It doesn't tell us what those topics are about
• We have to try to figure that out from the words associated with the topic
• As a way of extracting the key themes from large volumes of text it can be
quite impressive
Measuring Culture Across Organizations
Topic Modelling

Application
• Idea is that you can take whatever open text you have
• Pulse surveys — employee responses to open ended questions
• Use this technique to distill it into the key themes
• Can then look at how these themes emerge over time or distribute
across departments
AI Applications in People Management
AI and Attrition

Matthew Bidwell, Associate Professor of Management

Attrition Overview

In trying to figure out which of our employees might leave, what information
would we look at?
• Personal circumstances and where they are in their career
• Current job
• Some roles have high attrition
• Signals of engagement
• Whether things are going well at work
• How withdrawn are they?
Building ML Models of Attrition

• Use machine learning to combine these characteristics in a systematic way

• Using past data on which people stayed or left
• Idea is to understand how each of these characteristics might feed in to a
prediction model
• Advantages
• Accuracy
• Doesn’t get distracted — keeps monitoring when we have moved on
AI Applications in People Management
Building ML Attrition Models

Matthew Bidwell, Associate Professor of Management

Building ML Models of Attrition

What are some of the factors that matter?

• Demographics, job role
• Transaction level data
• Data on specific events that have happened to people that also shape
their probability of leaving
Transaction Level Data — What Changed?

• Receiving a signal that they are not being valued as highly within the
organization
• Changes in performance evaluations
• Reduced raises and bonuses
• Turned down for promotion
• Rejection from other roles within the company
• Disruptions in their environment
• Colleagues are leaving — “turnover contagion”
• Loss of manager
• Reorganizations
Building ML Models of Attrition

What are some of the factors that matter?

• Demographics, job role
• Transaction level data
• Data on specific events that have happened to people that also shape
their probability of leaving
• Behavioral data
• What are people actually doing day to day?
Behavioral Data

• Sudden declines in performance or productivity

• Information from how people are interacting with others
• Those who have more contacts are more likely to stay
• Those who have fewer contacts are more likely to leave
• Also evidence that the quality of those contacts matters
Quality of Interactions

• Study by a group of researchers at Stanford and Berkeley looking at what

people said in their emails to each other
• Similar language suggest a stronger relationship and deeper
understanding of culture
• Divergence in language predicted that a person would leave the
organization
Social Media

• Push factors — how people are being “pushed” out of an organization

• Pull factors matter as well — are people being “pulled” by the prospect of a
better job elsewhere
• Most people tend to find jobs through their networks
• Online networking has made those searches visible
• Can track if people are searching for new jobs — if they are, they may be
more likely to leave
AI Applications in People Management
The Value of Attrition Models

Matthew Bidwell, Associate Professor of Management

Why Model Attrition?

More insight about the levers that affect turnover

• Experiences at the beginning of employment influence attrition
• Whether or not a manager scheduled a 1-on-1 meeting with a new
employee in their first week
• If an employee’s new team sent someone down to collect them from new
hire orientation vs. when a new hire was expected to find their own way
Allow for early intervention
• If we know more about who is a flight risk, then we know who to focus on
• Many organizations will do a "stay interview”
• Proactive interventions based on flight risk models can be effective
Why Model Attrition?

Planning for replacements

• We can’t prevent all attrition
• Not all attrition is regrettable
• If we can predict when people are likely to leave, we can plan ahead to
minimize disruption
• Manage succession internally
• Planning how many people we’re going to need to hire
AI Applications in People Management
AI and Careers

Matthew Bidwell, Associate Professor of Management

Why Manage Careers?

• Why do organizations care about the management of their employees’

careers?
• Companies care about attrition, and people often leave firms to advance
their careers
• If we make it easier for people to advance their careers within the
organization, they are less likely to leave us
A Bit of History

• Traditionally, organizations had really well-defined career ladders

• Part of the expected contract for workers
• Things started to change over the last 20-30 years
• Organization structures changed — from hierarchical bureaucracies to
flatter, more fluid organizations
• Organizations drew back from paying attention to their employees careers
• Starting to see more organizations making attempts to build clearer career
paths internally
• Can be quite complicated
• Machine learning is an opportunity for organizations in rebuilding these
career paths
Predicting Job Paths

Make it easier for people to find jobs

• Helping people understand what jobs are open to them
• Making it easier for recruiters to find existing employees who would be good
candidates for those roles

Helping people understand the longer term career paths

• Helps them direct their search for the next job
• Gives them confidence that they can grow their career within the firm
• Guide development
Predicting Job Paths

Early example: IBM Watson Career Coach

• Tool that employees could interact with to explain their preferences for their
careers
• Watson would take their preferences and combine it with data on possible
career paths to provide guidance, with the goal of increasing retention and
engagement
Approaches for Developing Career Recommendations

Analyze the career paths of those who have come before

• Works well for common roles
• In a fast moving environment, just relying on prior roles might not be that
effective
Approaches for Developing Career Recommendations

Draw career paths based on skills

• Look for a match between skills and the skill demands of different jobs
• Identifying new career paths that people haven't taken yet
• Paths in and out of new jobs
• The main barrier to using data on skills is that we often don’t have very
good data on skills
• We need to know the skill demands of different jobs and what skills
people actually possess
AI Applications in People Management
Analyzing Skills

Matthew Bidwell, Associate Professor of Management

How Do We Measure Skills?

Ask the employees

• Ask them to create a list of skills they possess
• Ask them to continually update that profile as they continue to learn new
things
• One of the challenges is getting people to fill out the profiles and keep them
up to date
How Do We Measure Skills?

Create strong incentives for people to report their skills

• Can we create internal profiles and get people to fill in their skill data?
• Often suffer from a chicken and egg problem
How Do We Measure Skills?

Make identifying skills part of the formal performance appraisal process

• Requires getting employees to take the exercise seriously
• May be uses for AI here
• Suggest skills that employees might have based on data such as:
• The projects that they have worked on
• The job they’re in
• Even documents that they have written over the year
How Do We Measure Skills?

Infer employees’ skills from the jobs that they held in the past
• Need to know what skills are associated with each job
• O*Net
• Took 900 occupations and coded the skills and abilities that people are
expected to have in those occupations
• Very simple
• Challenge is it assumes that every job within each occupation requires
the same skills — they don’t
• Common approach for organizations getting a first understanding of skills
present in the organization based on the jobs people are doing
How Do We Measure Skills?

Infer employees’ skills from the jobs that they held in the past
• Job descriptions
• Formal job analysis
• Job postings