B üggemann, Immo; Buse, S ephan; Pa uschke, Janin; Villa eal, Nohemi
Wo king Pape
Compe i i e ad an ages h ough gene a i e AI: Expe ise
as he key o implemen a ion
Wo king Pape , No. 118
P o ided in Coope a ion wi h:
Hambu g Uni e si y o Technology (TUHH), Ins i u e o Technology and Inno a ion Managemen
Sugges ed Ci a ion: B üggemann, Immo; Buse, S ephan; Pa uschke, Janin; Villa eal, Nohemi (2025) :
Compe i i e ad an ages h ough gene a i e AI: Expe ise as he key o implemen a ion, Wo king
Pape , No. 118, Hambu g Uni e si y o Technology (TUHH), Ins i u e o Technology and Inno a ion
Managemen (TIM), Hambu g,
h ps://doi.o g/10.15480/882.14322
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/313610
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/
Wo king Pape
Compe i i e ad an ages h ough gene a i e AI:
expe ise as he key o implemen a ion
Immo B üggemann, D . S ephan Buse, Janin Pa uschke
and Nohemi Villa eal
Janua y 2025
Wo king Pape 118
Hambu g Uni e si y o Technology (TUHH)
Ins i u e o Technology and
Inno a ion Managemen
Am Schwa zenbe g-Campus 4
D-21073 Hambu g, Ge many
[email p o ec ed]
www. uhh.de/ im
1
This a icle was o iginally published in 2025 by he Ins i u e o Technology and Inno a ion Managemen
in coope a ion wi h CREATUM GmbH Hambu g as pa o he OpenInnoT ain P ojec
1
Copy igh in o ma ion: This wo k including all i s pa s is p o ec ed by copy igh . The wo k is licensed unde he C ea i e
Commons A ibu ion 4.0 In e na ional License (CC BY 4.0, h ps://c ea i ecommons.o g/licenses/by/4.0/legalcode.de).
Excluded om he abo e license a e pa s, images and o he hi d-pa y ma e ial, i ma ked o he wise.
DOI: h ps://doi.o g/10.15480/882.14322
ORCID:
S ephan Buse: h ps://o cid.o g/0000-0002-4308-1707
1
This esea ch has ecei ed unding om he Eu opean Union’s Ho izon 2020
esea ch and inno a ion p og amme, wi hin he OpenInnoT ain p ojec
unde he Ma ie Skłowdowska-Cu ie g an ag eemen no. 823971. The
con en o his publica ion does no e lec he o icial opinion o he Eu opean
Union. Responsibili y o he in o ma ion and iews exp essed in he
publica ion lies en i ely wi h he au ho (s).
1
Compe i i e ad an ages h ough gene a i e AI:
expe ise as he key o implemen a ion
Abs ac
The P oblem
In o de o unde s and gene a i e AI and i s mechanisms o ac ion, playe s need a high le el o
abs ac ion (Sai a and Zucke , 2013). Howe e , many companies lack he necessa y expe ise o
he echnology. In addi ion o cul u al ac o s, his lack o expe ise is he main ba ie o a esul s-
o ien ed implemen a ion o he echnology (Campos Zabala, 2023).
The Solu ion
To iden i y aluable ields o applica ion, companies need o unde s and he expe ise in he
decision-making p ocesses o hei own ac i i ies. The a icle shows conc e e examples and s eps
ha companies can use o es and apply new echnological possibili ies o hei own use cases.
Immo B üggemann
Immo.b ueggemann@c ea um.io
CREATUM GmbH
Am Sand o kai 32, 20457 Hambu g, Deu schland
D . S ephan Buse
[email p o ec ed]
Ins i u e o Technology and Inno a ion Managemen , Hambu g Uni e s iy o Technology (TUHH)
Am Schwa zenbe g-Campus 1, 21073 Hambu g, Deu schland
Janin Pa uschke
Janin.pa uschke@c ea um.io
CREATUM GmbH
Am Sand o kai 32, 20457 Hambu g, Deu schland
Nohemi Villa eal
Nohemi. illa eal@c ea um.io
CREATUM GmbH
Am Sand o kai 32, 20457 Hambu g, Deu schland
2
Con en s
1
In oduc ion 3
2
New echnological possibili ies
3
3
Mul i-agen AI and g aph neu al ne wo ks: e olu ionizing he way we
deal wi h complexi y and knowledge
4
4
The impac o new echnological oppo uni ies on business ac i i ies
5
5
Applica ion examples o gene a i e AI
7
5.1
Use case I: Moni o ing and assessmen o machine s a es ..........................................7
5.2
Use case II: Mul i-agen AI in ma ke ing .....................................................................8
6
Iden i ica ion o a eas o he implemen a ion o mul i-agen AI in he
company
9
7
Conclusion
10
Bibliog aphy 11
Wo king Pape No. 117
B
ü
ggemann, Buse, Pa uschke and Villa eal
3
1
In oduc ion
Many companies a e discussing he po en ial applica ions o gene a i e a i icial in elligence (AI).
Howe e , in o de o unde s and gene a i e AI and i s mechanisms o ac ion, he ac o s in ol ed
need o ha e he app op ia e echnological expe ise. Howe e , many companies o all sizes lack
p ecisely his (Campos Zabala, 2023). This a icle p o ides an o e iew o selec ed echnological
possibili ies o gene a i e AI, which help companies o ca y ou alue c ea ion ac i i ies mo e
e icien ly and e ec i ely.
Howe e , echnological expe ise alone is no enough o achie e compe i i e ad an ages wi h
gene a i e AI. In o de o iden i y aluable ields o applica ion, companies need o unde s and
he expe ise in he decision-making p ocesses o hei own ac i i ies. In addi ion o he new
echnological possibili ies, he a icle shows conc e e examples and s eps ha companies can use
o es and apply hese o hei own use cases.
2
New echnological possibili ies
Technological ad ances in he a eas o compu ing powe (Moo e, 1965), ansmission speed
(Bu e ) and memo y (K yde , 2005) ollow exponen ial g ow h cu es. The ac ha echnological
possibili ies a e con inuously and signi ican ly imp o ing on his basis is he e o e no hing new.
The p oblem is ha humans a e no able o in ui i ely unde s and exponen ial g ow h. A ac ha
is imp essi ely desc ibed by he ice g ain legend su ounding he c ea ion o he chessboa d in
India. The impac o gene a i e AI on he consciousness and s a egic conside a ions o decision-
make s in all indus ies was co espondingly g ea .
One obse a ion is ha he e olu iona y na u e o he publica ion o new echnological
capabili ies is o en ollowed by a mo e e olu iona y phase o op imiza ion o his echnology.
Ag awal, Gans and Gold a b desc ibe a di ec consequence o his ela i ely silen and con inuous
imp o emen in hei book “P edic ion Machines” as ollows: “This is simple economics: when he
cos o some hing alls, we do mo e o i . [...] Mo e signi ican ly, because i [a i icial in elligence]
is becoming cheape i is being used o p oblems ha we e no adi ionally p edic ion p oblems.”
Technological p og ess is he e o e no only making he echnology mo e capable, bu also
conside ably cheape . The esul is ha new ields o applica ion a e being opened up in which he
use o a i icial in elligence was p e iously oo complex o simply oo esou ce-in ensi e. (Ag awal,
Gans and Gold a b, 2018)
Companies a e asking hemsel es how hey should deal wi h gene a i e AI, i.e. how hey should
use i .
Dealing wi h gene a i e AI
Acco ding o McKinsey (Lama e, 2024), companies can choose and combine h ee basic
a che ypes (Make , Take and Shape ) o in eg a e gene a i e AI in o hei business models:
I.
Make :
Make s a e in es ing hea ily in he de elopmen o hei own AI echnologies and
pla o ms. A p ocess model ha is likely o exceed he esou ce a ailabili y o mos
companies many imes o e and is he e o e p ima ily pu sued by la ge echnology
companies such as OpenAI o Google.
II.
Take :
Take s use and in eg a e exis ing, widely a ailable AI echnologies in o hei
business p ocesses. While his o e s a mo e cos -e ec i e way o implemen AI, i a ely
enables sus ainable medium- e m compe i i e ad an ages o be achie ed as he same
echnologies a e also a ailable o compe i o s.
Wo king Pape No. 117
B
ü
ggemann, Buse, Pa uschke and Villa eal
4
III.
Shape :
Shape s use basic models and modi y hem wi h hei own company's da a and
speci ica ions in ields o applica ion ha a ec he company's co e p ocesses. This
app oach o e s he g ea es po en ial o companies o achie e signi ican compe i i e
ad an ages.
Be o e he ques ion o how companies can use gene a i e AI in selec ed alue c ea ion ac i i ies
is answe ed in chap e s 5 and 6, he echnological ounda ions a e b ie ly ou lined in he ollowing
chap e .
3
Mul i-agen AI and g aph neu al ne wo ks: e olu ionizing he
way we deal wi h complexi y and knowledge
The au osapien model o AI applica ions
A common mis ake when o mula ing he goal o an AI p ojec is ha AI solu ions a e designed
wi h he aim o p o iding a inal “ igh ” solu ion and au oma ing decision-making. The
assump ion ha gene a i e AI comple ely elie es decision-make s o hei wo k and
independen ly p esen s decisions ha a e undoub edly co ec is de ini ely w ong. AI is an
auxilia y ool ha can suppo manage s in he decision-making p ocess.
The concep o “au osapien ” AI, as desc ibed in he Ha a d Business Re iew a icle “Leading
in a Wo ld Whe e AI Wields Powe o I s Own”, b ings wi h i a di e en pe spec i e on he
o mula ion o goals o AI sys ems. Au osapien sys ems a e designed o lea n au onomously,
con inuously imp o e and in e ac wi h human ac o s. (Heimans and Timms,2024) This can ake
place a di e en le els wi h a ying deg ees o complexi y.
Fou le els o AI models
In o de o be e unde s and he ans o ma ion o AI sys ems, i is necessa y o look a he
de elopmen om simple models o complex mul i-agen sys ems. Fou le els can be
dis inguished, which co ela e wi h inc easing ask complexi y. (Guo e al. 2024; Pa hasa a hy
e al. 2024):
I. Simple models
These a e basic language models (LLMs) ha p ocess and gene a e na u al language. They
a e sui able o gene al in o ma ion que ies and simple decision-making p ocesses. LLMs
can analyze la ge amoun s o ex and p o ide simple answe s. Howe e , hei capabili ies
a e limi ed o p ocessing s a ic in o ma ion.
II. Speci ically ained models
These models a e ailo ed o speci ic asks o domains and p o ide mo e de ailed and
con ex ual insigh s based on indus y o company-speci ic da a. The ocus is on op imized
ou pu ha goes beyond gene ic answe s and includes speci ic expe ise.
III. AI agen s
An AI agen can pe o m complex, specialized asks au onomously and in e ac wi h hi d-
pa y sys ems ia in e aces. I ac s as an ad anced decision-making assis an o a speci ic
a ea o applica ion.
IV. Mul i-agen AI
These sys ems consis o se e al specialized agen s ha wo k oge he o pe o m highly
complex asks wi h a high deg ee o eliabili y. Each agen con ibu es i s own pa ial
expe ise, esul ing in comp ehensi e and coo dina ed decision-making.
Each agen can be based on di e en echnological ounda ions. G aph neu al ne wo ks in
Wo king Pape No. 117
B
ü
ggemann, Buse, Pa uschke and Villa eal
5
conjunc ion wi h g aph da abases, which a e explained in he necessa y dep h below, a e
pa icula ly ele an o he cen al idea o his a icle, which is o ans e company expe ise in o
AI models.
G aph neu al ne wo ks and g aph da abases
G aph neu al ne wo ks (GNN) a e a o m o deep lea ning speci ically designed o p ocess complex
da a s uc u es in he o m o g aphs. A g aph consis s o nodes (en i ies) and edges ( ela ionships)
ha ep esen connec ions be ween he nodes. GNNs make i possible o unde s and and p ocess
bo h indi idual nodes and he ela ionships be ween hem, which dis inguishes hem om
con en ional neu al ne wo ks. The unde lying g aph da a can come om a a ie y o sou ces,
including social ne wo ks, molecules, o e en g aph da abases speci ically designed o s o e and
que y g aph s uc u es. (Khemani e al., 2024)
A speci ic o m o g aph da abase is knowledge g aphs, which accumula e domain-speci ic
knowledge and make i explici ly accessible o AI applica ions. (Peng e al., 2023) By in eg a ing
knowledge g aphs, especially deep in o ma ion, AI models can be imp o ed in hei pe o mance.
(Elnaga and Weis o e , 2019) Chap e 5 g aphically illus a es a speci ic use case.
How he inno a i e echnological possibili ies ou lined in his chap e can be used o suppo
he execu ion o alue c ea ion ac i i ies is desc ibed below.
4
The impac o new echnological oppo uni ies on business
ac i i ies
Businesses a e d i en by ac i i ies ha a e ca ied ou wi hin co e, suppo and managemen
p ocesses. Acco ding o Ag awal, Gans and Gold a b (2018), hese ac i i ies ollow a gene ic
a chi ec u e, which is illus a ed in Figu e 1:
Figu e 1: Ana omy o an ac i i y acco ding o Ag awal, Gans, and Gold a b (2018)
A he cen e o he a chi ec u al model is he p edic ion elemen , in which he si ua ion is
assessed and possible ou comes a e p edic ed. I is he basis o human judgmen and subsequen
ac ion. The p edic ion elemen can be ca ied ou by bo h humans and machines, explici ly by
p edic i e AI models. The book published in 2018 desc ibes cases as sui able o AI suppo i hey
in ol e la ge amoun s o his o ical da a, epea able, consis en si ua ions and clea , measu able
Wo king Pape No. 117
B
ü
ggemann, Buse, Pa uschke and Villa eal
6
goals (Ag awal, Gans and Gold a b, 2018).
The a ailabili y o gene a i e AI in he o m o la ge language models, especially in conjunc ion
wi h g aph da abases, expands he ypes o possible use cases in which he p edic ion module can
be execu ed by machine. Inc easingly complex, knowledge-based ques ions can be p ocessed
explici ly. (Elnaga and Weis o e , 2019)
A dis inc ion is i s made be ween expe ience and expe ise based on Malho a and Baze man
(2008). Expe ience desc ibes he equency wi h which an ac i i y is ca ied ou and he e o e has
a di ec e ec on e iciency, bu only an indi ec e ec on he quali y o he ac i i y. Expe ise, on
he o he hand, comp ises he me hodological componen o a decision, speci ically he explici
knowledge in he assessmen o a si ua ion. Expe ise has a co esponding e ec on he quali y o
he esul and he e ec i eness o he ac i i y.
This dis inc ion be ween expe ience and expe ise is pa icula ly ele an when i comes o
he expansion o machine-aided decision-making. In his con ex , wo ypes o cases can be
iden i ied, which a e explained in mo e de ail in he ollowing chap e using speci ic applica ion
examples.
I. Cases in which he expe ise is no explici ly a ailable in he company
This expe ise can be de eloped using gene a i e AI on he basis o uns uc u ed in e nal
and ex e nal da a. To do his, he uns uc u ed da a is con e ed in o a knowledge g aph
using a language model, which is hen used in he p edic i e model and imp o es he esul
(Elnaga and Weis o e , 2019). Speci ic sou ces can be p ocesses om a cus ome suppo
icke sys em, machine manuals o email co espondence.
II. Cases in which he e a e no measu able esul s o he ac i i ies
In ac i i ies ha do no ocus on classic p edic i e issues, he in eg a ion o expe ise in
language models can be used o suppo decision-making. Speci ically, o example,
expe ise on s a egic managemen me hods can be in eg a ed in o au osapien AI
assis an s ha suppo decision-make s in s a egy de elopmen and implemen a ion.
(Csasza e al., 2024)
The consequence o companies
Companies mus unde s and hese new oppo uni ies and e lec hem on hei ac i i ies in o de
o de elop compe i i e ad an ages and new o ms o alue c ea ion by shaping he echnology (see
chap e 2). Two key ques ions a ise explici ly:
I.
How g ea is he po en ial o expe ise: Wha expe knowledge is a ailable in he company
in an uns uc u ed o incomple e o m and can be made accessible and in eg a ed in a
scalable way using o ms o AI?
II.
How can he new echnological possibili ies be used o con e exis ing expe ise in o new
o ms o alue c ea ion?
The ollowing chap e desc ibes wo examples o use cases ha illus a e he ad an ages o he
inno a i e p ocess app oach.
