PR ED IC TIVE A N A LY T I C S
The privacy pickle
Hewlett-Packard’s prediction of employee behavior.
BY ERIC SIEGEL
H
Hewlett-Packard (HP) knows
there are two sides to every coin. The company has
achieved new power by predicting employee behavior, a profitable practice that may raise eyebrows among some
of its staff. HP tags its more than 330,000
workers with a so-called Flight Risk score.
40
|
A N A LY T I C S - M A G A Z I N E . O R G
This simple number foretells whether each
individual is likely to leave his or her job.
With the advent of predictive analytics, organizations gain power by predicting potent yet – in some cases – sensitive
insights about individuals. These predictions are derived from existing data, almost as if creating new information out
W W W. I N F O R M S . O R G
of thin air. Examples include HP inferring
an employee’s intent to resign, retailer
Target deducing a customer’s pregnancy, and law enforcement in Oregon and
Pennsylvania foretelling a convict’s future repeat offense.
By predicting which of its staff are
likely to leave, HP can focus its efforts on
retaining them, thereby reducing the high
cost associated with finding and training
replacements. Managers “drive decisions
with the support of the [predictive] report
and the story it tells for each of their employees,” says Gitali Halder, who leads
this prediction project’s team at HP and
holds a master’s in economics from the
University of Delhi.
HP credits its capacity to predict with
helping decrease its workforce turnover
rate for a specialized team that provides
support for calculating and managing the
compensation of salespeople globally. The
roughly 300-member team’s turnover rates,
which were a relatively high 20 percent in
some regions, have decreased to 15 percent and continue to trend downward.
FLIGHT RISK PROMISES BIG
SAVINGS
Beyond this early success, HP’s
Flight Risk prediction capability promises
$300 million in estimated potential savings with respect to staff replacement
and productivity loss globally. The Flight
A NA L Y T I C S
Risk scores adeptly inform managers
where risk lurks: the 40 percent of HP
employees assigned highest scores includes 75 percent of those who will quit
(almost twice that of guessing).
HP, which literally started in the proverbial garage, came in as the 27th
largest employer of 2011; $127 billion
in revenue places it among the top few
technology companies globally.
Halder and her teammate Anindya Dey
first broke this ground in 2011, mathematically scrutinizing the loyalty of each one of
their 330,000 colleagues. The two crackerjack scientists, members of HP’s group
of 1,700 analytics workers in Bangalore,
built this prognostic capability with predictive analytics, technology that learns from
the experience encoded in big data to
form predictive scores for individual workers, customers, patients or voters.
To prepare learning material, they
pulled together two years of employee
data such as salaries, raises, job ratings
and job rotations. Then they tacked on, for
each employee record, whether the person had quit. Compiled in this form, the
data served to train a Flight Risk detector
that recognizes combinations of factors
characteristic to likely HP defectors.
The results surprised Halder and Dey,
revealing that promotions are not always
a good thing. While promotions decrease
Flight Risk across HP as a whole, the effect
N O V E M B E R / D E C E M B E R 2 013
|
41
PR ED IC TIVE A N A LY T I C S
Promotions are not
always a good thing.
While promotions
decrease flight risk across
HP as a whole, the effect
is reversed within the
sales compensation team:
Those promoted more
times are more likely to
quit, unless they have
also experienced a more
significant pay hike.
42
|
is reversed within the sales compensation team: Those
promoted more times are more likely to quit, unless they
have also experienced a more significant pay hike.
NEWFOUND POWER, THREATS
The analysis confirmed that Flight Risk also depends on things one might expect. Employees with
higher salaries, more raises and increased performance ratings are less prone to quit. Job rotations
also keep employees on board by introducing change,
given the rote, transactional nature of some compensation staff activities.
HP’s newfound power brings with it a newfound
threat in the eyes of some employees. A novel element is emerging within staff records: speculative
data. Beyond standard personal and financial data
about employees, this introduces an estimation of future behavior, and so speaks to the heart, mind and
intentions of the employee. The concerned ask: What
if your Flight Risk score is wrong, unfairly labeling you
as disloyal and blemishing your reputation?
Most at HP don’t know they are being predicted.
“The employees are not aware of this model,” Halder
says. “[It] is not designed to penalize the employees
but to make necessary adjustments that would result
in lower probabilities of them leaving us.”
HP deploys this newly synthesized, sensitive isotope with care. The Flight Risk scores are securely
delivered to only a select few high-level managers.
“We are keeping this data very much confidential,”
assures Halder.
Despite concerns, predictive analytics does not itself
invade privacy. Although sometimes referred to with the
broader term data mining, its core process doesn’t “drill
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
PR ED IC TIVE A N A LY T I C S
down” to inspect individuals’ data. Instead,
this analytical learning process “rolls up,”
discovering broadly-applicable patterns
by way of number crunching across multitudes of individual records.
However, after this analysis, the application of what’s been learned to render
predictions may divulge unvolunteered
truths about an individual. Here workers
and consumers fear prediction “snoops”
into their private future. It’s not about misusing or leaking data. Rather, it’s the corporate deduction of private information.
It’s technology that speculates, sometimes
against the wishes of those speculated.
Like many technologies, predictive
analytics can enact both good and evil
– like a knife. Outlawing it is not viable,
and would be akin to forbidding deduction. Predicting human behavior truly
helps the world. Less paper is expended
when direct mail is predictively targeted and consumers receive fewer “junk
mail” items. Patients receive improved
healthcare. Police patrol more effectively by predicting crime, and fraud is
similarly detected. Nonprofits boost fundraising and more adeptly select the
beneficiaries of their services. Movie
and music recommendations improve.
CIVIL LIBERTIES AN ISSUE
But embracing predictive analytics
challenges the world with an unprecedented dilemma: How do we safely harness a predictive machine that foresees
job resignation, pregnancy (as predicted
by retailer Target) and crime without putting civil liberties at risk?
In light of Target’s prediction of customer pregnancy, activists point out that
divulging pregnancy can wreak havoc.
Per one online pundit, imagine a pregnant
woman’s “job is shaky, and your state
disability isn’t set up right yet and ... to
have disclosure could risk the retail cost
of a birth ($20,000), disability payments
during time off ($10,000 to $50,000), and
even her job.”
As crime-predicting computers emerge
in the law enforcement world, the risk
comes not when prediction is right but
when it’s wrong; a convict’s future now
rests in nonhuman hands. Granted power
as the trusted advisor to judges determining sentences and parole boards deciding
Request a no-obligation INFORMS Member Benefits Packet
For more information, visit: http://www.informs.org/Membership
44
|
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
on criminals’ release, the computer can
commit an error previously reserved for
humans: injustice. This is often compared
to the prognostic law enforcement in the
dystopic science fiction movie “Minority
Report.” If you prevent a crime by incarcerating the would-be offender, how can you
verify it was ever going to happen?
Predictive analytics promises yet
forebodes. Spider-Man’s wise uncle
paraphrased the Bible and Voltaire
when he said, “With great power comes
great responsibility.” The agreement we
A NA L Y T I C S
collectively come to for predictive analytics’ position in the world is central to
the massive cultural shifts we face as we
fully enter the information age. ❙
Eric Siegel, Ph.D., is the founder of Predictive
Analytics World (www.pawcon.com) – coming
in 2014 to Boston, Chicago, San Francisco,
Toronto, Washington, D.C., Berlin and London –
and executive editor of the Predictive Analytics
Times (www.predictiveanalyticstimes.com). For
more information about predictive analytics, see
the “Predictive Analytics Guide.” This article was
adapted with permission of the publisher, Wiley,
from Siegel’s book, “Predictive Analytics: The
Power to Predict Who Will Click, Buy, Lie, or Die.”
© 2013 Eric Siegel. All rights reserved.
N O V E M B E R / D E C E M B E R 2 013
|
45
Purchase answer to see full
attachment