Value creation opportunities of generative AI: A case study

Sake, Alexande
A icle
Value c ea ion oppo uni ies o gene a i e AI: A case s udy
Junio Managemen Science (JUMS)
P o ided in Coope a ion wi h:
Junio Managemen Science e. V.
Sugges ed Ci a ion: Sake, Alexande (2025) : Value c ea ion oppo uni ies o gene a i e AI: A case
s udy, Junio Managemen Science (JUMS), ISSN 2942-1861, Junio Managemen Science e. V.,
Planegg, Vol. 10, Iss. 3, pp. 631-656,
h ps://doi.o g/10.5282/jums/ 10i3pp631-656
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/326968
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by/4.0/
Junio Managemen Science 10(3) (2025) 631-656
Junio Managemen Science
www.jums.academy
ISSN: 2942-1861
Edi o :
DOMINIK VAN AAKEN
Ad iso y Edi o ial Boa d:
FREDERIK AHLEMANN
JAN-PHILIPP AHRENS
THOMAS BAHLINGER
MARKUS BECKMANN
SULEIKA BORT
ROLF BRÜHL
KATRIN BURMEISTER-LAMP
CATHERINE CLEOPHAS
NILS CRASSELT
BENEDIKT DOWNAR
KERSTIN FEHRE
MATTHIAS FINK
DAVID FLORYSIAK
GUNTHER FRIEDL
MARTIN FRIESL
FRANZ FUERST
WOLFGANG GÜTTEL
NINA KATRIN HANSEN
ANNE KATARINA HEIDER
CHRISTIAN HOFMANN
SVEN HÖRNER
STEPHAN KAISER
NADINE KAMMERLANDER
ALFRED KIESER
ALEKSANDRA KLEIN
NATALIA KLIEWER
DODO ZU KNYPHAUSEN-AUFSESS
SABINE T. KÖSZEGI
ARJAN KOZICA
CHRISTIAN KOZIOL
MARTIN KREEB
WERNER KUNZ
HANS-ULRICH KÜPPER
MICHAEL MEYER
JÜRGEN MÜHLBACHER
GORDON MÜLLER-SEITZ
J. PETER MURMANN
ANDREAS OSTERMAIER
BURKHARD PEDELL
ARTHUR POSCH
MARCEL PROKOPCZUK
TANJA RABL
SASCHA RAITHEL
NICOLE RATZINGER-SAKEL
ASTRID REICHEL
KATJA ROST
THOMAS RUSSACK
FLORIAN SAHLING
MARKO SARSTEDT
ANDREAS G. SCHERER
STEFAN SCHMID
UTE SCHMIEL
CHRISTIAN SCHMITZ
MARTIN SCHNEIDER
MARKUS SCHOLZ
LARS SCHWEIZER
DAVID SEIDL
THORSTEN SELLHORN
STEFAN SEURING
VIOLETTA SPLITTER
ANDREAS SUCHANEK
TILL TALAULICAR
ANN TANK
ANJA TUSCHKE
MATTHIAS UHL
CHRISTINE VALLASTER
PATRICK VELTE
CHRISTIAN VÖGTLIN
BARBARA E. WEISSENBERGER
ISABELL M. WELPE
HANNES WINNER
THOMAS WRONA
THOMAS ZWICK
Volume 10, Issue 3, Sep embe 2025
JUNIOR
MANAGEMENT
SCIENCE
Johannes Wi e ,P edic ing S ock Re u ns Wi h Machine
Lea ning: Global Ve sus Sec o Models
Robin Roskosch, Bewa e o Bullshi –A Quali a i e S udy on
Young Adul s’ Sus ainabili y Awa eness o Online
Se ices
Nadhilla Mazaya,Boa d Gende Di e si y: E idence F om
Indonesia
Alexande Sake, Value C ea ion Oppo uni ies o Gene a i e
AI –A Case S udy
Jus us Olb ich, The E ec o Changes in In e nal Con ol
Sys ems on Audi Risk
Jan Oli e Ho s mann, Manda o y ESG Disclosu e and Fi m
Value –A Quan i a i e Analysis o he E ec o
Di ec i e 2014/95/EU on Fi m Value
Me e Anna Gläse , Go e nmen In e en ions Du ing he
COVID-19 Pandemic, Cul u e, and Co po a e Cos
Beha iou
Zewei Shi, Modeling he Impac o Emission C edi Sys ems on
Au omo i e P oduc Po olios: A Ma hema ical
Analysis o Policy E ec s in Eu ope, China, and he
U.S. Unde Di e en Demand Scena ios
Hagen Alexande Höne loh, Nume ical S udies o he
Scheduling o Con inuous Annealing Lines
Lea Wedel, KPIs o Sus ainabili y: De ining he S a egy o a
Sus ainable Fu u e in he Insu ance Indus y
561
582
609
631
657
677
715
748
781
810
Published by Junio Managemen Science e.V.
This is an Open Access a icle dis ibu ed unde he e ms o he CC-BY-4.0
(A ibu ion 4.0 In e na ional). Open Access unding p o ided by ZBW.
ISSN: 2942-1861
Value C ea ion Oppo uni ies o Gene a i e AI – A Case S udy
Alexande Sake
Technical Uni e si y o Munich
Abs ac
The ans o ma i e po en ial o Gene a i e AI p omises no el capabili ies wi hin business en i onmen s. This s udy examines
he alue c ea ion po en ial o Gene a i e AI wi hin a la ge mul ina ional co po a ion. A single case s udy app oach a Siemens
was employed, combining ex ensi e obse a ions, in e iews, and he applica ion o exis ing AI amewo ks. Findings e eal
di e se use cases demons a ing alue c ea ion po en ial, pa icula ly h ough sma assis an s and ligh house p ojec s. This
hesis p oposes a no el amewo k o Gene a i e AI adop ion, emphasizing he dis inc i e explo a ion phase made possible
by he echnology’s accessibili y o non- echnical domain expe s, while also ou lining essen ial scaling s a egies. This s udy
o e s aluable insigh s in o a company’s app oach o Gene a i e AI, p o ides p ac ical implica ions, and expands ongoing
esea ch on AI-d i en alue c ea ion.
Keywo ds: a i icial in elligence; explo a ion; gene a i e AI; scaling; echnology adop ion; use cases; alue c ea ion
1. In oduc ion
The elease o Cha GPT in No embe 2022 ma ked a sig-
ni ican momen whe e he public could, o he i s ime, di-
ec ly engage wi h he la es ad ances in a i icial in elligence
(AI). Since i s launch, he ool’s impac and capabili ies ha e
been a opic o di e se discussions among jou nalis s, scien-
is s, manage s, and go e nmen s, oscilla ing be ween p aise
and cau iona y no es. This hesis in ends o shed ligh on he
la es ad ances o he echnology, o desc ibe he alue c e-
a ion oppo uni ies ha a ise om i and o help companies
iden i ying equisi es and capabili ies o a success ul in o-
duc ion in o hei o ganiza ion.
Many s udies ha e been conduc ed on he impac o AI
on companies and socie y. Following echnological ad ance-
I would like o exp ess my since e g a i ude o my supe iso , Tim Läm-
me mann, M.Sc., o his aluable suppo and guidance h oughou my
mas e hesis. I also ex end my special hanks o P o . D . Thomas
Hu zschen eu e , Chai o S a egic and In e na ional Managemen , o
his inspi ing eaching and expe ise. I am deeply g a e ul o all he in e -
iew pa ne s who con ibu ed o my wo k wi h hei aluable insigh s
and expe iences. A big hank you goes o my manage s o allowing me
o pu sue his p ojec alongside my job. Las ly, I would like o hank my
amily and iends – you unde s anding and suppo du ing he in ense
pe iod o he mas e hesis we e in aluable.
men s, hese s udies o en p edic ed new alue c ea ion op-
po uni ies a ising om he success ul implemen a ion o AI
solu ions in o ganiza ions. Missing ou on AI, esul s e-
quen ly indica ed a signi ican gap in u u e pe o mance and
compe i i eness. A s udy highligh ing hese aspec s, was con-
duc ed by Accen u e (Reilly e al., 2019). Based on a global
su ey among 1,500 C-sui e execu i es, hey disco e ed ha
84% o leade s belie e o achie e hei g ow h ambi ions only
wi h he help o AI. A he same ime, 76% s uggle o scale
he echnology ac oss hei o ganiza ion.
The ise o Gene a i e AI was accompanied by many s ud-
ies as well, e alua ing eal-wo ld use cases and analyzing he
impac on o ganiza ional p ocesses and he u u e o wo k.
In pa icula , when he GPT-4 model was eleased, ques ions
a ose how he ool will ans o m and po en ially eplace ac-
i i ies o knowledge wo ke s (Dwi edi e al., 2023, p. 7).
A s udy by BCG among 750 consul an s e ealed a pe o -
mance inc ease o 40% when using GPT-4 o ypical consul -
ing ac i i ies compa ed o a con ol g oup (Dell’Acqua e al.,
2023, p. 17). This was accompanied by an inc ease in speed
o 25% wi h a posi i e impac ac oss all skill le els. Focus-
ing on he pe o mance o cus ome suppo agen s, B yn-
jol sson e al. (2023) also examined he impac o Gene a-
DOI: h ps://doi.o g/10.5282/jums/ 10i3pp631-656
© The Au ho (s) 2025. Published by Junio Managemen Science.
This is an Open Access a icle dis ibu ed unde he e ms o he CC-BY-4.0
(A ibu ion 4.0 In e na ional). Open Access unding p o ided by ZBW.
A. Sake /Junio Managemen Science 10(3) (2025) 631-656632
i e AI. Based on da a om 5,176 agen s, hey ound ha
AI was able o inc ease p oduc i i y by 14%. In hei su ey,
di e ences in ega d o skill-le els became appa en : While
new o low-skilled wo ke s’ pe o mance inc eased by 34%,
highly skilled wo ke s ba ely bene i ed om he ool (2023,
p. 1). While hese examples show he po en ial added alue
in business se ings, o he s udies ocus on he impac on job
oles and he global economy. Findings by Goldman Sachs
Resea ch sugges ha Gene a i e AI could ha e a “p o ound”
e ec on he wo ld economy and socie y, po en ially aising
global GDP by 7% o e a 10-yea pe iod (Goldman Sachs,
2023). A epo by Eloundou e al. (2023) ocusing on he
US labo ma ke see 80% o he wo k o ce impac ed by Gen-
e a i e AI by a leas 10% while o almos 20% o he wo k-
o ce, hal o he asks a e seen o be exposed o AI (2023,
p. 1).
2. Resea ch Objec i e
Gene a i e AI ep esen s a signi ican ad ancemen in
a i icial in elligence, cha ac e ized by i s abili y o p oduce
new con en h ough pa e n ecogni ion in exis ing da a
(Feue iegel e al., 2023, p. 1). This is a shi om adi-
ional AI, which p ima ily deal wi h asks such as iden i-
ica ion and classi ica ion (LeCun e al., 2015, p. 436). A
majo de elopmen in Gene a i e AI was he in oduc ion o
he T ans o me a chi ec u e in 2017 (Vaswani e al., 2017),
which enhanced he AI’s capabili y o comp ehend and p o-
cess ex ensi e in o ma ion, he eby imp o ing i s con en
gene a ion abili y. Fu he ad ancemen s we e achie ed wi h
he in oduc ion o he Gene a i e P e- ained T ans o me
(GPT) model by Rad o d and Na asimhan (2018), ep e-
sen ing a signi ican s ep o wa d in he ield o AI-d i en
ex gene a ion. T ained on di e se da ase s, he model was
capable o p oducing ex ha is no only cohe en bu also
con ex ually ele an .
The p ima y Resea ch Ques ion (RQ1) in ends o del e
in o he ad ancemen s in oduced by Vaswani e al. (2017)
and Rad o d and Na asimhan (2018). I aims o in es iga e
how Gene a i e AI di e ges om p e ious AI models in e ms
o lea ning, in o ma ion p ocessing, and con en c ea ion.
G asping hese a ibu es o Gene a i e AI is key o unde -
s anding i s ans o ma i e impac . To add ess hese consid-
e a ions, he esea ch ques ion is o mula ed as ollows:
RQ1: How does he a chi ec u e and unc ionali y
o Gene a i e AI di e om p e ious AI models?
The ou lined s udies conduc ed by B ynjol sson e al.
(2023) and Dell’Acqua e al. (2023) p o ide ini ial insigh s
in o he po en ial o Gene a i e AI o enhance wo k o ce
p oduc i i y. Howe e , u he esea ch is necessa y o gain
a deepe unde s anding o he echnology and i s po en ial
bene i s, pa icula ly o la ge o ganiza ions. The dis up-
i e na u e o AI has long been a subjec o in es iga ion,
ye many companies s uggle o ansla e AI’s po en ial in o
angible business alue (Shollo e al., 2022, p. 1) To ad-
d ess his gap, i is c ucial o analyze he cha ac e is ics o
Gene a i e AI and explo e how i can bene i la ge co po-
a ions on a b oade scale. This explo a ion is essen ial o
iden i ying and unde s anding he oppo uni ies ha Gen-
e a i e AI p esen s o c ea ing alue wi hin la ge co po a e
en i onmen s. By doing so, he hesis aims o con ibu e o
he ongoing discussion on he alue c ea ion po en ial o AI.
Exp essed in Resea ch Ques ion 2:
RQ2: To wha ex en does Gene a i e AI open up
new oppo uni ies o alue c ea ion in companies?
Based on he po en ial alue-add, he ques ion a ises how
companies can o ganize and d i e he implemen a ion o AI
in hei o ganiza ion. P e ious esea ch has shown ha mul-
iple dimensions need o be conside ed o adop AI success-
ully (U en & Edwa ds, 2023). Howe e , depending on he
ocus a ea o he esea ch, di e en equi emen s ha e been
p oposed. While some s udies ocus on echnological aspec s
o AI (Da enpo & Ronanki, 2018, p. 52), o he s highligh
he impo ance o s a egic conside a ions (B ock & Wangen-
heim, 2019, p. 7). P o iding insigh s om a case s udy could
help o e i y he a ious aspec s o hese s udies, s eng hen
he ou lined models, and add o he ongoing discou se in he
ield o science. A icula ed in Resea ch Ques ion 3:
RQ3: Wha conside a ions should companies make
in o de o exploi he alue c ea ion po en ial o
Gene a i e AI?
The hesis aims o p o ide aluable insigh s o bo h
schola s and p o essionals h ough explo ing he de ined e-
sea ch ques ions. Fi s ly, i seeks o e alua e exis ing models
in eal-wo ld scena ios, helping o unde s and hei p ac i-
cal use and how ea lie esea ch ela es o Gene a i e AI.
Secondly, i aims o o e p ac i ione s use ul insigh s in o
po en ial uses and s a egies o scaling Gene a i e AI in
hei o ganiza ions.
In concluding his chap e , i is i al o highligh ha
a comp ehensi e in e p e a ion o he e m Gene a i e AI
should be main ained h oughou he hesis. Gi en he ex-
plo a o y cha ac e o he case s udy and he no el y o he
echnology, in he in e iews, a clea dis inc ion be ween di -
e en echnological aspec s was no always made. This holds
in pa icula ue o he e ms “Gene a i e AI”, “Cha GPT”
and “La ge Language Model (LLM)”, which ha e been e-
quen ly used. As his wo k is abou iden i ying alue c ea ion
oppo uni ies om a business pe spec i e a he han de ail-
ing exac echnical mechanisms, his should be conside ed an
accep able inaccu acy.
3. Concep ual and Theo e ical Backg ound
The nex chap e del es in o he concep ual and heo e i-
cal amewo k essen ial o he hesis. I begins by highligh -
ing key de elopmen al s ages ha ha e led o Gene a i e AI.
A. Sake /Junio Managemen Science 10(3) (2025) 631-656 633
Following ha , i in oduces heo e ical concep s ha shed
ligh on how AI can enhance alue c ea ion and he essen ial
capabili ies equi ed by companies o achie e his.
3.1. E olu ion o A i icial In elligence and Machine Lea n-
ing
The idea o a i icial in elligence da es back o he 1950s
when he ma hema ician and compu e scien is Alan Tou -
ing asked himsel how he in elligence o compu e s could be
measu ed (Tu ing, 1950). Ins ead o asking whe he a ma-
chine is in elligen , he in oduced he “Imi a ion Game” (now
known as he Tu ing Tes ) o ask whe he machines can im-
i a e human esponses con incingly (Tu ing, 1950, p. 433).
Tu ing desc ibed machines o digi al compu e s as complex
sys ems capable o a wide ange o asks, simila o a human
ollowing ins uc ions (Tu ing, 1950, p. 436). He sugges ed
ha hese compu e s, due o hei as capabili ies, could po-
en ially pass as human in his es (Tu ing, 1950, p. 442).
This idea was a signi ican s ep in unde s anding machine in-
elligence, p oposing a p ac ical way o measu e i and laying
he g oundwo k o he ield o a i icial in elligence.
The e m “A i icial In elligence” i sel only became a col-
lec i e e m o a a ie y o di e en concep s in 1956. Mc-
Ca hy, Assis an P o esso o Ma hema ics a Da mou h Col-
lege, chose he e m o a wo kshop on he opic and is he e-
o e ega ded oday, alongside Ma in Minsky, Allen Newell,
He be Simon as one o he “ a he s o AI” (Nilsson, 2013,
p. 80). McCa hy was among he i s who ou lined he e-
qui emen s o a sys em o e ol e human-like in elligence in
mo e de ail. In his pape “‘p og ams wi h common sense”
(1958), he discussed ea u es which would be equi ed o a
machine o e ol e in elligence. These included he capabili y
o ep esen all beha io s, simply exp ess in e es ing beha -
io al changes, imp o e mos aspec s o beha io , unde s and
pa ial success in complex p oblems, and c ea e imp o able
sub ou ines (McCa hy, 1958, p. 5).
Building on McCa hy’s ounda ional wo k, a ious de i-
ni ions o AI ha e eme ged o e ime. Simmons and Chappell
(1988, p. 14) de ine AI as he beha io o a machine which,
i a human beha es in he same way, would be conside ed
in elligen . Luge and S ubble ield (1998, p. 1) de ine AI as
a b anch o compu e science ocused on he au oma ion o
in elligen beha io . In con as , Russell and No ig (2021,
p. 2) ollow an a ional agen app oach, de ining AI as he
s udy and cons uc ion o in elligen agen s. Based on he
scien i ic wo k, he ollowing de ini ion shall be applied in
his hesis: AI is de ined as he abili y o machines o com-
pu e s o lea n and pe o m asks ha a e ypically a ibu ed
o human in elligence.
The idea o eaching machines o lea n can be aced back
o Samuel (1959). Using he game o checke s, he was able
o p og am a compu e ha was able o play he game and
ou pe o m a human playe based on a udimen a y se o
pa ame e s and ules (Samuel, 1959, p. 535). His no el con-
cep o machines ha could imp o e hei pe o mance o e
ime h ough expe ience, is o en ci ed as one o he ea lies
wo k o “machine lea ning” (McCa hy & Feigenbaum, 1990,
p. 10).
Wha does “lea ning” ac ually mean when discussing
compu e p og ams? In his book “Machine Lea ning” (1997),
T. Mi chell o e s he ollowing de ini ion: “A compu e p o-
g am is said o lea n om expe ience E wi h espec o some
class o asks T and pe o mance measu e P, i i s pe o mance
a asks in T, as measu ed by P, imp o es wi h expe ience E”
(Mi chell, 1997, p. 2). He also uses he game o checke s as
an example o desc ibe a compu e p og am ha imp o es
i s abili y o win by gaining expe ience. This example can
be applied o mo e han simple games. Mo e complex ules
and sequences o ins uc ions, be e known as algo i hms,
now enable compu e p og ams o lea n a wide a ie y o
asks. Speech ecogni ion, d i ing au onomous ca s and he
classi ica ion o new as onomical s uc u es a e examples
whe e machine lea ning is used in p ac ice oday (Mi chell,
1997, p. 3).
Be o e del ing in o he speci ics o machine lea ning, Fig-
u e 1p esen s key e ms ha a e bene icial o subsequen
explo a ion o he opic. Machine lea ning, which can be seen
as a sub ield o a i icial in elligence, encompasses di e se
lea ning pa adigms. These will be ou lined in he ollowing
sec ion, p eceding in-dep h explo a ions o deep lea ning and
Gene a i e AI in la e chap e s.
The i s ype o lea ning is called supe ised lea ning
(Good ellow e al., 2016, p. 103). In his me hod, he ma-
chine o algo i hm is p o ided wi h a da ase whose exam-
ples a e labeled. When new, unlabeled da a is p esen ed, he
machine a emp s o p edic he label based on he pa e ns
and p ope ies o he known da a. The esul is hen com-
pa ed wi h he ac ual label. The aining o he algo i hm
consis s o educing he e o be ween he es ima ed and he
ac ual labels, enabling he machine o ecognize new inpu
and classi y i co ec ly (LeCun e al., 2015, p. 436). The
second ype is called sel - o unsupe ised lea ning. Ins ead
o p o iding he machine wi h labeled da a, i lea ns o iden-
i y pa e ns and p ope ies om he da a i sel . This way, i
lea ns abou he p obabili y dis ibu ion o he en i e da ase
(Good ellow e al., 2016, p. 103). While his ype o lea ning
has been explo ed o a long ime (see Hin on and Sejnowski,
1985), i gained popula i y wi h he in oduc ion o T ans-
o me a chi ec u e, la ge da ase s, and he a ailabili y o
compu ing powe (Rad o d e al., 2019, p. 10). A hi d ype is
called ein o cemen lea ning, which in oduces a eedback
loop in o he lea ning p ocess. Du ing aining, he model
uses i s expe iences o imp o e pe o mance h ough a e-
wa d and penal y sys em, aiming o inc ease i s cumula i e
ewa ds o e ime (Su on & Ba o, 2018, p. 2).
3.1.1. F om Tu ing o Deep Lea ning
O e he pas ew decades, esea ch in he ield o AI has
made subs an ial p og ess. New me hodologies, models, and
a chi ec u es ha e been de eloped, as ly su passing ini ial
concep ions. These ad ancemen s enable AI o pe o m asks
such as w i ing poems, de eloping so wa e, and composing
music (Feue iegel e al., 2023, p. 1). A pi o al echnique
A. Sake /Junio Managemen Science 10(3) (2025) 631-656634
Figu e 1: Rela ionship be ween AI e ms which will be discussed in his hesis.
ha has been cen al o hese capabili ies is deep lea ning
(Good ellow e al., 2016, p. 5), which in ol es decomposing
complex ela ionships in o simple , in e connec ed concep s.
These simple concep s can hen be ep esen ed by e en mo e
undamen al concep s (Good ellow e al., 2016, p. 1). In
his way, deep lea ning enables compu e s o in e complex
ela ionships om simple ones (Good ellow e al., 2016, p.
2). The a chi ec u e o hese models comp ises nume ous
hie a chical laye s, which is why i is e e ed o as “deep
lea ning”.
Deep lea ning is based on he concep o a i icial neu-
al ne wo ks (ANN), ha is, sys ems inspi ed by he human
b ain’s s uc u e (Good ellow e al., 2016, p. 165). They
a e e med ne wo ks because hey ypically comp ise a la ge
numbe o in e connec ed nodes, e e ed o as neu ons, dis-
ibu ed ac oss a ious laye s (Good ellow e al., 2016, p.
164). Figu e 2illus a es he simpli ied s uc u e o an ANN.
The inpu laye ecei es he ini ial in o ma ion ( o exam-
ple, aw da a), which is hen ansmi ed o he subsequen
laye . Each node o neu on p ocesses his inpu h ough a
simple compu a ion and o wa ds i s ou pu o he nex laye
o neu ons. The a i icial ne wo k “lea ns” by modi ying he
pa ame e s o he connec ions be ween indi idual neu ons,
known as weigh s, based on expe ience and pe o mance.
This adjus men is made using a unc ion ha calcula es he
disc epancy be ween he ac ual ou pu and he desi ed ou -
pu , hen al e s he weigh s o minimize his e o (LeCun e
al., 2015, p. 436) This p ocess, known as back-p opaga ion,
was i s de ailed in an a icle by Rumelha e al. (1986, p.
533) and enables he p og am o sel -o ganize and e ine i s
in e nal s uc u e. The laye s be ween he inpu and ou pu
a e called “hidden laye s” because hei ac i i ies a e no di-
ec ly obse able om he inpu da a. Ins ead, he model
mus de e mine which pa e ns a e signi ican o explaining
he inpu da a (Good ellow e al., 2016, p. 6).
The e a e di e en ypes o a i icial neu al ne wo ks.
Two o hese a e ecu en neu al ne wo ks (RNN) and con-
olu ional neu al ne wo ks (CNN). While RNNs specialize in
he p ocessing o sequences, such as ex (G a es, 2012, p.
1), he s eng h o CNNs lies in he p ocessing o as e in-
o ma ion, such as images (Good ellow e al., 2016, p. 367).
A weak poin o he RNN and CNN a chi ec u es is he lim-
i ed con ex window, i.e., he amoun o in o ma ion ha he
sys em can s o e o e a longe pe iod o ime (Hoch ei e
e al., 2001, p. 11). To mi iga e his ulne abili y, Hoch e-
i e and Schmidhube (1997) de eloped a no el RNN a chi-
ec u e called Long Sho -Te m Memo y (LSTM). Thei cus-
omiza ion made i possible o sol e la ge and mo e complex
asks ha we e no easible wi h he s anda d RNN a chi ec-
u e (Hoch ei e & Schmidhube , 1997, p. 2).
Howe e , he p ocessing o long sequences emained a
majo challenge. This became pa icula ly appa en in he
a ea o machine ansla ion, as i was di icul o RNNs o
ecognize wo d dependencies o e long dis ances in sen-
ences o pa ag aphs (Bahdanau e al., 2014, p. 6; Kim
e al., 2017, p. 2). One idea was no o encode he en i e
sequence, bu a he o ocus on indi idual sec ions. The
so-called “a en ion mechanism” allows he model o dynam-
ically ocus on di e en pa s o he inpu (Kim e al., 2017,
p. 2). To simpli y, his can be compa ed o eading a scien-
i ic a icle and s umbling o e a di icul sec ion. Ins ead o
ying o comp ehend he en i e con en , i is help ul o ocus
on indi idual wo ds o o look back a he p e ious sec ions.
In his way, a en ion is shi ed o he key aspec s, helping
o b eak down he sec ion o sequence in o smalle , mo e
manageable pa s and o unde s and how each pa connec s
o he o he s.

A. Sake /Junio Managemen Science 10(3) (2025) 631-656 635
Figu e 2: Simpli ied s uc u e o an A i icial Neu al Ne wo k (ANN)
3.1.2. Eme gence o A en ion Mechanisms and T ans o m-
e s
The a en ion mechanism desc ibed abo e was e y e ec-
i e, especially o machine ansla ion asks, and is he e o e
egula ly used in deep lea ning models (Kim e al., 2017, p.
1). Howe e , he sequen ial na u e o RNNs emained he
limi ing ac o in e ms o pa alleliza ion and scalabili y o
hese models (Peng e al., 2023, p. 1). In 2017, Vaswani e
al. p oposed a new a chi ec u e in hei a icle “A en ion Is
All You Need”. Called he “T ans o me ”, his new model is
based solely on he a en ion mechanism and no longe on
an RNN o CNN s uc u e. I s s eng h lies in i s abili y o
e e ence in ini ely long sequences wi hou he p e ious lim-
i a ions o RNNs (Vaswani e al., 2017, p. 5), as isualized in
Figu e 3.
I he model is gi en he ask o answe a ques ion, i can
e e o he en i e ex wi h he help o he a en ion mecha-
nism. This is pa icula ly help ul when ex ual ela ionships
only become appa en upon iewing an en i e pa ag aph.
RNNs, wi h hei limi ed con ex window, we e es ic ed in
his ega d. In he example, an RNN model would “ o ge ”
ha he ques ion e e ed o a ca .
In 2015, he esea ch g oup OpenAI was o med wi h he
goal o ad ancing in he ield o AI (OpenAI, 2015). One o
he i s a icles published by OpenAI’s esea che s ela ed o
he T ans o me a chi ec u e was “Imp o ing Language Un-
de s anding by Gene a i e P e-T aining” (2018). In he a i-
cle, Alec Rad o d and Ka hik Na asimhan ocus on enhanc-
ing na u al language p ocessing (NLP) ia a semi-supe ised
lea ning app oach ha combines unsupe ised p e- aining
and supe ised ine- uning (2018, p. 1). The au ho s p opose
a no el app oach ha u ilizes la ge unlabeled ex co po a
o p e- aining a language model, ollowed by ask-speci ic
ine- uning. Thei app oach aims o o e come he limi a ions
o supe ised models, which o en equi e ex ensi e labeled
da a ha is sca ce o expensi e o ob ain.
In hei model, Rad o d and Na asimhan (2018, p. 2) em-
ploy he T ans o me a chi ec u e due o i s e icien handling
o long- e m dependencies in ex sequences. Thei aining
me hod consis s o wo s ages: In he p e- aining phase, hey
use a language modeling objec i e on an unsupe ised co -
pus, es ablishing he ini ial pa ame e s o he model. Lan-
guage modeling is he p ocess o aining a model o p edic
he nex wo d based on p e ious wo ds in a sen ence (Voi a
e al., 2019, p. 3). Figu e 4shows a simpli ied example us-
ing a Google sea ch, whe e he model sugges s he nex wo d
based on i s p edic ed p obabili y.
Applying his p inciple o aining da a, a se o 7,000
unpublished books, he model was able o ecognize co e-
la ions and gain knowledge no only abou indi idual sen-
ences bu also abou he na u e o language in gene al
(Rad o d & Na asimhan, 2018, p. 8). In he ine- uning
s age, he lea ned co ela ions and knowledge we e adap ed
o speci ic language unde s anding asks such as ques ion
answe ing, common sense easoning, seman ic simila i y
analysis, and ex classi ica ion. Subsequen ly, hey used
di e en es s o measu e he pe o mance in hese ca e-
go ies, achie ing s a e-o - he-a esul s in 9 o 12 ca ego ies
(Rad o d & Na asimhan, 2018, p. 8).
3.1.3. Ad ancemen s in GPTs and Founda ion Models
Rad o d and Na asimhan’s (2018) wo k laid he basis
o u he de elopmen o he GPT a chi ec u e. In an a -
icle published in 2019, a nex -gene a ion GPT model, GPT-
2, was in oduced (Rad o d e al., 2019). Ins ead o using
cu a ed da ase s, hey based hei aining on publicly a ail-
able da a om he in e ne . Wi h he in en ion o c ea ing
a la ge and di e se co pus o na u al language ex co e ing
as many domains as possible, hey ga he ed o e 8 million
documen s, o aling 40 GB o da a (Rad o d e al., 2019, p.
3). When es ing o na u al language p ocessing asks, such
as ques ion answe ing, ansla ions, eading comp ehension,
o summa iza ion, hey obse ed he model’s capabili y o
lea n hese asks wi hou explici supe ision (Rad o d e al.,
A. Sake /Junio Managemen Science 10(3) (2025) 631-656636
Figu e 3: Compa ison o con ex windows
Figu e 4: Google Sea ch: Example o wo d p edic ons
2019, p. 1). This can be seen as an signi ican b eak h ough,
as p e ious models equi ed subs an ial amoun s o labeled
and cu a ed da a o lea n e ec i ely (Rad o d & Na asimhan,
2018, p. 1). Du ing hei es s, hey expe imen ed wi h di -
e en model sizes and showed ha pe o mance imp o ed
wi h inc easing model capaci y. Thei la ges model, wi h 1.5
billion pa ame e s, achie ed s a e-o - he-a esul s in mul i-
ple se ings (Rad o d e al., 2019, p. 10).
The po en ial o pe o mance gains h ough inc eased
model capaci y was u he explo ed by B own e al. (2020).
Thei a icle in oduced he GPT-3 model, con aining 175 bil-
lion pa ame e s and ep esen ing a signi ican ad ancemen
o e p e ious models by a ac o o 10. Con i ming he e-
sea che s’ hypo heses, he model ou pe o med i s p edeces-
so , GPT-2, and e en i aled he pe o mance o s a e-o - he-
a ine- uned sys ems (B own e al., 2020, p. 9). They ound
ha addi ional compu a ional powe di ec ly co ela ed wi h
inc eased pe o mance, laying he g oundwo k o u he
esea ch. Mo eo e , since hei wo k ocused on c ea ing a
ask-agnos ic model, hey an icipa ed ha u u e ine- uning
would u he enhance hei model’s pe o mance (B own e
al., 2020, p. 2).
Fu he de elopmen s occu ed o e he nex yea s, ad-
ancing he model and in oducing ine- uning o boos pe -
o mance (Ouyang e al., 2022, p. 1). This culmina ed in
OpenAI’s elease o he now well-known Cha GPT (OpenAI,
2022). Wi h i s unexpec ed capabili y o gene a e ex indis-
inguishable om human w i ing and engage in belie able
human-like con e sa ions, i caugh many esea che s o
gua d (Dwi edi e al., 2023, p. 4). Technologically, he ad-
ances s em om wo p ima y aspec s. Fi s ly, an upda ed
GPT-3 model wi h a modi ied da ase se es as he ounda-
ion o Cha GPT (GPT-3.5). Secondly, a echnique called
Rein o cemen Lea ning om Human Feedback (RLHF) was
used o ine- une he model. RLHF can be de ined as a
me hod whe e he model is imp o ed based on human e al-
ua ions o i s ou pu s, guiding i owa d desi ed beha io s
and esponses (Ziegle e al., 2019, p. 1). The p ocess u i-
lizes a ewa d model ha is ained using eedback om
human labele s, who e alua e and ank he model’s answe s
om bes o wo s . This ewa d model is hen used wi hin
he main model, employing ein o cemen lea ning o sel -
op imiza ion. This signi ican ly enhances he model’s abili y
o p o ide answe s aligned wi h human p e e ences and con-
side p e ious inpu in con e sa ional se ings (Feue iegel e
al., 2023, p. 4). Alongside he echnological leap, OpenAI’s
decision o g an un es ic ed public access o Cha GPT and
i s simpli ied use in e ace ueled i s explosi e wo ldwide
adop ion (Feue iegel e al., 2023, p. 5). Figu e 5shows he
landing page o Cha GPT 3.5, o e ing examples o ques ions
and a p omp ba .
In Ma ch 2023, jus i e mon hs a e Cha GPT’s elease,
OpenAI un eiled he nex gene a ion o hei GPT model
(OpenAI, 2023). Unlike p e ious models ocused solely on
ex ual inpu , GPT-4 could p ocess bo h ex and images as
inpu , gene a ing ex ual ou pu s. While de ailed speci ica-
ions emain limi ed, i s pe o mance exceeds ha o i s p e-
decesso , GPT-3.5, in nume ous a eas, achie ing human-le el
esul s on mul iple benchma ks (OpenAI, 2023, p. 6). As
p io esea ch demons a ed a co ela ion be ween pe o -
mance and compu a ional powe , anonymous epo s sug-
ges he model may possess 1.76 illion pa ame e s (Bas-
ian, 2023). While his in o ma ion should be ea ed wi h
cau ion, i gi es an indica ion o he apid echnological de-
elopmen in he a ea o Gene a i e AI. F om 1.5 billion pa-
ame e s in 2019, o e 175 billion in 2020 o po en ially 1.76
illion in 2023.
Wi h he ad ances in he ield o AI, a shi in he de-
elopmen o AI models has eme ged. While p e ious mod-
els we e ained on labeled da a and ine- uned o speci ic
asks, newe models such as GPT-3 and 4 di e signi ican ly
in wo a eas. Fi s ly, ins ead o labeled da a, hey a e ained
A. Sake /Junio Managemen Science 10(3) (2025) 631-656 637
Figu e 5: Cha GPT Landing Page.
on b oad, unlabeled in o ma ion using sel -supe ised lea n-
ing. Secondly, he shee scale allows he models o pe o m
asks hey ha e ne e been explici ly ained on (B own e
al., 2020, p. 9). This gene al unde s anding o language and
o he pa e ns, os e ed by he inc ease in a ailable compu-
a ional powe , led o he need o c ea e a new class o hese
models. In hei scien i ic a icle, Bommasani e al. (2021)
he e o e p oposed he e m “ ounda ion model” o encapsu-
la e hese cha ac e is ics and desc ibe he eme gence and ap-
plica ion o hese models in de ail. Acco ding o he au ho s,
ounda ion models build on he idea o la ge language mod-
els, which gain an unde s anding o na u al language based
on ex ual da ase s (B own e al., 2020, p. 9). Expanding
on his concep , ounda ion models a e mul imodal – besides
ex , hey a e able o p ocess da a such as images, ideos, 3D
signals, and o he s. As wi h la ge language models, hei as
da ase s allow hem o gain a gene al unde s anding o he
unde lying s uc u es and p ope ies applicable o di e en
downs eam asks (Bommasani e al., 2021, p. 6). A p omi-
nen example is GPT-4, wi h i s capabili y o p o ide answe s
based no only on ex bu also on images (OpenAI, 2023).
3.1.4. Gene a i e AI: Scope and Capabili ies
In he p e ious sec ion, a ious de elopmen s eps and
echnological inno a ions ha e been ou lined, om simple
ML algo i hms o deep lea ning echniques, culmina ing in
GPTs, la ge language and ounda ion models. One ea u e
ha was men ioned in connec ion wi h GPTs is hei abili y
o gene a e new con en . While he concep o GPTs has i s
o igin in he publicly a ailable wo k by Vaswani e al. (2017)
on he T ans o me a chi ec u e, he e m i sel was p ima -
ily coined by OpenAI’s ad ances in he ield (in pa icula
wi h GPT-3 and 4). Google, Me a, and o he companies ha e
since published hei own models based on he T ans o me
model ( o example Google’s Gemini o Me a’s Llama 2). To
main ain a neu al pe spec i e, he e m “Gene a i e AI” is
he e o e used h oughou his hesis.
Gene a i e AI, om a echnical s andpoin , is oo ed in
gene a i e modeling, which seeks o unde s and he join
p obabili y dis ibu ion P(X,Y)whe e X ep esen s he da a
and Ydeno es he labels. The objec i e is o g asp how da a
is gene a ed in o de o c ea e new da a poin s (Ng & Jo -
dan, 2001, p. 1). This app oach di e s om disc imina i e
modeling, which ocuses on modeling he condi ional p ob-
abili y P(X,Y), ep esen ing he p obabili y o he label Y
gi en he inpu da a X(Ng & Jo dan, 2001, p. 1). This di -
e en ia ion is c ucial om an applica ion s andpoin : Gene -
a i e AI models enable he gene a ion o new da a ins ances
based on obse ed p obabili y dis ibu ions wi hin a gi en
da ase (Good ellow e al., 2016, p. 716), whe eas disc im-
ina i e modeling exhibi s supe io pe o mance in classi ica-
ion asks (Ng & Jo dan, 2001, p. 1).
3.2. AI Value C ea ion and Business Impac
O e he las ew decades, a la ge numbe o scien i ic
a icles ha e ocused on he in luence o AI on companies.
The ques ion o how and in which a eas AI can be used and
wha p e equisi es companies need o ul il in o de o do his
success ully is a ecu ing one. Selec ed heo e ical concep s
will he e o e be p esen ed and desc ibed in he ollowing
sec ion. Cane and Bha i’s (2020, p. 182) concep ual ame-
wo k can be used as a s a ing poin . Based on an ex ensi e
li e a u e e iew, he au ho s p opose six pe spec i es om
which he s a egic dimension o AI can be iewed:
A. Sake /Junio Managemen Science 10(3) (2025) 631-656638
1. Capabili ies and Limi a ions o AI
2. Business Func ions and AI
3. Tasks, Jobs, and In elligence
4. Economy and AI
5. AI and Law, Regula ions, Go e nance
6. Indus ies and AI
This hesis ocuses on he i s h ee aspec s, as hey a e
especially use ul o unde s anding how AI impac s indi id-
ual companies. The i s aspec co e s he capabili ies and
limi a ions o AI, as discussed in Chap e 3.1. I explo es
he echnical aspec s o AI, including i s a ious applica ions,
echnologies, and limi a ions. Unde s anding AI’s s eng hs
and weaknesses is c ucial in a business con ex o se ealis ic
expec a ions and iden i y p ac ical deploymen op ions (Da -
enpo & Ronanki, 2018, p. 110). While some cons ain s
iden i ied by Cane and Bha i (2020), like da a labeling and
lea ning gene alizabili y, ha e been add essed wi h echno-
logical p og ess ( e e o OpenAI, 2023), challenges such as
biases and lack o explainabili y pe sis .
The aim o he second dimension business unc ions and
AI is o ca ego ize he ields o applica ion o AI. Cane and
Bha i base his on an IBM su ey and a s udy by Da enpo
and Ronanki (2018) (Cane & Bha i, 2020, p. 186). They
di e en ia e be ween wo subjec a eas: Fi s ly, he a eas in
which AI is used, e.g., in in e nal p ocesses o cus ome se -
ices. Secondly, he way in which AI is used, e.g., o au oma e
en i e p ocesses o gain new insigh s om da a.
The hi d dimension, asks, jobs, and in elligence, de-
sc ibes he di e en s ages in which AI can be used. Based
on Rao’s (2017) amewo k, a dis inc ion is made be ween
h ee le els. Fi s ly, “Assis ed In elligence” includes all p o-
cesses ha a e conduc ed in a con en ional manne and a e
suppo ed by AI. Secondly, “Augmen ed In elligence” e e s o
a eas whe e AI akes o e a la ge pa o he alue c ea ion
( he a eas men ioned a e, o example, au oma ic ansla ion
and au oma ic analysis o legal documen s). Thi dly, “Au-
onomous In elligence” comp ises hose p ocesses ha can
be conduc ed comple ely wi hou human in e en ion in he
u u e.
3.2.1. AI in S a egic Business Con ex
The p og ess ha has been made in he ield o a i icial
in elligence epea edly aises he ques ion o how his ech-
nology can be s a egically implemen ed and u ilized. In pa -
icula , he ques ion o he sou ces o alue c ea ion is being
esea ched in dep h (Bo ges e al., 2021; Ki sios & Kama i-
o ou, 2021; T unk e al., 2020). This aligns wi h he hesis’s
objec i e, namely, o in es iga e he ex en o which he ex-
is ing concep s can be applied o Gene a i e AI. Bo ges e al.
(2021) ex ensi e li e a u e e iew p o ides he i s insigh s
in o his a ea. Based on 41 s udies, hey de ine ou sou ces
o alue c ea ion om AI:
1. Decision Suppo
2. Cus ome and Employee Engagemen
3. Au oma ion
4. New p oduc s and se ices
Decision suppo e e s o he abili y o AI o suppo hu-
mans in s a egic and ope a ional business decisions (Bo ges
e al., 2021, p. 11). This occu s, o example, when deep
lea ning echniques a e used o de ec pa e ns in da a, sub-
sequen ly guiding he decision-making p ocesses. Conse-
quen ly, decisions can be execu ed mo e swi ly and wi h
g ea e eliabili y (Bo ges e al., 2021, p. 12). Howe e ,
scien is s highligh he necessi y o u he esea ch in his
apidly e ol ing ield, pa icula ly conce ning he in e ac-
ion be ween humans and AI and i s impac on o ganiza-
ional pe o mance (Lich en hale , 2019, p. 8). This hesis
p esen s an oppo uni y o examine hese dynamics, espe-
cially in he con ex o Gene a i e AI. I will be insigh ul
o de e mine whe he and how he echnology can op imize
decision-making p ocesses and he eby con ibu e o alue
c ea ion.
The second iden i ied sou ce o alue c ea ion is called
cus ome and employee engagemen . The aim is o use AI
o imp o e he cus ome expe ience and in e nally o a ac
employees o he new echnology. Al hough he a icle anal-
yses a ious academic pape s, he au ho s no e ha u he
esea ch is needed as some examples canno be gene alized.
Gene a i e AI, wi h he abili y o c ea e con en ha is indis-
inguishable om human inpu (Feue iegel e al., 2023, p.
1), could o e new oppo uni ies in his con ex .
Au oma ion ep esen s he nex sou ce o alue c ea ion,
allowing cos s o be educed and e iciency o be inc eased
h ough he s a egic use o AI echnologies. The au ho s a -
gue ha a compe i i e ad an age can also be c ea ed i p o-
cesses can be au oma ed mo e quickly han is possible o
compe i o s (Bo ges e al., 2021, p. 12).
The ou h sou ce o alue c ea ion, new p oduc s and
se ices, deals wi h he abili y o gene a e new business
ideas h ough AI. Bo ges e al. (2021) sugges AI’s po en ial
in d i ing inno a ion and c ea ing new p oduc s and se ices,
howe e hey could only iden i y and p o ide limi ed empi -
ical e idence. Consequen ly, he au ho s a gue ha u he
esea ch is needed o unde s and how AI can be used s a e-
gically o c ea e new p oduc s and solu ions (Bo ges e al.,
2021, p. 12).
The sou ces o alue c ea ion lis ed by Cane and Bha i
as well as Bo ges e al. can also be ound in simila o m in
o he li e a u e e iews and scien i ic a icles. Fo example,
in hei li e a u e e iew o 81 a icles in he con ex o AI and
business s a egy, Ki sios and Kama io ou (2021) iden i ied
“AI and Machine Lea ning in o ganiza ions”, “AI, knowledge
managemen and decision-making” and “AI, se ice inno a-
ion and alue” as sou ces o alue c ea ion.
An a ea ha was no men ioned in ea lie sou ces and
which hey disco e ed was he alignmen o AI ools and In-
o ma ion Technology (IT) wi h o ganiza ional s a egy.
A. Sake /Junio Managemen Science 10(3) (2025) 631-656 645
Figu e 7: Value C ea ion Oppo uni ies
Gene a i e AI con e s use commands in na u al language
in o SQL commands o e ie e in o ma ion. One in e ie-
wee emphasized ha his could enable as e , mo e indi id-
ualized que ies in he u u e, dec easing he need o c ea e
new dashboa ds o end use s (IP-4, Domain Expe ). O he
examples include combining Gene a i e AI wi h Robo ic P o-
cess Au oma ion (IP-3, P oduc Owne ), using Gene a i e AI
alongside SAP inpu s (IP-7, Domain Expe ), o e alua ing
applica ion documen s in he Siemens-wide job po al using
Gene a i e AI (IP-1, P og am Manage ).
The da a e ealed ha he main ocus o assis an sys ems
is on use cases ha ac i ely suppo use s in hei asks. These
coaches & helpe s accompany he use along he alue chain
in a ious p ocess s eps and in luence he desi ed esul s.
Added alue is expec ed in wo a eas in pa icula . In he use
o ex e nal ools such as Mic oso Copilo and in he de elop-
men o in e nal, domain-speci ic assis an s, such as a cha -
bo o se ice echnicians in he a ea o Sma In as uc u e
(IP-21, P og am Manage ). Fu he applica ion examples in-
clude he de elopmen o cha bo s o p epa e and p ac ice
nego ia ions in pu chasing (IP-7, Domain Expe ), he use o
Gene a i e AI in in e iew p ocesses (IP-9, Technology Ex-
pe ) o suppo o p ocess e iews as pa o in e nal audi s
(IP-12, Domain Expe ).
While he majo i y o he iden i ied use cases en isage he
abo e-men ioned assis ance ole o Gene a i e AI, indi id-
ual in e iew pa ne s see he possibili y o using he ech-
nology o au oma e en i e p ocesses. This usually includes
he in e ac ion o se e al so wa e applica ions and sys ems,
in he con ex o which Gene a i e AI ep esen s he in e ace
o he use . Illus a ing his poin , an in e iewee men ioned
a Gene a i e AI agen ha makes i possible o classi y cus-
ome inqui ies au oma ically, ex ac he necessa y in o ma-
ion, upda e he linked sys ems, and w i e a esponse o he
cus ome (IP-10, P og am Manage ). I was poin ed ou ha
while his was al eady possible in heo y, in eali y he deci-
sions would s ill be made by humans due o he ma u i y o
he use case, he speci ic ea u es o he echnology and he
associa ed isks.
Ano he la ge clus e comp ises use cases ha can be
summa ized as coding suppo . As a echnology company,
Siemens employs so wa e in a wide a ie y o o ms and a -
eas. The abili y o Gene a i e AI o gene a e code in di e en
p og amming languages was he e o e seen as a g ea le e
o inc ease p oduc i i y and imp o e exis ing so wa e solu-
ions. One example men ioned in he in e iews and commu-
nica ed ex e nally by Siemens as a ligh house p ojec e ol es
a ound he Siemens Indus ial Copilo , as desc ibed in de ail
below.
Example: Siemens Indus ial Copilo
Siemens o e s a a ie y o solu ions o ac o y
au oma ion, including con ol sys ems o ma-
chines and p ocesses. In his use case, Gene a-
i e AI is deployed o w i e PLC (P og ammable
Logic Con olle ) code o hese machines. A dis-
inc i e ea u e men ioned is he abili y o ans-
la e and imp o e code om o he p og amming
languages. Addi ionally, he so wa e le e ages
Gene a i e AI o iden i y bugs and sugges solu-
ions, signi ican ly educing p og amming e o
and po en ial down ime in ac o ies.
Ano he example o Gene a i e AI used wi hin Siemens’
so wa e solu ions is i s implemen a ion in he Mendix low-
code pla o m. In collabo a ion wi h ex e nal pa ne AWS,
Siemens ex ended he p og am o u he suppo use s in he

A. Sake /Junio Managemen Science 10(3) (2025) 631-656646
de elopmen , alida ion, and op imiza ion o Mendix appli-
ca ions. In addi ion o hese wo cus ome - acing solu ions,
many in e iewees epo ed added alue in de eloping in-
e nal applica ions. One example men ioned by a espon-
den was Gene a i e AI assis ing in he de elopmen o an
au oma ion sc ip o a building’s ene gy supply (IP-19, Do-
main Expe ). A common heme om he in e iews was
ha he coding en i onmen ’s ocus was less on indi idual
use cases and mo e on ans o ming he way p og amming is
done. Many esponden s see g ea added alue in he abili y
o gene a e and imp o e code, as well as in using assis an s
like Gi Hub Copilo (IP-6, P oduc Owne ).
The las clus e can be desc ibed as con en c ea ion,
whe e he ocus is on gene a ing new insigh s and da a wi h
he help o Gene a i e AI. Common examples include i s use
in ma ke ing o w i ing news a icles o in human esou ces
o gene a ing job ad e isemen s. In addi ion o hese gen-
e al applica ions, speci ic Siemens use cases we e iden i ied.
Fo example, one in e iew pa ne epo ed he use o Gen-
e a i e AI as pa o he inno a ion p ocess o ind new al-
e na i e ma e ials o p oduc s (IP-7, Domain Expe ). An-
o he pa icipan epo ed using Gene a i e AI o ansla e 3D
models o me al pa s in o code o milling machines (IP-18,
Technology Expe ).
In summa y, he collec ed use cases indica e a b oad
ange o applica ions o Gene a i e AI. In addi ion o sim-
ple models o in o ma ion e ie al, he echnology also
o e s added alue o complex in e nal p ocesses as well as
cus ome - acing applica ions. In his con ex , he iden i ied
alue c ea ion clus e s p o ide a aluable s a ing poin o
be e unde s anding he di e se po en ial o hese use cases.
5.3. Value C ea ion D i e s
The da a analysis e ealed wo main d i e s o Gene a i e
AI use cases, as highligh ed in Figu e 8.
Figu e 8: Value C ea ion D i e s
The i s g oup can be b oadly summa ized as s a egic
di e en ia ion. These use cases ocus on gaining a unique
compe i i e ad an age by le e aging esou ces exclusi e o
he company, enabling di e en ia ion om he compe i ion.
These esou ces include in e nal company da a (such as cus-
ome in o ma ion), p ocess knowledge, and he IT in as-
uc u e i sel . This g oup also encompasses he de elop-
men o new p oduc s o he expansion and applica ion o
Gene a i e AI capabili ies wi hin exis ing p oduc s. Exam-
ples include he Indus ial Copilo , which Siemens uses o
s eng hen i s own so wa e applica ions, and he in eg a ion
o Gene a i e AI in o he Mendix low-code pla o m.
In e nally, his can in ol e de eloping echnology s acks
and pla o ms o accele a e he ollou and dis ibu ion o
speci ic use cases. Fo ins ance, he Global Sha ed Se ices
di ision de eloped a Gene a i e AI pla o m ha enables
depa men s wo ldwide o adop simple use cases wi hou
needing o in es in he echnology hemsel es (IP-10, P o-
g am Manage ). This c ea es syne gy e ec s, as he in e nal
solu ion equi es de elopmen only once and can subse-
quen ly be olled ou ac oss he en i e o ganiza ion.
The second d i e is p oduc i i y imp o emen . The da a
and in e iews e ealed ha he majo i y o cu en use cases
all in o his ca ego y, less ocused on unique solu ions and
mo e on le e aging he echnology e ec i ely. This includes
making solu ions b oadly accessible o as many use s as pos-
sible o imp o e ope a ional e iciency on a la ge scale. Pa -
icipan s equen ly men ioned as e in o ma ion e ie al
as one o he bigges le e s. O he s lis ed quali y imp o e-
men s, as e execu ion o p ocesses, and isk minimiza ion
as p oduc i i y-boos ing ac o s.
5.4. Value C ea ion A eas
One inding ha eme ged du ing he case s udy is he
company’s need o clus e po en ial use cases no only ac-
co ding o i s echnological backg ound, bu in pa icula ac-
co ding o company-speci ic segmen a ion. This was pa icu-
la ly no iceable when analyzing company p esen a ions and
pa icipa ing in mee ings on he possible a eas o applica ion
o Gene a i e AI. Depending on he ocus, di e en pe spec-
i es can be a o ed in he a ious company di isions. Ta-
ble 3shows h ee dis inc iews on use cases which a e used
wi hin he company. The i s iew is buil a ound unc ional
a eas wi hin he company and allows in e es ed pa ies o
di e en ia e Gene a i e AI use cases acco ding o an es ab-
lished company-wide used segmen a ion. The second iew is
buil a ound business capabili ies which desc ibe wha busi-
nesses do o achie e a speci ic pu pose. D i en by p ocesses
and no by o ganiza ional s uc u es o echnology, his iew
is o en used in IT o s a egic analysis. Las ly, clus e ing in
oppo uni y a eas p o ides a mo e gene ic iew which is in
pa icula sui able o gaining a as e unde s anding o Gen-
e a i e AI applica ion a eas. O iginally p oposed by Ga ne
(2023), Siemens adop ed he model o i s own o ganiza ion.
I dis inguishes be ween on and back-o ice asks, p oduc ,
and se ices as well as co e capabili ies o a company.
5.5. Explo a ion and Scaling
While ocusing on he alue c ea ion oppo uni ies o
Gene a i e AI, he case s udy iden i ied se e al supplemen-
a y elemen s. Many in e iews p o ided insigh s in o he
cu en s a e o Gene a i e AI a Siemens and ac o s ha
should be conside ed o a success ul company-wide ollou .
Analyzing he da a, wo majo pa e ns eme ged:
Fi s ly, Siemens’ cu en s a e ega ding Gene a i e AI can
be desc ibed as an explo a ion phase. This is d i en by he
echnology’s no el y and Siemens’ app oach o he opic o e
he las yea . Figu e 9shows he key aspec s o his explo-
a ion phase ha we e men ioned du ing he in e iews and
obse ed du ing he case s udy.
A. Sake /Junio Managemen Science 10(3) (2025) 631-656 647
Table 3: Value C ea ion A eas
Func ions Business Capabili ies Oppo uni y A eas
Legal S a egic & In o ma ion Managemen P oduc /Se ices
Cus ome Se ice Sales & Ma ke ing F on O ce
Ope a ions Inno a ion & Li ecycle Managemen Back o ice
Finance & S a egy Supply Chain Managemen Co e Capabili ies
P oduc and R&D Manu ac u ing and P oduc ion
Sales & Ma ke ing Se ice Managemen
Quali y Managemen P ojec Managemen
Human Resou ces En e p ise Se ices
Supply Chain Managemen
Figu e 9: Key aspec s o he explo a ion phase
A consis en pa e n ha eme ged ac oss in e iews was
c ea ing space o inno a ion. In e iew pa ne s e-
quen ly no ed ha hey we e gi en he esou ces and ime
necessa y o explo e his new echnology in dep h. While
his seems logical o echnology depa men s p e iously in-
ol ed in NLP de elopmen , many business a eas we e also
g an ed he esou ces o pa icipa e in he ea ly idea ion
s age. Rema kably, many pa icipan s, pa icula ly domain
expe s, highligh ed ha hei explo a ion o Gene a i e AI
was d i en by hei own in insic in e es . Based on his
pe sonal mo i a ion, hey we e gi en he eedom o explo e
he opic wi hin a company con ex (IP-13, Domain Expe ).
The second heme con ibu ing o he explo a ion phase is
knowledge c ea ion. Findings sugges ha many use cases
a e a om deploymen o a e al eady obsole e due o he
ield’s apid echnological ad ances o e he las yea . How-
e e , as emphasized by many in e iew pa ne s, he knowl-
edge gained du ing hei de elopmen is a aluable asse o
he u u e. Addi ionally, se e al pa icipan s ou lined he
need o in es in he a ea o Gene a i e AI o s ay compe -
i i e. As IP-15 pu s i :
[...]we need o s a building up esou ces as
quickly as possible, because once he opic akes
o , e en wi h compe i o s, i will be all he mo e
di icul o ec ui new people (IP-15, CIO).
Las ly, wo mo e hemes eme ged in se e al in e iews.
Unde s anding he cu en se up summa izes eedback ha
companies should ca e ully analyze hei cu en pain poin s
o apply new echnologies such as Gene a i e AI e ec i ely.
Pa icipan s poin ed ou ha his analysis should conside
bo h he quali y o exis ing da a and po en ial bo lenecks (IP-
5, IT P o essional), as well as he business p ocesses whe e
imp o emen o e s he g ea es added alue (IP-10, P og am
Manage ). Finally, acknowledging limi a ions highligh s
he ac ha he in oduc ion and explo a ion o new ech-
nologies comes wi h a ious unce ain ies. One unce ain y
is he apid de elopmen o Gene a i e AI i sel , making i di -
icul o commi and in es in a speci ic solu ion, as i could
become obsole e in a ew mon hs. Technology-inhe en lim-
i a ions such as hallucina ions, he limi ed explainabili y o
i s ou comes as well as egula o y and legal conside a ions
we e named as addi ional cons aining ac o s (IP-16, Head
o Technology).
Besides he ou lined a ea o explo a ion, in e iew pa -
ne s exp essed a ious equi emen s and challenges which
can be summa ized unde he e m scaling. I became appa -
en ha o la ge co po a ions, he pa h om indi idual pi-
lo s o widely accep ed app oaches and p ocesses is a c i ical
hu dle ha mus be o e come o achie e sus ainable added
alue om new echnologies like Gene a i e AI. To acili a e
a be e unde s anding o he a ious aspec s, he esul s a e
ca ego ized in o he ollowing dimensions:
• People
• P ocesses
• Technology
• Da a
• O ganiza ion
• S a egy
• Communica ion
• Timing
A. Sake /Junio Managemen Science 10(3) (2025) 631-656648
While sligh o e laps a e possible be ween hese dimen-
sions (such as People and O ganiza ion o Timing and S a -
egy), he e a e dis inc di e ences. This is also e idenced
by he ac ha o e 500 code segmen s om he in e iews
could be assigned o he a ea o scaling. The main indings
om he analysis a e he e o e p esen ed below.
The people dimension deals wi h he in luence o Gen-
e a i e AI on employees wi hin he company. A dominan
heme ha eme ged epea edly was he belie ha Gene -
a i e AI will ha e a majo impac , equi ing eskilling and
upskilling o he wo k o ce (IP-1, P og am Manage ). Pa -
icipan s no ed challenges s emming om he echnology’s
apid pace o e olu ion. While some colleagues adap quickly,
in e iewees exp essed conce n ha o he s, o en om he
olde gene a ions, migh s uggle o keep up. This could os-
e a nega i e a i ude owa ds Gene a i e AI, u he ueled
by ea s o job displacemen (IP-2, P og am Manage ).
Ano he hu dle iden i ied was he lack o expe s wi h
bo h business knowledge and deep unde s anding o he
echnology, making alue-adding applica ions di icul o
iden i y (IP-3, P oduc Owne ). Addi ionally, due o widesp ead
media co e age and heigh ened expec a ions, main aining
employee mo i a ion as ini ial ideas u n in o conc e e use
cases becomes challenging (IP-5, IT P o essional).
Se e al key ac o s o o e coming hese hu dles we e
iden i ied. P ima ily, awa eness mus be os e ed h ough
aining, wo kshops, o coaching, accompanied by clea ex-
pec a ion managemen . One in e iew pa ne emphasized
ha he psychological impac , and hus he human ac o ,
canno be unde es ima ed (IP-8, CFO). Cul u al di e ences
mus be conside ed o p e en nega i e consequences om
he in oduc ion o Gene a i e AI. A simple bu illus a i e ex-
ample is he AI-suppo ed eco ding and ansc ip ion unc-
ion in Mic oso Teams, which summa izes mee ing con en ,
analyzes i , and de ines ac ion i ems. This could lead o em-
ployees eeling uncom o able and con ibu ing less o u u e
mee ings.
Finally, alen managemen is c ucial o success ul scal-
ing wi hin he o ganiza ion. Iden i ying key in e nal e-
sou ces is impo an o ecognize gaps and, i necessa y, ill
hem wi h ex e nal candida es. In his con ex , inc eased
coope a ion wi h uni e si ies was men ioned as a way o
a ac young alen o he company (IP-15, CIO).
In e ms o p ocesses, he da a analysis e ealed ha he
in oduc ion o Gene a i e AI c ea es challenges simila o
hose aced wi h o he echnologies. This can be a ibu ed
o he complex and he e ogeneous p ocess landscape wi hin
companies like Siemens. The di e se business a eas, speci ic
equi emen s, and some imes his o ically g own p ocesses
limi he company-wide scaling o new echnologies. In-
e iews indica ed ha pa icipan s iew mo e uni o m p o-
cesses and s anda dized app oaches as majo le e s o scal-
ing (IP-14, P og am Manage ). In his con ex , pa icula im-
po ance was placed o new so wa e solu ions whose se up
would o e he possibili y o conside he equi emen s o
Gene a i e AI a an ea ly s age (IP-5, IT P o essional).
In e ms o he echnology i sel , se e al in e es ing in-
sigh s eme ged. I became clea ha many use cases ailed o
mee ini ial expec a ions, p ima ily due o he echnology’s
cu en ma u i y and limi a ions o complex business p o-
cesses (IP-1, P og am Manage ). A key aspec is he p e-
iously men ioned issue o hallucina ions. Unlike p e ious
models ha p o ided de e minis ic answe s, Gene a i e AI
esponses a e p obabilis ic (Rad o d e al., 2019, p. 2; Ope-
nAI, 2023). I gene a es ou pu s by calcula ing p obabili-
ies based on i s aining da a. The echnology’s ma u i y
also p esen s a signi ican o ganiza ional hu dle. Companies
mus iden i y and e alua e di e en p o ide s and solu ions
be o e app o ing hem o in e nal use. This p ocess equi es
ime and esou ces and can become a bo leneck due o he
apid pace o he echnological de elopmen . The da a e-
ealed se e al elemen s o add ess hese challenges: Fi s ,
Companies should s i e o emain endo and model in-
dependen . Open AI, Mic oso , Google, AWS, and open-
sou ce p o ide s cons an ly elease newe and be e Gen-
e a i e AI solu ions. As men ioned by IP-6, in e nally de el-
oped pla o ms should be as independen as possible o ben-
e i om hese enhancemen s. Second, echnological s acks
should be sha ed wi hin he company o a oid duplica e ap-
p oaches and educe ini ial de elopmen e o s (IP-3, P od-
uc Owne ). Thi d, companies mus ca e ully weigh whe he
in e nal de elopmen is necessa y o i solu ions can be pu -
chased ex e nally. This includes cases like in-house aining
o la ge language models o use cases ha migh be co e ed
by u u e Mic oso Copilo unc ionali ies (IP 16, Head o
Technology).
The da a dimension plays a signi ican ole in he success-
ul implemen a ion and scaling o Gene a i e AI. This s ems
om he expec a ion ha companies can achie e a compe i-
i e ad an age h ough le e aging hei own da a. Howe e ,
many pa icipan s exp essed he need o s uc u e and con-
nec da a wi hin he company o meaning ul use (IP-5, IT
P o essional). Addi ionally, insu icien da a quali y in ce -
ain a eas poses a challenge. The indings sugges ha a clea
da a s a egy add essing he a o emen ioned aspec s is neces-
sa y o success ul scaling o he echnology. As one in e iew
pa ne men ioned, i could be bene icial o appoin a Chie
Da a Manage o emphasize he impo ance o he opic and
guide i in he igh di ec ion (IP-8, CFO). The Siemens Da a
Cloud, an exis ing app oach o da a sha ing and s eamlin-
ing da a p ocessing, was ci ed as a posi i e example (IP-1,
P og am Manage ). O e all, all pa icipan s emphasized he
impo ance o his dimension.
Conside ing he complexi y o la ge companies, he o ga-
niza ional impac o g oundb eaking echnologies like Gen-
e a i e AI can c ea e signi ican challenges. This is he case
o Siemens, as e idenced by he da a analysis. A ecu ing
challenge equen ly add essed by he in e iew pa ne s is
he po en ial o a silo men ali y, whe e mul iple eams migh
wo k on ela ed opics independen ly (IP-22, P og am Man-
age ). While decen alized s uc u es o e indi idual busi-
ness uni s eedom, hey can hinde he e icien in oduc-
ion o new echnologies. The e o e, clea s uc u es, a com-
mon s a egy, and he sha ing o echnological app oaches
A. Sake /Junio Managemen Science 10(3) (2025) 631-656 649
a e conside ed essen ial o success ully scaling Gene a i e
AI use cases (IP-4, P ocu emen ).
Du ing he explo a ion phase men ioned a he begin-
ning, he aim is o build up knowledge and y ou all ace s
o he new echnology. Howe e , con e ing hese indings
in o added alue o he company equi es a clea s a egy.
Findings sugges his is a challenge ega ding Gene a i e AI,
as he e is a lack o clea me ics o measu e he success
o indi idual use cases. Especially ega ding p oduc i i y-
enhancing use cases, e alua ion in absolu e igu es becomes
di icul as poin ed ou by one in e iew pa ne (IP-10, P o-
g am Manage ). Findings sugges ha when de eloping Gen-
e a i e AI solu ions, p io i izing business a eas whe e he
echnology has he g ea es po en ial impac seems bene i-
cial. This aligns wi h ecommenda ions om in e iew pa -
ne s. Fo example, he enginee ing a ea a Siemens Mobil-
i y was iden i ied as a po en ial ocus due o i s la ge sha e
in p ojec execu ion and lack o skilled ailway enginee s.
P oduc i i y imp o emen s in his a ea could he e o e yield
subs an ial imp o emen po en ial (IP-2, P og am Manage ).
This aligns wi h a key inding when analyzing indi idual use
cases. Gene a i e AI o e s a my iad o possible applica ions
in a company like Siemens. Consequen ly, i was emphasized
ha a conside ed app oach o esou ce alloca ion, including
delibe a e decisions on indi idual use cases, is essen ial o
op imal alue (IP-8, CFO).
The as a en ion su ounding Gene a i e AI sugges s
he need o clea in e nal communica ion wi hin he com-
pany. In e iew pa ne s poin ed ou ha widesp ead awa e-
ness is key o e ec i e c oss-depa men al use o he echnol-
ogy. An in e es ing pa e n eme ged du ing he case s udy: In
he ini ial mon hs, he desi e o exchange and communica-
ion solidi ied in o c oss-depa men al communi ies. These
communi ies used in e nal pla o ms like Mic oso Vi a En-
gage (Appendix 4) and Teams (Appendix 5), hos ing egula
in o ma ion e en s. Exis ing AI s uc u es we e expanded,
and hei communica ion e o s we e s eng hened o mee
he immense hi s o knowledge among employees and
manage s. One example is he “AI A ack” communi y, an
IT g oup ocused on s eamlining he de elopmen and op-
e a ion o AI solu ions. A e Cha GPT’s elease, a sepa a e,
secu e in e nal Gene a i e AI Pla o m was es ablished in
pa ne ship wi h Mic oso (see Figu e 10).
This pla o m enables employees o use he echnology
wi hou ea o exposing con iden ial da a and was accom-
panied by an in o ma ion campaign, signi ican ly con ibu -
ing o he company-wide dis ibu ion. O he communi ies
o med in a eas such as human esou ces, u he con ibu -
ing o knowledge ans e and idea exchange.
In ecen mon hs, a new end has been obse ed ha
i s in wi h he p e ious indings, namely he es ablishmen
o c oss-sec o ini ia i es. These e lec he need and desi e
o bundle use cases, c ea e mo e anspa ency and d i e he
opic o wa d s a egically. Due o he size o he company,
hese we e no only se up a co po a e le el bu also in he
indi idual business uni s in o de o ake accoun o indi id-
ual needs and p io i ies. One pa e n, exp essed by almos all
in e iew pa icipan s, was he desi e o c ea e anspa ency
a ound indi idual use cases. This e lec s a p e-exis ing need
wi hin he company, as e o s o ga he and showcase AI use
cases we e al eady unde way. The eme gence o Gene a i e
AI u he ampli ied his need, leading o he c ea ion o new
use case collec ions ac oss di e en a eas o he company.
One example a he co po a e le el is he so-called “Inno a-
ion Rada ”, as shown in Figu e 11. The ada shows use
cases om all a eas o he company and s ages o de elop-
men , anging om ini ial ideas o ligh house p ojec s. The
in e ac i e display allows he use o il e by di e en iews,
such as unc ions, objec i es, o applica ion clus e s. As pa
o he case s udy, his o e iew was used alongside o he
a ea-speci ic use case collec ions o e i y in e iew esul s
and gain a be e unde s anding o use case examples.
Finally, a heme ha can also be seen as a sub opic o he
o he dimensions is iming. This dimension encompasses all
obse a ions ega ding he app op ia e sequencing and ap-
p oach o he in oduc ion o Gene a i e AI. In e iew da a
e ealed di e se pe spec i es on his poin , which can be di-
ided in o wo a eas. While he company o en employs a
well-de ined s uc u e and miles ones o pilo ing and in o-
ducing indi idual p ojec s, such as he Siemens Indus ial
Copilo , i aces challenges in implemen ing he echnology
a scale. As men ioned p e iously, he complex sys ems and
p ocesses we e equen ly ci ed as easons o his. Wi h e-
ga d o in e nal p oduc i i y inc eases, in e iewees poin ed
ou ha a ew impac ul use cases could gene a e widesp ead
employee en husiasm o he echnology. This aligns wi h he
au ho ’s obse a ions: while some employees a e p oac i ely
in eg a ing Gene a i e AI in o hei daily wo k, he majo i y
a e no ye ac i ely using i (Appendix 6).
5.6. Addi ional Obse a ions
Fu he insigh s can be ex ac ed om he con e sa ions
and da a collec ed. A majo opic a ea e ol es a ound he
isks associa ed wi h Gene a i e AI and possible mi iga ion
s a egies. The Siemens legal and compliance depa men is-
sued guidelines a an ea ly s age ega ding he use o ex e -
nal se ices such as Cha GPT. These included ules ela ing
o sensi i e in o ma ion, illegal o une hical use, he p o ec-
ion o p ope y igh s, da a p i acy, and expo con ol. Po-
en ial disc imina ion, biases, epu a ional isks, and ques-
ions abou accoun abili y o decisions made by Gene a i e
AI. Addi ionally, some pa icipan s exp essed conce ns abou
po en ial nega i e e ec s om an icipa ed inc eases in p o-
duc i i y. While many poin o he skilled wo ke sho age
and he need o s eamline ou ine p ocesses, o he s o esee
mo e a - eaching consequences. As IP-1 pu s i :
“I like he e iciency gain o i , bu I also dislike he
e iciency gain o i (IP-1, P og am Manage )”
Va ious mi iga ion measu es we e men ioned o minimize
isks. The mos equen was “keeping a human in he loop”,
meaning ha c i ical decisions should con inue o be made
o o e seen by humans. Using Gene a i e AI as a co-pilo
A. Sake /Junio Managemen Science 10(3) (2025) 631-656650
Figu e 10: Siemens’ in e nal Gene a i e AI pla o m
Figu e 11: Siemens’ Inno a ion Rada o Gene a i e AI
( hough his e m has been adop ed by Mic oso and o h-
e s o hei own solu ions) was equen ly men ioned in his
con ex (IP-16, Head o Technology). F om he in e iewees’
pe spec i e, he use o syn he ic da a can educe he isk o
copy igh lawsui s. And while he EU A i icial In elligence
Ac is seen as a u he challenge, i also p o ides guidance
o u u e de elopmen s in he a ea o Gene a i e AI.
Finally, wo addi ional aspec s om he in e iews should
be men ioned. Fi s ly, in e iewees ag ee ha Siemens can
be seen as a pionee in implemen ing Gene a i e AI wi hin
i s indus y. This is a ibu ed o he ea ly p o ision o in-
as uc u e (such as he in e nal Gene a i e AI pla o m),
which allowed employees om all business uni s o de elop
use cases. The associa ed knowledge building is conside ed
aluable in ini ia ing discussions and de eloping he ame-
wo k o u he implemen a ions. Addi ionally, op manage-
men quickly emb aced he opic, leading o egula discus-
sions a in e nal e en s. Ex e nal p esen a ions, such as he
Indus ial Copilo a he Hanno e Messe o Roland Busch’s
CES keyno e speech, unde line he impo ance o he ech-
nology om a co po a e pe spec i e. Las ly, ano he insigh
eme ged: many in e iew pa ne s see a kind o hype a ound
Gene a i e AI. This is pa icula ly due o he low ba ie o
en y, hough implemen ing alue-adding applica ions o he
company emains challenging. As IP-5 pu s i :

A. Sake /Junio Managemen Science 10(3) (2025) 631-656 651
“I hink we all s ill ha e some way o go. On he
one hand, lea ning how i wo ks, bu also a ce ain
ole ance o us a ion. [...] ha will de ini ely
come when he ini ial hype has died down and you
ealize ha i ’s a bi mo e complica ed o gene a e
all his s u . (IP-5, IT P o essional)
6. Discussion
This case s udy explo es he alue c ea ion po en ial
o Gene a i e AI wi hin a la ge, mul ina ional co po a ion.
Spanning a one-yea obse a ion pe iod and d awing on
23 in-dep h in e iews wi h expe s and manage s ac oss
a ious business a eas, i o e s a comp ehensi e iew o
he p ocesses and challenges associa ed wi h in oducing
his ans o ma i e echnology. The ollowing sec ion will
del e in o he key indings and hei implica ions o bo h
esea che s and p ac i ione s. This analysis will add ess he
cen al esea ch ques ions and highligh he s udy’s heo e -
ical and p ac ical con ibu ions. The chap e will conclude
wi h a c i ical e lec ion, acknowledging he s udy’s limi a-
ions and ou lining po en ial a enues o u u e esea ch.
6.1. Technological Inno a ion and Usabili y
When OpenAI launched Cha GPT in No embe 2022, he
public was able o expe ience he ad ances in he ield o
a i icial in elligence o he i s ime. Based on he ou -
lined wo k by B own e al. (2020), Rad o d and Na asimhan
(2018), and Vaswani e al. (2017) and many o he s, Cha -
GPT showed wha was possible wi h he help o “a i icial
in elligence”. Some s udies poin ed o he possible signi i-
can in luence o LLMs on di e en occupa ional g oups ( o
example, Eloundou e al., 2023, p. 11) while o he s ecog-
nized he abili y o Gene a i e AI o ake on c ea i e asks
ha p e iously could only be pe o med by humans (Feue -
iegel e al., 2023, p. 1). Explo ing Resea ch Ques ion 1 on
how Gene a i e AI di e s om p e ious models in e ms o
a chi ec u e and unc ionali y, wo key indings eme ge om
a business pe spec i e. Fi s ly, he echnological inno a ion
enables companies like Siemens o explo e new applica ion
a eas and hus oppo uni ies o gene a e addi ional business
alue. This is pa icula ly impo an o exis ing p oduc s
and so wa e solu ions ha can be imp o ed wi h he help
o Gene a i e AI. Examples ha ha e been ou lined in he
case s udy a e he Siemens Indus ial Copilo in eg a ed in
Siemens’ au oma ion en i onmen o Siemens’ low-coding
pla o m Mendix ha inco po a es Gene a i e AI capabili-
ies. While hese a e cus ome - acing p oduc s, he adap -
abili y o Gene a i e AI appea s o be use ul o in e nal ap-
plica ions as well. Use cases can be ealized wi h signi i-
can ly less e o , i.e., wi h a lowe in es men olume and in
less ime compa ed o p e ious models, hus enabling mo e
business a eas o deploy AI solu ions o s eamline hei p o-
cesses. Al hough long- e m p oduc i i y imp o emen s a e
no ye demons able, he case s udy has shown ha e i-
ciency gains a e an icipa ed ac oss all business. This s a s
wi h in o ma ion e ie al sys ems which educe he ime
spen o sea ching speci ic in o ma ion and con inues wi h
assis an sys ems which ac i ely suppo he use and help
imp o e quali y in p ede ined asks. F om he company’s
poin o iew, ex e nal solu ions in pa icula can o e added
alue alongside in-house de elopmen s. Building on pa ne -
ships wi h so wa e p o ide s, such as Mic oso , companies
like Siemens can bene i di ec ly om new AI unc ionali ies.
This aligns wi h co po a e goals, as he widesp ead adop ion
o ex e nal so wa e solu ions maximizes he each o po en-
ial imp o emen s. In he case o Mic oso , Siemens was
able o e alua e he Mic oso Copilo a an ea ly s age and
make i a ailable o selec ed employees as pa o an in e -
nal pilo . This allowed he company o assess he Gene a i e
AI unc ionali ies and build knowledge wi hin he o ganiza-
ion, laying he ounda ion o u u e implemen a ion. Ad-
d essing he cen al esea ch ques ion, ano he key inding is
he no able usabili y o Gene a i e AI. I s abili y o execu e
asks based on na u al language p omp s o e s a signi ican ly
s eamlined use expe ience o in e ac ing wi h AI sys ems.
This, coupled wi h he accessibili y o pla o ms like OpenAI’s
Cha GPT (and subsequen in e nal solu ions), expands he
scope o po en ial applica ions. Fo la ge co po a ions, his
p esen s a signi ican shi , empowe ing employees wi hou
specialized AI knowledge o di ec ly u ilize he echnology.
6.2. Eme ging Pa e ns o Value C ea ion
Based on he case s udy esul s, he alue c ea ion po-
en ial o Gene a i e AI has been ou lined in Chap e 5. Use
cases we e clus e ed in o b oade schemes and key d i e s
we e iden i ied. Mo eo e , applica ion a eas we e disco e ed
which help companies like Siemens o s uc u e and p io i-
ize i s use cases. Wi h his, he case s udy suppo s closing
he gap be ween scien i ic concep s and eal-wo ld applica-
ion. I p o ides an answe o Resea ch Ques ion 2, namely
in which ways Gene a i e AI p esen s new a enues o alue
c ea ion. Based on he echnological inno a ion and usabili y
as desc ibed abo e, he clus e ing o use cases in pa icula
helps o unde s and he po en ial added alue. Con as ed
wi h he ea lie wo k on alue c ea ion mechanisms, such
as he p ocess model o ML alue c ea ion by Shollo e al.
(2022) o he ou sou ces o alue c ea ion by Bo ges e al.
(2021), simila i ies bu also di e gences become isible (see
Table 4). Mos no iceable, he use o AI o in o ma ion e-
ie al can be seen as a new a ea o alue c ea ion which was
no co e ed be o e. This can be aced back o he echno-
logical ad ances o Gene a i e AI and i s capabili y o p ocess
and p o ide answe s om la ge da a se s.
Ano he segmen which was no ouched on be o e, is
coding suppo . While i can be a gued ha his could be
seen as a subse o he clus e assis an s, i s a - eaching im-
plica ions can se e as a eason o p esen ing i sepa a ely.
Wi h hei knowledge o all majo p og amming languages,
Gene a i e AI models can help o gene a e, adjus , es , and
op imize code au oma ically, changing he way p og amme s
deal wi h asks and business p oblems. These unique capabil-
i ies o Gene a i e AI could no be obse ed by esea che s in
A. Sake /Junio Managemen Science 10(3) (2025) 631-656652
Table 4: Compa ison o case s udy esul s wi h p e ious esea ch
Case S udy Clus e s Shollo e al. (2022): Shi ing ML
alue c ea ion mechanisms
Bo ges e al. (2021): The s a egic
use o a i icial in elligence in he
digi al e a
Assis an s - In o ma ion Re ie al N/A N/A
Da a Syn hesis Knowledge C ea ion Decision Suppo
Assis an s - Coaches & Helpe s Task Augmen a ion Cus ome and Employee
Engagemen
Au onomous Assis an s Au onomous Agen s Au oma ion
Con en C ea ion N/ANew P oduc s and Se ices
Coding Suppo N/A N/A
2021 and 2022 espec i ely and he e o e ep esen a alu-
able supplemen o exis ing esea ch.
6.3. S a egic and Ope a ional Deploymen
Besides he ou lined a enues o alue c ea ion, suppo -
ing s a egies and conside a ions ha e been iden i ied. To
add ess Resea ch Ques ion 3, which explo es ways o ha -
ness Gene a i e AI’s ull alue c ea ion po en ial, he ol-
lowing sec ion discusses hese indings in de ail. Based on
he insigh s gained ega ding explo a ion and scaling, a new
amewo k is p oposed (Figu e 12). I illus a es he connec-
ion be ween hese phases, along wi h he necessa y asks
o e ec i ely exploi Gene a i e AI’s alue c ea ion po en-
ial. The hou glass shape symbolizes he c i ical in e ace be-
ween he wo phases. I emphasizes he need o a business
impac assessmen o success ully scale use cases ac oss he
o ganiza ion. This assessmen equi es weighing he cos s
and bene i s o each use case indi idually. Since companies
seek he bes possible use o hei limi ed esou ces, no all
ideas om he explo a ion phase will necessa ily be imple-
men ed. In o de o de e mine he added alue, subsequen
ques ions should be add essed, namely who coo dina es he
implemen a ion, how much de elopmen e o is equi ed,
which accompanying change managemen measu es a e nec-
essa y and who is esponsible o he ca e, main enance, and
cos s o he new solu ion.
The business impac assessmen goes hand-in-hand wi h
s a egic conside a ions. Fo la ge companies, i is essen ial
o iew AI ini ia i es wi hin a b oade con ex and align hem
wi h co po a e s a egy (Ki sios & Kama io ou, 2021, p. 6).
This is especially impo an since he added alue o Gen-
e a i e AI use cases migh no be immedia ely measu able.
Companies should consciously de ine co e a eas o s ee e-
sou ces e ec i ely. In his con ex , Gene a i e AI p esen s
a new challenge o manage s and s a egis s. While pas AI
ini ia i es we e mo e manageable and clea ly de inable, ech-
nological inno a ion now o e s a my iad o po en ial alue
c ea ion oppo uni ies. E alua ing and il e ing hese, along
wi h de ining a clea s a egy, becomes an essen ial ask o
success ully u ilizing Gene a i e AI. The amewo k sugges s
main asks o each phase. In he explo a ion phase, hese in-
clude c ea ing awa eness, sha ing knowledge, building com-
muni ies, and p o iding a common echnology s ack. To sup-
po he business impac assessmen and s a egic alignmen ,
key asks in ol e c ea ing anspa ency, minimizing mac o-
isks, and alloca ing esou ces o iden i ied use cases. Finally,
o success ully scale solu ions, companies should ocus on
ha monizing hei da a landscape, adjus ing p ocesses, and
o e ing aining and guidance o employees.
6.4. Theo e ical Con ibu ions
The esul s o his case s udy con ibu e o exis ing e-
sea ch in he ields o alue c ea ion, AI adop ion, and ca-
pabili ies. This hesis o e s a aluable con ibu ion by an-
alyzing p oposed amewo ks by Bo ges e al. (2021) and
Shollo e al. (2022) wi hin he speci ic con ex o Gene a i e
AI. The analysis iden i ies a eas whe e hese models could
be u he expanded. Addi ionally, he amewo k by U en
and Edwa ds (2023), wi h i s ou lenses People, P ocesses,
Technology, and Da a p o ed highly applicable. This case
s udy expands upon hei model by de ailing speci ic con-
side a ions wi hin each ca ego y ele an o Gene a i e AI
implemen a ion. Fu he mo e, new dimensions eme ged o
analysis, including he impo ance o iming, o ganiza ional
se up, s a egy, business, alue and communica ion. The case
s udy aligns wi h he indings o Cane and Bha i (2020),
highligh ing he mul iple pe spec i es om which AI can be
iewed. Fo example, he esul s sugges ha while Gene a-
i e AI has o e come ce ain limi a ions like da a labeling o
gene alizabili y o lea ning, o he s, such as biases and non-
explainabili y, emain a challenge. Ano he in e es ing a -
enue o u u e esea ch could be o in es iga e u he ap-
plica ion a eas o Gene a i e AI, as i s p ima y use is cu en ly
in augmen a ion asks, wi h au onomous sys ems seeming
less p e alen . In conclusion, his hesis adds o he exis -
ing body o knowledge by e ining and ex ending es ablished
models in he con ex o Gene a i e AI. This de ailed explo-
a ion o e s aluable insigh s ha can ad ance heo e ical
unde s anding and guide he de elopmen o mo e comp e-
hensi e amewo ks o AI adop ion.
A. Sake /Junio Managemen Science 10(3) (2025) 631-656 653
Figu e 12: P oposed amewo k o explo a ion and scaling o Gene a i e AI
6.5. P ac ical Implica ions
Me ging he esul s om he case s udy and heo e ical
concep s on AI alue c ea ion and capabili ies, p ac ical im-
plica ions eme ge. In he ollowing sec ion, hese should be
ou lined in mo e de ail, p o iding companies wi h a clea e
unde s anding o Gene a i e AI and he equi ed ac ions o
c ea e added alue.
Fi s ly, companies should c ea e oom o inno a ion.
Gene a i e AI is unique in ha i s co e unc ionali ies (i.e.,
summa iza ion, ansla ion, and con en c ea ion) can be
used by almos any employee. This empowe s mo e peo-
ple, e en hose wi hou a backg ound in IT o da a science,
o de elop business ideas wi hin hei espec i e domains.
Companies should emb ace his oppo uni y and ac i ely
suppo he idea ion and explo a ion o new Gene a i e AI
solu ions.
An impo an p e equisi e o his is he p o ision o he
equi ed in as uc u e. Wi h ega d o Gene a i e AI, his
comp ises company-in e nal pla o ms and so wa e pack-
ages ha employees can use sa ely and secu ely, wi hou he
isk o exposing sensi i e company in o ma ion ex e nally. A
he same ime, i is impo an o highligh he limi a ions o
he echnology o a oid exagge a ed expec a ions.
Secondly, companies should igh o anspa ency.
La ge co po a ions wi h hei decen alized s uc u es end
o de elop simila solu ions in mul iple business a eas. Wi h
Gene a i e AI, his challenge becomes e en mo e appa en
as use cases can be de eloped wi hin days o weeks. While
explo a ion should be encou aged, c ea ing anspa ency
o e all ini ia i es is key o companies; o he wise, esou ces
isk being was ed on edundan solu ions. Employees and
depa men s should he e o e be encou aged o sha e hei
knowledge and solu ions wi h a b oade audience. This
should no be aken ligh ly: ac i e managemen o po en-
ial use cases and clea guidance o employees a e c ucial o
educe isola ed solu ions, ic ional losses, and un ul illed ex-
pec a ions. I possible, op managemen should en o ce he
needed s uc u es o p o ide a company-wide amewo k.
Thi dly, a clea AI s a egy is needed o s ee he com-
pany’s esou ces in he igh di ec ion. As ou lined be o e,
Gene a i e AI o e s mul i old oppo uni ies o in eg a ion
in o p ocesses and enhancemen o exis ing p oduc s. How-
e e , companies should ca e ully e alua e each use case and
i s con ibu ion o he o e all company s a egy. Rega ding
cus ome - acing solu ions, he inhe en echnological isks
need o be analyzed in de ail. Fo in e nal p ocesses, he
added alue o Gene a i e AI mus be weighed agains in-
c eased complexi y and consequen ial cos s. To guide eams
e ec i ely, companies should p o ide guidance o hei mid-
dle managemen , enabling hem o e alua e new ideas wi hin
a b oade con ex .
Finally, companies should ocus on b idging he gap be-
ween Gene a i e AI knowledge and ope a ional business
uni s. Since Gene a i e AI use cases can inc easingly be
d i en di ec ly by hese uni s wi hou ex ensi e IT o da a
science in ol emen , i is wise o empowe hem. Companies
should conside s eng hening exis ing depa men s (such as
inno a ion, ope a ional, o business excellence) o es ablish-
ing dedica ed Gene a i e AI esou ces ac oss he o ganiza-
ion o maximize he echnology’s po en ial.
6.6. S udy Limi a ions
Al hough p o iding aluable insigh s in o he alue c e-
a ion oppo uni ies o Gene a i e AI, his case s udy is no
wi hou i s limi a ions. The i s limi a ion lies in he de-
sign o he s udy. As a single case s udy, i o e s unique and
comp ehensi e insigh s in o he objec o in es iga ion; how-
e e , i s indings canno be b oadly gene alized. The second
limi a ion is oo ed in he au ho ’s ac i e in ol emen in he
A. Sake /Junio Managemen Science 10(3) (2025) 631-656654
explo a ion o Gene a i e AI in he company. While his al-
lowed access o in o ma ion ha would no mally no ha e
been accessible ( o example, he pa icipa ion in ex e nal
and in e nal con e ences o he egula exchange wi h da a
scien is s on echnical aspec s o Gene a i e AI), biases and
pe sonal pe cep ions could impac he inal esul s, despi e
he au ho ’s bes e o s o a oid unin ended in e e ences.
The hi d limi a ion conce ns he da a used o he case s udy.
Due o he no el y o Gene a i e AI, exis ing concep s o AI
alue c ea ion, adop ion, and capabili ies we e used, con-
cep s ha migh no ully e lec he la es echnological ad-
ances. The selec ion o concep s was based on a s uc u ed
app oach; howe e , i canno be uled ou ha o he uncon-
side ed scien i ic wo k migh ha e p o ided addi ional alue.
Rega ding he da a collec ed in he company, i should be
no ed ha he selec ion o in e iew pa ne s migh ha e in-
luenced he esul s. An a emp was made o co e as many
a eas and unc ions as possible, bu a selec ion was necessa y
due o he subs an ial numbe o po en ial con ac s.
In conclusion, his wo k ep esen s a snapsho in ime.
The ield o Gene a i e AI is e ol ing apidly. While he ocus
a he beginning o he case s udy was on ex ual unde s and-
ing, new models now gene a e bo h images and ideos. De-
spi e he desc ibed limi a ions, his wo k aims o con ibu e
o ongoing esea ch on he alue c ea ion po en ial o Gen-
e a i e AI.
6.7. Fu u e Resea ch Oppo uni ies
Building on he ou lined limi a ions, u u e esea ch
could e i y and expand he indings o his case s udy.
Fi s ly, ano he case s udy in a simila se up could e i y
and s eng hen he esul s, po en ially adding o he p o-
posed amewo k o explo a ion and scaling. Addi ionally,
quan i a i e da a would help deepen he unde s anding o
a company’s equi emen s and an icipa ed challenges when
in oducing Gene a i e AI. Su eys among manage s and em-
ployees could p o ide aluable insigh s in o c i ical success
ac o s. Mo eo e , mo e scien i ic wo k is needed o explain
he unique capabili ies o Gene a i e AI compa ed o ea lie
machine lea ning models, and o in eg a e hese indings
in o exis ing concep ual amewo ks. Finally, u u e esea ch
should in es iga e he impac and alue c ea ion po en ial o
he la es echnological ad ances. Wi h new models such as
Google’s Gemini, p o iding a con ex window o up o one
million okens, ecen ly de eloped solu ions o in o ma ion
e ie al could become obsole e. This and o he ad ances
such as OpenAI’s ex - o ideo model So a con inually open
new a enues o esea ch in his ield.
7. Conclusion
In his s udy, he alue c ea ion oppo uni ies o Gene a-
i e AI we e explo ed. Based on a single case s udy design,
Siemens was chosen as he objec o in es iga ion as i o -
e ed a unique oppo uni y o s udy he phenomenon in
dep h. O e a pe iod o one yea , a conside able amoun
o ime was spen obse ing and pa icipa ing in nume ous
e en s o achie e a be e unde s anding o he echnol-
ogy and Siemens’ app oach o in oducing and exploi ing
i s alue c ea ion po en ial. This was supplemen ed by 23
comp ehensi e in e iews wi h key s akeholde s, p o iding
a comple e iew o use cases, ac i i ies, pe spec i es, and
conside a ions.
The indings sugges ha Gene a i e AI p esen s new
alue c ea ion oppo uni ies, d i en by echnological ad-
ances and usabili y ha b ing AI close o domain expe s.
Use cases ha e been iden i ied ac oss a ious a eas and clus-
e ed in o ou main g oups. Sma assis an s wi h a ious
le els o complexi y o m he majo i y o obse ed pilo s,
along wi h ligh house p ojec s in di e en business a eas.
While many applica ions o e po en ial added alue, he
s udy acknowledges ha a conside able numbe o p ojec s
a e s ill in hei ea ly s ages.
Con ibu ing o he ongoing discou se in AI esea ch on
alue c ea ion and equi ed capabili ies, di e en ame-
wo ks we e applied o explain he case s udy indings. Fo
example, add essing he capabili ies o he success ul im-
plemen a ion o AI, he ou lined dimensions om p e ious
esea ch eappea ed in he da a. O he concep s could no
be ully con i med by he indings, opening oom o u -
he esea ch. Fo example, exis ing wo k on alue c ea ion
mechanisms was no able o explain all alue c ea ion clus-
e s unco e ed in he case s udy.
A key con ibu ion o he hesis can be seen in i s long-
e m obse a ion o a company in he ea ly adop ion phase
o a new echnology. S a ing in he ea ly disco e y s age
only a ew mon hs a e he elease o Gene a i e AI o he
public, Siemens’ s eps o in oduce, manage, and p o i om
he echnology could be obse ed o e a yea . Building on
he ga he ed knowledge, a new amewo k is p oposed o
explain he ac ions and s a egies e ealed in he case s udy.
I in oduces an explo a ion phase as an impo an ime pe-
iod in he adop ion o Gene a i e AI. Mo eo e , i iden i ies
essen ial asks o capi alize on he oppo uni ies p esen ed in
his phase and ou lines he necessa y s eps o a success ul
ansi ion om explo a ion o scaling.
To conclude, he hesis p o ides p ac ical implica ions
and highligh s po en ial a eas o u u e esea ch. Acknowl-
edging i s limi a ions, he hesis o e s unique insigh s in o a
company’s app oach o Gene a i e AI and con ibu es o he
ongoing discou se on he po en ial alue c ea ion oppo uni-
ies o he echnology.
Re e ences
Akki aju, R., Sinha, V., Xu, A., Mahmud, J., Gundecha, P., Liu, Z., Liu, X., &
Schumache , J. (2018). Cha ac e izing machine lea ning p ocess:
A ma u i y amewo k. h ps://doi.o g/10.48550/a Xi .1811.0
4871
Bahdanau, D., Cho, K., & Bengio, Y. (2014). Neu al Machine T ansla ion by
Join ly Lea ning o Align and T ansla e. h ps://doi.o g/10.485
50/a Xi .1409.0473
Bas ian, M. (2023). GPT-4 has mo e han a illion pa ame e s. Re ie ed
Ma ch 25, 2023, om h ps:// he-decode .com/gp -4-has-a- il
lion-pa ame e s/