COMMUNITY BIG (HR) DATA, BIG (HR) CHALLENGES · engine can continue spinning out faulty and...
Transcript of COMMUNITY BIG (HR) DATA, BIG (HR) CHALLENGES · engine can continue spinning out faulty and...
COMMUNITY‘BIG (HR) DATA, BIG (HR) CHALLENGES’POOLSTOK
26/02/2019 – De Punt Gent(brugge)
Welkomstwoord Vincent Van Malderen
01
2
3
COMMUNITY
Coachingtrajecten onder de loep
Leuven - Stadhuis
15 juni 2017
Evaluatie²
Gent - Zebra
24 oktober 2017
Welzijn, pick (y)our brain
Mechelen - Lamot
13 maart 2018
Reflecties over Flexwerk
Antwerpen - Havenhuis
13 september 2018
Big (HR) Data, Big (HR) Challenges
Gent - De Punt
26 februari 2019
5
6
7
AI
Machine learning / neural networks
Algoritmes/ predictiveanalytics
Blockchain in Gov
RPA / Robotics / ...
NU
TOEKOMST
ETHIEK
PRIVACY
GDPR
EVIDENCE-BASED
8
9
10
11
12
13
Powered by …
14
Poolstok community:
50% discount!!
De code 'OpenBeliever_Poolstok' geeft 50% korting op een regular
ticket. (Geldig van dinsdagochtend tot woensdagavond)
15
1JAAR
16
NIEUWE WEBSITE! ➔ Nieuws, blog, community, …
17
13u30 - 13u40 Welkomstwoord door Vincent Van Malderen, algemeen directeur, Poolstok
13u40 - 14u20 HR analytics: een brug over troebel waterBernie Caessens, Resolved
14u20 - 14u50 Ethische aspecten van big dataKatleen Gabriels, assistant professor Maastricht university
14u50 - 15u20 Pauze
15u20 - 15u50 Een brug slaan tussen wetenschap en data, voer voor HR innovatieCédric Velghe, The VIGOR Unit
15u50 - 16u20 The Pros and Cons of Digital Footprint Analysis in HRVesselin Popov, Business Development Director for the University of Cambridge Psychometrics Centre
16u20 – 16u40 Debat
16u40 – 18u30 ReceptiePowered by Rooffood
18
Dank u
19
RESOLVEDdata-science for humans
HR ANALYTICSEEN BRUG OVERTROEBEL WATERBERNIE CAESSENS | MANAGING PARTNERPOOLSTOK COMMUNITY | 26.02.19 | GENT
RESOLVEDdata-science for humans
THE UNIVERSALPROBLEM-SOLVING
ALGORITHM
WRITE DOWN THE PROBLEM
WRITE DOWN THE SOLUTION
THINK REALLY, REALLY HARD ABOUT IT
1
2
3
RESOLVEDdata-science for humans
THE ANALYTICSVERSION
TAKE YOUR HR DATA
WRITE DOWN THE SOLUTION
1
2
3
APPLY HR ANALYTICS TO UNDERSTAND AND SOLVE
RESOLVEDdata-science for humans
AITO THERESCUE
TAKE YOUR HR DATA
WRITE DOWN THE SOLUTION
1
2
3
USE AI: LET THE COMPUTER SOLVE IT
RESOLVEDdata-science for humans
DO NOTSTART WITH DATATHINK SMALL
DATA
INSIGHT
STORYDATA
INSIGHT
QUESTION
THE EMPIRICAL METHOD
RESOLVEDdata-science for humans
[2]
WHAT IS HRANALYTICS
UNDERSTANDING HOW HUMAN BEHAVIORSYSTEMATICALLY INFLUENCES THE RESULTS OFAN ORGANISATION THROUGH THE USE OFMEASURABLE (QUANTIFIABLE) INDICATORS
RESOLVEDdata-science for humans
A FAMILIAREXAMPLE
CULTURE IS KING
Improving our culture will improve our company
Improve productivity
More applicants
Outpace competition
A two-step approach to improving culture
Release satisfaction survey
Develop plan
1
2
How we’ll measure success
70% Survey completion
10% Increase in satisfaction
20% Increase in applicants
5x more referrals
RESOLVEDdata-science for humans
THE PROBLEM
IN HR
A STATEMENT, NOT A PROBLEMBY THE WAY, WHAT DO YOU MEAN WITH CULTURE?
SOUNDS LOGICAL, BUT IS THERE A REAL CAUSAL LINK WITH ALL THREE CONCEPTS?
IMPLICITLY ASSUMES A LINK BETWEEN CULTURE AND SATISFACTION
NO SUCCESS FOR PRODUCTIVITY?ONLY KPI’S WITHIN HR
RESOLVEDdata-science for humans
[1]
HC BRIDGEFRAMEWORK
IMPACT EFFECTIVITY EFFICIENCYWHAT IS THE IMPACT ON OUR STRATEGIC GOALS OF INCREASING THE QUALITY OR AVAILABILITY OF OUR TALENT POOL?
HOW STRONGLY DO OUR HR PROGRAMS AND PROCESSES INFLUENCE THE CAPACITY, ACTIONS AND INTERACTIONS AMONGST OUR TALENT POOLS
HOW MUCH HR PROGRAMS AND PROCESSES CAN WE GET FOR OUR INVESTMENTS IN TIME AND MONEY
HOW HIGH IS OUR ANNUAL TURNOVER?
HOW MUCH OF ANNUAL TURNOVER IS REGRETTED LOSS?
HOW MUCH DID REGRETTED LOSS IMPACT OUR CAPACITY TO CLOSE DEALS?
RESOLVEDdata-science for humans
ANALYTICSTHE FAILURE
LACK OF A FRAMEWORK
LOGIC
LACK OFPROPER
DATA
RESOLVEDdata-science for humans
LACK OF AFRAMEWORK
IMPROVE PRODUCTIVITY
MORE APPLICANTS
COMPANY CULTURE
RESOLVEDdata-science for humans
SOURCES FOR
LOGIC
DECOMPOSE THE PROBLEM AND
CONNECT TO YOUR KNOWLEDGE
SCIENTIFICLITERATURE
RESOLVEDdata-science for humans
ASK QUESTIONSTO BUILD AFRAMEWORK
Which values, beliefs and norms guide individual’s and group behavior within our organization
Overall, more than 25% of our projects are over budget and over time. Is there anything related to our culture that plays a part in this?
TEAM
VALUESCOMPETENCIES
PROJECT OK?
RESOLVEDdata-science for humans
EXPAND AND
REFINEFRAMEWORK
PROJECT ON TIME/BUDGET
TEAM MEMBER VALUES
SUPPLIERS
Attract applicants with the right set of values
BETTER PROJECT
ESTIMATES
PM SKILLS
TEAM MEMBER MOTIVATION
Attract the right PM skills
RESOLVEDdata-science for humans
TEAM COMPOSITON
ANALYTICS &LOGIC GUIDEACTION
PROJECT ON TIME/BUDGET
TEAM MEMBER VALUES
SUPPLIERS
Attract applicants with the right set of values
EMPLOYER BRAND &RECRUITMENT PROCESS
LEARNING &DEVELOPMENT
PERFORMANCE & REWARD
RESOLVEDdata-science for humans
SUPPORTSDECISIONS
TEAM COMPOSITON
PROJECT ON TIME/BUDGET
TEAM MEMBER VALUES
Attract applicants with the right set of values
EMPLOYER BRAND &RECRUITMENT PROCESS
LEARNING &DEVELOPMENT
BUDGETTALENTSKNOW-HOWEXPECTED IMPACT
RESOLVEDdata-science for humans
LACK OFPROPERDATA
HR DATA ARE EVERYWHERE
HR DATA ARE MESSY
HR DATA ARESPARSE & DISCRETE
MANY INTANGIBLES
THE EMPIRICAL METHOD
RESOLVEDdata-science for humans
[3]
THE STATE
OF AI
PATTERN RECOGNITION ON PAR WITH HUMANS
WITH MASSIVE DATA FOR TRAINING
RESOLVEDdata-science for humans
[4]
THE HOW
OF AILOW LEVELRAW DATA
NETWORKRESPONSE
CORRECTRESPONSECOMPARE
ADJUST TO MINIMIZE ERROR
LEARNING
RESOLVEDdata-science for humans
LEVEL 2 | USE MACHINE LEARNING TO MEASURE BEHAVIOR AND CAPTURE REAL-TIME DATA
LEVEL 1 | USE MACHINE LEARNING TO CLEAN OR LABEL DATA AUTOMATICALLYAPPLICATION
IN HR?
HR DATA ARE EVERYWHERE HR DATA ARE MESSY
HR DATA ARESPARSE & DISCRETE MANY INTANGIBLES
LEVEL 3 | USE MACHINE LEARNING TO MEASURE, TRACK & PREDICT
DATA-CAPTURE
DECISION-MAKING
RESOLVEDdata-science for humans
EXAMPLESFOR HR
LEVEL 2
LEVEL 1
LEVEL 3
AUTOMATIC CV-SCREENING: EXTRACTING STRUCTURED INFORMATION FROM UNSTRUCTURED TEXT
AUTOMATIC CLASSIFICATION OF E-MAIL COMPLAINTS INTO CATEGORIES
EXTRACTING PERSONALITY PROFILES FROM SOCIAL MEDIA
EXTRACTING JOB POSITION FROM TWEETS
E-MAIL SENTIMENT ANALYSIS
PREDICTING PERSON-JOB FIT FROM CV-ANALYSIS
RESOLVEDdata-science for humans
AUGMENTEDINTELLIGENCE
DATA
INSIGHT
QUESTION
LEVEL 1 | ACCESS TO RAW DATA
LEVEL 2 | ENRICHED DATA
DATA
STORY
LEVEL 3 DECISION MAKING
MORE RESEARCH NEEDED
RESOLVEDdata-science for humansCONCLUSION
WRITE DOWN THE BUSINESS PROBLEM
IMPLEMENT ACTIONS AND START OVER
WORK REALLY, REALLY HARD ON IT WITH THE PROPERTOOLS & METHODS
1
2
3
ANALYTICS AND AI OFFER ADDED VALUE FOR HR BUTNEITHER IS A COMMODITY (YET) NOR AUTOMAGIC
RESOLVEDdata-science for humans
THANK
YOUSOURCES
IMAGESAll Icons by FlatIconSlide 1 | Photo by Jonas Verstuyft on UnsplashSlide 2 | Photo Feynman Wikimedia CommonsSlide 3 | Photo by PixelRaw on UnsplashSlide 4 | Photo by Priscilla Du Preez on Unsplash
[1] Boudreau, J.W. & Ramstad, P.M. (2005). Talentship and the Evolution of Human Resource Management: From “Professional Practices” To “Strategic Talent Decision Science” Human Resource Planning Journal. 28 (2) 17-26.
[2] van den Heuvel, S. & Bondarouk, T. (2016). The Rise (and Fall) of HR Analytics. Paper presented at the 2nd HR Dvision International Conference, Sydney.
[3] Goodfellow, I. et al. (2014). Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks, arXiv: 1312.6082v4
[4] Radford, A., Jozefowicz, R. & Sutskever, I. (2017). Learning to Generate Reviews and Discovering Sentiment, arXiv: 1704.01444v2
https://www.washingtonpost.com/opinions/its-okay-to-be-paranoid-someone-is-watching-
you/2018/03/27/1a161d4c-2327-11e8-86f6-
54bfff693d2b_story.html?noredirect=on&utm_term=.8a457c337d5d
https://www.academicforecast.org/about
https://www.academicforecast.org/about
• Big data ethics = critical thinking
- Correlation is not causation
www.facebook.com/depoorterdries
• Big data ethics = critical thinking
- Not everything that counts can be counted
Outline
‛Ethical implications of:
‛1. Tracking employees (IoT)
‛2. Machine learning at work (AI)
Quantifiedworkplace.eu
Quantified workplace
• Presented as self-tracking but: also other-tracking
• Surveillance and coveillance
• Neo-Taylorism / Digital Taylorismo ‘The corporeal turn’
• Privacy and physical integrity
• Control versus Self-Determination Theoryo ‘Success’, ‘efficiency’, ‘productivity’
‛Gabriels, K., & Coeckelbergh, M. (2019, in press). ‘Technologies of the self and other’: How self-tracking
technologies also shape the other. Journal of Information, Communication and Ethics in Society 17 (2).
https://www.businessinsider.com/amazon-patents-bracelet-that-tracks-workers-2018-2?r=US&IR=T
Humanyze.com
https://blogs.nvidia.com/blog/2016/07/29/whats-difference-artificial-intelligence-machine-
learning-deep-learning-ai/
Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., and Thrun, S. (2017). Dermatologist-level
classification of skin cancer with deep neural networks. Nature 542, pp. 115-118. Doi: 10.1038/nature21056.
‛“Algorithms are opinions embedded in code” (Cathy O’Neil)
https://www.theverge.com/2018/10/10/17958784/ai-recruiting-tool-bias-amazon-report
• Analyses (and correlations) with ‘bad’ datasets
• Algorithm is not inherently objective
• “statistical systems require feedback” (p. 7)- “Without feedback, however, a statistical
engine can continue spinning out faulty and
damaging analysis while never learning from
its mistakes” (p. 7)
• The model you are applying must be trained with a lot of data, should be transparent, and should be updated regularly
O’Neil, C. (2016/2017). Weapons of Math Destruction. How Big Data Increases Inequality and Threatens Democracy.
Broadway Books, New York.
‛“you could argue that WMDs are no worse than the humannastiness of the recent past. (…) But human decisionmaking, while often flawed, has one chief virtue. As humanbeings learn and adapt, we change, and so do ourprocesses. Automated systems, by contrast, stay stuck in timeuntil engineers dive in to change them. (…) Big Dataprocesses codify the past. They do not invent the future.Doing that requires moral imagination, and that’s somethingonly humans can provide. We have to explicitly embed bettervalues into our algorithms, creating Big Data models thatfollow our ethical lead. Sometimes that will mean puttingfairness ahead of profit”‛(O’Neil, 2017, pp. 203-204)
‛1. Codes of ethics and professional conduct• “the Hippocratic Oath ignores the on-the-ground pressure that data
scientists often confront when bosses push for specific answers”
(O’Neil, 2017, p. 206)
‛2. Strong regulation• “To disarm WMDs, we also need to measure their impact and
conduct algorithmic audits” (O’Neil, 2017, p. 208)
‛3. More research• Example: https://unbias.wp.horizon.ac.uk/
• No blind faith in data set and algorithms
• Does the problem need a big data solution?
- If yes, look for an evidence based one- Importance of education and critical thinking- Importance of long term vision- Stakeholders have to be informed
https://www.fatml.org/resources/principles-for-accountable-algorithms
Een brug slaan tussen wetenschap en dataVoer voor HR innovatie
Cédric Velghe@Poolstok Community
26 februari 2019
“Today, I will play the bad cop and shoot my own foot.”
(“HR analytics” OR “people analytics” OR “human capital analytics” OR “workforce analytics”)
● To date, this query delivers only 76 hits in Web of Science
● There is no generally accepted definition and conceptualization of what HR Analytics is (not)?
. Assumption 1 “HR Analytics is hard science.”
The research on the alleged benefits of HR analytics remains very scant (Rasmussen & Ulrich, 2015)
● Multiple pitfalls threaten to prevent HR analytics from delivering on its promises.
● The primary issue is that HR analytics too often seems to become an end in itself.
○ Governed by expanding data-availability ↔
starting from organizational challenges
○ Preoccupation with revealing statistical
associations ↔ testing substantiated theories
○ Dustbowl empiricism ↔ delivering
value to HR decision-making
. Assumption 2 “Organizations benefit from adopting HR analytics”
“Garbage in, garbage out.”
● Digital footprints are often incomplete, not necessarily accurate or representative, and often
irrelevant to the job.
○ → Collect digital data through controlled and standardized environments, e.g. digital
interviewing
○ ! Text analysis is gaining maturity, but, be wary of
psychometric claims based on image-, sound- or
video-analysis as we still lack substantiated theories
. Assumption 3 “Cybervetting using machine learning algorithms improves hiring decisions.”
“The query, (game* AND "personnel selection"), results in only 19 hits in Web of Science.”
● Statistical associations between in-game behavior and personal characteristics or outcomes are not
sufficient evidence for the validity, reliability and utility of recruitment games.
● Theories of how specific personal characteristics/outcomes are manifested through in-game
behavior need to be validated through controlled experiments
. Assumption 4 “Games are better at hiring than traditional psychometric tests”
Build HR analytics projects on the extensive collection of substantiated theories and meta-analytical findings in the academic literature (van der Togt & Rasmussen, 2017)
● ≠ cherry picking scientific studies
● = Systematic Reviews, Rapid Evidence Assessments (REA), Critically Appraised Topics
. Solution “Bridging science and practice“
“Consulting science can prevent organizations from losing time over avenues that have been extensively researched in the literature.”
● For instance, pay is a weak predictor of employee voluntary turnover
○ → Allocate data-analytical resources to other hypotheses, e.g. leadership
. Benefit 1 “Asking the right questions and focussing the efforts“
“Science can also inform organizations on the most adequate conceptualization, measurement and operationalization of the variables that they wish to analyze.”
● The distance between home and work is not a well suited metric for analyzing the impact of
commuting on employee voluntary turnover
● Commuting time can be estimated with the help of Google maps
. Benefit 2 “Adequately conceptualizing and operationalizing the variables“
“A literature review provides organizations with the opportunity to explicitly assess the potential of the analytical methods that have so far been used by academics and make well-informed choices on what methods to adopt.”
● E.g. in turnover research researchers have been overly reliant on cross-sectional and static cohort
designs (Allen, Hancock, Vardaman, 2014)
○ → Include temporal considerations, e.g. survival analysis with time dependent covariates and
time series analysis.
. Benefit 3 “Choosing the right analytical methods“
“One of the biggest risks with HR analytics is misjudging the statistical output (Wenzel & Van Quaquebeke, 2017). ”
● Whenever you obtain findings which do not align with the state-of-the-science, be careful
● Applying the effect-sizes from published meta-analyses as the prior belief in Bayesian estimations,
will reduce the risk for capitalizing on chance.
. Benefit 4 “Critically appraising the findings and improving prediction accuracy“
“Organizations can obtain valuable insights and recommendations from science which could never be obtained from the available data.”
. Benefit 5 “Consulting the scientific evidence is a core aspect of evidence-based HR“
“Science reviews and insights from organizational data provide a systematic and judicious overview of what is (not) known about your problem.”
● Avoid losing resources over reinventing the wheel
● Avoid losing money over investments that could have been expected to be ineffective
● Substantiated conclusions and recommendations that can convince your various stakeholders to
engage with the suggested innovation
● Identifying and prioritizing potential angles of approach for inventing innovative solutions
(employee* OR worker* OR staff OR workforce) AND (turnover OR retention OR attrition OR leave OR
stay OR quit OR withdrawal) → 38.783 hits in WoS
. Benefit 6 “Knowledge and expertise are key to successful (HR) innovation“
Questions?
The Pros and Cons of Digital Footprint Analysis in HR
OBJECT IVE 2 OF 7
To harness methodologies from psychometrics and big data analytics in predicting and
understanding human behaviour in the online environment
Disruption in context
72% feel it’s important to embrace AI, but only 31% feel ready to address it
A. 24% automating routine tasksB. 16% augmenting human skills
C. 7% restructuring work entirely
Disruption in context
•
•
•
•
•
•
•
•
•
•
•
•
•
•
PREDICT PSYCHOLOGY
MAKE A DECISION
CREATE DIALOGUE
PROVIDE FEEDBACKGET MORE DATA
PREDICT PSYCHOLOGY
MAKE A DECISION
RECORD OUTCOME
Obtaining a psychological signal with consent
Using tests Using social media
Data shared with 80+ Universities worldwide
Honest feedback was the only incentive
45 peer-reviewed articles since 2011
30 validated psychometric tests
All data collected through opt-in
6 million volunteers’ psych and social media profiles
myPersonality project (2007-)
March 2013
January 2015
45 peer-reviewed publications using our data since 2011
October 2017
Political Views Religious Views Financial Risk
+ Use of addictive substances, parents’ relationship status, profession, sexuality, ethnicity, gender, age and more
Intelligence Life SatisfactionBIG5 Personality
Kosinski, Stillwell & Graepel. PNAS 2013
Predictions from social media data
Computers assess personality better than we do
Number of Facebook Likes (log scaled)
Acc
ura
cy (
self
-oth
er a
gre
emen
t)
See Youyou W., Kosinski & Stillwell. PNAS 2015
Principles for AI implementation
Based on Psychometrics Centre survey of 34,267 respondents globally
PREDICT PSYCHOLOGY
MAKE A DECISION
Communications
AI in Communications
Pros
• Automation reduces costs
• Summary and transcription tasks
• Chatbots can generate interview questions and handle onboarding
• Detect emerging employee concerns
• Ongoing satisfaction measurement
• More relevance and engagement
• Personalised content more persuasive
Cons
• Can feel impersonal
• Breed linguistic determinism
• Closed vocab tools lack nuance
• Out-of-the-box products not sensitive to company culture
• Passive analysis tools can inhibit natural conversations
Same newspaper, same date
Psycholinguistic Tailoring
Park et. Al JPSP 2015
Predicting interpersonal deviance
Predicting interpersonal devianceD
.Pre
oti
uc-
Pie
tro
, J
.Ca
rpe
nte
r, S
. G
iorg
i, L
. U
ng
ar.
CIK
M 2
016
pp
.76
1-7
70
Study design
O C E A N
O Congruent
C Congruent
E Congruent
A Congruent
N Congruent
Dance like no one’s watching - but they totally are
Beauty doesn’t have to shout
Ad
Va
ria
nt
Target Group
Introverts Extraverts
Intr
overt
ed
Extr
avert
ed
Study design
Ad Variants:
Personality-matched content is 2 x as persuasive
RETURN ON INVESTMENT (%)
Introverts Extraverts
200%
400%
0
Dance like no one’s watching - but they totally are
Beauty doesn’t have to shout
Matz, Kosinski, Nave and Stillwell. PNAS 2017
The Psychometrician’s Dilemma
The Psychometrician’s Dilemma
Perfect Match App
•
•
•
•
Perfect Match App
Machine bias
AI in Communications - Compromise
CREATE DIALOGUE
PROVIDE FEEDBACKGET DIGITAL FOOTPRINT
PREDICT PSYCHOLOGY
MAKE A DECISION
RECORD OUTCOME
AI&
HR