BEGIN:VCALENDAR
PRODID;X-RICAL-TZSOURCE=TZINFO:-//com.denhaven2/NONSGML ri_cal gem//EN
CALSCALE:GREGORIAN
VERSION:2.0
BEGIN:VTIMEZONE
TZID;X-RICAL-TZSOURCE=TZINFO:Europe/Berlin
BEGIN:DAYLIGHT
DTSTART:20260329T020000
RDATE:20260329T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
DTEND;TZID=Europe/Berlin;VALUE=DATE-TIME:20260616T113000
DTSTART;TZID=Europe/Berlin;VALUE=DATE-TIME:20260616T110000
CATEGORIES:BUSINESS
DESCRIPTION: Despite structured interviews\, psychometric testing and AI-
supported screening\, personnel selection continues to produce a high nu
mber of false positives and false negatives. Organizations systematicall
y hire people who later fail — and reject candidates who would have perf
ormed well.\nThis talk examines why this problem persists\, even under “
best practice” conditions. Drawing on evidence from industrial-organizat
ional psychology\, meta-analyses on selection validity\, and real-world
hiring data\, I argue that the core issue is not poor execution\, but st
ructural limits of selection methods themselves.\nStatistical validity d
oes not translate directly into decision accuracy at the individual leve
l. Base rates\, construct underrepresentation\, context specificity of j
ob performance\, and non-orthogonal predictors lead to systematic decisi
on errors — regardless of whether selection is human-driven or algorithm
ic.\nThe session critically discusses why AI and matching algorithms inh
erit the same structural weaknesses as traditional selection tools\, and
why more data or better models alone will not solve the problem. Instea
d\, we need to rethink what selection can realistically achieve — and wh
ere its explanatory power ends.\nThe goal of this talk is not to promote
a specific tool\, but to establish a more realistic\, evidence-based un
derstanding of personnel selection limits — especially relevant for tech
-driven organizations building AI-supported HR systems.\n
URL:https://www.empfehlungsbund.de/events/150
SUMMARY:20th Silicon Saxony Days - a Session with Jörg Klukas: Why person
nel selection still fails
ORGANIZER;CN=Silicon Saxony e.V. / Empfehlungsbund.de:https://www.empfehl
ungsbund.de/events/150
ATTACH:https://www.empfehlungsbund.de/events/150
LOCATION:Silicon Saxony e.V.\, International Airport Dresden
X-ALT-DESC: Despite structured interviews, psychometric testing and AI-supported sc
reening, personnel selection continues to produce a high number of false
positives and false negatives. Organizations systematically hire people
who later fail — and reject candidates who would have performed w
ell.
This talk examines why
this problem persists, even under “best practice” conditions
. Drawing on evidence from industrial-organizational psychology, meta-an
alyses on selection validity, and real-world hiring data, I argue that t
he core issue is not poor execution, but structural limits of selection
methods themselves.
Statisti
cal validity does not translate directly into decision accuracy at the i
ndividual level. Base rates, construct underrepresentation, context spec
ificity of job performance, and non-orthogonal predictors lead to system
atic decision errors — regardless of whether selection is human-dr
iven or algorithmic.
The ses
sion critically discusses why AI and matching algorithms inherit the sam
e structural weaknesses as traditional selection tools, and why more dat
a or better models alone will not solve the problem. Instead, we need to
rethink what selection can realistically achieve — and where its
explanatory power ends.
The
goal of this talk is not to promote a specific tool, but to establish a
more realistic, evidence-based understanding of personnel selection limi
ts — especially relevant for tech-driven organizations building AI
-supported HR systems.
END:VEVENT
END:VCALENDAR