Business Horizons (2020) 63, 215e226

Available online at www.sciencedirect.com

ScienceDirect
w w w. j o u r n a l s . e l s e v i e r. c o m / b u s i n e s s - h o r i z o n s

AI-enabled recruiting: What is it and how

a
INSEAD, United States
b
AUT Business School, Auckland University of Technology, Auckland, New Zealand

KEYWORDS Abstract AI-enabled recruiting systems have evolved from nice to talk about to
AI-enabled recruiting; necessary to utilize. In this article, we outline the reasons underlying this develop-
Artificial intelligence; ment. First, as competitive advantages have shifted from tangible to intangible as-
Digital recruiting sets, human capital has transitioned from supporting cast to a starring role. Second,
technology; as digitalization has redesigned both the business and social landscapes, digital re-
Human resources cruiting of human capital has moved from the periphery to center stage. Third,
recent and near-future advances in AI-enabled recruiting have improved recruiting
efficiency to the point that managers ignore them or procrastinate their utilization
at their own peril. In addition to explaining the forces that have pushed AI-enabled
recruiting systems from nice to necessary, we outline the key strategic steps man-
agers need to take in order to capture its main benefits.
ª 2019 Kelley School of Business, Indiana University. Published by Elsevier Inc. All
rights reserved.

1. Human capital: From supporting cast assets accounting for roughly 65% of the average
to a starring role firm’s value (Black, 2019), people went from cogs
in the wheel of tangible assets to the engines
“People are our most important asset.” This is not driving value. For the majority of intangible assets,
just a nice-sounding platitude, it is a competitive people either make up their sum and substance, or
imperative born of a seismic shift in where firm they are the direct drivers. What would happen if
value and competitive advantage are found. From you separated people from intangible assets like
the turn of the 20th century until the early 1980s, customer service, customer insight, or innova-
70%e90% of firm value was tied to tangible assets tiondwhat would you have left? The answer is not
such as plant, property, and equipment (Lev, much (Paschen, Pitt, & Kietzmann, 2020).
2000). By 2000, this had flipped. With intangible As a consequence of this shift in the source of
value and competitive advantage, recruitment has
evolved from an important HR activity to a top
* Corresponding author strategic concern for CEOs. Over the last several
E-mail addresses: [email protected] (J.S. Black), years, CEOs listed attraction, selection, and
[email protected] (P. van Esch)

https://doi.org/10.1016/j.bushor.2019.12.001
0007-6813/ª 2019 Kelley School of Business, Indiana University. Published by Elsevier Inc. All rights reserved.
216 J.S. Black, P. van Esch

retention of human capital as their top strategic media, but it had to settle for a few lines of text
concern (Conference Board, 2018). about the job because it was too expensive to
also add information richness. Conversely, a
company could achieve high richness via its em-
2. Digital recruiting: From periphery to ployees, who could describe the company and
center stage open positions in great detail to their friends and
family. However, reach was limited to the people
Against the backdrop of the ascending role of in the employees’ circle. A company could ach-
human capital, the technological context of how ieve reasonable richness and reach through
companies recruit people has also changed. search firms, but the cost of doing so was high
Today, companies are at the beginning of what enough that firms could only afford it for a
we term Digital Recruiting 3.0. At the heart of limited number of position openings. The same
this transformation is the use of artificial intelli- general trade-off between information reach and
gence (AI) in recruiting activities. Computers can richness was true for prospective employees as
now perform tasks and make decisions that nor- well. Simultaneously knowing all the potential
mally require human intelligence. Some key po- jobs out there and knowing a lot about those jobs
tential advantages include the ability to more was prohibitively expensive for prospective
effectively identify, attract, screen, assess, employees.
interview, and coordinate with job candidates. Within the confines of this recruiting frontier,
These advantages come from AI’s ability to pro- humans were tasked with attracting candidates,
cess information and make decisions at volumes filing applications, screening applicants, and
and speeds that far exceed human capacity and determining who should move on to and who
the availability of AI-enabled recruiting tools and should be dropped from subsequent steps in the
systems that overcome common cognitive biases process. But human recruiters could only read so
that hurt the reliability and validity of human fast and process so much information. On top of
judgment in recruiting activities. In subsections this, they needed periodic rest in order to keep
2.1e2.4., we examine three key stages that fatigue from degrading their results. In addition,
brought us to the early days of Digital Recruiting humans are beset with cognitive biases that
3.0. limited the reliability and validity of their judg-
ments in the screening process (Judge, Cable, &
2.1. Analog recruiting Higgins, 2000), including:

Until the mid-to-late 1990s, recruiting was an  Anchoring bias, in which the information a
analog process in which people were the primary recruiter sees or hears first unduly shapes or
mechanism for recruiting new employees. Often, influences how he or she interprets subsequent
job candidates had to physically go to job boards in
search of opportunities, or read about them in Figure 1. Analog information reach and richness
newspapers and other print media. Once they frontier
found a job they were interested in, they typically
had to physically go to the company offering the
job, get a paper job application, and manually fill
it out and turn it in. The analog nature of the
process made it tiresome. Firms wanted to reach
as many qualified candidates as possible to provide
them with both the rich details the job and the
context of the opportunity, but maximizing both
was prohibitively expensive. As a consequence,
firms had to make trade-offs, which created what
we refer to as the analog reach and richness
frontier (see Figure 1).
For example, if a company wanted to optimize
reach, it could advertise its jobs in a national
newspaper or some other widely distributed print
AI-enabled recruiting 217

information (Levashina, Hartwell, Morgeson, & and universities. With digital recruiting, they were
Campion, 2014); able to reach a diverse set of prospective candi-
dates across thousands of campuses.
 Confirmatory bias, in which recruiters seek out The leap in value that Digital Recruiting 1.0
and note information that confirms initial judg- brought to both employers and prospective em-
mentsdpositive or negativedabout job candi- ployees was such that for the next decade, new
dates (Windschitl, Scherer, Smith, & Rose, companies and job boards proliferated, providers
2013); and consolidated, and the overall digital recruiting
market soared. Monster.com’s revenue increased
 Similarity bias, in which recruiters uncon- from $162.6 million in 1996 to $1.1 billion in 2006.
sciously favor candidates who are similar to
them independent of whether those similarities
are good predictors of the candidates’ subse- 2.3. Digital Recruiting 2.0
quent performance (Sacco, Scheu, Ryan, &
Schmitt, 2003). Digital Recruiting 2.0 emerged 10 years after the
start of Digital Recruiting 1.0, and it was driven
As a consequence of these and other cognitive largely by two key developments. The first was the
biases, multiple studies have shown that an un- ability to aggregate jobs across multiple individual
structured interviewdthe most widely used job boards, which led to the emergence of firms
traditional analog selection mechanismdis only such as Indeed. This meant candidates could
about 14% accurate in identifying candidates who essentially access all the unique jobs that existed
subsequently stay with the firm and perform well across multiple job boards without having to visit
(Hunter & Schmidt, 1998). Even when these in- and search within each of them, and companies
terviews are structured, their validity is only about could reach unique job candidates across all the
30% (Huffcutt, Culbertson, & Weyhrauch, 2013). job platforms without having to list their jobs on
each one individually.
2.2. Digital Recruiting 1.0 The second major development was the advent
of digital professional and social network plat-
The digitalization of both job and candidate in- forms. One of the earliest and most successful
formation via the internet in the mid-to-late 1990s professional network platforms is LinkedIn.
broke through the original reach and richness Launched in 2003, LinkedIn allows people to form
frontier. Early digital job boards such as Monster. professional networks and communities of interest,
com (established in 1994) could take rich de- exchange information, and endorse the people in
scriptions of jobs and convey them to thousands of their networks as well as receive endorsements
prospective employees at minimal cost because it from people in their networks. One of the earliest
did not need to print or ship newspapers and incur and most successful digital social platforms is
all the attendant costs. Likewise, candidates did Facebook, which launched in 2004. Facebook al-
not have to scour print ads or spend time hand lows individuals to build their social network by
delivering and mailing hundreds of résumés and ‘friending’ others and keeping friends updated by
job applications. Prospective employees could posting activities, videos, pictures, etc., and com-
simply go to a digital job board site and in minutes menting on friends’ posts. While on the surface,
freely search and filter through thousands of jobs these digital networks and exchange platforms
to identify the ones that were the best fit. The might not seem to have much bearing on digital
internet also enabled companies to reach thou- recruiting, nothing could be further from the truth
sands of prospective employees via corporate (Bizzi, 2018). They provide consolidated digital
websites. They could include as much static in- space where firms can digitally post their job op-
formation (e.g., words on a page) and dynamic portunities efficiently. In addition, the platforms
information (e.g., videos) as they desired and felt provide information that helped companies better
would be effective. target job ads and opportunities to prospective job
The network effects were exponential and self- candidates. As was true for Digital Recruiting 1.0,
reinforcing. The more jobs that Monster.com could the network effects were exponential and rein-
list, the more job seekers it could attract. The forcing. As a consequence, in just 5 short years,
more job seekers it could attract, the more jobs it Facebook went from a few thousand users to 608
could persuade companies to list. As a conse- million by the end of 2010; similarly, over the same
quence, employers began opting out of the analog period, LinkedIn went from approximately 5 million
practice of sending recruiters to select colleges to nearly 100 million users.
218 J.S. Black, P. van Esch

2.4. Digital Recruiting 3.0 to hire armies of screeners to sift through the
avalanche of digital applications.
As Digital Recruiting 2.0 matured from 2010 to The second contextual element underlying Dig-
2015, Digital Recruiting 3.0 transitioned out of ital Recruiting 3.0 was the widespread acceptance
conversations at conferences into commercial ap- of the criticality of human capital by top execu-
plications. The principal new element of Digital tives, including CEOs. Although the shift in the
Recruiting 3.0 was the introduction of AI (Kaplan & source of firm value and competitive advantage
Haenlein, 2018). However, before examining in had largely occurred by 2000, it took CEOs a
detail the potential advantages and applications of decade or more before they generally acknowl-
AI to recruiting, as well as various challenges, it is edged the shift and recognized the role of human
important to highlight two outcomes generated by capital in driving intangible assets and firm value
Digital Recruiting 1.0 and 2.0 that became critical (Conference Board, 2018). When people were just
background elements for this new phase. cogs in the wheel, selecting the best cogs made a
The first contextual element was the avalanche marginal difference, but once people became the
in applications per position that Digital Recruiting engines, effective recruiting became mission-
1.0 and 2.0 generated. They did this primarily critical. Research also began to demonstrate the
through the reduction and near elimination of difference that quality talent could make when
friction in the process of jobs finding people and intangible assets were the primary source of firm
people finding jobs (Maurer & Liu, 2007). Although value (Paschen, Wilson, & Ferreira, in press). A
we know of no definitive study of the increase, one study of 600,000 researchers, entertainers, politi-
estimate found that toward the end of Digital cians, and athletes found that the very best of
Recruiting 2.0, each online job was generating 250 them were more than 400% more productive than
applications (Sullivan, 2013). Various company the average among them (Herman & O’Boyle,
cases illustrate this growth, including: 2012). In another study, McKinsey found that for
complex jobs, the impact on performance was an
 In 2013, Walmart, the largest private employer astonishing 800% higher for top performers
on the planet, received on average 23,000 ap- compared to the average performer (Keller &
plications for 600 positions when it opened a Meaney, 2017). This performance gap proved that
new store (Lutz, 2013); finding the right needles in an avalanche of hay
was critical, and CEOsdnot just HR execu-
 In 2017, Johnson & Johnson generated over 1 tivesdwanted to identify, develop, and deploy
million applications for 28,000 positions every possible technical and technological advan-
(McIlvaine, 2018); and tage in the recruiting race for talent (Desouza,
Dawson, & Chenok, 2020).
 In 2017, Google generated an estimated 2
million applications for just 14,500 jobs (Torres,
2017), meaning that it was nearly 10 times more 3. AI-enabled recruiting: From nice to
difficult to get a job at Google than to get into necessary
Harvard University.
This background and context turned AI-enabled
Ironically, as digitalization lowered friction costs recruiting from nice-to-have to necessary-to-employ.
and the number of candidates per position AI-enabled recruiting tools have primarily been
increased, it also drove the number of unqualified employed across four general sets of activities:
candidates higher. Estimates vary, but toward the outreach, screening, assessment, and coordination. In
end of Digital Recruitment 2.0, between 75%e88% the outreach stage, firms try to identify candidates and
of all job applicants were unqualified for the po- get job opportunities in front of them in ways that will
sition for which they applied (Ideal, 2018). The prompt them to apply. Applying for a position might
reason for this is straightforward. If it costs a involve filling in a digital application or electronically
candidate next to nothing in terms of time and transmitting a résumé. Once candidates submit these
money to apply for a job, then why would candi- applications, the employer has the daunting task of
dates not apply for more rather than fewer posi- screening them. As mentioned, many companies now
tions and even apply for positions in which they receive between 20e200 applications for every open-
were interested but not truly qualified? With job ing, and screening them efficiently is no small task. For
application costs plunging and the number of ap- those candidates who pass the initial screening, em-
plicants per job soaring, firms were forced either ployers need to assess and evaluate candidates to
to take longer to review all the new applicants or determine which are most appropriate for the job. This
AI-enabled recruiting 219

stage may involve more than one round or means of The ideal candidate pool for companies consists
assessment, but the ultimate objective is to identify the of both active job and passive job candidates.
best candidates who will then receive job offers. AI can However, today it is not just the breadth or size of
be used to coordinate with candidates all along the the pool that has expanded, but the depth of in-
process. Based on a recent survey (Deloitte, 2018), only formation about candidates in the pool has deep-
about 38% of firms use AI-enabled recruiting tools across ened to almost unfathomable levels. In 2018,
these core recruiting activities. LinkedIn had nearly 600 million users, and each
one had literally hundreds of unique data points in
their profiles. Thee discrete data points for each
3.1. Outreach profile yields volumes of data that are nearly
incomprehensible. Sifting through that many pro-
Given the importance of finding the right people, files and that much data intelligently and effi-
companies need to be as broad as possible but also ciently without AI would be impossible, even if a
as targeted as possible in their outreach efforts. firm hired an army of humans to do the work.
Clearly, firms want to reach all of the right active AI can not only help firms increase the total
candidates as possible (i.e., people who are number of applicants, but it can also target more
deliberately taking actions to find a job). However, appropriate candidates. For example, Unilever
the majority of people are not actively looking for partnered with AI hiring provider Pymetric to
a job and are therefore passive candidates. Eighty target candidates for its 200 key internships
percent of people not actively looking for a (Feloni, 2017). This effort more than doubled ap-
different job would nonetheless consider an plications from 15,000 to 30,000, or 150 applicants
appropriate job opportunity if it were presented to per position, and also dramatically increased the
them (Smith & Kidder, 2010); the number of pas- diversity of the candidate pool. Specifically, Uni-
sive candidates is roughly three times larger than lever noted that it was able to broaden the base of
active candidates. applicants from 840 universities to 2,600 univer-
Intelligently identifying both active and passive sities. Even more dramatic, in 2017 L’Oréal used AI
job candidates is critical for companies to create to not only present its opportunities to active
the best possible candidate pool (Guinan, Parise, & candidates but to identify passive candidates as
Rollag, 2014). Companies such as Pandologic, Tal- well. As a consequence, it received 2 million ré-
enya, and HireScore use AI to scrape data from sumés for only 5,000 positions, a stunning 400 ap-
LinkedIn, Facebook, Instagram, Pinterest, Twitter, plicants per job (Sharma, 2018).
XING, Ryze, Beyond, and MeetUp and then match These examples illustrate how AI helped push
candidates to the job (Campbell, Sands, Ferraro, back the reach-richness frontier even further than
Tsao, & Mavrommatis, 2020). Over time, the AI Digital 1.0 and Digital 2.0 achieved. Using AI, the
tools learn what methods work best with each type reach portion of the frontier has been pushed back
of candidate. More precisely, the AI connects the because firms can now reach not only thousands of
right presentation methods (e.g., banner ads, active candidates for a given position, but it can
email, text) with the best candidates. The system also identify more passive candidates who are
learns and places job opportunities via banners, likely even better matches for an opening. AI also
popups, email, texts, etc. for the optimal uptake has the potential to push the richness frontier back
and response by candidate profile. just as far, but this opportunity has not yet been
This learning can be applied not just to the fully explored or developed. Specifically, just as AI
delivery form of the job opportunity but also the has been used to determine which aspects of a
exact wording and description of the job. For candidate make them more or less of a fit for the
example, Textio can use AI to adjust the wording in position, AI has the potential to determine which
ads and track the effect of those changes on the aspects of the companydits culture, results,
number of applicants and various demographic leadership, technology, etc.dshould be presented
dimensions of applicants, helping clients to to candidates in order to generate the most posi-
improve outreach impact. Johnson & Johnson used tive responses.
Textio to adjust its job description opportunity As important as it is for companies to generate
language and increased female qualified hires by these pools of candidates, most already have pools
13% (McIlvaine, 2018). L’Oréal was able to use AI to of past candidates that typically lie dormant and
remove previous gender bias wording with a unleveraged: past rejected candidates (Kakatkar,
resulting even split between male and female Bilgram, & Füller, 2020). While it may not be
candidates, which the organization had never obvious that these rejected candidates should be
achieved before (Sharma, 2018). examined for current positions, the fact that past
220 J.S. Black, P. van Esch

candidates were not a match for a previous jobs In addition to the recruiting speed and effi-
does not mean that they are not a match for a ciency gains that are possible with AI, there are
current opening. However, because these appli- also potentially impressive effectiveness gains as
cants and applications often sit in different for- well. For example, recent research found that AI-
mats and in different places (i.e., in on-premises enabled tools were at least 25% superior to humans
servers, in third-party digital storage, in the in screening applicants even when humans took a
cloud), it is too costly to manually examine this reasonable amount of time to evaluate an appli-
pool. AI tools have the ability to screen these cation or résumé (Kuncel, Klieger, & Ones, 2014).
candidates regardless of the format of the appli- Today, AI-enabled screening has moved beyond
cation. As a consequence, companies such as just looking for keywords in applications and ré-
Engage Talent use AI-enabled tools to examine sumés to inferring capabilities not stated in spe-
past applicants and match them to current open cific words. For example, persistence might be a
positions. characteristic required in a particular job. Today,
instead of just scanning for that term or common
3.2. Screening synonyms, AI-enabled screening tools can infer
persistence from natural language sentences that
describe not quitting when facing an obstacle or
It does little good to reach more active candidates,
overcoming resistance when implementing a new
identify and attract more passive candidates, and
process.
reactivate past candidates if firms cannot effec-
tively screen them. The evidence that AI-enabled
screening tools save time is somewhat anecdotal 3.3. Assessing
but worth noting. For example, Ideal provides AI-
enabled screening tools and claims that across its Once companies have screened candidates and
clients, time-to-hire has dropped from an average have eliminated 50%e80% of them, AI-enabled
of 24 to 9 daysda 62.5% decline. While more assessments can help narrow the field even
research and study are needed to establish the further. These assessments can come in a variety
impact of AI on time-to-hire, specific company of forms. Some involve the gamification of tests
case studies suggest that AI can help achieve sig- that provide insight into skills, capability, and even
nificant reductions in lead time. Hilton Hotels & personality. For example, Unilever used Pymetric
Resorts implemented an AI-enabled screening tool to create 12 neuroscience-based games that can-
and saw its time-to-hire drop from 42 days to just didates complete in just 20 minutes (Feloni, 2017).
5dan 88% decline (McLaren, 2018). L’Oréal used One of the games measured risk-taking. Candi-
AI-enabled screening tools and the time to review dates had 3 minutes to collect as much money as
a résumé dropped from 40 minutes to 4 minutesda they could by clicking ‘pump’ to inflate a digital
reduction of 90% (Sharma, 2018). balloon with air and money. Each click added 5
The strategic human capital implications of re- cents. At any point, the candidate could choose to
ductions in time-to-hire are potentially game- collect money to add the amount to his or her total
changing for some firms. Take the case of Hilton and start with a new balloon. However, if the
referenced above. According to the U.S. Bureau of candidate waited too long and the balloon popped,
Labor Statistics, the annual turnover rate for ho- the candidate collected no money from that
tels is over 70% (Bureau of Labor Statistics, n.d.). balloon. Candidates could collect money about as
Thus, hotel companies such as Hilton are fast by clicking early and frequently or waitingdas
constantly trying to find and hire staff. If Hilton long as they didn’t wait too long. The point of the
can make an offer to a housekeeping job candidate game was not really about the amount of money
in 5 days and its competitor takes 42 days, will collected but identifying the individual’s risk
Hilton or its competitor likely win the battle for propensity.
that housekeeping candidate? What are the odds Obviously, before implementing this assessment
that candidate would wait 37 days after receiving game in the recruiting process, Unilever needed to
an offer from Hilton to see if he/she will also understand the relationship between risk pro-
receive an offer from the competitor? Clearly, the pensity and job success for certain positions (e.g.,
odds are near zero. Thus, AI’s ability to reduce product managers). In fact, Unilever found an
time-to-hire represents not just an efficiency gain inverted-U relationship between risk propensity
but also potentially a strategic advantage in the and job performance. Specifically, moderate to
battle for human capital, especially in industries in moderately high levels of risk propensity were
which there is high turnover. positively related to job performance, while low
AI-enabled recruiting 221

and very high levels had negative relationships The AI system subsequently analyzed and
with job performance. compared the content of candidates’ answers with
The top 33% of candidates who completed the the answers given by high-performing L’Oréal
12 game assessments were subsequently asked to employees. It also analyzed the sentence structure
also participate in a video-recorded interview with and vocabulary used in the answers and, in com-
AI-enabled technology and analytics provided by bination with the content analysis, created an
HireVue. During the interview, the AI system asked overall score for each candidate. Only then did the
candidates various questions and candidates sub- team of 145 global recruiters interview candidates
mitted their recorded responses. The questions and make final selection decisions.
were based on an analysis of successful and
average employees in those internship positions in 3.4. Facilitating across stages
the past. Based on this research, Unilever deter-
mined what capabilities and characteristics were Given the high volume of applications that AI-
most likely to lead to success and which questions enabled outreach can generate, most companies
HireVue’s AI system would ask. The system are in the rejecting business rather than the hiring
analyzed not only the content of candidates’ re- business. Nonetheless, it is in their own self-
sponses but their word choice, tone of voice, and interest to ensure that the experience is a posi-
microfacial movements and correlated them to tive one for all candidates, especially those
those of Unilever’s successful employees. Candi- rejected, for at least three reasons. First, rejected
dates were able to participate in the virtual candidates today may be good-fit candidates
interview on any day or at any time convenient to tomorrow. Candidates who were once rejected by
them within a several-day window. This not only a company are more likely to be open to a subse-
saved countless hours in scheduling but also gave quent opportunity if they had a positive experi-
candidates more control over the experience. ence when they were rejected (Swider,
Unilever wanted this increased sense of control Zimmerman, & Barrick, 2015). Second, the posi-
because it understood what research has consis- tive or negative experience of rejected candidates
tently showndpeople have more positive attitudes informs the positive or negative word-of-mouth
toward experiences in which they feel they have comments to friends and family. This makes an
more control over the process (Hamilton & important difference when rejected candidates
Davison, 2018). These AI-enabled interviews and can constitute as much as 99% of the total candi-
evaluations narrowed the original pool of over dates. In today’s connected, social-media infused
45,000 candidates for internships in the United world, positive or negative comments can ripple
States to 300 finalists. Unilever made offers to 240 through family, friends, and followers with a
finalists. This was approximately a 25% higher offer breadth and speed not seen or perhaps even
rate than in the past. Of these, 82% accepted the imagined in the past, and those positive and
offersda yield rate that was significantly higher negative ripples can, in turn, drive up or down the
than the previous rate of 65% (Feloni, 2017). sentiments of individuals who might be the com-
L’Oréal used a similar AI-enabled interview tool pany’s future job candidates (Van Esch & Mente,
from Mya Systems to assess candidates who passed 2018). Therefore, structuring a positive experi-
initial screening and verification (less than 10%). ence so that rejected candidates still provide
Maya, an AI chatbot, interactively asked candi- positive word-of-mouth is simply smart business.
dates three questions (Sharma, 2018): Third, at the end of the day, companies not only
want to select the best candidates, they also want
1. Tell us about a project that you worked on and need those selected candidates to say yes to
which failed. What did you learn from that an offer. After all, companies can only employ
project? those who say yes to an offer. Therefore, com-
panies need the candidates they want to want
2. Tell us about a project where you were working them. The more positive the candidate’s recruiting
with the multi-cultural teams and what expe- experience, the higher the likelihood the candi-
rience did you have? date will say yes to the offer at the end (Jarrahi,
2018).
3. Tell us about a situation where you were Before diving into the ways in which AI can
convinced about your idea, but your seniors enhance an applicant’s recruiting experiencedeven
were not. How did you convince them? for those who get rejecteddit is helpful to paint a
222 J.S. Black, P. van Esch

picture of the less-than-ideal current state of affairs systems predefined in order to facilitate storage,
for candidates. First, according to one study (Talent search, and retrieval. Digital Recruiting 3.0 AI
Board, 2017), 41.3% of companies do not survey or systems do not require this step. In fact, AI sys-
gather data from job candidates about their expe- tems now do not even require candidates to fill out
riences. This is unfortunate because 77% of candi- an application or submit a résumé. For example,
dates who had a positive experience and 61% who Unilever only asked applicants to submit their
had a negative experience share their experiences LinkedIn profile and the AI system intelligently
with friends and family. Obviously, if a company does combed through the candidates’ profiles and filled
not know how good or bad its candidates’ experi- in the application for them.
ences are, it cannot deliberately leverage the posi- Once candidates submit their applications, AI-
tive or change the negative. This matters because enabled chatbots can take over to smooth and
the same Talent Board (2017) survey found that 81% improve the process in three ways. First, chatbots
of rejected candidates who had a strongly positive can proactively let applicants know where they are
experience were still willing to recommend or refer in the system and elucidate the next steps. At any
people they knew to the company. More responses point in the process, AI chatbots can answer can-
from the survey (Talent Board, 2017) include: didates’ questions about the process, such as:
“Should I wear a suit for the HireVue video inter-
 Of those who had a very negative experience, view?” Second, AI chatbots can ask candidates
84% would definitely not or likely not recom- questions in order to fill in any missing or unclear
mend the company to others; bits of information, including the candidate’s po-
tential start date. Third, chatbots can answer
 Of those who had a positive experience, 51% candidates’ questions about the company or the
posted comments about their positive experi- job that candidates might have, such as salary
ence on one or more of their social or profes- range or education reimbursement benefits, on a
sional media accounts; 24/7/365 basis. Thus, AI has the ability to enhance
the efficiency of the coordination of candidates
 Of those who had a negative experience, 31% across the core recruiting activities.
put comments about their negative experience
on one or more of their social or professional
media accounts; and 4. Recommendations for managers

 One of the strongest causes of a negative The early academic research, practitioner surveys,
experience was simply receiving little informa- and company experiences have all indicated that
tion during the processd52% of candidates fell Digital Recruiting 3.0 via AI-enabled systems is
into this camp. providing significant improvements in recruiting
efficiency. However, we would argue that given
The positive or negative ripple effects from the volume of applications triggered by Digital
rejected candidates via their social media ac- Recruiting 1.0 and 2.0 and the fact that that genie
counts can reverberate with impact unseen in the cannot be put back in the bottle, the continued
analog days of the past (Van Esch, Overton, & Van growth of Digital Recruiting 3.0 is inevitable, but
Esch, 2014). not without challenges or potential concerns
Fortunately, AI-enabled systems can make the (Kietzmann & Pitt, 2020).
job application experience a smooth and positive One of the challenges of AI-enabled recruiting is
one, even for the vast majority of candidates who simply the cost of creating the tools and systems.
are rejected by companies. As evidence, 92% of The complexity involved in creating AI-enabled
the nearly 2 million candidates L’Oréal rejected outreach, screening, assessment, and coordination
indicated that they were satisfied to very satisfied tools is significant. Unless a company has a large
with the process (Sharma, 2018). Part of the key number of hires each year and can amortize the
for L’Oréal’s success was thinking of candidates as cost of developing and deploying those tools, it
customers and trying to make the digital experi- may make economic sense to buy tools from
ence as seamless and positive for candidates as it external providers. In addition, even if a company
had done for its online customers. Creating a could justify the development of the AI recruiting
positive experience can start when candidates tools, the limited supply of talent who could un-
apply for the job. In Digital Recruiting 2.0, appli- dertake this challenge and the exploding demand
cant tracking systems (ATS) required candidates to for these people may make internal development
put their résumés in a format that the structured simply impractical.
AI-enabled recruiting 223

Another challenge is privacy. If laws are passed cost-effectively apply AI recruiting tools to candi-
or if people in general start to significantly restrict date pools that will result in 200 or 5,000 hires.
data in their social media or professional network
profiles, this could significantly hinder the effec- 4.2. Take care to corral
tiveness of AI-enabled outreach tools. As another
example, if laws are passed that do not allow firms By its nature, an AI system does not automatically
to keep information on past candidates, AI tools know what a bias is or is not, and cannot deter-
specifically built to mine this particular pool of mine if it is learning one. This is important, espe-
candidates would have limited value (Canhoto & cially when companies design AI tools to look at
Clear, 2020). existing high performers in order to determine key
The possibility that HR employees will see AI- capabilities and characteristics of future em-
enabled recruiting tools as a threat to their jobs is ployees (Neubert & Montañez, 2020). If there has
another challenge and, as a consequence, will been gender, age, race, education, or other biases
stymy or even sabotage the implementation of in the past and if those emerge in the current high
such tools. Whether recruiters in HR see AI as a performers in the company who serve as bench-
threat or opportunity will likely depend on which marks, the algorithms will simply learn those pat-
tasks AI is applied to first and whether it is applied terns and perpetuate the biases (Lee & Shin,
as a complement or substitute. For example, if AI 2020). Amazon learned this the hard way (Dastin,
is applied as a substitute for on-campus recruiting, 2018) when it hired 77,000 net new people in
there is likely a good number of HR recruiters who 2014 (meaning that the total number of hires was
would view AI as a threat. However, if AI is used to much higher due to turnover) and they projected
generate candidates as a complement to those hiring of an additional 110,000 net new employees
reached by on-campus recruiters, recruiters may in 2015. As a consequence, Amazon had a high
not view AI as a threat at all. If AI is used heavily in incentive to develop an AI tool that would help this
tasks that recruiters find routine and often fati- task be more efficient and effective. The tool
guingdsuch as screening 8,000 resumes per day in looked back over the previous 10 years to deter-
the case of IBMdAI could be viewed by HR staff as mine the capabilities and characteristics of in-
a welcome relief from lesser-valued tasks and as a dividuals who were high performers at Amazon.
means of opening up time to engage in higher- The majority of these benchmark individuals were
value tasks, such as workforce planning or candi- men. Thus, soon the tool learned to penalize ré-
date interviews. sumés that had female designations such as
Despite these challenges and concerns, the ‘women’s basketball team captain’ and to dis-
need for AI and the potential of AI in recruiting are count applicants who came from two different
significant enough that practicing managers are women’s colleges. By 2015, key executives were
looking for guidance on how to best work with and frustrated enough with these results that they
deploy AI-enabled recruiting tools. To this end, we scrapped the project. However, Amazon may have
present five strategic-level recommendations. thrown the baby out with the bathwater because it
is possible to code algorithms to be neutral on di-
4.1. Identify the critical positions mensions such as gender, ethnicity, race, and
religion. The key point is that if there are biases
For the majority of companies today, AI-enabled that may have unwittingly been at work in the
recruiting systems are new and unfamiliar. Trying past, it is necessary to deliberately neutralize
to implement them from the very start across all them and to corral the AI system and not allow it
categories and levels of employees may be taking freely learn. Moreover, companies need to place
on too much too quickly. Consequently, companies an emphasis not on old patterns but on new inputs
should exercise some caution given the vast to ensure AI tools are aligned with the organiza-
research that has found that 60%e80% of large tion’s human capital strategy going forward
organizational change initiatives, in general, and (Weinstein, 2012).
digital transformations, in particular, fail (Black,
2014). Therefore, companies might do well to 4.3. Build an integrated system
identify important categories of talent, as Unilever
did relative to interns, and apply AI-enabled tools Because AI-enabled recruiting is new enough and
to those limited groups of job candidates. The challenging enough, firms providing services have to
good news is that, even though they are not free, some extent focused on particular activities within
AI-tools are much more efficient on a per- the overall recruiting process, such as HireVue’s
candidate basis than humans. Companies can focus on interviews. However, the risk is that
224 J.S. Black, P. van Esch

companies will look for the shiniest AI objects out and culture of the company that will employ them
there and assemble an array of tools in which the and the people they will work with (Kristof, 1996).
whole is less than the sum of the parts, especially As a consequence, companies have to keep in mind
from the perspective and experience of candidates. that as much as they are selecting employees,
It may be some time before there are end-to-end employees are selecting them. Companies can
solutions that are not just efficient and effective but implement all the best procedures and use AI to
are seamless and enjoyable for candidates. Even enhance recruiting, but it will only matter if the
though most companies by volume of activity are in candidates to whom the company ultimately makes
the rejection rather than the hiring business, the offers to say yes. Thus, final interviews are not now
consequences of the candidate’s experience are and should never be handed over to AI-enabled
wide-ranging and potentially long-lasting. The more toolsdnot because AI systems could not make
companies take a candidate-as-customer perspec- effective final evaluations, but because candidates
tive and build convenient and integrated experi- could not effectively evaluate the company without
ence, the more candidates are willing to both use interacting with the humans in the company. This is
and promote AI recruiting technology (Baum, 2017; why humans need to conduct final interviews with
Brahmana & Brahmana, 2013). candidates. Candidates do not need an opportunity
to determine if they like and want to spend more
4.4. Be transparent and upfront time with chatbots, but if they like to chat and work
with real people in the organization. They want and
Our research suggests that AI-enabled recruiting need a chance to determine if they like the com-
systems are less biased and more objective than pany culture in which they will be working and the
humans (Van Esch, Black, & Ferolie, 2019). In people with whom they will be working.
addition, the majority of candidates are motivated
to engage with AI-enabled systems because they 5. Casting a wider net
perceive them as novel, empowering, and conve-
nient. Using and acknowledging the use of AI in The new primary source of competitive advantage
recruiting results in candidates perceiving a com- and firm valuedhuman capitaldand the greater
pany’s brand as being cutting-edge (McIlvaine, volume in candidates per job seem here to stay. As
2018; Miles & McCamey, 2018). All of these fac- a consequence, Digital Recruiting 3.0 and the role
tors lead not only to more positive perceptions of of AI in recruiting are likely to only grow. Firms and
AI-enabled recruiting systems, but also to a higher executives that do not quickly embrace AI-enabled
likelihood that candidates complete the applica- recruiting may find that the 60%e80% of their
tion process (Van Esch et al., 2019). This same employees whom they considered engaged based
principle of transparency applies to chatbots. The on internal surveys are nonetheless vulnerable to
majority of applicants feel comfortable with targeted, proactive, and customized outreach ef-
chatbots. In a recent survey by Montage (2018), forts by their rivals. They may find that if they do
82% of respondents indicated that it was important not quickly fully embrace and create solid and
for the hiring organization to inform candidates integrated AI-enabled systems, they will see either
that they were interacting with a chatbot and not the quality of their pool of candidates suffer or
a real person even though the language fluency of their yields decline, or both, as their competitors
chatbots is reaching such a level that candidates outdo them in the continuing Digital Recruiting 3.0
might not detect the difference. battles for talent. Conversely, if firms and their
executives embrace AI-enabled recruiting systems,
4.5. Be human they have the potential to attract ever broader,
more diverse, and higher quality pools of talents.
At the end of the day, even individuals hired to work They can take the time and money saved to ensure
on AI technology will spend most of their time that the company and its culture are ever more
working with other humans and not with AI chat- attractive so that the people who executives want
bots. As a consequence, despite all the recruiting also want them.
tasks that AI can facilitate, it cannot be the be-all
and end-all of recruiting. Whatever the contribu- References
tion of AI in reaching, screening, scheduling,
interviewing, and evaluating candidates might be, Baum, S. D. (2017). On the promotion of safe and socially
what matters most when candidates decide to take beneficial artificial intelligence. AI & Society, 32(4),
the job or not is what they think of the environment 543e551.
AI-enabled recruiting 225

Bizzi, L. (2018). The hidden problem of Facebook and social Judge, T. A., Cable, D. M., & Higgins, C. A. (2000). The
media at work: What if employees start searching for other employment interview: A review of recent research and
jobs? Business Horizons, 61(1), 23e33. recommendations for future research. Human Resource
Black, J. S. (2014). It starts with one: Changing individuals changes Management Review, 10(4), 383e406.
organizations. Upper Saddle River, NJ: Pearson Education. Kakatkar, C., Bilgram, V., & Füller, J. (2020). Innovation ana-
Black, J. S. (2019). Competing for and with human capital: It is lytics: Leveraging artificial intelligence in the innovation
not just for HR anymore. Milton Park, UK: Taylor and process. Business Horizons, 63(2), 171e181.
Francis. Kaplan, A., & Haenlein, M. (2018). Siri, Siri, in my hand: Who’s the
Brahmana, R. K., & Brahmana, R. (2013). What factors drive job fairest in the land? On the interpretations, illustrations, and
seekers’ attitude in using e-recruitment? South East Asian implications of artificial intelligence. Business Horizons, 62(1),
Journal of Management, 7(2), 123e134. 15e25.
Bureau of Labor Statistics. (n.d.). Table 16: Annual total sep- Keller, S., & Meaney, M. (2017). Attracting and retaining the
arations rates by industry and region, not seasonally right talent. McKinsey & Company. Available at https://
adjusted. Available at https://www.bls.gov/news.release/ www.mckinsey.com/business-functions/organization/our-
jolts.t16.htm insights/attracting-and-retaining-the-right-talent
Campbell, C., Sands, S., Ferraro, C., Tsao, J., & Kietzmann, J., & Pitt, L. (2020). Artificial intelligence and
Mavrommatis, A. (2020). From data to action: How mar- machine learning: What managers need to know. Business
keters can leverage AI. Business Horizons, 63(2), 227e243. Horizons, 63(2), 131e133.
Canhoto, A., & Clear, F. (2020). Artificial intelligence and ma- Kristof, A. L. (1996). Person-organization fit: An integrative
chine learning as business tools: A framework for diagnosing review of its conceptualizations, measurement, and impli-
value destruction potential. Business Horizons, 63(2), cations. Personnel Psychology, 49(1), 1e49.
183e193. Kuncel, N. R., Klieger, D. M., & Ones, D. S. (2014). hiring, al-
Conference Board. (2018). C-Suite challenge. Available at gorithms beat instinct. Harvard Business Review, 92(5), 32.
https://www.conference-board.org/publications/ Lee, I., & Shin, Y. J. (2020). Machine learning for enterprises:
publicationdetail.cfm?publicationidZ7691 Applications, algorithm selection, and challenges. Business
Dastin, J. (2018, October 9). Amazon scraps secret AI tool that Horizons, 63(2), 157e170.
showed bias against women. Reuters. Available at https:// Lev, B. (2000). Intangibles: Management, measurement, and
www.reuters.com/article/us-amazon-com-jobs- reporting. Washington, DC. Brookings Institution Press.
automation-insight/amazon-scraps-secret-ai-recruiting- Levashina, J., Hartwell, C. J., Morgeson, F. P., & Campion, M. A.
tool-that-showed-bias-against-women-idUSKCN1MK08G (2014). The structured employment interview: Narrative and
Deloitte. (2018). The rise of the social enterprise. Available at quantitative review of the research literature. Personnel
https://www2.deloitte.com/content/dam/insights/us/ Psychology, 67(1), 241e293.
articles/HCTrends2018/2018-HCtrends_Rise-of-the-social- Lutz, A. (2013, November 19). Applicants for jobs at the new D.C.
enterprise.pdf Walmart face worse odds than people trying to get into Harvard.
Desouza, K. C., Dawson, G. S., & Chenok, D. (2020). Designing, Business Insider. Available at https://www.businessinsider.
developing, and deploying artificial intelligence systems: com/wal-mart-receives-23000-applications-2013-11
Lessons from and for the public sector. Business Horizons, Maurer, S. D., & Liu, Y. (2007). Developing effective e-recruiting
63(2), 205e213. websites: Insights for managers from marketers. Business
Feloni, R. (2017, June 29). Consumer-good giant Unilever has Horizons, 50(4), 305e314.
been hiring employees using grain games and artificial McIlvaine, A. (2018). Data in the driver’s seat. Human Resource
intelligencedand it’s a huge success. Business Insider Executive. Available at https://hrexecutive.com/talent-
Australia. Available at https://www.businessinsider.com. acquisitions-leaders-use-ai-to-improve-hiring/
au/unilever-artificial-intelligence-hiring-process-2017-6 McLaren, S. (2018 May 24). How Hilton, Google, and more have
Guinan, P. J., Parise, S., & Rollag, K. (2014). Jumpstarting the dramatically reduced time to hire. LinkedIn Talent Blog.
use of social technologies in your organization. Business Available at https://business.linkedin.com/talent-
Horizons, 57(3), 337e347. solutions/blog/recruiting-strategy/2018/how-4-companies-
Hamilton, R. H., & Davison, H. K. (2018). The search for skills: reduced-time-to-hire
Knowledge stars and innovation in the hiring process. Busi- Miles, S. J., & McCamey, R. (2018). The candidate experience: Is it
ness Horizons, 61(3), 409e419. damaging your employer brand? Business Horizons, 61(5),
Herman, A., & O’Boyle, E. (2012). The best and the rest: 755e764.
Revisiting the norm of normality in individual performance. Montage. (2018). State of AI in talent acquisition. Modern Hire.
Personnel Psychology, 65(1), 79e119. Available at https://learn.modernhire.com/ai-in-ta
Huffcutt, A. I., Culbertson, S. S., & Weyhrauch, W. S. (2013). Neubert, M. J., & Montañez, G. D. (2020). Virtue as a framework
Employment interview reliability: New meta-analytic esti- for the design and use of artificial intelligence. Business
mates by structure and format. International Journal of Horizons, 63(2), 105e204.
Selection and Assessment, 21(3), 264e276. Paschen, U., Pitt, C., & Kietzmann, J. (2020). Artificial intelli-
Hunter, J. E., & Schmidt, F. L. (1998). The validity and utility of gence: Building blocks and an innovation typology. Business
selection methods in personnel psychology: Practical and Horizons, 63(2), 147e155.
theoretical implications of 85 years of research findings. Paschen, J., Wilson, M., & Ferreira, J. (2020). Collaborative
Psychological Bulletin, 124(2), 262e274. intelligence: How human and artificial intelligence create
Ideal. (2018). AI for recruiting: A definitive guide for HR pro- value along the B2B funnel. Business Horizons. https:
fessionals. Available at https://ideal.com/ai-recruiting/ //doi.org/10.1016/j.bushor.2020.01.003.
Jarrahi, J. (2018). Artificial intelligence and the future of work: Sacco, J. M., Scheu, C. R., Ryan, A. M., & Schmitt, N. (2003). An
Human-AI symbiosis in organizational decision making. investigation of race and sex similarity effects in interviews:
Business Horizons, 61(4), 577e586.
226 J.S. Black, P. van Esch

A multilevel approach to relational demography. Journal of Torres, M. (2017). 2 million candidates are desperate to work at
Applied Psychology, 88(5), 852e865. Google. Why? Ladders. Available at https://www.
Sharma, A. (2018 August 16). How AI reinvented hiring practice theladders.com/career-advice/2-million-job-candidates-
at L’Oréal. People Matters. Available at https://www. google
peoplematters.in/article/techhr-2018/how-the-worlds- van Esch, P., Black, S. J., & Ferolie, J. (2019). Marketing AI
largest-cosmetic-company-transformed-its-hiring-practice- recruitment: The next phase in job application and selec-
with-ai-19006 tion. Computers in Human Behavior, 90, 215e222.
Smith, W. P., & Kidder, D. L. (2010). You’ve been tagged! (Then van Esch, P., & Mente, M. (2018). Marketing video-enabled so-
again, maybe not): Employers and Facebook. Business Ho- cial media as part of your e-recruitment strategy: Stop
rizons, 53(5), 491e499. trying to be trendy. Journal of Retailing and Consumer
Sullivan, J. (2013). Why you can’t get a job . Recruiting Services, 44, 266e273.
explained by the numbers. ERE Recruiting Intelligence. van Esch, P., Overton, L. R. A., & Van Esch, L. J. (2014). Mass
Available at https://www.ere.net/why-you-cant-get-a-job- media social marketing campaigns: A review. International
recruiting-explained-by-the-numbers/ Business Research, 7(6), 1e17.
Swider, B. W., Zimmerman, R. D., & Barrick, M. R. (2015). Weinstein, D. (2012). The psychology of behaviorally-focused
Searching for the right fit: Development of applicant person- résumés on applicant selection: Are your hiring managers
organization fit perceptions during the recruitment process. really hiring the ‘right’ people for the ‘right’ jobs? Business
Journal of Applied Psychology, 100(3), 880e893. Horizons, 55(1), 53e63.
Talent Board. (2017). Talent board North American candidate Windschitl, P. D., Scherer, A. M., Smith, A. R., & Rose, J. P.
experience research report. Available at https://www. (2013). Why so confident? The influence of outcome desir-
thetalentboard.org/press-releases/talent-board-2017- ability on selective exposure and likelihood judgment.
north-american-candidate-experience-research-report- Organizational Behavior and Human Decision Processes,
now-available/ 120(1), 73e86.