F uhwi h, Michael; Pamme -Schindle , Vik o ia; Thalmann, S e an
A icle
Knowledge leaks in da a-d i en business models?
Explo ing di e en ypes o knowledge isks and
p o ec ion measu es
Schmalenbach Jou nal o Business Resea ch (SBUR)
P o ided in Coope a ion wi h:
Schmalenbach-Gesellscha ü Be iebswi scha e.V.
Sugges ed Ci a ion: F uhwi h, Michael; Pamme -Schindle , Vik o ia; Thalmann, S e an (2024) :
Knowledge leaks in da a-d i en business models? Explo ing di e en ypes o knowledge isks
and p o ec ion measu es, Schmalenbach Jou nal o Business Resea ch (SBUR), ISSN 2366-6153,
Sp inge , Heidelbe g, Vol. 76, Iss. 3, pp. 357-396,
h ps://doi.o g/10.1007/s41471-024-00189-z
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/312594
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/
ORIGINAL ARTICLE
h ps://doi.o g/10.1007/s41471-024-00189-z
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
Knowledge Leaks in Da a-D i en Business Models?
Explo ing Di e en Types o Knowledge Risks and
P o ec ion Measu es
Michael F uhwi h · Vik o ia Pamme -Schindle ·
S e an Thalmann
Recei ed: 16 Feb ua y 2022 / Accep ed: 17 June 2024 / Published online: 30 July 2024
© The Au ho (s) 2024
Abs ac Da a-d i en business models imply he in e -o ganisa ional exchange o
da a o simila alue objec s. Da a science me hods enable o ganisa ions o disco e
pa e ns and e en ually knowledge om da a. Fu he , by aining machine lea ning
models, knowledge is ma e ialised in hose models. Thus, o ganisa ions migh isk
he exposu e o compe i i e knowledge by sha ing da a- ela ed alue objec s, such
as da a, models o p edic ions. Al hough knowledge isks ha e been s udied in a-
di ional business models, li le esea ch has been conduc ed in he di ec ion o da a-
d i en business models. In his explo a i e quali a i e s udy, we conduc ed 28 expe
in e iews in h ee ounds ( wo explo a o y and one e alua o y) and iden i ied i e
ypes o isks along he h ee basic ypes o alue objec s: da a, models and p edic-
ions. These isks depend on he con ex , i.e., when compe i i e knowledge could be
disco e ed om sha ed alue objec s. We ound ha hose isks can be mi iga ed by
echnology, con ac ual egula ions, us ed ela ionships, and adjus ing he business
model design. In his s udy, we show ha he isk o knowledge leakage is a ele-
an isk ac o in da a-d i en business models. O e all, knowledge isks should be
Michael F uhwi h · Vik o ia Pamme -Schindle
Know-Cen e GmbH, Sandgasse 34/II, 8010 G az, Aus ia
Ins i u e o In e ac i e Sys ems and Da a Science, Facul y o Compu e Science and Biomedical
Enginee ing, G az Uni e si y o Technology, Sandgasse 36/III, 8010 G az, Aus ia
E-Mail: michael. uhwi h@s uden . ug az.a ; ik o ia.pamm[email p o ec ed]
P esen Add ess:
Michael F uhwi h
Silicon Aus ia Labs GmbH, Sandgasse 34/IV, 8010 G az, Aus ia
E-Mail: michael. uhwi h@silicon-aus ia.com
S e an Thalmann
Business Analy ics and Da a Science-Cen e (BANDAS-Cen e ), School o Business, Economics and
Social Sciences, Uni e si y o G az, Uni e si ä ss aße 15 Building F/III, 8010 G az, Aus ia
E-Mail: s e an. halmann@uni-g az.a
K
358 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
conside ed al eady du ing business model design, and hei managemen equi es
an in e disciplina y app oach ia a balanced assessmen . The le el o knowledge
p o ec ion om a echnology pe spec i e highly depends on compu e science inno-
a ions and hus is a mo ing a ge . As an ou look, we sugges ha knowledge isk
will become e en mo e ele an wi h he ex ensi e usage o machine lea ning and
a i icial in elligence in da a-d i en business models.
Keywo ds Business model inno a ion · Da a analy ics · Da a-d i en business
models · Knowledge isks · Risk managemen · Value objec s
1 In oduc ion
De elopmen s in big da a echnologies and a i icial in elligence (AI), as well as
he a ailabili y o la ge da a se s, hold he oppo uni y o he de elopmen o
new p oduc s, se ices, and business models (Gün he e al. 2017; Woe ne and
Wixom 2015), so-called da a-d i en business models (DDBMs) (Ha mann e al.
2016; Wiene e al. 2020). Such business models o en imply he exchange o da a
and simila da a- ela ed alue objec s. Fu he , in such business models, sensi i e
in o ma ion and compe i i e knowledge a e ma e ialised in da a o models. A he
same ime, da a science me hods allow ex ac ing in o ma ion o knowledge om
ine-g anula , he e ogeneous da a, leading o po en ial isks when da a is sha ed.
Whe eas be o e, knowledge needed o be ep esen ed in a much mo e explici man-
ne . Thus, i is challenging o o ganisa ions o e alua e wha knowledge could be
disco e ed om sha ed da a se s (Zei inge and Thalmann 2020). Fo ins ance, sim-
ply “looking a he da a” (i.e., a he heade s o a da abase o desc ip i e s a is ics
o e a single da ase ) is no enough o assess which knowledge could be d awn om
he da a. Sha ing da a implies he isk—which we e e o as knowledge isks— ha
compe i i e knowledge could leak and spill o e o o he o ganisa ions.
Fo example, we ound such isks in a case s udy wi h an indus ial company
(F uhwi h e al. 2019). In his case, no el knowledge o a eal-wo ld physical
phenomenon (i.e., p edic ing he esidual li e ime o a physical componen ) was
gene a ed om da a and ma e ialised in a model. Building new DDBMs a ound his
model (i.e., o e ing he model) could imply he isk o leaking co e knowledge, as
one wo kshop wi h manage s o his company showed. Fu he , he willingness o
sha e da a is o en a p e equisi e o a DDBM, bu po en ial knowledge leakages neg-
a i ely in luence his willingness. Thus, DDBMs equi e balancing be ween sha ing
and p o ec ing knowledge. Fu he , IP migh be sha ed o could be e-enginee ed
when o e ing machine lea ning (ML) models h ough an API (Applica ion P o-
g amming In e ace) (Hanzlik e al. 2021).
Knowledge isks ha e been s udied in s a egic alliances (He nandez e al. 2015;
Jiang e al. 2016;Kalee al.2000) and adi ional business models (Al-Aali and
Teece 2013). Howe e , as shown abo e, DDBMs imply new ypes o isks, pa ic-
ula ly ha knowledge may spill o e o compe i o s ia sha ing da a and simila
alue objec s. Al hough such isks exis , li le has been w i en abou how di e en
ypes o o e ings o DDBMs, o exchanged alue objec s in pa icula , ela e o
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 359
knowledge isks. The e o e, we add ess he ollowing esea ch ques ion in his pa-
pe : Wha knowledge isks a e associa ed wi h sha ing di e en ypes o da a- ela ed
alue objec s in da a-d i en business models, and wha a e p o ec ion measu es?
To answe his esea ch ques ion, we in e iewed 28 expe s om indus y and
academia o explo e cases o knowledge isks. We s uc u ed di e en ypes o
isks, con ex ual ac o s and p o ec ion measu es based on he h ee basic ypes o
alue objec s: da a, models and p edic ions. Based on ou indings, we sugges h ee
ields o ac ion o mi iga e knowledge isks in DDBMs: using echnology, adjus ing
he business model design and es ablishing us ul ela ionships and con ac ual
egula ions. Managing knowledge isks in DDBMs equi es a balanced iew and
in e disciplina y app oach al eady du ing he design o a DDBM.
2 Backg ound
2.1 Da a-D i en Business Models
Da a-d i en business models (DDBMs) ha e a concep ual ocus on alue c ea ion
om da a (Guggenbe ge e al. 2020). A business model is a concep ual ool ha
allows a simpli ied desc ip ion o how o ganisa ions c ea e, deli e and cap u e alue
(Os e walde and Pigneu 2010; Os e walde e al. 2005; Teece 2010).
Fi ms wi h a DDBM u ilise da a as a key esou ce o new business (Ha mann
e al. 2016). They gene a e cus ome alue h ough da a analy ics and machine lea n-
ing (Schü i z e al. 2017b). Da a analy ics and machine lea ning echniques a e used
o disco e insigh s om da a (Kühne and Böhmann 2019). These insigh s a e de-
li e ed as da a analy ics-based ea u es, p oduc s, o se ices and suppo cus ome s
in hei decision-making p ocess (Schü i z e al. 2019) and enable he gene a ion
o new e enue s eams (Schü i z e al. 2017a). Thus, da a in e media ion is he
cen al alue p oposi ion (Do e 2016). De eloping a DDBM equi es business and
echnological capabili ies (S ahl e al. 2023).
Li e a u e s a ed o analyse and ca ego ise DDBMs om di e en pe spec i es.
Two common app oaches a e o di e en ia e based on he ype o da a sou ces
used (e.g., in e nal exis ing o sel -gene a ed da a s ex e nally acqui ed, cus ome -
p o ided o ee a ailable da a; see, e.g., Ha mann e al. 2016) o he ype o
analy ics used (e.g., desc ip i e, diagnos ic, p edic i e, s p esc ip i e; see, e.g.,
Hunke e al. 2019). As da a in e media ion is he cen al alue p oposi ion (Do e
2016), i is also wo hwhile o dis inguish DDBMs based on he ype o alue
p oposi ion and o e ings. Fo ins ance, Schü i z e al. (2019) di e en ia e be ween
da a, insigh s, and ac ions as o e ings. Dehne e al. (2021) u he di e en ia e
be ween da a, in o ma ion/knowledge, ac ions and non-da a p oduc s and se ices
in DDBMs. Hi and Kühl (2018) desc ibe Model-as-a-Se ice and P edic ion-as-a-
Se ice as wo o he ypes o o e ings.
These o e ings can be di e en ia ed by he ype o exchange o alue objec s
(Leski e al. 2021). A alue objec , as desc ibed in he e-3 alue on ology, “is
o alue o one o mo e ac o s. Ac o s may alue an objec di e en ly and subjec-
i ely, acco ding o hei own alua ion p e e ences” (Go dijn and Akke mans 2003,
K
360 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
Da a-D i en Business Models
Da a-sha ing
Business Models
Model-sha ing
Business Models
P edic ion-sha ing
Business Models
Fig. 1 Sub ypes o DDBMs based on exchanged alue objec s
p. 120). Conce ning DDBMs, such a alue objec can be da a (e.g., Dehne e al.
2021), models (e.g., Hi and Kühl 2018)o p edic ions o insigh s in gene al (e.g.,
Schü i z e al. 2019).
By da a, we unde s and a adeable collec ion o “codi ied obse a ion[s] ixed
in a angible medium” (Thomas e al. 2023, p. 256). Sha ed da a can be in he
o m o speci ic da a poin s, whole da a se s (o da a s eams) o agg ega ed da a
(e.g., ia desc ip i e s a is ics). By model, we unde s and a p og am o unc ion
ha can iden i y pa e ns o p o ide p edic ions based on p e iously unseen inpu
da a. A model is a esul o applying a machine lea ning algo i hm o a se o
( aining) da a. A model consis s o i s code and con igu a ion. Hi and Kühl (2018)
di e en ia e be ween base models speci ic o one pa icula p oblem and ans e
models ha can be applied o ans e ed o a se o simila p oblems. The ype
o p edic ion encompasses iden i ying pa e ns, p edic ing e en s o a ibu es, o
ecommending ac ions based on incoming da a applied o a lea ning model (Hi
and Kühl 2018). P edic ions also ep esen a ge -speci ic insigh s ha a e sha ed o
sol e a speci ic (decision) p oblem o he cus ome , c ea e cus ome bene i , and, in
e u n, gene a e e enue.
As Fig. 1illus a es, da a-, model, and p edic ion-sha ing business models can
be unde s ood as h ee sub ypes o DDBMs. Di e en ia ing DDBMs based on ex-
changed alue objec s is s ill unde - ep esen ed in he DDBM li e a u e, bu a ea-
sonable di e en ia ion when i comes o knowledge isks: We assume ha sha ing
di e en ypes o alue objec s leads o di e en ypes o isks.
Examples o DDBMs ha p o ide da a as an exchanged alue objec a e API-
based da a-sha ing business models in logis ics (e.g., Mölle e al. 2020). In such
da a-sha ing business models (Schweiho e al. 2023) o “da a-as-a-se ice” busi-
ness models (Chen e al. 2011), he business model owne g an s o he pa ies access
o his own da a se in exchange o compensa ion (Schweiho e al. 2023; Vesselko
e al. 2019). One majo obs acle o da a sha ing in o ganisa ions is he conce n abou
exposing sensi i e da a and gi ing compe i o s a compe i i e ad an age (Gelhaa and
O o 2020; Schweiho e al. 2023). Thus, secu i y aspec s, such as usage es ic ions
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 361
o c yp og aphy, need o be implemen ed in such business models (Schweiho e al.
2023).
Examples o DDBMs ha p o ide models as an exchanged alue objec a e Lan-
guage-Model-as-a-Se ice (Sun e al. 2022). In such a model-as-a-se ice business
model, he use p o ides o uploads da a o he se ice p o ide who builds ( ains)
a model based on his aining da a and his own human and/o machine in elligence
(Hi and Kühl 2018).
Examples o DDBMs ha p o ide p edic ions as an exchanged alue objec a e
p edic ion APIs (San hosh e al. 2019). In a “p edic ion-as-a-se ice” o mo e gene al
“analy ics-as-a-se ice” business model, he p o ide applies a (machine lea ning)
model o he inpu da a p o ided by he cus ome o gene a e a p edic ion o e en s,
ecommenda ions o o iden i y pa e ns and inally o suppo decisions o au oma e
ac ions o he cus ome (Hi and Kühl 2018; Schü i z e al. 2019). We subsume
hese di e en e ms unde he e m p edic ion o he con ex o his pape .
2.2 Knowledge Risks Eme ging om Da a Sha ing
DDBMs in ol e new ypes o isks. La ge-scale da a sha ing can cause leakage
o compe i i e knowledge and in ellec ual p ope y (Zei inge and Thalmann 2020;
Zeng e al. 2012). This isk is called knowledge isk and comp ises po en ial knowl-
edge a i ion, loss, leakage o spill-o e o knowledge ha could ad e sely a ec
he o ganiza ion’s s a egic ad an age (Du s and Zieba 2017; Pe o 2007).
Compe i i e knowledge o a i m can be disco e ed om sha ed da a se s using
ad anced analy ics me hods (Il onen e al. 2018). Fu he , i is di icul o i ms
o e alua e which knowledge could be disco e ed by ex e nal ac o s om sha ed
da a (Zei inge and Thalmann 2020). Known app oaches o ex e nal acquisi ion o
compe i i e knowledge ha endange a i m’s in ellec ual p ope y a e in o ma ion
leakage in supply chains (Zhang e al. 2012), indus ial/da a espionage (Thiel and
Thiel 2015), o da a b eaches (Khan e al. 2021). An ad e sa ial ac o could also
ob ain aluable knowledge by e e se-enginee ing he i mwa e o a physical p oduc
o econs uc an embedded algo i hm (Thiel and Thiel 2015). Fo ins ance, i is
echnically possible o e e se-enginee black-box neu al ne wo ks (e.g., Oh e al.
2019), o o s eal machine lea ning models ia API access (e.g., T amè e al. 2016).
The desc ibed a acks can lead o unin ended leakage o spill-o e o knowledge,
deno ed as knowledge isk (Il onen e al. 2018; Zei inge and Thalmann 2020).
A knowledge isk is he “measu e o he p obabili y and se e i y o ad e se e ec s
o any ac i i ies engaging o ela ed somehow o knowledge ha can a ec he unc-
ioning o an o ganisa ion on any le el” (Du s and Zieba 2018, p. 2). Knowledge
isks can be analysed by he ac o s ha cause hem and he p e en i e measu es
o ganisa ions can ake (Du s and Zieba 2017). Managing knowledge isks in e ms
o knowledge p o ec ion is one co e s a egy o knowledge managemen (Loebbecke
e al. 2016). I is c ucial o o ganisa ions as knowledge is essen ial o compe i i e
ad an age (Jennex and Zyngie 2007). The e o e, knowledge p o ec ion p e en s
unwan ed knowledge leakage o non-au ho ized people and o ganisa ions (Manha
and Thalmann 2015). Exis ing knowledge p o ec ion li e a u e ocuses on o mal and
explici knowledge. I does no conside aci knowledge in o ganisa ions (Manha
K
362 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
and Thalmann 2015) and he knowledge ha can be disco e ed om da a s eams
(Il onen e al. 2018). While explici knowledge (e.g., ma e ialised in da a- ela ed
alue objec s) could quickly lea e a company, aci knowledge is mo e di icul o
ans e and in o mal knowledge p o ec ion p ac ices a e needed (Thalmann e al.
2024).
Finally, de eloping business models can be unde s ood as a se “o conc e e
choices and he consequences o hese choices” (Casadesus-Masanell and Rica
2010, p. 198). Manage s mus balance expec ed isks and es ima ed e u ns when
deciding be ween di e en business model design op ions (Casadesus-Masanell and
Rica 2010; Tesch e al. 2017). Such isks can h ea en he p o i abili y o he
business model o e en he i m’s alue (B illinge 2018), making i necessa y o
manage he isks. Risk managemen gene ally in ol es iden i ying, assessing and
moni o ing isks (B illinge e al. 2020; Hallikas e al. 2004). Risks a e usually
e alua ed by assessing he p obabili y o a isk e en and i s impac on he business
model (Hallikas e al. 2004; B illinge e al. 2020). The p oblem wi h assessing non-
inancial isks, such as cybe secu i y isks, is ha li le quan i a i e in o ma ion is
a ailable, especially no eliable p obabili y dis ibu ions (F anke 2020). Despi e
his, iden i ying and deciding how o deal wi h isks al eady in he business model
design is c ucial (Gi o a and Ne essine 2011). A e iden i ying and being awa e o
isks, manage s can adap he business model design as a isk managemen measu e
(B illinge e al. 2020).
Ou Conclusion om he Li e a u e DDBMs can be di e en ia ed based on he
o e ing o , in pa icula , exchanged alues. Based on he li e a u e, we ha e s a ed
ha o e ings in DDBM can be dis inguished by h ee ypes o alue objec s: da a,
models and p edic ions. Fu he , knowledge p o ec ion li e a u e ecognises da a
sha ing as a knowledge isk in gene al and ha ex ac ing knowledge om sha ed
da a is possible ia da a science me hods. We al eady ha e he i s e idence om
p e ious esea ch ha exchanging da a- ela ed alue objec s can lead o knowledge
isks (F uhwi h e al. 2019). Ne e heless, he ela ionship be ween knowledge isks
and exchanged da a- ela ed alue objec s in DDBMs has no been s udied, and his
connec ion has no been made by p e ious li e a u e.
3 Resea ch Me hod
Ou s udy aims o explo e knowledge isks speci ic o DDBMs due o he speci ic
na u e o alue objec s. Gi en he no el y o he p oblem and lack o unde s anding
o how and i knowledge isks occu in DDBMs, we applied an explo a o y, quali a-
i e esea ch design ha is app op ia e o in es iga ing why a ce ain phenomenon
occu s (Yin 2009). The esea ch design is quali a i e, as we analysed in e iew da a
(see da a collec ion sec ion below), and explo a o y, as we used a bo om-up da a
analysis me hod as in o med induc i e coding (see da a analysis sec ion below).
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 363
Table 1 O e iew o ou da a collec ion p ocess
In e iew Round 1 In e iew Round 2 In e iew Round 3
In e iew
pa ici-
pan s
16 In e iews 7 In e iews 5 In e iews
7 Resea che s (R1–R7), 9 Indus-
y Expe s (I1–I9) ac i e in da a-
d i en se ices, business model
inno a ion and knowledge isks
3 Resea che s (R8–R10),
4 Indus y Expe s
I10–I13 ac i e in da a-
d i en se ices
5 Indus y Expe s
(I14–I18) ac i e in
da a-d i en se ices
Du a ion 35–75min 38–59min 40–59min
Goal,
main
ques ions
and con-
en
Focus on knowledge isks in
DDBMs in gene al
Focus on knowledge isks
om sha ing da a- e-
la ed alue objec s (da a,
models and p edic ions)
E alua ion o esul s
P esen a ion o
5 ypes o knowl-
edge isks
Main
ou comes
Knowledge isks di e i da a,
models o p edic ions a e sha ed
Iden i ied i e ypes o
isks based on he h ee
ypes o alue objec s
Sub ypes o isks o
each ype o sha ed
alue objec s &
con ex ual ac o s
3.1 Da a Collec ion
Due o he aci and sensi i e na u e o he opic o o ganisa ions, we decided
on expe in e iews in h ee ounds as ou p ima y da a sou ce (see Table 1), as
in e iews allow comp ehensi e discussions (Yin 2009). As in e iew pa ne s, we
selec ed 28 expe s, 18 om indus y (I1–I18) and 10 om esea ch ins i u ions
(R1–R10) (see Table 3in Appendix A).
We ollowed a pu posi e sampling s a egy (E ikan 2016) and, in pa icula , an
expe sampling s a egy ha is use ul “when in es iga ing new a eas o esea ch”
and in pa icula when “ he e is cu en ly a lack o obse a ional e idence” (E ikan
2016, p. 3). As i was challenging o iden i y sui able cases (i.e., o ganisa ions) whe e
knowledge isks ha e o could occu , as such in o ma ion is no publicly a ailable,
we also selec ed consul an s and esea che s as in o man s who epo ed such cases.
Academic expe s epo ed on hei expe ience and cases o knowledge isks in
DDBMs based on hei collabo a ion wi h indus y (e.g., as pa o esea ch o
consul ing p ojec s). We selec ed expe s based on hei knowledge and expe ience
in de eloping DDBMs o suppo ing o ganisa ions in ha p ocess. Fo academic
expe s, we conside ed hei ecen publica ions on DDBM as an addi ional selec ion
c i e ion. The selec ion o expe s in he ini ial in e iew ound was b oade : we
also selec ed expe s in business model inno a ion and knowledge isks in gene al
o explo e he opic. We sea ched o expe s in ou immedia e ne wo k and h ough
an ex ended ne wo k on he LinkedIn pla o m (2nd o de con ac s).
We conduc ed he in e iews as ace- o- ace mee ings o ia digi al communica-
ion so wa e and audio- eco ded hem. Appendix A p o ides a de ailed desc ip ion
o he expe s who we e in e iewed.
The scope o he i s in e iew ound was e y b oad, se ing as a s a ing poin
o explo e knowledge isks in DDBMs. A e ini ial da a analysis, we ound ha
di e en ia ing and analysing knowledge isks in DDBMs based on exchanged alue
objec s is in e es ing and easonable. The e o e, we conduc ed se en addi ional and
K
364 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
mo e ocused in e iews wi h addi ional expe s. In his second in e iew ound,
we p esen ed and discussed he h ee da a- ela ed alue objec s (da a, model, and
p edic ions) and asked abou cases and hei ela ion o knowledge isks. In he i s
ound, no all alue objec s we e co e ed in each in e iew as he insigh s eme ged
o e ime. Fu he , we in es iga ed mo i a ions and p ac ices in he design phase
o a DDBM in de ail, as we could now ask mo e ocused ques ions in he second
ound.
A he beginning o ou semi-s uc u ed in e iew guideline, we p esen ed wo k-
ing de ini ions o cen al concep s and an abs ac p oblem de ini ion, illus a ed
wi h a case example. The in e iew was di ided in o wo pa s: The i s wo- hi ds
o he in e iew ocused on explo ing he p oblem o knowledge isks in DDBM.
The las one- hi d (only in in e iew ound 1) ocused on discussing equi emen s
o ICT ools iden i ying and desc ibing knowledge isks in DDBMs (F uhwi h e
al. 2021). We asked he in e iew pa ne s o eal examples om hei con ex o
conc e ise and g ound he discussion as much as possible wi hin hei expe ience.
The guideline was es ed wi h a PhD s uden om he same subjec (wi h p ac i-
cal expe ience) and me hodological knowledge ( aining) ega ding he guideline’s
comp ehensibili y, ques ion low and s uc u e. We adjus ed ou in e iew guideline
o he second se o in e iews h ough de ailed ques ions (e.g., ega ding p o ec ion
measu es) and a sho p esen a ion o ou in e im esul s. We p esen ed each ype o
alue objec sho ly and asked he expe s how hey pe cei ed he knowledge isk
ela ed o each alue objec .
To alida e ou esul s, we conduc ed a hi d in e iew ound wi h i e addi ional
indus y expe s in da a-d i en se ices and da a analy ics. The in e iews las ed
be ween 40 and 59min. We again p esen ed ou p oblem de ini ion, he concep s
om he da a analysis s ep a e he wo p e ious ounds and he i e ypes o isks
iden i ied. Fu he , we p o ided one slide pe ype o isk wi h a sho desc ip ion
and one example om he ini ial expe in e iews. The h ee guiding ques ions o
he e alua ion in e iews we e: 1) Do you pe cei e hese isks as ele an o you
business? 2) A e he e any o he ypes o isks missing in ha con ex ? 3) Is he
desc ip ion o each isk easonable o you?. Table 1summa izes ou da a collec ion
p ocess.
3.2 Da a Analysis
In e iews we e ully ansc ibed and cleaned. Quo es used in his publica ion om
in e iews conduc ed in Ge man we e ansla ed in o English (ma ked wi h a “*”)
and e iewed by a second esea che . We analysed his da a ollowing a quali a i e
con en analysis app oach ia in o med induc i e coding (May ing 2015)using
MAXQDA V.11.
Fo analysing he i s ound o in e iews, he dimensions o analysis we e hemes
ha co esponded o he leading in e iew ques ions and de eloped a p o isional
coding scheme o s uc u e he da a. The majo hemes om he in e iew guideline
ha e been “causes o knowledge isks”, “consequences o knowledge isks”, and
examples. Fo he heme o he causes, we gene a ed “in luencing ac o s” and
“mechanism” as ou majo ca ego ies. We dis inguished be ween “ ype o knowledge
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 371
“On he o he hand, when I alk abou da a ha di ec ly ela es o he p oduc ,
wi h which i is possible o d aw conclusions abou he a chi ec u e and ech-
nological speci ics. He e, o cou se, he si ua ion is di e en and he sensi i i y
o he in o ma ion is highe .” (I17*, Manage Da a Analy ics Semiconduc o
Company)
On he o he hand, he manage also men ioned ha sha ing ope a ional da a
om hei p oduc ion machines o main enance o op imisa ion was pe cei ed as
less c i ical, as no conclusions on compe i i e knowledge a e possible.
The isk o knowledge leakage h ough da a sha ing depends on he con ex . I
da a is sha ed ha ela es o compe i i e knowledge, i.e., abou hei p oduc s o co e
p ocesses, ha allows an ex e nal pa y o make conclusions on he a chi ec u e o
echnology used, hen i is pe cei ed as c i ical. I he da a ela es o a mo e common
con ex , such as he main enance o machines, sha ing da a was pe cei ed as less
c i ical. Thus, wha is compe i i e knowledge is e y speci ic o he company and
depends on i s business model.
One in e iewee, he e o e, poin ed o he di ec ion ha in e nal balancing is
necessa y, i.e., a wha s age is he e ie al o knowledge no accep able o he
company anymo e? They need o ake measu es:
“The in e nal discussions ha e o be held abou when we ha e eached a le el
whe e d awing conclusions abou he da a o , o example, he ehicle’s con ig-
u a ion, he p oduc ion, he de elopmen , is no longe accep able o us, and we
he e o e ha e o do some hing else.” (I14*, Consul an Da a-D i en Se ices)
4.1.4 Knowledge P o ec ion Measu es
As we ha e seen abo e, knowledge isks in DDBMs a e e y con ex ual, i.e., i
he sha ed da a ela es o compe i i e knowledge. One p o ec ion measu e ha ou
in e iewees men ioned was o classi y he da a sou ces and o decide i his da a
can be sha ed o no , as one manage om he semiconduc o indus y men ioned:
“And you ha e o ha e business p ocesses in place. Tha ’s wha we ha e a
ou company in place, whe e you e alua e he da a acco ding o ca ego ies,
om public o s ic ly con iden ial, o example.” (I17*, Manage Da a Science
Semiconduc o Company)
Ano he mechanism o ackle knowledge isks and enable da a sha ing is o
in ol e a da a pla o m. I media es he da a exchange be ween ac o s wi h echnical
measu es implemen ed in he pla o m while p ese ing he p o ide ’s knowledge.
The au omo i e manage u he men ioned he e:
“Tha ’s why he e a e all hese da a-sha ing pla o m ini ia i es, [enabling] da a
exchange unde he p emise o knowledge p ese a ion. So, I can e ain my
knowledge bu s ill sha e da a. Howe e his may wo k, i ’s a ask ha p obably
needs o be sol ed so ha i eally akes o .” (I4*, Manage Da a Analy ics)
K
372 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
The in e iewee highligh ed ha knowledge p o ec ion conce ns seem o be one o
he main mo i a ions o he ise o da a pla o ms. Howe e , he also acknowledges
ha p o ec ion conce ns mus be add essed p ope ly be o e implemen ing a DDBM.
The e a e also echnical measu es ega ding secu e echnologies, like enc yp ing
o decen alising da a when pe o ming da a analy ics and hus applying me hods
such as mul i-pa y compu a ion o homomo phic enc yp ion. Ano he app oach
men ioned was o sha e only syn he ic da a, i.e., da a gene a ed by gene a i e AI
wi h simila p ope ies necessa y o sha ing.
Ou in e iewees equen ly also men ioned using con ac s such as NDAs (Non-
Disclosu e Ag eemen s) o ackle his isk. Ne e heless, hey canno p e en knowl-
edge leakage when he con ac is b eached. Fu he , ou in e iewees equen ly
men ioned us ed ela ionships as a measu e o mi iga e knowledge isks. One p ac-
ical app oach men ioned was o begin sha ing smalle and less c i ical da a se s and
o in ensi y he ela ionship o e ime.
Fi ms and cus ome s migh be o e -cau ious and o e -p o ec i e and, he e o e,
unwilling o sha e hei da a o ea o knowledge isks. This would imply ha he
DDBM is no implemen ed. This is especially he case as he e is cu en ly much
awa eness o da a- ela ed isks. Ou in e iewees epo ed he ea ha o he s could
bene i mo e om sha ing and, he e o e, as a consequence, decided no o sha e he
da a. This is pe cei ed as a ba ie o DDBMs, as one da a science manage in he
au omo i e indus y men ioned:
“Because all he companies in he [supply] chain a e so a aid o losing know-
how, hey don’ sha e he da a. [...] This leads o he ac ha i is some imes
di icul in he da a en i onmen o me o do business.” (I4*, Manage Da a
Analy ics)
No ealising a DDBM is he mos ex eme knowledge p o ec ion measu e which
is chosen i he pe cei ed ( ague) isks ou weigh he pe cei ed bene i s o he
DDBM. The e o e, ou in e iewees sugges ed balancing he expec ed bene i s and
possible isks:
“And hen he e is also he ques ion o he bene i : How much in o ma ion can I
gain when I gi e ou da a o u he p ocessing, e sus he isk, wha am I
gi ing away?” (I18*, Managing Di ec o Da a Se ice Company)
Thus, he isk can be educed by unning a da a se ice o p edic ion model on-
p emise, i.e., locally a he cus ome ’s p emise, so ha he da a does no ha e o be
sha ed. Ano he app oach would be o use ede a ed lea ning a chi ec u es,whe e
he da a s ays local and only ( ans e ) models a e sha ed o he weigh s o a neu al
ne wo k.
A u he knowledge p o ec ion measu e is sha ing models ins ead o da a. Mod-
els a e exchanged o p o ec he unde lying da a and allow a bidi ec ional low o
in o ma ion wi hou exposing compe i i e knowledge, as one da a science p o esso
explained:
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 373
“To build a model in o de no o sha e he da a. The model is al eady a isk mi -
iga ion me hod. Wi h he goal, hough, ha you hen ha e a low o in o ma ion
in bo h di ec ions.” (R4*, P o esso o Da a Science)
The impo an aspec men ioned he e is ha exchanging models is a isk mi iga-
ion s a egy, which is pa o DDBMs.
4.2 Sha ing Models
4.2.1 The Risk o Leaking Compe i i e Knowledge om Sha ed Model
Compe i i e knowledge migh be leaked by sha ing models, as knowledge om
expe s (e.g., enginee s) is in oduced o he model in he p ocess o c ea ing o
aining (e.g., enginee ing knowledge abou he ageing beha iou o a ce ain ech-
nical componen ). Models could also e eal in o ma ion hey ha e lea ned bu no
in ended o be sha ed. I he model is sha ed in a whi e-box-like manne (i.e., sha -
ing he code wi h pa ame e s and con igu a ion), compe i i e knowledge is likely
sha ed, leading o a knowledge isk. Fo ins ance, models a e deli e ed as pa o
a consul ing o enginee ing p ojec o suppo he cus ome in de eloping a DDBM,
as one in e iewee epo ed:
“We a e a se ice p o ide o model de elopmen and algo i hms, and we sell
hose di ec ly o ou cus ome s, hen we always sell a bunch o knowledge oo.”
(I4*, Manage Da a Analy ics)
The in e iewee highligh s ha , wi h he model, a huge amoun o knowledge is
ans e ed o he cus ome . Thus, ou in e iewees acknowledge ha compe i i e
knowledge could spill o e o o he ac o s i models a e sha ed. The in e iewed
manage is al eady awa e o his p oblem and men ioned la e ha he e a e ha dly
any o ganisa ional guidelines o ensu e ha sha ed models a e no misused ega ding
knowledge leakage.
Recons uc he Pa ame e s o Con igu a ion om a Black-Box Model E en
i models a e sha ed as black boxes, i.e., he con igu a ion and pa ame e s o he
model a e hidden, he e is also he isk ha knowledge can be e ie ed h ough e-
enginee ing o he model h ough speci ic da a science me hods om a heo e ical
poin . O e all his isk was pe cei ed as low compa ed o whi e-box models. One
da a analy ics consul an epo ed he e on one case:
“In gene al, you can e-enginee nea ly e e y model i you know he inpu and
he ou pu . Then he e a e also algo i hmic me hods o decompose analy ics
models. The e a e me hods om explainable AI o unde s and hem. [...] We
see his mo e, and mo e equen ly, ou cus ome s y o be e unde s and how
ou models wo k.” (I6*, Manage Da a Analy ics Consul ing)
This example shows ha business cus ome s a e al eady ying o unde s and and
e-enginee models and ha p o ide s a e awa e o his ac . Howe e , simila o
o he secu i y mechanisms, i is a ques ion o e o .
K
374 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
Needing o Explain how he Model Comes o Ce ain Decisions o P edic ions
The equi emen o ai , accoun able, and anspa en AI (FAT AI) c ea es a demand
o explain how models come o a ce ain decision o ecommenda ion. One p o es-
so in Business Analy ics sees his as a challenging end om he pe spec i e o
knowledge p o ec ion and epo ed on one case om an indus y p ojec :
“And he e is a p essu e he e om he cus ome o he p o ide . Because you
ha e o explain how a cha bo comes up wi h ha conclusion. So, in ha way,
you a e kind o exposing he algo i hm behind his. [...] he openness o he
algo i hm means ha you also expose knowledge.” (R3, P o esso o Business
Analy ics)
This example shows ha p o ide s could be o ced o expose hei unde lying
models and algo i hms, and he eo knowledge could be e ie ed om he exposed
model. Thus, FAT AI-complian models o explainable AI app oaches could educe
he p o ec i e e ec o models in DDBM.
Leaking he Model o a Thi d Pa y, e.g., when Collabo a ing wi h a S a up o
Build he Model A knowledge isk om sha ing models could also a ise when
a model is join ly de eloped wi h a pa ne (e.g., an AI s a -up) and he model is
leaked he e o a hi d pa y. One manage , o ins ance, men ioned one po en ial
scena io:
“Le ’s say I ha e a ans o me model ha knows exac ly how I make a chip
a ou company. I I lose some hing like ha ou o my hands, o example
by coope a ing wi h a s a up o a pa ne company, whe he i ’s small o la ge.
Then I lose all know-how a he push o a bu on.” (I17*, Manage Da a Science
Semiconduc o Company)
4.2.2 The Risk o In e ence o he Unde lying T aining Da a
Fu he , da a science me hods, such as model in e sion a acks, allow someone o
in e he o iginal da a used o ain he model. Compe i i e knowledge migh spill
o e when he model use can econs uc he o iginal aining da a om a sha ed
model, in pa icula , o in e he s uc u e o he da a (e.g., pa icula da a ields) o
he s uc u e in he da a (e.g., p ope ies o he sample and he bias in he da a). So-
called model in e ence is echnically possible in pa icula cases, acco ding o da a
science li e a u e (e.g., F ed ikson e al. 2015). Ou expe s men ioned ha his can
happen i a model is o e i ing. This is pa icula ly impo an o gene a i e models,
whe e no he o iginal aining da a is gene a ed, bu only simila da a. One o ou
in e iewees men ioned he e one hypo he ical example whe e his model in e ence
could happen:
“[...] Then he e is he isk ha you a e e ealing in o ma ion abou you own
da a wi h he models. [...] Le ’s assume we ake wo insu ance companies. They
wan o imp o e aud de ec ion. They exchange me a-in o ma ion o ain
models oge he o do ha . F om ha , you can ge he s uc u e o he da a
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 375
used o aining. And ha unde lying s uc u e can al eady gi e one insu ance
company, which o cou se is a compe i o , a lo o in o ma ion abou he o he .”
(I9*, CEO/Co-Founde Da a Science Company B)
4.2.3 Con ex ual Fac o s
The isk o knowledge leakage om sha ing models depends on he con ex , i.e., i
compe i i e knowledge can be de i ed om a sha ed model. Especially in consul ing
and enginee ing, p ese ed domain knowledge om expe s is leaked when whi e-
box models a e sha ed, as one in e iewee epo ed:
“I I ake hese models and gi e hem away, hen I’ e aken he knowledge
ha I’ e disco e ed om people, om hei ac ions, om hei labelling, om
hei inpu , p ese ed i in he model, and sold i o he ou side wo ld. Tha ’s
a emendous isk.” (I4*, Manage Da a Analy ics)
This case shows ha expe knowledge om employees is ma e ialised in o mod-
els. As pa o an enginee ing business, models a e sha ed wi h hei cus ome s.
Mo eo e , h ough sha ing he model, he ma e ialised knowledge o hei expe s
could spill o e o hei cus ome s.
One expe om he semiconduc o indus y (I17) also men ioned a u u e exam-
ple in e ms o gene a i e AI and ans o me models ha could explain how o build
a echnical sys em (e.g., a mic ochip). This could be a huge isk i such a model was
ained wi h company-speci ic da a and leaked (e.g., h ough a collabo a ion wi h
a s a -up).
The isk depends on how easily he model can be applied and ans e ed o o he
applica ion scena ios, as one manage men ioned:
“I i [ he model] is e y speci ic o a p oblem, I’m no a aid. [...] I he model is
e y gene ic and easily ans e able o di e en ypes o p oblems, o a di e en
da a se , o a di e en con ex , [...], hen we ha e o be ca e ul.” (I4*, Manage
Da a Analy ics)
The in e iewe men ions, “I am no a aid” and “we ha e o be ca e ul”. Bo h
ph ases clea ly indica e ha his is a well-e alua ed decision. Beyond abs ac ans-
e abili y, ano he o ganisa ion also needs he capabili y and knowledge o apply
he model. Fu he , he a ailabili y o app op ia e da a se s whe e he model can be
applied in luences he isk, as one consul an men ioned:
“Wi hou he aw da a, he algo i hm is less use ul o me. [...] has he o he
pa y also he same aw da a o o he da a wi h simila o ma s? I yes, hen ha
is a big isk. [...] And he highes isks a e in cases in which when he algo i hm
is leaked, and he aw da a is a ailable o ep oducible.” (I6*, Manage Da a
Analy ics Consul ing)
The in e iewee poin s o he s a egy o keeping he aining da a in he back and
jus sha ing he model. This is especially impo an , as many success ul DDBMs
ely on unique dynamic da a se s gene a ed h ough using he se ice (e.g., loca ion
K
376 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
da a o a ic pa icipan s o p edic a ic jams). Thus, he model only has alue i
i is used in combina ion wi h his a ailable da a. Ano he in luencing ac o is he
ola ili y o he model: The isk inc eases i he model is alid o a longe pe iod.
Whe eas he isk is lowe i a dynamic model is cons an ly adjus ed and upda ed.
Models implici ly con ain he knowledge ep esen ed by he da a used o ain he
model. Building a model also comp ises knowledge o how o c ea e alue-added
in o ma ion om aw da a, as one consul ing manage explained:
“[You need] a combina ion o knowledge o he da a scien is who jus looks a
he aw da a, a he g aph, e y simply speaking, and he enginee who knows
exac ly how he machine wo ks, who knows exac ly wha i means when he e’s
a p essu e d op in he hyd aulic a m o he welding obo .” (I6*, Manage Da a
Analy ics Consul ing)
This s a emen shows ha , on he one hand, domain- ela ed knowledge, e.g., om
enginee ing, is needed o ain a model. On he o he hand, knowledge om he da a
science discipline is also needed. Domain (expe ) knowledge abou a eal-wo ld
phenomenon can add alue o he model, such as speci ic casual ies o ela ion-
ships ha canno be disco e ed om he da a i sel bu need addi ional con ex ual
knowledge on he domain. Da a science knowledge in ol es he labelling, p epa -
ing, and agg ega ing o he da a and subsequen analy ics and algo i hmics and hei
combina ion.
4.2.4 Knowledge P o ec ion Measu es
Ou in e iewees men ioned ha adi ional legal p o ec ion mechanisms o IPR
(e.g., pa en s) do no wo k o models. As he knowledge is only implici ly con ained
in he model, a lawsui o con ic he guil y seems e y challenging. The e o e, he
owne o he know-how and IP should be de ined in con ac s, e.g., he IP ega ding
he model c ea ion emains a he p o ide .
Fu he , ou in e iewees men ioned de ining and iden i ying wha in o ma ion
should be e ealed by he model and which no o build he model acco dingly and
ensu e ha he model is only used as in ended. Models should be designed so
ha hey only disclose he in ended minimum amoun o in o ma ion (e.g., only
he ans e unc ion wi hou e ealing he in luencing pa ame e s (e.g., I17)). This,
again, equi es alignmen and balance be ween sha ing and p o ec ing knowledge.
The isk also depends on he balance be ween gene a ed e u ns and he es ima ed
isk. Fo ins ance, one expe men ioned ha he mone a y alue o selling a model
would be signi ican ly highe han only sha ing p edic ions, in pa icula , i he code
o he model can be accessed. Thus, he isk can also be mi iga ed by adjus ing he
business model,o mo ep ecisely, hep icing model.
Thus, p o ec ing knowledge in DDBMs is cu en ly mainly pe o med ia echni-
cal measu es. One simple knowledge p o ec ion mechanism is o sha e only black-
box models:
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 377
“Fo example, i I sha e he sou ce code, whe e I can see e e y pa ame e o
he [decision] ee, hen i ’s clea ha I’m selling c i ical knowledge. Bu in
con as , i I make p edic ions black box-like, hen I would ind i di icul o
econs uc he pa ame e s.” (I2*, Da a Scien is Au omo i e Company A)
Ou in e iewees sugges applying da a science me hods o p e en model in-
e sion a acks, such as andomisa ion in aining, di e en ial p i acy, o o he
anonymiza ion me hods. Fo example, ou in e iewees men ioned using di e en
loss unc ions o syn he ic da a o model aining.
Ano he p o ec ion mechanism is o keep he model wi hin he o ganisa ion’s
knowledge bounda y and o e he model as a se ice ia a pla o m o an API.
Howe e , he use o he model has o sha e his da a now wi h he se ice p o ide ,
which could c ea e a knowledge isk o he use . The p o ide sha es only he
esul s.
4.3 Sha ing P edic ions
4.3.1 The Risk o Recons uc ing a Model om Sha ed P edic ions
Compe i i e knowledge migh spill o e when plen y o p edic ions a e sha ed, and
he ecei e can econs uc he model o pa s o he model based on hese p e-
dic ions. Acco ding o compu e science li e a u e, econs uc ing models based on
p edic ions is echnically possible in pa icula cases (e.g., T amè e al. 2016). How-
e e , such a acks can be mi iga ed easily by es ic ing he numbe o p edic ions
o he alue ange. Thus, his isk was pe cei ed as low.
One way o disco e knowledge is o econs uc he unde lying model by p o ok-
ing lo s o p edic ions. Mo eo e , he model allows in e ences abou he ma e ialised
knowledge. One academic expe in knowledge p o ec ion poin ed o he p oblem:
“I you sell many ou comes, yeah, hen i would be e en hen possible o e-en-
ginee he algo i hm i sel . I you a e looking a wha kind o esul s a e c ea ed
by wha kind o da a.” (R5, Senio Resea che Knowledge Managemen )
Howe e , he in e iewee e e s o “wha kind o da a”, and ano he in e iewee,
a da a scien is , speci ies his in mo e de ail:
“I you ake a look a he p edic ions now, you’ll p obably see a ew ea u es and
check o which g oup i ’s wo king be e o wo se. You’ll be able o econs uc
some hing he e.” (I2*, Da a Scien is Au omo i e Company A)
As he says, “ o econs uc some hing he e”, he acknowledges he big challenge
o disco e ing compe i i e knowledge om a p edic ion-based alue p oposi ion.
Howe e , ou in e iewees pe cei ed he isk o knowledge leakage h ough sha ing
p edic ions as low as, o ins ance, one in e iewee said:
K
378 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
“Fo example, he cus ome only ge s he esul s back. In ha case, I hink
he isk is e y low ha any knowledge will d ain om he p o ide because
he cus ome doesn’ ha e access o ha knowledge.” (R8*, Resea che Da a-
D i en Se ices)
This s a emen shows ha he knowledge is hidden and ha he cus ome has no
di ec access o he model and he ma e ialised knowledge in he model. Fu he ,
ou expe s (e.g., I17) no ed ha i is no so easy o de i e clea conclusions— he
in e ence is subjec o p obabili ies.
4.3.2 Con ex ual Fac o s
Recons uc ing he model om p edic ions is possible om a heo e ical poin o
iew. Howe e , in eali y, his is no i ial and equi es some p io in o ma ion
on he model a ailable. How easy i is o econs uc he model also depends on
i s complexi y and he inpu da a a iabili y, as one expe in he ield o DDBMs
explained:
“The hea o a good model is he a iance o he inpu ac o s. And i I jus
o e an API, whe e I only p o ide a esul o ce ain inpu alues, bu he inpu
da a ha ha e led o ha model has mo e a ie y han I’m allowing h ough he
API, I can ac ually [p e en ha well].” (I5*, Di ec o Digi al Business)
As his quo e shows, e-enginee ing a model based on lo s o “ esul s” ha we
call p edic ions depends on he a iance o he inpu da a i i co e s he whole inpu
space. The knowledge is hidden and is ma e ialised in he p edic ion model i sel .
The single p edic ion hus o e s only a small and sca e ed glimpse o he model.
Many p edic ions need o be collec ed o e en p o oked in a sys ema ic a ack o
e-enginee knowledge:
“I you send enough di e en que ies, you can al eady [ econs uc ] wha
knowledge is ma e ialised in he model. Depending on he complexi y o he
p oblem, his migh be a ask a he momen , which do no allow model e-
enginee ing due o he complexi y.” (I1*, Da a Analy ics Consul an )
This quo e shows ha econs uc ing knowledge is possible bu equi es signi -
ican e o and expe ise. I insigh s abou he model a e success ully collec ed,
knowledge could be disco e ed. One men ioned example o knowledge ha could
be econs uc ed is he bias ha he model has lea ned. Fu he , one mus balance
he e o i i is wo h i o he a acke .
4.3.3 Knowledge P o ec ion Measu es
When p edic ions a e sha ed h ough access o a p edic ion model, one simple p o-
ec ion measu e is o con ol he access in e ms o he numbe o allowed eques s,
he minimum ime span be ween wo eques s, and he ange o inpu alues. Lim-
i ing he numbe o eques s p e en s b u e o ce a acks o econs uc ion and
also denial o se ice a acks. Po en ial a acks could be ecognised h ough a ypi-
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 379
cal eques s, e.g., uni o mly dis ibu ed ac oss he inpu space, as aining and e-
enginee ing a model equi es a b oad ange o inpu da a. Ou in e iewed expe
con inued:
“Fi s o all, when someone pene a es me and asks me ques ions o e he en i e
ec o space, hen I no ice ha his is a ypical. Tha would be a uni o m dis i-
bu ion in he que y, which is o ally a ypical o such a hing, he e you a he
ha e a no mal dis ibu ion in he que ies.” (I5*, Di ec o Digi al Business)
One p o ec ion measu e is o build a p edic ion se ice ha elies on dynamic
da a, such as eal- ime ehicle loca ion da a, gene a ed h ough se ice usage and
no sha ed wi h o he ac o s. E en i he p edic ion model could be econs uc ed
based on many p edic ions, he knowledge canno be applied as one malicious ac o
canno access he necessa y da a. Ano he p o ec ion measu e lies in he design o
he business model: in p edic ion-as-a-se ice business models, pay-pe -use e enue
models a e o en used, which means ha eques ing lo s o p edic ions ge s expen-
si e, and by ha , e en i some hing could be econs uc ed om he model, i was
compensa ed mone a ily.
5 Discussion o Resul s
5.1 Discussion o P oblem and Risk Rele ance
In ou in e iews, we ound ha knowledge isks a e a ele an opic in da a-d i en
business models. Fo he h ee ypes o alue objec s da a, model and p edic ion, we
iden i ied i e ypes o isks ha a ise when hey a e exchanged in a DDBM: The isk
o leaking compe i i e knowledge om sha ed da a, he isk o exposing compe i i e
knowledge by using a da a se ice; he isk o leaking compe i i e knowledge om
a sha ed model; he isk o in e ence o he unde lying aining da a; and he isk o
econs uc ing a model om sha ed p edic ions.
The alida ion in e iews con i med he i e ypes o isks, i.e., no addi ional ypes
we e sugges ed o eme ged, and he desc ip ion o he exis ing ones was su icien .
The isk o exposing compe i i e knowledge by using a da a se ice was pe cei ed
as he mos ele an isk in he alida ion in e iews, as one expe b ough i o he
poin : “I hink ha is he bigges , bu also e y ha d o g asp, h ea o ea ha he
managemen in he indus y has now” (Indus y Expe 14*). One p oblem is ha
he isk is e y di icul o g asp. The e o e, he e is some imes a lo o ea , and as
a consequence, companies a e e y cau ious, and DDBMs may no be ealised.
Knowledge Risks in DDBMs Depend on Con ex ual Fac o s o he DDBM I sel
The isk depends on he a ea o he company om which da a- ela ed alue objec s
a e sha ed. Fo ins ance, i da a is sha ed o op imise an ancilla y p ocess (e.g.,
main enance o p oduc ion machines), he isk is pe cei ed as less c i ical. Whe eas,
i da a om hei co e p ocess allows in e ence on hei co e p ocesses, e.g., he
design and con igu a ion o p oduc s, he knowledge isk was pe cei ed as c i ical.
Thus, i mus always be assessed i he (po en ial) leaked knowledge is compe i i e
K
380 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
and business-c i ical. Ou da a also sugges ha knowledge isks a e pa icula ly
ele an in knowledge-in ensi e businesses ha wan o inno a e owa ds DDBMs
in addi ion o hei exis ing business model. In such business models, domain ex-
pe knowledge (e.g., enginee ing) is ma e ialised in models ha a e sha ed wi h
cus ome s and pa ne s as pa o a DDBM. Thus, compe i i e knowledge migh be
pu a isk. Fu he , in business models wi h complex sys ems and high compe i ion
(e.g., he au omo i e o semiconduc o indus y), o ganisa ions a e e y es ic i e
abou da a sha ing, as co po a e sec e s migh be sha ed wi h he da a.
Knowledge Risks in DDBM Di e om Knowledge Risks Associa ed wi h T a-
di ional Business Models As mo e a eas o an o ganisa ion a e digi ised, he e
is a isk ha mo e compe i i e knowledge is ma e ialised in (AI) models. These,
howe e , a e easy o ans e compa ed o adi ional business models, whe e engi-
nee s om he compe i ion need o be headhun ed o a p oduc needs o be e e se
enginee ed. In DDBMs, leaking a model could be su icien o knowledge leakage.
Wi h he sp ead o gene a i e AI and ans o me models, we assume his aspec
will become e en mo e impo an in he upcoming yea s (cp. T edinnick and Lay-
ba s 2023). Thus, he ques ion o how o p o ec knowledge and IP in DDBMs will
become mo e impo an .
5.2 Discussion o P o ec ion Measu es—How o Deal wi h he Risk?
We ound ha knowledge isk in DDBMs can be mi iga ed by echnology (which
migh be as changing), by business model design op ions, and by ensu ing ans-
pa ency, building us and con ac ual egula ions. As a syn hesis o hese h ee
a eas o ac ion, one majo s a egic implica ion o ou wo k is ha knowledge isk
mi iga ion in DDBMs needs a di e en ia ed and balanced assessmen o whe he
he pe cei ed isk has a nega i e economic impac o is accep able compa ed o he
expec ed e u n in he DDBM.
5.2.1 Technology o Mi iga e Knowledge Risks
A knowledge isk can o en be educed up on by echnology. Compu e science
li e a u e discusses se e al echnical a acks o e ie e some hing om da a and
models (see, e.g., Kaissis e al. 2020). Such a acks encompass aining da a leak-
age, model s ealing, e e se enginee ing o membe ship in e ence (Hanzlik e al.
2021). P e en ing such a acks o exace ba ing he knowledge disco e y p ocess
can be done by echnical measu es ha ela e o con empo a y compu e science
esea ch (see, e.g., Kaissis e al. 2020). P i acy-p ese ing echnologies ailo ed o
he con ex o big da a analy ics ensu e he con iden iali y o he da a (e.g., Yak-
oubo e al. 2014). Examples o such p i acy-p ese ing echnologies a e mul i-
pa y compu a ion (e.g., A che e al. 2018), da a anonymiza ion (Zei inge e al.
2024), homomo phic enc yp ion (e.g., Alabdula i e al. 2020), wa e ma king (e.g.,
Regazzoni e al. 2021) o me a- and ans e machine lea ning (e.g., Hi and Kühl
2018), which we e also men ioned by ou expe s as echnical p o ec ion measu es.
Such echnology, like mul i-pa y compu a ion, has al eady been ound o be a p o-
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 387
7 Appendix
7.1 Appendix A
Table 3 Lis o In e iewed Expe s
Round ID Type o Posi ion Indus y Du a ion
(min)
Language
1 R1 P o esso o Digi al Business Resea ch 36 EN
1 R2 P o esso o Business Model Inno a-
ion
Resea ch 61 DE
1 R3 P o esso Business Analy ics Resea ch 35 EN
1 R4 P o esso o Da a Science Resea ch 56 DE
1 R5 Senio Resea che o Knowledge Man-
agemen
Resea ch 67 EN
1 R6 P o esso o Knowledge Managemen Resea ch 60 DE
1 R7 P o esso o Knowledge Managemen Resea ch 59 EN
1 I1 Consul an Da a Analy ics Consul ing 39 DE
1 I2 Da a Scien is Au omo i e 52 DE
1 I3 CEO/Co-Founde Cybe Secu i y 63 DE
1 I4 Manage Da a Analy ics Au omo i e 62 DE
1 I5 Di ec o Digi al Business In o ma ion
Technology
76 DE
1 I6 Manage Da a Analy ics Consul ing 67 DE
1 I7 Manage Digi al Business Au omo i e 48 DE
1 I8 CEO/Co-Founde Da a Science 70 DE
1 I9 CEO/Co-Founde Da a Science 45 DE
2 R8 Resea che Da a-D i en Se ices Resea ch 50 DE
2 R9 Resea ch G oup Leade Da a Analy ics Resea ch 53 DE
2 R10 Senio Resea che Da a-D i en Se ices Resea ch 38 DE
2 I10 Manage Da a Analy ics Consul ing 55 DE
2 I11 Manage Da a Science Da a-D i en
Se ice
45 DE
2 I12 Managing Di ec o Da a-D i en
Se ice
48 DE
2 I13 Consul an Business Model Inno a ion Consul ing 59 DE
3 I14 Consul an Da a-D i en Se ices In o ma ion
Technology
59 DE
3 I15 Consul an Da a P o ec ion, A i icial
In elligence
Consul ing 43 DE
3 I16 Founde and Managing Di ec o Consul ing 50 DE
3 I17 Manage Da a Science Semiconduc o 42 DE
3 I18 Managing Di ec o Da a-D i en
Se ice
40 DE
K
388 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
7.2 Appendix B: In e iew Guideline
7.2.1 Guiding Ques ions In e iew Round 1
P oblem desc ip ion
Do you see his as a ele an p oblem? And do you know any simila examples?
Wha o he causes o isk could you imagine in his con ex in da a-d i en business
models?
Wha consequencies do you see based on hese isks?
P esen a ion o exempla y consequencies (knowledge loss, knowledge leakage,
knowledge spill-o e )
How do you assess each o hese consequences as a possible/ ele an p oblem in
da a-d i en business models? Fo each, do you know any example?
Wha o he consequences could a ise om such knowledge isks?
Wha examples om he p ac ice o companies do you known o you whe e he
opic o knowledge isks in da a-d i en business models is, was o could be ele-
an ?
2nd pa (no he scope o his pape ): P esen a ion o a ool and e alua ion ques-
ions
7.2.2 Guiding Ques ions In e iew Round 2
P oblem desc ip ion
How do you assess his p oblem o knowledge isks jus desc ibed as a ele an
p oblem in you business model/in gene al?
Can you ell an examples om p ac ice you awa e o whe e he issue o knowledge
isks in da a-d i en business models is, was o could be ele an ?
Wha po en ial mechanisms can you hink o o econs uc o access knowledge
in he h ee ypes as cus ome and a acke a he same ime?
Wha would be he po en ial consequences o such knowledge isks o you /an
o ganiza ion?
Wha ac o s in luence he isk o knowledge leakage h ough he exchange o
da a, models o p edic ions?
Wha p o ec ion measu es ha e you implemen ed o a oid o p e en such knowl-
edge isks?
7.2.3 Guiding Ques ions In e iew Round 3
P esen a ion o p oblem knowledge isks in DDBMs
P esen a ion o o in e im esul s (main concep s, 5 ypes o isks, o each a sho
desc ip ion and exempla y quo es om he in e iews)
Do you pe cei e hese isks as ele an o you business?
A e he e any o he ypes o isks missing in ha con ex ?
Is he desc ip ion o each isk easonable o you?
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 389
7.3 Appendix C
Table 4 Coding scheme wi h main ca ego ies a e in e iew ound 2
Ca ego y Desc ip ion
Mo i es o sha -
ing
This ca ego y desc ibes mo i es why a ype o alue objec is sha ed wi h o he
s akeholde s
Type o knowledge This ca ego y desc ibes di e en ypes o knowledge ha can be disco e ed om
da a- ela ed alue objec s
Knowledge e-
ie al mechanism
This ca ego y desc ibes mechanisms o how he knowledge can be disco e ed
om he da a- ela ed alue objec by ano he pa y leading o a knowledge leak-
age
In luencing ac o s This ca ego y desc ibes he ci cums ances ha make knowledge e ie al and,
hus a, knowledge leakage possible. These ac o s in luence he p obabili y o he
isk
Knowledge p o ec-
ion measu es
This ca ego y desc ibes measu es o how echnical o business model design
measu es could p e en such knowledge leakage
K
390 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
7.4 Appendix D
Table 5 Snapsho o he coding scheme and exempla y ex segmen s
Tex segmen Code Type o
alue
objec
Ca ego y
“Ich mach das Ganze dann als So wa e-as-a-Se -
ice. Das wä e so die bes e Mi iga ion.” (I10)
O e Model-
as-a-Se ice as
a p o ec ion
measu e
Model Knowledge
p o ec ion
measu e
“Wenn man das Modell nu als API zu Ve ügung
s ell , dann kann jemand zwa An agen s ellen, da
kann jemand das Modell abe noch nich ekons u-
ie en.” (I9)
“... dass man e schlüssel e Da en ü so eine Dien-
s leis ung e wende .” (I1)
Using
enc yp ed da a
Da a
“Da en soll en au jeden Fall e schlüssel übe a-
gen we den.” (I11)
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 391
Acknowledgemen s The esea ch based on his pape has ecei ed unding om he Aus ian COMET
P og am—Compe ence Cen e s o Excellen Technologies—unde he auspices o he Aus ian Fede al
Minis y o T anspo , Inno a ion and Technology, he Aus ian Fede al Minis y o Digi al and Economic
A ai s and by he S a e o S y ia. COMET is managed by he Aus ian Resea ch P omo ion Agency (FFG).
Con lic o in e es M. F uhwi h, V. Pamme -Schindle and S. Thalmann decla e ha hey ha e no
known compe ing inancial in e es s o pe sonal ela ionships ha could ha e appea ed o in luence he
wo k epo ed in his pape .
Open Access This a icle is licensed unde a C ea i e Commons A ibu ion 4.0 In e na ional License,
which pe mi s use, sha ing, adap a ion, dis ibu ion and ep oduc ion in any medium o o ma , as long as
you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Com-
mons licence, and indica e i changes we e made. The images o o he hi d pa y ma e ial in his a icle
a e included in he a icle’s C ea i e Commons licence, unless indica ed o he wise in a c edi line o he
ma e ial. I ma e ial is no included in he a icle’s C ea i e Commons licence and you in ended use is no
pe mi ed by s a u o y egula ion o exceeds he pe mi ed use, you will need o ob ain pe mission di ec ly
om he copy igh holde . To iew a copy o his licence, isi h p://c ea i ecommons.o g/licenses/by/4.
0/.
Re e ences
Agaha i, Wi awan, Hosea O e, and Ma k de Reu e . 2022. I is no (only) abou p i acy: How mul i-pa y
compu a ion ede ines con ol, us , and isk in da a sha ing. Elec onic Ma ke s 32(3):1577–1602.
h ps://doi.o g/10.1007/s12525-022-00572-w.
Al-Aali, Abdul ahman Y., and Da id J. Teece. 2013. Towa ds he (s a egic) managemen o in ellec ual
p ope y: e ospec i e and p ospec i e. Cali o nia Managemen Re iew 55(4):15–30. h ps://doi.o g/
10.1525/cm .2013.55.4.15.
Alabdula i , Abdula i , Ib ahim Khalil, and Xun Yi. 2020. Towa ds secu e big da a analy ic o cloud-
enabled applica ions wi h ully homomo phic enc yp ion. Jou nal o Pa allel and Dis ibu ed Com-
pu ing 137:192–204. h ps://doi.o g/10.1016/j.jpdc.2019.10.008.
A che , Da id W., Dan Bogdano , Yehuda Lindell, Liina Kamm, Ku Nielsen, Jakob I. Pag e , Nigel
P. Sma , and Rebecca N. W igh . 2018. F om keys o da abases— eal-wo ld applica ions o secu e
mul i-pa y compu a ion. The Compu e Jou nal 61(12):1749–1771. h ps://doi.o g/10.1093/comjnl/
bxy090.
Azkan, Can, F ede ik Molle , Lenna Iggena, and Bo is O o. 2022. Design p inciples o indus ial da a-
d i en se ices. IEEE T ansac ions on Enginee ing Managemen 71:2379-2402 h ps://doi.o g/10.
1109/TEM.2022.3167737.
Bai d, Aa on, and Likoebe M. Ma uping. 2021. The nex gene a ion o esea ch on IS use: a heo e ical
amewo k o delega ion o and om agen ic IS a i ac s. Managemen In o ma ion Sys ems Qua e ly
45(1):315–341. h ps://doi.o g/10.25300/MISQ/2021/15882.
B illinge , Anne-Sophie. 2018. Mapping business model isk ac o s. In e na ional Jou nal o Inno a ion
Managemen 22(05):1840005. h ps://doi.o g/10.1142/S1363919618400054.
B illinge , Anne-Sophie, Ch is ian Els, Bjö n Schä e , and Bea e Bende . 2020. Business model isk and
unce ain y ac o s: owa d building and main aining p o i able and sus ainable business models.
Business Ho izons 63(1):121–130. h ps://doi.o g/10.1016/j.busho .2019.09.009.
Casadesus-Masanell, Ramon, and Joan E. Rica . 2010. F om s a egy o business models and on o ac ics.
Long Range Planning 43(2–3):195–215. h ps://doi.o g/10.1016/j.l p.2010.01.004.
Chen, Ying, Je ey K eulen, Mu ay Campbell, and Ca l Ab ams. 2011. Analy ics ecosys em ans o ma-
ion: a o ce o business model inno a ion. 2011 Annual SRII Global Con e ence., 11–20. h ps://
doi.o g/10.1109/SRII.2011.12.
Coombs, C ispin, Donald Hislop, S animi a K. Tane a, and Sa ah Ba na d. 2020. The s a egic impac s o
in elligen au oma ion o knowledge and se ice wo k: an in e disciplina y e iew. The Jou nal o
S a egic In o ma ion Sys ems 29(4):101600. h ps://doi.o g/10.1016/j.jsis.2020.101600.
Dehne , Maik, Alexande Gleiss, and Reiss F ede ik. 2021. Wha makes a da a-d i en business model?
A consolida ed axonomy. P oceedings o he Twen y-Nin h Eu opean Con e ence on In o ma ion
Sys ems (ECIS 2021).
Do e , Lau a. 2016. Da enzen ische Geschä smodelle als neue Geschä smodell ypus in de Elec onic-
Business-Fo schung: Konzep ionelle Bezugspunk e, Klassi ika ion und Geschä smodella chi ek u .
K
392 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung 68(3):307–369. h ps://doi.o g/10.
1007/s41471-016-0014-9.
Du s , Susanne, and Malgo za a Zieba. 2017. Knowledge isks— owa ds a axonomy. In e na ional Jou -
nal o Business En i onmen 9(1):51–63. h ps://doi.o g/10.1504/IJBE.2017.084705.
Du s , Susanne, and Malgo za a Zieba. 2018. Mapping knowledge isks: owa ds a be e unde s anding o
knowledge managemen . Knowledge Managemen Resea ch & P ac ice 17(1):1–13. h ps://doi.o g/
10.1080/14778238.2018.1538603.
E ikan, Ilke . 2016. Compa ison o con enience sampling and pu posi e sampling. Ame ican Jou nal o
Theo e ical and Applied S a is ics 5(1):1–4. h ps://doi.o g/10.11648/j.aj as.20160501.11.
Fa ayola, Oluwa oyin A., A. Abdul Adekunle, Blessing O. I abo , and E elyn C. Okeleke. 2023. Inno a-
i e business models d i en by AI echnologies: a e iew. Compu e Science & IT Resea ch Jou nal
4(2):85–110. h ps://doi.o g/10.51594/csi j. 4i2.608.
Fassnach , Ma cel, Ca ina Benz, Daniel Heinz, Jasmin Leims oll, and Ge ha d Sa zge . 2023. Ba ie s o
da a sha ing among p i a e sec o o ganiza ions. In P oceedings o he 56 h Annual Hawaii In e na-
ional Con e ence on Sys em Sciences, Janua y 3–6, 2023., ed. Tung X. Bui, 3695–3704. Honolulu:
Depa men o IT Managemen Shidle College o Business Uni e si y o Hawaii.
F anke, Gün e . 2020. Managemen nich - inanzielle Risiken: eine Fo schungsagenda. Schmalenbachs
Zei sch i ü be iebswi scha liche Fo schung 72:279-320. h ps://doi.o g/10.1007/s41471-020-
00096-z.
F ed ikson, Ma , Somesh Jha, and Thomas Ris enpa . 2015. Model in e sion a acks ha exploi con-
idence in o ma ion and basic coun e measu es. CCS ’15: P oceedings o he 22nd ACM SIGSAC
Con e ence on Compu e and Communica ions Secu i y., 1322–1333. h ps://doi.o g/10.1145/
2810103.2813677.
F uhwi h, Michael, Vik o ia Pamme -Schindle , and S e an Thalmann. 2019. To Sell o No o Sell:
Knowledge Risks in Da a-D i en Business Models. In P oceedings o he 2019 P e-ICIS SIGDSA
Symposium. 11.
F uhwi h, Michael, Ch is iana Ropposch, and Vik o ia Pamme -Schindle . 2020. Suppo ing da a-d i en
business model inno a ions: a s uc u ed li e a u e e iew on ools and me hods. Jou nal o Business
Models 8(1):7–25.
F uhwi h, Michael, Vik o ia Pamme -Schindle , and S e an Thalmann. 2021. A Ne wo k-based Tool o
Iden i ying Knowledge Risks in Da a-D i en Business Models. In P oceedings o he 54 h Hawaii
In e na ional Con e ence on Sys em Sciences, ed. Tung X. Bui, 5218-5227.
Gelhaa , Joshua, and Bo is O o. 2020. Challenges in he eme gence o da a ecosys ems. Twen y-Thi d
Paci ic Asia Con e ence on In o ma ion Sys ems. UAE, 2020, Dubai.
Gieß, Anna, F ede ik Mölle , Tho s en Schoo mann, and Bo is O o. 2023. Design op ions o da a spaces.
Thi y- i s Eu opean Con e ence on In o ma ion Sys ems (ECIS 2023).
Gi o a, Ka an, and Se guei Ne essine. 2011. How o build isk in o you business model. Ha a d Business
Re iew 89(5):100–105.
Go dijn, Jaap, and Hans Akke mans. 2001. Designing and e alua ing e-business models. IEEE In elligen
Sys ems 16(4):11–17. h ps://doi.o g/10.1109/5254.941353.
Go dijn, Jaap, and J.M. Akke mans. 2003. Value-based equi emen s enginee ing: explo ing inno a i e
e-comme ce ideas. Requi emen s Enginee ing 8(2):114–134. h ps://doi.o g/10.1007/s00766-003-
0169-x.
Guggenbe ge , Mo i z T., F ede ik Mölle , Ka im Boualouch, and Bo is O o. 2020. Towa ds a uni ying
unde s anding o digi al business models. Twen y-Thi d Paci ic Asia Con e ence on In o ma ion Sys-
ems, UAE, 2020, Dubai.
Guggenmos, Flo ian, Bjö n Häckel, Philipp Ollig, and Bas ian S ahl. 2022. Secu i y i s , secu i y by de-
sign, o secu i y p agma ism—s a egic oles o IT secu i y in digi aliza ion p ojec s. Compu e s &
Secu i y 118:102747. h ps://doi.o g/10.1016/j.cose.2022.102747.
Gün he , Wendy A., Mohammad H. Rezazade Meh izi, Ma leen Huysman, and F ans Feldbe g. 2017.
Deba ing big da a: a li e a u e e iew on ealizing alue om big da a. The Jou nal o S a egic
In o ma ion Sys ems 26(3):191–209. h ps://doi.o g/10.1016/j.jsis.2017.07.003.
Hallikas, Jukka, I is Ka onen, U ho Pulkkinen, Veli-Ma i Vi olainen, and Ma kku Tuominen. 2004.
Risk managemen p ocesses in supplie ne wo ks. In e na ional Jou nal o P oduc ion Economics
90(1):47–58. h ps://doi.o g/10.1016/j.ijpe.2004.02.007.
Hanzlik, Lucjan, Yang Zhang, Ka h in G osse, Ahmed Salem, Maxmilian Augus in, Michael Backes, and
Ma io F i z. 2021. MLCapsule: gua ded O line deploymen o machine lea ning as a se ice.P o-
ceedings o he IEEE/CVF Con e ence on Compu e Vision and Pa e n Recogni ion (CVPR) Wo k-
shops, 3300–3309.
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 393
Ha mann, Philipp M., Mohamed Zaki, Niels Feldmann, and Andy Neely. 2016. Cap u ing alue om
big da a—a axonomy o da a-d i en business models used by s a -up i ms. In e na ional Jou nal o
Ope a ions & P oduc ion Managemen 36(10):1382–1406. h ps://doi.o g/10.1108/IJOPM-02-2014-
0098.
He nandez, Exequie, G. Sande s, and Anja Tuschke. 2015. Ne wo k de ense: p uning, g a ing, and
closing o p e en leakage o s a egic knowledge o i als. Academy o Managemen Jou nal
58(4):1233–1260. h ps://doi.o g/10.5465/amj.2012.0773.
Hi , Robin, and Niklas Kühl. 2018. Cogni ion in he e a o sma se ice sys ems: in e -o ganiza ional
analy ics h ough me a and ans e lea ning. In P oceedings o he 39 h In e na ional Con e ence on
In o ma ion Sys ems—B idging he In e ne o People, Da a, and Things. F ancisco., ed. Jan P ies-
Heje, Sudha Ram, and Michael Rosemann
Hunke, Fabian, Ch is ian Engel, Ronny Schü i z, and Philipp Ebel. 2019. Unde s anding he ana omy
o analy ics-based se ices: a axonomy o concep ualize he use o da a and Analy ics in se ice.
In P oceedings o he 27 h Eu opean Con e ence on In o ma ion Sys ems—In o ma ion Sys ems o
a Sha ing Socie y. S ockholm Uppsala., ed. Jan Vom B ocke, Shi ley G ego , and Oli e Mülle
Il onen, Ilona, S e an Thalmann, Ma kus Manha , and Ch is ian Sillabe . 2018. Reconciling digi al ans-
o ma ion and knowledge p o ec ion: a esea ch agenda. Knowledge Managemen Resea ch & P ac-
ice 16(2):235–244. h ps://doi.o g/10.1080/14778238.2018.1445427.
Jennex, Mu ay E., and Suzanne Zyngie . 2007. Secu i y as a con ibu o o knowledge managemen suc-
cess. In o ma ion Sys ems F on ie s 9(5):493–504. h ps://doi.o g/10.1007/s10796-007-9053-4.
Jiang, Xu, Bao Yongchuan, Yan Xie, and Shanxing Gao. 2016. Pa ne us wo hiness, knowledge low in
s a egic alliances, and i m compe i i eness: a con ingency pe spec i e. Jou nal o Business Resea ch
69(2):804–814. h ps://doi.o g/10.1016/j.jbus es.2015.07.009.
Kaise , Rene, S e an Thalmann, and Vik o ia Pamme -Schindle . 2021. An in es iga ion o knowledge p o-
ec ion p ac ices in in e -o ganisa ional collabo a ion: p o ec ing specialised enginee ing knowledge
wi h a p ac ice based on g ey-box modelling. VINE Jou nal o In o ma ion and Knowledge Manage-
men Sys ems 51(5):713–731. h ps://doi.o g/10.1108/VJIKMS-11-2019-0180.
Kaissis, Geo gios A., Ma cus R. Makowski, Daniel Rücke , and Rickme F. B a en. 2020. Secu e, p i-
acy-p ese ing and ede a ed machine lea ning in medical imaging. Na u e Machine In elligence
2(6):305–311. h ps://doi.o g/10.1038/s42256-020-0186-1.
Kale, P ashan , Ha bi Singh, and Howa d Pe lmu e . 2000. Lea ning and p o ec ion o p op ie a y asse s
in s a egic alliances: building ela ional capi al. S a egic Managemen Jou nal 21(3):217–237.
Kanbach, Dominik K., Louisa Heiduk, Geo g Bluehe , Maximilian Sch ei e , and Alexande Lahmann.
2023. The GenAI is ou o he bo le: gene a i e a i icial in elligence om a business model inno a-
ion pe spec i e. Re iew o Manage ial Science h ps://doi.o g/10.1007/s11846-023-00696-z.
Khan, F eeha, Jung H. Kim, La s Ma hiassen, and Robin Moo e. 2021. Da a b each managemen : an
in eg a ed isk model. In o ma ion & Managemen 58(1):103392. h ps://doi.o g/10.1016/j.im.2020.
103392.
Kühne, Babe , and Tilo Böhmann. 2019. Da a-d i en business models: building he b idge be ween da a
and alue. In P oceedings o he 27 h Eu opean Con e ence on In o ma ion Sys ems—In o ma ion
Sys ems o a Sha ing Socie y. S ockholm Uppsala., ed. Jan Vom B ocke, Shi ley G ego , and Oli e
Mülle
Leski, Flo ian, Michael F uhwi h, and Vik o ia Pamme -Schindle . 2021. Who Else do you need o
a da a-d i en business model? Explo ing oles and exchanged alues. In 34 h bled econ e ence digi al
suppo om c isis o p og essi e change. June 27–30, 2021., ed. And eja Puciha , Mi jana Kljaji´
c
Bo š na , Roge Bons, Helen C ipps, Anand Sheomba , and Do o eja Vidma , 365–378.
Loebbecke, Claudia, Paul C. an Fenema, and Philip Powell. 2016. Managing in e -o ganiza ional knowl-
edge sha ing. The Jou nal o S a egic In o ma ion Sys ems 25(1):4–14. h ps://doi.o g/10.1016/j.jsis.
2015.12.002.
Manha , Ma kus, and S e an Thalmann. 2015. P o ec ing o ganiza ional knowledge: a s uc u ed li e a u e
e iew. Jou nal o Knowledge Managemen 19(2):190–211. h ps://doi.o g/10.1108/JKM-05-2014-
0198.
May ing, Philipp. 2015. Quali a i e Inhal sanalyse: G undlagen und Techniken, 12 h edn., Weinheim:
Bel z.
Mio andi, Daniele, Sab ina Sica i, F ancesco de Pelleg ini, and Im ich Chlam ac. 2012. In e ne o hings:
Vision, applica ions and esea ch challenges. Ad Hoc Ne wo ks 10(7):1497–1516. h ps://doi.o g/10.
1016/j.adhoc.2012.02.016.
Mölle , F ede ik, Maleen S achon, Ch is ina Ho mann, Hen ik Bauhaus, and Bo is O o. 2020. Da a-
d i en business models in logis ics: a axonomy o op imiza ion and isibili y se ices. In P oceed-
K
394 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
ings o he 53 d Annual Hawaii In e na ional Con e ence on Sys em Sciences (HICSS2020), ed. Tung
Bui, 5379–5388.
Mu ay, Alex, Jen Rhyme , and Da id G. Si mon. 2021. Humans and echnology: o ms o conjoined
agency in o ganiza ions. Academy o Managemen Re iew 46(3):552–571. h ps://doi.o g/10.5465/
am .2019.0186.
Oh, Seong J., Max Augus in, Be n Schiele, and Ma io F i z. 2019. Towa ds e e se-enginee ing black-
box neu al ne wo ks. In Explainable AI: In e p e ing, Explaining and Visualizing Deep Lea ning,ed.
Wojciech Samek, G égoi e Mon a on, And ea Vedaldi, La s Kai Hansen, and Klaus-Robe Mülle ,
121–144. Cham: Sp inge .
Op iel, Sebas ian, F ede ik Mölle , U e Bu kha d , and Bo is O o. 2021. Requi emen s o usage con ol
based exchange o sensi i e da a in au omo i e supply chains. In P oceedings o he 54 h Annual
Hawaii In e na ional Con e ence on Sys em Sciences, ed. Tung Bui, 431–440.
Os e walde , Alexande , and Y es Pigneu . 2010. Business model gene a ion: a handbook o isiona ies,
game change s, and challenge s, 1s edn., Hoboken: Wiley.
Os e walde , Alexande , Y es Pigneu , and Ch is ophe L. Tucci. 2005. Cla i ying business models: o i-
gins, p esen , and u u e o he concep . Communica ions o he Associa ion o In o ma ion Sys ems
h ps://doi.o g/10.17705/1CAIS.01601.
Pe o , B uce E. 2007. A s a egic isk app oach o knowledge managemen . Business Ho izons
50(6):523–533. h ps://doi.o g/10.1016/j.busho .2007.08.002.
Rashed, Faisal, Paul D ews, and Mohamed Zaki. 2022. A e e ence model o da a-d i en business model
inno a ion ini ia i es in incumben i ms. P oceedings o he Thi ie h Eu opean Con e ence on In-
o ma ion Sys ems (ECIS 2022), Timi,
soa a.
Regazzoni, F ancesco, Paolo Palmie i, Fe hulah Smailbego ic, Rosa io Camma o a, and Ilia Polian. 2021.
P o ec ing a i icial in elligence IPs: a su ey o wa e ma king and inge p in ing o machine lea n-
ing. CAAI T ansac ions on In elligence Technology 6(2):180–191. h ps://doi.o g/10.1049/ci 2.12029.
San hosh, Gau ham, Fab izio de Vi a, Da io B uneo, F ancesco Longo, and An onio Pulia i o. 2019. To-
wa ds us less p edic ion-as-a-se ice. 2019 IEEE In e na ional Con e ence on Sma Compu ing
(SMARTCOMP)., 317–322. h ps://doi.o g/10.1109/SMARTCOMP.2019.00068.
Schä e , Fabian, Heiko Gebaue , Ch is oph G öge , Oli e Gassmann, and Felix Wo mann. 2023a. Da a-
d i en business and da a p i acy: Challenges and measu es o p oduc -based companies. Business
Ho izons 66(4):493–504. h ps://doi.o g/10.1016/j.busho .2022.10.002.
Schä e , Fabian, Je emy Rosen, Ch is ian Zimme mann, and Felix Wo mann. 2023b. Unleashing he po-
en ial o da a ecosys ems: es ablishing digi al us h ough us -enhancing echnologies. Thi y-
i s Eu opean Con e ence on In o ma ion Sys ems (ECIS 2023).
Schü i z, Ronny, S e an Seebache , and Rebecca Do ne . 2017a. Cap u ing alue om da a: e enue mod-
els o da a-d i en se ices. In P oceedings o he 50 h Hawaii In e na ional Con e ence on Sys em
Sciences. Waikoloa Village., ed. Tung Bui, 5348–5357.
Schü i z, Ronny, S e an Seebache , Ge ha d Sa zge , and Lukas Schwa z. 2017b. Da a iza ion as he nex
on ie o Se i iza ion: unde s anding he challenges o ans o ming o ganiza ions. P oceedings
o he Thi y-Eigh h In e na ional Con e ence on In o ma ion Sys ems (ICIS), Seoul.
Schü i z, Ronny, Killian Fa ell, Ba ba a H. Wixom, and Ge ha d Sa zge . 2019. Value co-c ea ion in
da a-d i en se ices: owa ds a deepe unde s anding o he join sphe e. P oceedings o he Fo ie h
In e na ional Con e ence on In o ma ion Sys ems (ICIS), Munich.
Schweiho , Julia, Ilka Jussen, Valen in Dahms, F ede ik Mölle , and Bo is O o. 2023. How o sha e
da a Online ( as )—A axonomy o da a sha ing business models. P oceedings o he 56 h Hawaii
In e na ional Con e ence on Sys ems Sciences (HICSS).
Shollo, A isa, Kons an in Hop , Tiemo Thiess, and Oli e Mülle . 2022. Shi ing ML alue c ea ion
mechanisms: a p ocess model o ML alue c ea ion. The Jou nal o S a egic In o ma ion Sys ems
31(3):101734. h ps://doi.o g/10.1016/j.jsis.2022.101734.
S achon, Maleen, F ede ik Mölle , Mo i z T. Guggenbe ge , and Ma in Tomczyk. 2023. Unde s and-
ing da a us s. P oceedings o he Thi y- i s Eu opean Con e ence on In o ma ion Sys ems (ECIS
2023), K is iansand.
S ahl, Bas ian, Bjö n Häckel, Daniel Leu he, and Ch is ian Ri e . 2023. Da a o business i s ?—manu ac-
u e s’ ans o ma ion owa d da a-d i en business models. Schmalenbachs Zei sch i ü be iebs-
wi scha liche Fo schung 75:303–343. h ps://doi.o g/10.1007/s41471-023-00154-2.
S obel, Ge o, F ede ik Mölle , Tho s en Schoo mann, and Bo is O o. 2024. In oduc ion o he 2nd Mini-
T ack on Designing Da a Ecosys ems: Values, Impac s, and Fundamen als. In P oceedings o he 57 h
Annual Hawaii In e na8onal Con e ence on Sys em Sciences, ed. Tung X. Bui, 4236–4237.
K
Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396 395
Sun, Tianxiang, Shao Yun an, Huang Xuanjing, and Xipeng Qiu. 2022. Black-Box Tuning o Lan-
guage-Model-as-a-Se ice. P oceedings o he 39 h In e na ional Con e ence on Machine Lea ning,
20841–20855.
Teece, Da id J. 2010. Business models, business s a egy and inno a ion. Long Range Planning
43(2–3):172–194. h ps://doi.o g/10.1016/j.l p.2009.07.003.
Tesch, Jan, Anne-Sophie B illinge , and Dominik Bilge i. 2017. In e ne o hings business model in-
no a ion and he s age-ga e p ocess: an explo a o y analysis. In e na ional Jou nal o Inno a ion
Managemen 21(5):1740002-1–1740002-17. h ps://doi.o g/10.1142/S1363919617400023.
Thalmann, S e an, Ronald Maie , Ul ich Remus, and Ma kus Manha . 2024. Connec wi h ca e: in o mal
knowledge p o ec ion p ac ices o enhance knowledge sha ing in ne wo ks o o ganiza ions. VINE
Jou nal o In o ma ion and Knowledge Managemen Sys ems, Vol. ahead-o -p in No. ahead-o -p in .
h ps://doi.o g/10.1108/VJIKMS-02-2022-0051
Thiel, Ch is ian, and Ch is oph Thiel. 2015. Ha e and o oise: can indus y 4.0 win he ace agains coun-
e ei ing and pi acy? Da enschu z und Da ensiche hei 1(0):663–667.
Thomas, Llewellyn D.W., Aija Leiponen, and Pan elis Kou oumpos. 2023. P o i ing om da a p oduc s. In
Resea ch handbook on digi al s a egy, ed. Ca melo Cennamo, Gio anni Ba is a Feng Zhu Dagnino,
255–272. Chel enham: Edwa d Elga Publishing.
T amè , Flo ian, Fan Zhang, A i Juels, Michael K. Rei e , and Thomas Ris enpa . 2016. S ealing ma-
chine lea ning models ia p edic ion APis. P oceedings o he 25 h USENIX Secu i y Symposium,
601–618.
T edinnick, Luke, and Clai e Layba s. 2023. The dange s o gene a i e a i icial in elligence. Business
In o ma ion Re iew 40(2):46–48. h ps://doi.o g/10.1177/0266382123118375.
Vesselko , Alexand , Heikki Hämmäinen, and Juuso Töyli. 2019. Design and go e nance o mheal h da a
sha ing. Communica ions o he Associa ion o In o ma ion Sys ems 45(1):299–321. h ps://doi.o g/
10.17705/1CAIS.04518.
Ve e , Oli e A., Felix S. Ho mann, Luisa Pumplun, and Pe e Buxmann. 2022. Wha cons i u es a ma-
chine-lea ning-d i en business model? A axonomy o B2B s a -ups wi h machine lea ning a hei
co e. P oceedings o he Thi ie h Eu opean Con e ence on In o ma ion Sys ems (ECIS 2022),
Timi ,
soa a.
Webe , Michael, Mo i z Beu e , Jö g Weking, Ma kus Böhm, and Helmu K cma . 2022. AI s a up busi-
ness models. Business & In o ma ion Sys ems Enginee ing 64(1):91–109. h ps://doi.o g/10.1007/
s12599-021-00732-w.
Wiene , Ma in, Ca ol Saunde s, and Ma co Ma abelli. 2020. Big-da a business models: a c i ical li e a-
u e e iew and mul ipe spec i e esea ch amewo k: a c i ical li e a u e e iew and mul i-pe spec-
i e esea ch amewo k. Jou nal o In o ma ion Technology 35(1):66–91. h ps://doi.o g/10.1177/
0268396219896811.
Woe ne , S ephanie L., and Ba ba a H. Wixom. 2015. Big da a: ex ending he business s a egy oolbox.
Jou nal o In o ma ion Technology 30(1):60–62. h ps://doi.o g/10.1057/ji .2014.31.
Yakoubo , Sophia, Vijay Gadepally, Nabil Schea , Emily Shen, and A kady Ye ukhimo ich. 2014. Asu -
ey o c yp og aphic app oaches o secu ing big-da a analy ics in he cloud. 2014 IEEE High Pe o -
mance Ex eme Compu ing Con e ence (HPEC), 1–6. h ps://doi.o g/10.1109/HPEC.2014.7040943.
Yin, Robe K. 2009. Case s udy esea ch: design and me hods, 4 h edn., Los Angeles: SAGE.
Zei inge , Johannes P. 2021. Tackling knowledge isks in da a-cen ic collabo a ions: a ackling knowledge
isks in da a-cen ic collabo a ions: a li e a u e e iew li e a u e e iew. PACIS 2021 P oceedings.
Zei inge , Johannes P., and S e an Thalmann. 2020. Knowledge isks in digi al supply chains: a li e a-
u e e iew. In WI2020 Zen ale T acks: 15 h In e na ional Con e ence on Wi scha sin o ma ik.
Po sdam, Ma ch 9–11, 2020., ed. No be G onau, Mo een Heine, K. Pous cchi, and H. K asno a,
370–385. GITO Ve lag. 15. In e na ionale Tagung Wi scha sin o ma ik.
Zei inge , Johannes P., and S e an Thalmann. 2022. Knowledge sha ing and p o ec ion in da a-cen ic
collabo a ions: an explo a o y s udy. Knowledge Managemen Resea ch & P ac ice 20(3):436–448.
h ps://doi.o g/10.1080/14778238.2021.1978886.
Zei inge , Johannes P., S e an Thalmann, and Jü gen Fleiss. 2024. Da a anonymiza ion as ins umen o
manage knowledge isks in supply chains. In P oceedings o he 57 h Annual Hawaii In e na ional
Con e ence on Sys em Sciences, ed. Tung X. Bui, 5503–5512.
Zeng, Yong, Wang Lingyu, Deng Xiaoguang, Cao Xinlin, and Na isa Khundke . 2012. Secu e collabo a ion
in global design and supply chain en i onmen : P oblem analysis and li e a u e e iew. Compu e s in
Indus y 63(6):545–556. h ps://doi.o g/10.1016/j.compind.2012.05.001.
K
396 Schmalenbachs Zei sch i ü be iebswi scha liche Fo schung (2024) 76:357–396
Zhang, Da Yong, Cao Xinlin, Wang Lingyu, and Yong Zeng. 2012. Mi iga ing he isk o in o ma ion
leakage in a wo-le el supply chain h ough op imal supplie selec ion. Jou nal o In elligen Manu-
ac u ing 23(4):1351–1364. h ps://doi.o g/10.1007/s10845-011-0527-3.
Publishe ’s No e Sp inge Na u e emains neu al wi h ega d o ju isdic ional claims in published maps
and ins i u ional a ilia ions.
K
