Toward LLM-enabled business process coherence checking based on multi-level process documentation

Schul e, Ma ek; F anzoi, Sand o; Köhne, F ank; om B ocke, Jan
A icle — Published Ve sion
Towa d LLM-enabled business p ocess cohe ence
checking based on mul i-le el p ocess documen a ion
P ocess Science
Sugges ed Ci a ion: Schul e, Ma ek; F anzoi, Sand o; Köhne, F ank; om B ocke, Jan (2025) : Towa d
LLM-enabled business p ocess cohe ence checking based on mul i-le el p ocess documen a ion,
P ocess Science, ISSN 2948-2178, Sp inge In e na ional Publishing, Cham, Vol. 2, Iss. 1,
h ps://doi.o g/10.1007/s44311-025-00024-6
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/333235
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 p://c ea i ecommons.o g/licenses/by/4.0/
RESEARCH Open Access
© The Au ho (s) 2025. 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 Commons 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/.
Schul e e al. P ocess Science (2025) 2:22
h ps://doi.o g/10.1007/s44311-025-00024-6
*Co espondence:
Sand o F anzoi
sand o. anzoi@uni-muens e .de
1Uni e si y o Müns e , Müns e ,
Ge many
2Eu opean Resea ch Cen e o
In o ma ion Sys ems, Müns e ,
Ge many
3 iadee Un e nehmensbe a ung
AG, Müns e , Ge many
Towa d LLM-enabled business p ocess
cohe ence checking based on mul i-le el
p ocess documen a ion
Ma ekSchul e1, Sand oF anzoi1,2*, F ankKöhne3 and Jan om B ocke1,2
In oduc ion
De ia ions om expec ed p ocess beha io pe sis in i ually e e y o ganiza ion. As
such, unde s anding and add essing p ocess de iance is a c i ical conce n o bo h busi-
ness p ocess managemen (BPM) esea ch and p ac ice (König e al. 2019). In he con-
ex o oday’s dynamic business en i onmen s, de ia ions om es ablished p ocesses
a e no me ely challenges; hey ep esen po en ial oppo uni ies o signi ican p ocess
imp o emen s and inno a ions (Ba elheime e al. 2023). When e ec i ely iden i ied
and managed, hese de ia ions can enhance ope a ional e iciency and adap abili y, p o-
iding subs an ial alue o businesses (Se iawan and Sadiq 2013; Delias 2017).
Abs ac
In his pape , aP oCheCk, an Au onomous P ocess Cohe ence Checking me hod,
is de eloped. aP oCheCk le e ages la ge language models (LLMs) o enhance
he cohe ence checking o mul i-le el p ocess documen a ion wi hin business
p ocess managemen (BPM). This esea ch add esses he need o au oma ed ways
o managing incohe encies in p ocess documen a ion. The de elopmen o he
a i ac was guided by a design science esea ch app oach, which in ol ed i e a i e
de elopmen and e inemen . This was achie ed h ough expe in e iews wi h
esea che s and p ac i ione s, i e a i e expe imen al benchma king, and ocus
g oup alida ion based on demons a ions o a p o o ypical implemen a ion wi h
na u alis ic da a om di e se indus ies. aP oCheCk can dynamically analyze and
assess changes in BPM documen a ion, de ec incohe encies, and p o ide ac ionable
insigh s o main aining p ocess cohe ence. The indings e eal signi ican po en ial
o imp o ing ope a ional e iciency, educing manual e o , and de ec ing nega i e
and posi i e p ocess a ia ion ea ly o suppo con inuous p ocess inno a ion. This
esea ch con ibu es o he ield o BPM by in eg a ing LLMs in o he BPM li ecycle,
enhancing gene a i e AI-based applica ions wi hin BPM p ac ices, in oducing he
Business P ocess Change Classi ica ion F amewo k, and p o iding an open-sou ce
da ase ha can se e as a ounda ion o u u e esea ch and de elopmen .
Keywo ds La ge language model, P ocess cohe ence, A i icial in elligence, P ocess
de iance, Business p ocess managemen
P ocess Science
Page 2 o 33Schul e e al. P ocess Science (2025) 2:22
In he pas , inconsis encies in p ocess documen a ion we e ypically solely ega ded
as e o s ha needed o be ec i ied ( an de Aa e al. 2017). Howe e , he e is a g ow-
ing ecogni ion ha hese de ia ions can be iewed as oppo uni ies o imp o e-
men (Galpe in 2012; Me ens and Recke 2017). This is illus a ed in he shi owa d
an unde s anding ha BPM needs o balance he en o cemen o p ocess compliance
wi h he iden i ica ion o posi i e de iance (Se iawan and Sadiq 2013; Mendling e al.
2020). Iden i ying and analyzing hese de ia ions can p o ide insigh s ha lead o mo e
e ec i e business ou comes, inno a i e p ocess imp o emen s, o necessa y co ec i e
ac ions. This shi in pe spec i e unde sco es he need o sophis ica ed ools ha can
e icien ly de ec , analyze, and le e age bo h posi i e and nega i e de ia ions o d i e
business g ow h and inno a ion (König e al. 2019).
The eme gence o gene a i e a i icial in elligence (AI), pa icula ly la ge language
models (LLMs), signi ies a ans o ma i e de elopmen in his domain. LLMs and hei
capabili y o gene a e and unde s and human-like ex ende hem pa icula ly sui able
o in e p e ing complex business documen a ion associa ed wi h BPM (Vidgo e al.
2023; Fahland e al. 2024).
The po en ial ole o AI in de ec ing and managing p ocess de ia ions becomes c u-
cial when conside ing he au oma ion and enhancemen o hese asks (Weinzie l e al.
2024). In pa icula , LLMs a e capable o analyzing ex ensi e se s o p ocess documen a-
ion o iden i y inconsis encies and changes ha may indica e de ia ions (Vidgo e al.
2023). Recen esea ch shows ha LLMs a e also inc easingly capable o making com-
plex subjec i e decisions based on he gi en con ex (Binz and Schulz 2023; Mcin osh e
al. 2024). This cha ac e is ic enables he expansion o he scope om de ec ing quan i a-
i e de ia ions, o ins ance, based on e en logs and p ocess diag ams, o encompass
he mo e complex and subjec i e p oblem o checking o incohe encies in ex -based
business p ocess documen a ion.
By de ec ing de ia ions h ough he analysis o changes in he p ocess documen a ion,
LLMs enable o ganiza ions o no only main ain he in eg i y o hei p ocesses bu also
o ake ad an age o hese de ia ions o d i e inno a ion and imp o e e iciency (Vidgo
e al. 2023). This app oach add esses a c i ical esea ch gap iden i ied by Feue iegel e
al. (2024), which emphasizes he po en ial o gene a i e AI me hods o e eal oppo uni-
ies o p ocess inno a ion and suppo p ocess edesign ini ia i es.
This is especially impo an because o ganiza ions equen ly main ain mul iple, in e -
ela ed documen s, o en esul ing in incohe encies be ween hem – a p oblem magni-
ied by changes made o indi idual documen s ha a e no e lec ed consis en ly ac oss
all ela ed documen s (Ma in-To al e al. 2010). This issue is pa icula ly e iden in he
con ex o business p ocess documen a ion, as exp essed, o example, in he p esence
o con en inconsis encies be ween p ocess desc ip ions and he co esponding p o-
cess models ( an de Aa e al. 2017). This sugges s ha analyzing changes in p ocess
documen a ion and compa ing hem wi h o he ela ed p ocess documen s may e eal
incohe encies, which could po en ially be le e aged o de ec de iance and imp o e p o-
cesses as e and mo e e icien ly in o ganiza ions.
To add ess his p oblem, we o mula e he ollowing wo esea ch objec i es, which
aim o explo e he in eg a ion o LLMs wi hin p ocess cohe ence checking based on
p ocess documen a ion:
Page 3 o 33Schul e e al. P ocess Science (2025) 2:22
RO1: De ine he essen ial design objec i es and speci ica ions o an LLM-enabled
p ocess cohe ence checking me hod based on changes o p ocess- ela ed documen a-
ion.
RO2: De elop and e alua e a solu ion o employing LLMs o au oma icallye alu-
a e he consis ency and cohe ence ac oss di e en e sions o mul ipledocumen a-
ion ypes o business p ocesses.
Add essing hese esea ch objec i es, we ollowed a design science esea ch (DSR)
app oach and de eloped an a i ac called aP oCheCk, an ac onym o Au onomous
P ocess Cohe ence Checking. This a i ac ep esen s a me hod ha is designed o
le e age LLMs o enhance he enac men and e alua ion phases o he BPM li ecycle by
assessing he consis ency and cohe ence o mul i-le el p ocess documen a ion. The co e
unc ionali y o aP oCheCk is o au oma e p ocess cohe ence checking, he eby enabling
o ganiza ions o iden i y and manage bo h nega i e and posi i e de iance wi hin hei
p ocesses. Speci ically, aP oCheCk compa es wo en i ies o p ocess documen a ion
(e.g., p ocess model and p ocess desc ip ion o p ocess model and egula o y documen ,
among o he s) and hei change o e ime o iden i y incohe encies and subsequen ly
c ea e no i ica ions (see no i ica ion c ea ion in Fig.2), p esen ed in he o m o man-
agemen summa ies, o in o m p ocess owne s. We ins an ia e a so wa e p o o ype o
demons a e and e alua e he u ili y o ou a i ac in p ac ice. As such, ou esea ch
ma ks a undamen al s ep owa d cohe ence checking based on mul i-le el p ocess doc-
umen a ion. The iden i ica ion and assessmen o incohe encies allows o ganiza ions o
no only ec i y e o s bu also o le e age posi i e de ia ions as oppo uni ies o s a e-
gic ad ancemen and inno a ion.
Following es ablished guidelines on s uc u ing DSR (G ego and He ne 2013), we
p esen he emainde o he pape as ollows. Fi s , we in oduce he esea ch back-
g ound on p ocess cohe ence checking as well as gene a i e AI in BPM. Nex , we ou line
ou me hodological app oach and i s espec i e phases. Then we desc ibe ou de eloped
a i ac and p o ides in o ma ion on i s design and de elopmen . A e wa ds, we p esen
ou expe imen al and ocus g oup e alua ions. Finally, we discuss heo e ical and p ac i-
cal implica ions as well as he limi a ions o ou wo k be o e we conclude he pape .
Resea ch backg ound
To p o ide a comp ehensi e ounda ion o ou wo k, he esea ch backg ound sec ion
is s uc u ed in o wo subsec ions. Fi s , we in oduce he concep o p ocess cohe ence
checking, highligh ing he challenges o main aining seman ic consis ency ac oss mul i-
le el p ocess documen a ion and he limi a ions o adi ional app oaches. Second, we
shi he ocus o gene a i e AI in business p ocess managemen , ou lining how ecen
ad ances, pa icula ly in LLMs, a e opening up new oppo uni ies o add essing chal-
lenges in BPM.
P ocess cohe ence checking
P ocess cohe ence ep esen s a i al concep o ensu ing logical consis ency ac oss
business p ocesses and hei ela ed documen a ion. In he con ex o in o ma ion
sys ems, he an onym o cohe ence, incohe ence, p o ides a use ul poin o e e ence.
Sain -Dizie (2018), o ins ance, de ines incohe ence as linguis ic disc epancies be ween
documen s. Ma in-To al e al. (2008) de ine con en incohe ence as “[…] he weakness
Page 4 o 33Schul e e al. P ocess Science (2025) 2:22
o consis ency amongs ela ed documen s, o amongs di e en pieces o he same doc-
umen , o he lack o excess o in o ma ion in a documen ” (p. 283). In a b oade sense,
cohe ence is essen ial o p ese ing logical consis ency ac oss sys ems, p ocesses, and
documen a ion (Ma in-To al e al. 2010). I ensu es ha a ious componen s in e e-
la e cohe en ly, enabling clea communica ion and ope a ional e iciency. This p inciple
o cohe ence is pa icula ly impo an gi en he ex ensi e and complex documen a ion
ypical o business p ocesses. The concep o p ocess de iance (Me ens and Recke
2017) is closely ela ed. I e e s o use ac ions ha de ia e om he in ended p ocess
low and can be s udied h ough a ious me hods, including p ocess mining (Di F an-
cescoma ino e al. 2025). These de ia ions illus a e how use s in e ac wi h sys ems in
unp edic able o unin ended ways, leading o posi i e o nega i e e ec s (Delias 2017).
These use -d i en de ia ions may esul in upda es o ope a ional p ocess documen s,
such as aining ma e ials, o e lec ac ual p ac ices. Howe e , o he documen s ha a e
less use -cen ic, such as p ocess models, may no be upda ed synch onously, esul ing
in po en ial inconsis encies. In con as o p ocess de iance, incohe ence is mo e closely
associa ed wi h gene al inconsis encies wi hin he documen a ion o p ocesses, which
may occu independen ly o any use -d i en ac ions (Ma in-To al e al. 2010). Examples
o such incohe encies ha do no o igina e om use -d i en ac ions include changes o
o e a ching documen s such as egula ions o guidelines, o ganiza ional upda es such as
ole changes, and so wa e o ha dwa e upg ades (Rosemann e al. 2008).
The in oduc ion o mul i-le el p ocess documen a ion adds u he complexi y.
The a ious ypes o documen a ion, including p ocess models, ex ual desc ip ions,
guidelines, and policies, no only di e in o m bu also in pe spec i e, le el o abs ac-
ion, and s a e o knowledge (Poly yanyy e al. 2015; Rosemann and om B ocke 2015;
Nwankpa e al. 2022). Some documen s may add ess only speci ic aspec s o a p ocess,
whe eas o he s may deal wi h pa icula abs ac ion laye s, such as he da abase o s a-
egic abs ac ion laye s. Simila ly, in e nal o ex e nal egula ions, as well as modeling
con en ions, o en span mul iple p ocesses and documen a ion ypes, placing hem a a
highe abs ac ion le el. Ensu ing cohe ence ac oss hese di e se documen s is essen ial
o e ec i e BPM.
Based on hese insigh s, p ocess cohe ence in BPM can be de ined as seman ic con-
sis ency ac oss mul i-le el p ocess documen a ion. Main aining his cohe ence ensu es
ha documen s ela ed o a p ocess emain logically aligned, despi e di e ences in pe -
spec i e and g anula i y. This alignmen is c i ical o a oid miscommunica ion and ine -
iciencies and o suppo a cohesi e wo king en i onmen .
To add ess hese challenges, he deploymen o ad anced echnologies, pa icula ly
LLMs, enables o ganiza ions o au oma e he cohe ence checking o hei p ocess docu-
men a ion, he eby capi alizing on he echnological ad ancemen s o e exis ing me h-
ods. Fo example, Ma in-To al e al. (2008) in es iga ed he de ec ion o incohe encies
in a co pus o echnical and egula o y documen s using adi ional me hods such as
ex ac ion echniques in ol ing ex and documen mining. Recen ad ances, as exem-
pli ied by he wo k o Sai e al. (2023), highligh he impo ance o b idging he gaps
be ween o ganiza ional documen a ion and egula o y ex s h ough he applica ion o
ad anced NLP echniques. Thei s udy ocused on he compa ison o gene al business
documen s wi h egula o y amewo ks wi hou add essing he speci ics o BPM- ela ed
documen s o he ecen de elopmen s in LLM echnology. The in eg a ion o LLMs in

Page 5 o 33Schul e e al. P ocess Science (2025) 2:22
his con ex no only suppo s egula o y compliance bu also imp o es o e all p ocess
cohe ence and eliabili y.
Gene a i e AI in business p ocess managemen
The p og ession o gene a i e AI has ede ined he po en ial applica ions o AI in a
mul i ude o domains, including BPM (Kampik e al. 2024). By enabling mo e in ui i e,
adap i e, and c ea i e o ms o p ocess in e ac ion and design, gene a i e AI ex ends
he adi ional bounda ies o BPM beyond au oma ion owa d con e sa ional, au ono-
mous, and sophis ica ed p ocess capabili ies (Rosemann e al. 2024). Fo example, an
Dun e al. (2023) le e aged gene a i e ad e sa ial ne wo ks o help p ocess designe s in
he c ea ion o business p ocess imp o emen ideas. In ano he pape , Ha l e al. (2024)
showcase how gene a i e machine lea ning can be used o au oma ed business p ocess
edesign a un ime. Resea ch has also in es iga ed he oppo uni ies and challenges o
na u al language p ocessing o BPM ( an de Aa e al. 2018) and explo ed i s applica-
ion, o ins ance, in p edic i e business p ocess moni o ing (Teinemaa e al. 2016).
Mo e ecen ly, he e olu ion o ans o me a chi ec u es has ma ked ano he key
shi in he ield o gene a i e AI in BPM. He e, a key de elopmen was he eme gence
o ools such as Cha GPT, which employ LLMs ained on ex ensi e da ase s o gene -
a e con ex ually ele an esponses o use que ies (Vidgo e al. 2023). In he con ex
o u u e de elopmen s, he po en ial applica ions o gene a i e AI in BPM a e nume -
ous and di e se (Kampik e al. 2024). One a ea o signi ican g ow h is he de elopmen
o new gene a ions o p ocess guidance sys ems (Feue iegel e al. 2024). In con as
o adi ional sys ems ha ely on s a ic, manually c a ed knowledge bases, new sys-
ems powe ed by gene a i e AI can dynamically e ie e in o ma ion om a wide a ay
o s uc u ed and uns uc u ed da a sou ces, including emails, manuals, and co po a e
documen s (Mo ana e al. 2019; Feue iegel e al. 2024). Such sys ems can guide a wide
ange o business p ocess managemen asks, such as modeling (Kou ani e al. 2024),
knowledge managemen (F anzoi e al. 2025b), o analysis suppo in p ocess mining
(B ü zke e al. 2025). This enables he p o ision o eal- ime, con ex -sensi i e guidance,
making hese sys ems mo e adap i e and in elligen . This shi om s a ic o dynamic
guidance sys ems indica es a b oade mo e owa d mo e esponsi e and in elligen BPM
ools ha g ea ly enhance decision-making and op imiza ion ac oss o ganiza ional le -
els. As highligh ed by Feue iegel e al. (2024), he e is a con inuing need o explo e how
gene a i e AI can un eil new oppo uni ies o p ocess inno a ion. Thei wo k sugges s
ha u he in es iga ion in o he capabili ies o gene a i e AI could e eal ans o ma-
i e app oaches o BPM, ede ining ope a ional s a egies and compe i i e dynamics in
nume ous indus ies. As such, he e is s ill ample po en ial o employ LLMs o assess
p ocess de iance o p ocess cohe ence ac oss documen s.
Resea ch me hodology
To add ess ou esea ch objec i es, we employ a mul i ace ed me hodology, cen e ing
on he applicabili y o an eme ging echnology o add ess a p ac ical issue. The ounda-
ional me hodological app oach adop ed is DSR, which aims a iden i ying a p oblem
and de eloping an a i ac designed o mi iga e he issue wi hin p ede ined pa ame e s
(He ne e al. 2004).
Page 6 o 33Schul e e al. P ocess Science (2025) 2:22
Speci ically, we ollow he guidelines p oposed by Pe e s e al. (2007) and Tuunanen
e al. (2024), which p o ide a comp ehensi e amewo k o ca ying ou esea ch ha is
cen e ed a ound he c ea ion and e alua ion o IT a i ac s in ended o sol e iden i ied
p oblems. The p ocess p oposed by Pe e s e al. (2007) is di ided in o six key phases:
p oblem iden i ica ion, de ini ion o objec i es, and design and de elopmen as Build
phases, ollowed by demons a ion, e alua ion, and communica ion o esea ch indings
as E alua e phases. The app oach is illus a ed in Sec ion A in he online appendix1.
In adap ing he DSR amewo k o his esea ch, speci ic modi ica ions we e made
o ailo he app oach o he unique equi emen s and cons ain s o his s udy. The aim
was o enhance he ocus on he p ac ical applica ion o eme ging echnology in sol -
ing eal-wo ld p oblems. This cus omized app oach allowed o a mo e a ge ed de el-
opmen and e alua ion o he a i ac , ensu ing ha he esea ch ou comes a e bo h
p ac ical and heo e ically sound. Addi ionally, we also inco po a ed p inciples o an
i e a i e app oach, including equen e alua ion and adap a ion (Tuunanen e al. 2024).
Ou ins an ia ion o he DSR app oach is depic ed in Fig.1. The e alua ion phase in he
DSR app oach is c i ical o alida ing ha he de eloped a i ac mee s bo h heo e i-
cal expec a ions and p ac ical u ili y in eal-wo ld applica ions. This duali y ensu es ha
a i ac s a e obus in hei heo e ical unde pinnings and highly unc ional in p ac ical
scena ios, indica ing a success ul b idging o heo y and p ac ice (He ne e al. 2004).
Ou e alua ion s a egy employs a s uc u ed app oach, u ilizing a ious me hodologies
such as e iewing li e a u e, expe in e iews, ocus g oups, o expe imen al bench-
ma king (Sonnenbe g and om B ocke 2012). This s a egy is designed o align wi h
a ious e alua ion ypes as iden i ied in he DSR amewo k. Impo an ly, we emphasize
he i e a i e na u e o ou DSR app oach by highligh ing he ecu ing cycles be ween
design and de elopmen , as well as demons a ion and e alua ion. He e, phase 3.a, 3.b,
and 3.c ep esen design and de elopmen ac i i ies, and 4.a, 4.b, and 4.c ep esen dem-
ons a ion and e alua ion ac i i ies (see Fig.1). The ollowing pa ag aphs ou line he
speci ic s eps o ou app oach.
1 The ull online appendix wi h all ele an documen s is a ailable he e: h p s : / / g i h u b . c o m / i a d e e / p o c e s s - d o c u m e n
- c o h e e n c e - c h e c k e .
Fig. 1 Applied Design Science Resea ch App oach
Page 7 o 33Schul e e al. P ocess Science (2025) 2:22
Phase 1. Iden i y p oblem and mo i a e (E al 1)
In he ini ial phase, he unde lying p oblem is ca ed ou and discussed. Fo his pu -
pose, exis ing li e a u e is examined o iden i y ele an esea ch, es ablish he heo e i-
cal ounda ion o he wo k, and jus i y he p oblem s a emen .
Phase 2. De ine objec i es o a solu ion
Based on he iden i ica ion o he p oblem and he ele an esea ch iden i ied in he
p e ious phase, he scope o he a ge ed a i ac is de ined, and p o isional design
objec i es a e de i ed om he li e a u e.
Phase 3. Design and De elopmen
Phase 3.a design speci ica ions and ini ial p oo o concep (PoC)
Guided by he es ablished p o isional design objec i es, an exempla y applica ion is
composed u ilizing exis ing echnologies. Fu he mo e, p o isional design speci ica ions
a e de i ed om he design objec i es.
Phase 3.b design and de elopmen o p o o ype
In acco dance wi h he e ined design speci ica ions, an ini ial wo king p o o ype is
c ea ed.
Phase 3.c adap a ion o he p o o ype
The a i ac is e ined h ough an i e a i e p ocess based on he esul s o he expe i-
men al i e a ions.
Phase 4. Demons a ion and E alua ion
Phase 4.a demons a ion and e alua ion in Semi-s uc u ed expe in e iews (E al 2)
The PoC is demons a ed, and he design objec i es and speci ica ions a e discussed in
six semi-s uc u ed in e iews wi h h ee p ac i ione s in BPM- ela ed oles and h ee
esea che s in he BPM ield. This phase alida es he design objec i es and e ines he
speci ica ions, ocusing on unde s andabili y, easibili y, applicabili y, and ope a ionali y.
Phase 4.b expe imen al e alua ion o p o o ype (E al 3)
The applicabili y o he a i ac is demons a ed h ough i e a i e expe imen s based on
a da ase sou ced om li e a u e and en iched by a BPM expe . This phase ocuses on a
quan i a i e e alua ion o he a i ac ’s e iciency, e ec i eness, and obus ness.
Phase 4.c demons a ion and e alua ion in ocus g oups (E al 4)
The use ulness o he a i ac is e alua ed h ough demons a ions o an ins an ia ion in
wo ocus g oups, each comp ising ou o i e BPM and AI consul an s. Following he
demons a ion, a semi-s uc u ed discussion assesses p ac ical applicabili y, usabili y,
and eal-wo ld in eg a ion.
Phase 5. Communica ion
The con ibu ion o he knowledge base will be achie ed h ough he publica ion o he
esea ch.
Page 8 o 33Schul e e al. P ocess Science (2025) 2:22
This anspa en , s uc u ed, and i e a i e e alua ion app oach ensu es ha he a i ac
adhe es o igo ous academic s anda ds while enhancing BPM p ac ices in di e se ope -
a ional con ex s (He ne e al. 2024). Bo h o ma i e and summa i e e alua ion me hods
a e applied in a i icial as well as na u alis ic se ings (Venable e al. 2016), s eng hening
con idence in he o e all e alua ion me hodology ( om B ocke e al. 2020). By aligning
he in e es s and eedback o bo h esea che s and p ac i ione s, he a i ac is e ined
in o a obus ool ha o e s signi ican heo e ical and p ac ical con ibu ions o he
ield o business p ocess managemen (Sonnenbe g and om B ocke 2012).
A i ac desc ip ion
This sec ion ou lines he de eloped a i ac , aP oCheCk, which ep esen s a me hod
o LLM-enabled p ocess cohe ence checking. We i s p esen an o e iew o he inal
a i ac and i s h ee main s ages –p ep ocessing, con en compa ison, and cohe ence
checking– ollowed by he design a ionale, including objec i es, p oo o concep ,
expe e alua ion, and he Business P ocess Change Classi ica ion F amewo k. We hen
de ail he de elopmen and ope a ing logic, highligh ing p omp enginee ing echniques
and he implemen a ion s a egy.
A i ac o e iew: aP oCheCk
D awing on he p esen ed DSR app oach, his sec ion p esen s an o e iew o he inal
de eloped a i ac , aP oCheCk, which comp ises h ee p ima y s ages: (1) p ep ocessing,
(2) con en compa ison, and (3) cohe ence checking. Each s age can e mina e ea ly i he
documen s a e ound o be cohe en , he eby op imizing e iciency and educing hallu-
cina ion isks. Figu e2 depic s an o e iew o aP oCheCk, om iden i ying and inpu -
ing p ocess documen s o he op ional c ea ion o no i ica ions. The da ke g ey boxes
deno e LLM in e ac ions, while do ed lines indica e modi iable elemen s.
P ep ocessing
The p ep ocessing phase ocuses on noise educ ion by emo ing non- ele an isual
da a om XML ep esen a ions o BPMN iles, which we e iden i ied as unc i ical o
cohe ence checking h ough expe in e iews and es ing. An equali y check hen
de e mines i he p ep ocessed documen s a e iden ical; i so, he wo k low e mina es
ea ly, indica ing cohe ence and conse ing compu a ional esou ces.
Con en compa ison
In he con en compa ison phase, he con en o he wo p ocess documen e sions
is compa ed o iden i y changes. Iden i ied changes a e agg ega ed in o a JSON ele-
men and classi ied acco ding o he Business P ocess Change Dimensions: Task, Da a,
Fig. 2 O e iew o aP oCheCk
Page 15 o 33Schul e e al. P ocess Science (2025) 2:22
p ocess owne is p ima ily in e es ed in all changes o he con ol low, bu he p oduc ion
planne may be mo e in e es ed in he esou ce dimension”.
Sec ion C o he online appendix con ains a mo e de ailed explana ion o he Business
P ocess Change Dimensions and Change Rele ance Ca ego ies, p esen ing a amewo k
ha enhances he e iciency and accu acy o he cohe ence checking mechanism. An
o e iew o his amewo k is depic ed in Table1. By add essing he subjec i e na u e
o change classi ica ion, le e aging he decision-making capabili ies o LLMs, and main-
aining human-cen ic o e sigh , he a i ac aims o mo e e ec i ely manage p ocess
cohe ence and p o ide accu a e and ac ionable no i ica ions o use s.
A i ac de elopmen and ope a ing logic
This sec ion desc ibes he ansi ion om heo e ical design o p ac ical implemen a-
ion o aP oCheCk o LLM-enabled business p ocess cohe ence checking. Impo an ly,
while aP oCheCk cons i u es a gene al me hod o LLM-enabled p ocess cohe ence
checking based p ocess documen a ion, we also ins an ia e a so wa e a i ac o dem-
ons a e and e alua e aP oCheCk in a eal-wo ld scena io. The code o he ins an ia-
ion is a ailable in Sec ion G o he online appendix. The de elopmen phase is based
on he insigh s and e ined design speci ica ions gained om he expe in e iews. This
sec ion ocuses on p omp enginee ing echniques ha a e c i ical o op imizing he
capabili ies and pe o mance o he a i ac . Sec ion D o he online appendix p o ides
in-dep h desc ip ions o he echnology s ack unde lying he ins an ia ed a i ac , he
p o o ype de elopmen p ocess, and he so wa e a chi ec u e, including i s s uc u e,
componen s, and connec ions. The e, we also highligh he i e a i e na u e o he so -
wa e a chi ec u e and he a ious aspec s equi ed o build a obus , scalable a i ac
based on LLMs.
P omp enginee ing ep esen s a i al echnique o op imizing he u ili y o LLMs
ac oss a ange o domains, e iden in i s use in he con ex o BPM (Busch e al.2023).
Despi e i s ising signi icance, p omp enginee ing emains a ela i ely new a ea o
esea ch. Only ecen ly es ablished and alida ed echniques ha e begun o eme ge
in he li e a u e, signi ican ly inc easing he po en ial o he use o LLMs (Sahoo e
al.2024). The comp ehensi e axonomy p oposed by Schulho e al. (2024) classi ies 58
gene al ex -only p omp ing echniques in o six clus e s: Ze o-Sho , Few-Sho , Though
Gene a ion, Ensembling, Sel -C i icism, and Decomposi ion. Among hese app oaches,
Few-Sho P omp ing and Chain o Though p omp ing ha e p o en o be pa icula ly
e ec i e (Schulho e al.2024). Few-Sho p omp ing in ol es p o iding he model wi h
a limi ed numbe o examples om which o lea n and pe o m speci ic asks (Sahoo
e al.2024). This echnique is pa icula ly aluable in con ex s whe e ex ensi e ain-
ing da a is no a ailable, allowing he model o gene alize e ec i ely om minimal
Table 1 Business p ocess change classi ica ion amewo k o e iew

Page 16 o 33Schul e e al. P ocess Science (2025) 2:22
examples. Chain o Though p omp ing, he only echnique in oduced in he Though
Gene a ion ca ego y, encou ages he LLM o a icula e i s easoning p ocess s ep by s ep
be o e deli e ing a inal answe (Sahoo e al.2024). This encou ages esponses ha a e
mo e accu a e and logically s uc u ed. Bo h echniques a e in eg a ed in o he i e a i e
expe imen al op imiza ion p ocess desc ibed in he sec ion Expe imen Composi ion.
The ope a ing logic o he a i ac in ol es a wo-s age p ocess o logic-based cohe ence
checking ha se es as he ounda ion o he me hod, as illus a ed in Fig.4.
The i s s age o he p ocess is o compa e he con en o he modi ied and o igi-
nal documen s. Du ing his s age, indi idual seman ic changes a e iden i ied and ca -
ego ized acco ding o he Business P ocess Change Dimensions. The second s ep, he
cohe ence checking s ep, consis s o compa ing he p e iously iden i ied changes wi h
he con en o he ela ed documen . This compa ison includes he desc ip ion o any
equi ed changes o he ela ed documen ha a e necessa y o es o e p ocess cohe -
ence. He e, he iden i ied changes a e classi ied in o change ca ego ies as desc ibed in
he Business P ocess Change Classi ica ion F amewo k. The p ocess can be e mina ed
a any s age i cohe ence is con i med. The inal s ep, which is pa o ou ins an ia ion,
in ol es he gene a ion o he no i ica ion, which is cons uc ed in he o m o a ex ual
no i ica ion based on he s uc u ed ou pu . In addi ion, a no i ica ion i le and se e -
i y indica o a e added based on he esul s o he cohe ence check. Figu e4 shows a
concep ual b eakdown o he h ee s ages o he ope a ing logic a he han isola ed sub-
s eps. Since each s age is execu ed wi hin a single LLM p omp , measu ing he pe o -
mance o each s ep indi idually is no easible. The e o e, we e alua e he accu acy o he
cohe ence-checking logic holis ically, as de ailed in Appendix B as well as Sec ion E o
he online appendix.
Adhe ence o gene al p omp enginee ing bes p ac ices is c ucial in he de elopmen
o an LLM-based a i ac (Lo 2023; Ma in e al. 2024). Hence, h oughou he de el-
opmen o aP oCheCk, we inco po a ed es ablished p omp ing echniques o op imize
pe o mance. The p omp s we e s uc u ed o place key in o ma ion a he beginning,
ensu ing cla i y om he ou se and acili a ing p ocessing by he model. Clea s uc-
u al di isions wi hin p omp s allowed o logical low and cohe ence, while he s a egic
use o keywo ds highligh ed c i ical poin s and guided he model’s ocus. Backg ound
in o ma ion was p o ided a e he ask desc ip ion o p o ide he necessa y con ex
wi hou o e whelming he ini ial ins uc ions. S ep-by-s ep ins uc ions we e used o
main ain a logical sequence, wi h cons ain s lis ed a he end o a oid dis ac ion om
Fig. 4 Ope a ing Logic o he A i ac
Page 17 o 33Schul e e al. P ocess Science (2025) 2:22
ask pe o mance. Impo an ly, as pa o he expe imen al i e a ions, a sys em message
was se o assign he ole o an expe ienced BPM expe o he LLM, helping o imp o e
i s con ex ual unde s anding and decision accu acy. The inal p omp s as well as he
changes made in each expe imen al i e a ion a e a ailable in he code base in Sec ion G
o he online appendix.
E alua ion
The e alua ion o aP oCheCk was conduc ed in wo complemen a y s ages o ensu e
bo h echnical igo and p ac ical applicabili y. Fi s , an expe imen al e alua ion sys em-
a ically assessed he a i ac ’s e ec i eness, obus ness, and e iciency h ough an i e a-
i e e inemen p ocess. Second, a ocus g oup e alua ion examined an ins an ia ion o
he a i ac in na u alis ic se ings wi h eal asks, sys ems, and use s, p o iding con i -
ma o y insigh s in o i s u ili y, in eg a ion po en ial, and a eas o u he enhancemen .
Toge he , hese e alua ions es ablish he a i ac ’s eadiness o eal-wo ld applica ion
and iden i y di ec ions o u u e imp o emen s.
Expe imen al e alua ion
Expe imen composi ion
Expe imen al e alua ion is a c ucial s ep in he alida ion and benchma king o aP o-
CheCk, in line wi h he E al 3 phase o Sonnenbe g and om B ocke (2012). To accom-
plish his, we designed he e alua ion as a e e se abla ion s udy, ha is, we began wi h
a baseline p omp and inc emen ally added echniques in each i e a ion. This app oach
allowed us o isola e and measu e he indi idual impac o each echnique. The goal o
he e alua ion is o i e a i ely e ine he a i ac and igo ously assess i s pe o mance in
a con olled en i onmen , he eby ensu ing i s eadiness o p ac ical applica ion. Con-
duc ing such expe imen al e alua ions i e a i ely is impo an o he sys ema ic op imi-
za ion o he a i ac , closely ollowing he p inciples o DSR, which emphasizes i e a i e
de elopmen and con inuous imp o emen . Each i e a ion e ines a speci ic aspec o
he a i ac , ensu ing bo h p ac ical u ili y and heo e ical soundness. This dynamic
app oach is consis en wi h he p oblem-sol ing cycle inhe en in DSR, which in ol es
p oblem iden i ica ion, solu ion design, implemen a ion, e alua ion, and ongoing e ine-
men (Pe e s e al. 2007; Tuunanen e al. 2024). This amewo k ensu es ha he a i ac
e ol es p og essi ely, inco po a ing eedback and benchma king esul s o imp o e i s
pe o mance. The p ocedu e o ou expe imen i e a ions is isualized in Fig.5. The do -
ed a ows isualize he p ospec o olling back he changes made in he mos ecen
i e a ion i he benchma king esul s o he i e a ions a e no sa is ac o y.
Fig. 5 Expe imen I e a ions
Page 18 o 33Schul e e al. P ocess Science (2025) 2:22
Fo he empi ical alida ion, speci ically du ing he expe imen phase as de ined in
he DSR e alua ion p ocess, a comp ehensi e da ase om mul iple p ocess eposi o ies
was used. This no el P ocess Cohe ence Checking Da ase , which is published unde he
GNU Gene al Public License, is cen al o assessing he e ec i eness and obus ness o
LLMs in iden i ying and e i ying he cohe ence o mul i-le el p ocess documen a ion.
The ini ial da ase , de i ed om he esea ch o Sànchez-Fe e es e al. (2018),
includes a wide ange o p ocess models o iginally compiled by he BPM Academic Ini-
ia i e and u he de ailed by Eid-Sabbagh e al. (2012). These models, en iched wi h
ex ual desc ip ions compiled by expe p ocess modele s, o m he basis o a da ase o
checking he cohe ence o documen ed business p ocesses, as desc ibed in mo e de ail
in Sec ion E o he online appendix. The c i e ia used o he e alua ion o he expe i-
men a e also in oduced in de ail in Sec ion E o he online appendix.
The i s i e a ion ocuses on es ablishing a baseline using basic p omp ing guidelines.
This baseline uses s uc u ed p omp ing wi hou any ad anced p omp enginee ing o
ex ensi e da a p epa a ion. This simple app oach o ms he baseline agains which sub-
sequen i e a ions a e compa ed.
The second i e a ion aims o imp o e pe o mance by educing noise in he da a-
se . Impo an ly, his i e a ion ocuses solely on da a p ep ocessing. The p omp i sel
emains unchanged om he s uc u ed baseline p omp . Speci ically, i ele an isual
da a is emo ed om BPMN iles, signi ican ly educing he ile size and, he e o e,
he numbe o inpu okens equi ed o p ocessing. This s ep add esses he needle-in-
he-hays ack challenge by educing ex aneous da a ha may di e a en ion om he
incohe ence o he con en (Nelson e al. 2024). By p ese ing all he impo an connec-
ions and p ocess lows while emo ing he noise, he a i ac is expec ed o deli e mo e
accu a e esul s. In addi ion, an ex a check ensu es ha documen s ha a e iden ical
a e he noise educ ion s ep a e iden i ied be o ehand, sa ing esou ces and minimiz-
ing he isk o hallucina ions in he ou pu .
In he hi d i e a ion, ew-sho p omp ing is in oduced. This e inemen p o ides he
LLM wi h mo e comp ehensi e examples o inpu and ou pu scena ios, bo h speci ic
o BPMN models and gene al p ocess documen a ion, appended o he basic s uc-
u ed p omp ing empla e. The inclusion o mo e examples is in ended o be e guide
he LLM’s decision-making p ocess and ensu e consis en and accu a e ca ego iza ion
o changes. The en iched con ex is expec ed o imp o e he a i ac ’s abili y o eliably
de ec and ca ego ize inconsis encies.
The ou h i e a ion inco po a es ad anced p omp enginee ing echniques. A sys-
em message is in oduced o ame he LLM as an expe ienced BPM expe , which is
expec ed o imp o e i s con ex ual unde s anding and decision-making. In addi ion, he
chain o hough me hod is implemen ed, which p omp s he LLM o gene a e a de ailed
easoning p ocess be o e making ca ego iza ions and decisions. This app oach aims o
imp o e he easoning capabili ies o he a i ac , leading o mo e accu a e and con-
sis en ca ego iza ion o changes. The i e a i e e inemen p ocess ein o ces he p in-
ciples o DSR by con inuously enhancing he a i ac h ough empi ical e alua ion and
eedback in eg a ion. Each i e a ion builds on he p e ious one, inc emen ally imp o -
ing he unc ionali y and eliabili y o aP oCheCk. To ensu e ull anspa ency, he inal
p omp de eloped in his las i e a ion, including commen s on whe e con en was
Page 19 o 33Schul e e al. P ocess Science (2025) 2:22
added in p e ious i e a ions, is a ailable in he supplemen a y eposi o y in Sec ion G
o he online appendix. A highe le el ana omy o he inal p omp s used can be ound in
Appendix C.
Expe imen indings
The expe imen al esul s p o ide comp ehensi e insigh s in o he a i ac ’s pe o mance
ac oss mul iple e alua ion c i e ia, con i ming i s obus ness and eadiness o p ac ical
applica ion. Th ough a se ies o i e a ions, he accu acy, consis ency, and cos -e ec i e-
ness o aP oCheCk we e sys ema ically e alua ed and e ined. The esul s a e summa-
ized in Table2, whe e accu acy and consis ency a e exp essed as a alue om 0 o 1,
wi h 1 ep esen ing he op imal esul . Appendix B p o ides a de ailed explana ion o
how accu acy and consis ency a e calcula ed ( his is u he ex ended in Sec ion E o
he online appendix). The a e age API cos is exp essed on a mone a y scale, wi h he
lowes alues indica ing he bes esul s. The da a demons a es ha he mean accu acy
and consis ency ha e inc eased wi h each i e a i e e inemen , indica ing highly e ec-
i e pe o mance. Mo eo e , he cos s pe p ocess un we e educed by o e 50% om
he i s o he second i e a ion, indica ing a signi ican op imiza ion. Al hough he e
was a sligh inc ease in cos in he hi d and ou h i e a ions due o he inco po a ion o
addi ional sophis ica ed p omp ing echniques, he o e all cos -e ec i eness emained
signi ican ly be e han he baseline es ablished in he i s i e a ion. A mo e in-dep h
analysis o he esul s o he expe imen i e a ions, he s a is ical signi icance, and he
espec i e sco es o accu acy, obus ness, and e iciency can be ound in Sec ion E o
he online appendix.
O e all, aP oCheCk demons a ed excep ional pe o mance, gi en he complexi y and
subjec i i y o he p oblem domain. While pe ec accu acy is una ainable due o he
inhe en subjec i i y o he p oblem a hand, he s uc u ed and i e a i e op imiza ion
p ocess p o ed e ec i e in signi ican ly imp o ing obus ness, e iciency, and e ec i e-
ness, alida ing he a i ac ’s eadiness o eal-wo ld BPM applica ions.
Focus g oup e alua ion
The ocus g oup e alua ion is ancho ed in he h ee eali ies p oposed by Sun and
Kan o (2006): eal asks, eal sys ems, and eal use s. These dimensions a e cen al o
assessing he p ac ical applicabili y o he a i ac and ensu ing i s alignmen wi h he
equi emen s o eal business p ocesses. Real asks assess he pe o mance o he a i-
ac in he con ex o eal business p ocesses. Real sys ems e alua e he a i ac ’s in eg a-
ion in o es ablished BPM ecosys ems. Real use s highligh he p ac ical u ili y and use
expe ience o he a i ac , ensu ing alignmen wi h business needs. The use o na u al-
is ic da a om di e en indus ies unde lines aP oCheCk’s abili y o e ec i ely add ess
eal-wo ld scena ios. This da a no only p o ides a obus da ase o alida ion bu also
highligh s he e sa ili y and p ac icali y o he a i ac . By illus a ing he cohe ence
Table 2 Expe imen esul summa y
Expe imen I e a ion A e age Accu acy
(E ec i eness)
A e age Consis ency
(Robus ness)
A e age API Cos s pe P ocess Run (€)
(E iciency)
1 0,751 0,891 0,125
2 0,815 0,924 0,054
3 0,823 0,932 0,069
4 0,835 0,936 0,075
Page 20 o 33Schul e e al. P ocess Science (2025) 2:22
checking capabili ies o he a i ac ’s ins an ia ion in a a ie y o se ings, he eadiness
o use in eal-wo ld BPM en i onmen s is demons a ed.
Demons a ing he a i ac wi h na u alis ic da a
An ins an ia ion o aP oCheCk is demons a ed and e alua ed in a na u alis ic se ing,
ollowing he E al 4 phase in oduced by Sonnenbe g and om B ocke (2012). This na -
u alis ic demons a ion showcases he unc ionali y o he a i ac in a wide ange o
eal-wo ld scena ios wi h a p o o ypical implemen a ion. The na u alis ic demons a ion
employed o he ocus g oups includes checking he cohe ence o changes o a p ocess
model wi h a ela ed BPMN diag am and ice e sa. In addi ion, his demons a ion
in oduces an o e a ching documen cohe ence check, whe e p ocess models a e com-
pa ed agains upda ed p ocess modeling guidelines, and a cohe ence check on a chang-
ing p ocess diag am and a ela ed p ocess diag am o a a ian o he p ocess. I u he
gene alizes he applicabili y o aP oCheCk by inco po a ing di e en ile o ma s, such
as SVG o p ocess models.
The da a sou ces o he na u alis ic da a demons a ion consis o eal p ocess docu-
men s om h ee Ge man companies o di e se sizes and indus ies. These companies
ange om 200 o 300 employees o 10,000–15,000 employees, and he sec o s ep e-
sen ed include consul ing, indus ial manu ac u ing, and he ene gy and uel sec o s.
Table3 summa izes he demons a ion da a o he ou use cases. To ensu e con iden-
iali y, pa s o he documen s ha e been anonymized, and only he anonymized esul s
gene a ed by he aP oCheCk ool a e p o ided. The no i ica ion gene a ed by he ini ial
na u alis ic example p ocess is illus a ed in Fig.6. The emaining h ee examples can be
ound in Sec ion F o he online appendix, along wi h de ailed desc ip ions o he exem-
pla y use cases. These examples demons a e aP oCheCk’s abili y o iden i y and manage
changes ac oss mul iple le els o p ocess documen a ion. The a i ac e ec i ely ul ills
i s design speci ica ions by p o iding ac ionable insigh s, ensu ing p ocess cohe ence,
and ope a ing e icien ly in di e se eal-wo ld con ex s. These demons a ions unde line
he obus ness and applicabili y o he a i ac , highligh ing i s po en ial o signi ican ly
imp o e BPM p ac ices in na u alis ic scena ios.
Con i ma o y A i ac E alua ion
The con i ma o y e alua ion o he de eloped a i ac was conduc ed h ough wo
ocus g oups consis ing o 4 and 5 BPM IT consul an s, espec i ely. Each session las ed
app oxima ely one hou and aimed o alida e aP oCheCk in a na u alis ic se ing,
Table 3 Summa y o na u alis ic a i ac demons a ion da a
Indus y o
Company
P ocess Name Changed O iginal
Documen a ion
Rela ed P ocess
Documen
Consul ing Conduc ing Job In e iews BPMN Models
(XML Depic ion)
P ocess Desc ip ion
(Tex )
Ene gy/Fuel Sec o Regis e Incoming O de s (Va i-
an s: Manual and Au oma ed)
BPMN Model
(SVG Depic ion)
BPMN Model
(SVG Depic ion) o
P ocess Va ian
Ene gy/Fuel Sec o Regis e Incoming O de s (Va i-
an s: Manual and Au oma ed)
T aining Documen (P esen a ion
Tex )
BPMN Models
(SVG Depic ion) o
wo P ocess Va ian s
Insu ance Vehicle Insu ance P ocess O e a ching P ocess Modelling
Con en ion (Tex )
BPMN Model
(XML Depic ion)

Page 21 o 33Schul e e al. P ocess Science (2025) 2:22
employing he use cases in oduced in he p e ious sec ion. Bo h ocus g oups we e
conduc ed o he same pu pose and using he same me hodology. Conduc ing wo
sepa a e sessions enabled us o collec mo e ex ensi e eedback and alida e ou a i ac
mo e obus ly.
This app oach ollows he guidelines es ablished by T emblay e al. (2010) and is
aligned wi h he E al 4 phase o he e alua ion p ocess p oposed by Sonnenbe g and
om B ocke (2012). The ocus g oups we e designed o e alua e he a i ac in he con-
ex o he h ee eali ies p oposed by Sun and Kan o (2006): eal asks, eal sys ems,
and eal use s, ensu ing a comp ehensi e assessmen o he a i ac ’s pe o mance in
ealis ic condi ions. Pa icipan s we e asked o e alua e he a i ac based on h ee spe-
ci ic c i e ia de i ed om he E al 4 phase: ideli y wi h eal-wo ld phenomenon, impac
Fig. 6 aP oCheCk Demons a ion wi h Na u alis ic Da a
Page 22 o 33Schul e e al. P ocess Science (2025) 2:22
on a i ac en i onmen and use , and applicabili y (Sonnenbe g and om B ocke 2012).
Simila o he conduc ed expe in e iews, each c i e ion was accompanied by a Guid-
ing ques ion o ancho he discussions and allow o a comp ehensi e e alua ion, and
pa icipan s we e ins uc ed o a e each c i e ion on a Like scale om 1 o 7.
Fideli y wi h eal-wo ld phenomenon was explo ed wi h he guiding ques ion, “Could
he de eloped a i ac be used in a ealis ic wo king en i onmen ?” o assess i s applica-
bili y and p ac icali y in eal-wo ld se ings. This c i e ion assesses he a i ac ’s capaci y
o manage he complexi y o au hen ic BPM ope a ions and o in eg a e seamlessly in o
exis ing wo k lows.
Impac on a i ac en i onmen and use was e alua ed h ough he ques ion “How
do you assess he po en ial in luence o he a i ac on he wo king en i onmen and he
use ?” o unde s and he impac o he a i ac on he exis ing wo king en i onmen and
use in e ac ion. Pa icipan s a ed his c i e ion on a scale om 1, indica ing “no posi-
i e in luence a all”, o 7, indica ing “comple ely posi i e in luence”.
Applicabili y was assessed using he guiding ques ion, “Would he sys em’s no i ica-
ions be mo e o a bu den, o would you ind hem help ul?” o de e mine he unc ional
use ulness and p ac icali y o he a i ac . This c i e ion de e mines whe he he a i-
ac ’s no i ica ions a e ac ionable and bene icial in imp o ing p ocess cohe ence while
educing manual e o .
The boxplo diag am in Fig.7 depic s he dis ibu ion o sco es ac oss he speci ied
c i e ia, de i ed om he nine esponses p o ided by he pa icipan s. The esul s o he
e alua ion showed a consis en ly high a ing o he c i e ion o impac on a i ac en i-
onmen and use , e lec ing a s ong posi i e pe cep ion o he a i ac ’s po en ial in lu-
ence. Mo eo e , he ideli y wi h eal-wo ld phenomenon was a ed a o ably, al hough
some ese a ions we e exp essed ega ding he quali y o he da a p esen in p ac ice.
Such ac o s ha e he po en ial o impac he a i ac ’s pe o mance in eal-wo ld sce-
na ios. The applicabili y o he a i ac was also me wi h a o able esponses, al hough
some aised conce ns ega ding he e bosi y o he email no i ica ions.
Fig. 7 Focus G oup E alua ion Resul s
Page 23 o 33Schul e e al. P ocess Science (2025) 2:22
Key akeaways om he ocus g oups highligh a ious s eng hs and sugges ed
imp o emen s o aP oCheCk. The expe s consis en ly ound he a i ac imp essi e and
highly ele an , iden i ying nume ous use cases in hei espec i e clien o ganiza ions,
wi h I 9 (FG1) s a ing ha i is “Highly ele an , because […] in so many cus ome se ings
his issue somehow so quickly and easily leads o uncon olled documen a ion”. Focus
g oup pa icipan s emphasized he p ac ical applicabili y o he a i ac and p o ided
insigh s in o how i s unc ionali y could be enhanced. They sugges ed inco po a ing
deepe p ocess knowledge by embedding mo e o he o ganiza ion’s p ocess documen a-
ion in he LLM o iden i y in e ela ed changes ac oss di e en p ocesses, which could
signi ican ly enhance he use ulness o aP oCheCk. The expe s emphasized he consid-
e able po en ial o aP oCheCk in managing o e a ching p ocess documen s, pa icula ly
in educing he necessi y o manual wo k, exempli ied by I 11 (FG2) saying, “some hing
like his would ex emely educe he manual e o .“ They highligh ed ha when in e -
nal guidelines a e upda ed, he a i ac could au oma ically con i m cohe ence ac oss
all ela ed documen s. This capabili y elimina es he ime-consuming ask o manually
checking each documen indi idually, he eby ensu ing o ganiza ional consis ency and
compliance wi h new policies o egula ions. By au oma ing hese checks, he a i ac
signi ican ly enhances ope a ional e iciency and u ili y wi hin he o ganiza ion.
While acknowledging he cu en limi a ions, he expe s we e op imis ic ha u u e
i e a ions o LLMs would u he enhance he capabili ies o he a i ac . They sug-
ges ed ha u u e e sions could enable au oma ed p ocess model gene a ion and o he
ad anced ea u es. Da a quali y issues, p e alen in many o ganiza ions, we e iden i ied
as a challenge, bu he subjec i e decision-making capabili ies o LLMs based on gi en
con ex s we e seen as a p omising solu ion o main aining p ocess cohe ence.
In summa y, he ocus g oups p o ided aluable eedback highligh ing he signi i-
can po en ial o aP oCheCk, i s p ac ical u ili y, and a eas o u he imp o emen . By
inco po a ing hese insigh s in o u u e i e a ions, he a i ac can be e ined and op i-
mized o be e mee he needs o di e se BPM en i onmen s. This con i ma o y e alu-
a ion highligh s bo h he cu en s eng hs o aP oCheCk and he p omising di ec ions
o i s con inuous imp o emen .
Discussion
In o de o add ess ou esea ch objec i es, we employed a comp ehensi e and mul i-
ace ed DSR app oach in ol ing mul iple i e a ions and expe s om bo h esea ch
and p ac ice. Fi s , design objec i es we e de i ed om he exis ing li e a u e. Subse-
quen ly, p o isional design speci ica ions we e de eloped based on he design objec i es
and e alua ed and e ined h ough expe in e iews wi h esea che s and p ac i ione s.
The e alua ion was in o med by a p elimina y PoC demons a ion. aP oCheCk was
de eloped i e a i ely, e ined h ough expe imen al benchma king, and hen alida ed
h ough ocus g oups using na u alis ic da a and an ins an ia ion o he a i ac . In addi-
ion, a Business P ocess Change Classi ica ion F amewo k and an open-sou ce business
p ocess cohe ence checking da ase we e de eloped. As such, ou esea ch has impo -
an implica ions o esea ch and p ac ice.
Page 24 o 33Schul e e al. P ocess Science (2025) 2:22
Theo e ical implica ions
Ou wo k ep esen s a signi ican ad ancemen in he ield o BPM by applying gene a-
i e AI, speci ically LLMs, o imp o e p ocess managemen p ac ices. In pa icula , we
add ess he esea ch gap iden i ied by Feue iegel e al. (2024) conce ning he de ec ion
o posi i e p ocess de iance h ough he use o gene a i e AI. Ou esea ch makes se -
e al impo an con ibu ions:
Fi s , we con ibu e o he ield o business p ocess cohe ence checking by in oduc-
ing a no el app oach ha u ilizes LLMs o he dynamic analysis o di e se p ocess
documen a ion, wi h he objec i e o iden i ying incohe encies. The ield is cu en ly
domina ed by s a ic app oaches o inconsis ency de ec ion u ilizing s uc u ed p o-
cess documen a ion, such as e en logs (Ko and Comuzzi 2023). Exis ing esea ch on
uns uc u ed ex -based business p ocess documen s has la gely ocused on s a ic pa -
e n-ma ching app oaches (Ma in-To al e al. 2010; an de Aa e al. 2017). Ou LLM-
based app oach ad ances his line o wo k by enabling he au onomous iden i ica ion
o incohe encies in mul i-le el p ocess documen a ion, hus mo ing beyond adi ional
s a ic analyses.
Second, in eg a ing LLMs in o he BPM li ecycle, pa icula ly in he p ocess imple-
men a ion and he moni o ing phase, ans o ms adi ional me hods ha ely on s a ic
da a and manual e iews (Vidgo e al. 2023). While i s s udies ha e s a ed o in es i-
ga e he po en ial o LLMs in BPM (F anzoi e al. 2025b), speci ic applica ions, such as
p ocess cohe ence checking, emain sca ce. He e, ou wo k p o ides an impo an s a -
ing poin by igo ously de eloping and e alua ing an LLM-based a i ac o con inuously
assess he cohe ence o mul i-le el p ocess documen a ion. By enabling dynamic, AI-
d i en e alua ions, he a i ac acili a es he de ec ion o nega i e de ia ions indica ing
ine iciencies and posi i e de ia ions sugges ing inno a ion oppo uni ies, he eby sup-
po ing mo e con ex -sensi i e decision-making (F anzoi e al. 2025a) and demons a -
ing he impo ance o le e aging gene a i e AI o mo e om s a ic BPM me hods o
adap i e, p oac i e managemen sys ems (Feue iegel e al. 2024).
Thi d, we con ibu e o he ield by es ablishing de ailed design speci ica ions o
cohe ence checking based on mul i-le el p ocess documen a ion. These speci ica ions
balance unc ional equi emen s wi h he complexi ies o main aining BPM documen a-
ion cohe ence, p o iding a obus ounda ion o u u e esea ch.
Fou h, we in oduce he Business P ocess Change Classi ica ion F amewo k, which
comp ises Business P ocess Change Dimensions and Change Rele ance Ca ego ies.
De eloped h ough engagemen wi h es ablished BPM li e a u e and ex ensi e expe
in e iews, he amewo k deals wi h changes in ex -based, mul i-le el business p ocess
documen a ion. By sys ema ically ca ego izing and quan i ying changes, he amewo k
p o ides a s uc u ed mechanism o managing and in e p e ing he nuanced na u e o
BPM documen a ion. This ad ance imp o es he heo e ical unde s anding o change
managemen wi hin BPM. T adi ional me hods o en p o e inadequa e in add essing
hese complexi ies and subjec i i ies, whe eas he de eloped a i ac employs LLMs o
e ec i ely o e come hese challenges.
Fi h, we con ibu e a obus , open-sou ce da ase based on he es ablished wo k
o Sànchez-Fe e es e al. (2018), which p o ides u he suppo o empi ical BPM
esea ch. En iched wi h expe insigh s, his da ase is s uc u ed a ound he in oduced
Business P ocess Change Classi ica ion F amewo k, p o iding a aluable esou ce o
Page 31 o 33Schul e e al. P ocess Science (2025) 2:22
Table 6 S uc u e o con en compa ison and cohe ence check p omp s
P omp Elemen Sho Desc ip ion Example om So wa e
Ins an ia ion
4. JSON key- alue
pai s. - changed in
Reasoning S uc u ing
i e a ion o include
chain o hough
Lis s equi ed JSON keys o he ou pu
and de ails wha each should con ain.
“‘[…] ‘ echnical_compa ison’: A echnical
compa ison o he wo BPMN diag ams o
he same p ocess, including change IDs and
echnical de ail. The Change ID should s a a
c01 and hen coun up. […]”
5. Business P ocess
Change Cla i ica ion
F amewo k De ails
De ines ele an elemen s o he BP
Change Cla i ica ion F amewo k wi h
BPMN-speci ic examples.
“BPM Change Dimensions Cla i ica ions: Task:
Changes o he undamen al elemen s o he
wo k lows, including he in oduc ion o new
asks, he modi ica ion o exis ing asks o he
dele ion o obsole e asks. […]”
6. Gene al
Cla i ica ions
Lis s wha changes o dis ega d (e.g.
isual changes, syn ax changes, un e-
la ed pools, i ele an IDs).
“Changes ha a e only isual can be neglec -
ed, like he size o pools/swim lanes o he
coo dina es o elemen s do no ma e […]”
7. JSON Fo ma De ails
- changed in Reasoning
S uc u ing i e a ion
o include chain o
hough
Shows a comple e example JSON
ou pu in he co ec s uc u e.
“[…] “con en _compa ison “: [
{{ “id “: “c01 “, “de ail “: “con en compa i-
son de ail 1 “}},
{{ “id “: “c02 “, […] }”
8. Few Sho p omp -
ing examples -
in oduced in Da a
En ichmen i e a ion
P o ides conc e e change examples
wi h co ec classi ica ions o guide
he model.
“[…] Example 2: Two asks swap places and
he e o e all incoming and ou coming con-
ol lows need o be adop ed as well. Co ec
Classi ica ion: Only one ele an change o
ype ‘con ol low’ […]”
Table 7 S uc u e o no i ica ion c ea ion p omp
P omp
Elemen
Sho Desc ip ion Example om So wa e
Ins an ia ion
1. Task
desc ip ion
De ails he main pu pose o he
ask o c ea ing he managemen
summa y
“[…] Please i s desc ibe he changes made o he o ig-
inal documen and hen desc ibe, how hose changes
a e inconsis en wi h he ela ed documen . […]”
2. Task Da a Includes s uc u ed compa ison, e-
la ed p ocess documen , and mos
ecen BPMN e sion.
“[…] Co esponding ela ed p ocess documen only o
e e ence: { x _ ilename}: n{ x _con en }
[…] ”
3. Respond
in he
ollowing JSON
o ma
Re u n JSON wi h managemen
summa y and email i le including
u gency indica o .
“[…]’, ‘Email i le’: { ‘U gency and Impo ance Ra ing’: ‘
🟢
/
🟡
/
🟠
/
🔴
’, ‘Email Ti le’: ‘Sho and comp ehensi e
i le […]’ } }”
4. Task
cla i ica ion
Ou lines wha o include/omi , w i -
ing s yle, s uc u e, and o ma ing
ules.
“[…] c. W i e he ex in he same language as he
p ocess documen s. […] ”
Au ho s’ con ibu ions
We desc ibe he au ho s con ibu ions by desc ibing he espec i e con ibu ion oles acco ding o he CRediT
Taxonomy: Ma ek Schul e: Concep ualiza ion, Da a cu a ion, Fo mal analysis, In es iga ion, Me hodology, Valida ion,
Visualiza ion, W i ing – o iginal d a , W i ing – e iew & edi ing; Sand o F anzoi: Concep ualiza ion, In es iga ion,
Me hodology, P ojec adminis a ion, Valida ion, W i ing – o iginal d a , W i ing – e iew & edi ing; F ank Kühne:
Concep ualiza ion, Funding acquisi ion, Resou ces, Supe ision, Valida ion, W i ing – e iew & edi ing; Jan om B ocke:
Concep ualiza ion, Resou ces, Supe ision, Valida ion, W i ing – e iew & edi ing.
Funding
Open Access unding enabled and o ganized by P ojek DEAL. As pa o he Change.Wo kAROUND p ojec (p omo ion
sign 02J21C166), his esea ch was unded by he Ge man Fede al Minis y o Educa ion and Resea ch.
Da a a ailabili y
Supplemen a y ma e ials o he pape i led ‘Towa d LLM-Enabled Business P ocess Cohe ence Checking Based on
Mul i-Le el P ocess Documen a ion’ by Schul e, M.; F anzoi*, S.; Köhne, F.; om B ocke, J. submi ed o publica ion o he
jou nal P ocess Science can be accessed he e: h ps://gi hub.com/ ia dee/p ocess -documen -cohe ence-checke .

Page 32 o 33Schul e e al. P ocess Science (2025) 2:22
Decla a ions
Compe ing in e es s
The au ho s decla e no compe ing in e es s.
Recei ed: 30 Ap il 2025 / Accep ed: 7 Sep embe 2025
Re e ences
Ba elheime C, Wol V, Be e ungen D (2023) Wo ka ounds as gene a i e mechanisms o bo om-up p ocess inno a ion—
insigh s om a mul iple case s udy. In o m Sys J 33:1085–1150
Becke J, Be gene P, Del mann P, Egge M, Weiß B (2011) Suppo ing Business P ocess Compliance in Financial Ins i u ions - A
Model-D i en App oach. In: Be ns ein A (ed) P oceedings o he 10 h In e na ional Con e ence on Wi scha sin o ma ik:
16–18 Feb ua y 2011 Zu ich, Swi ze land, ol 10, Zü ich, pp 355–364
Binz M, Schulz E (2023) Using cogni i e psychology o unde s and GPT-3. P oc Na l Acad Sci U S A 120:1–10
Bose RPJC, an de Aals WMP, Žliobai ė I, Pechenizkiy M (2011) Handling concep d i in p ocess mining. In: Mou a idis H, Rol-
land C (eds) Ad anced in o ma ion sys ems enginee ing, ol 141, 23 d edn. Sp inge Be lin Heidelbe g, Be lin, Heidelbe g,
pp 391–405
B ü zke P, Killewald R, F anzoi S, om B ocke J (2025) AI-assis ed P ocess Mining o Con ex -sensi i e Analysis Suppo . P oceed-
ings o he Eu opean Con e ence on In o ma ion Sys ems (ECIS)
Busch K, Rochli ze A, Sola D, Leopold H (2023) Jus ell me: p omp enginee ing in business p ocess managemen . In: an de
Aa H, Bo k D, P ope HA, Schmid R (eds) Lec u e no es in business in o ma ion p ocessing. En e p ise, business-p ocess
and in o ma ion sys ems modeling, ol. 479. Sp inge Na u e, Swi ze land,pp 3–11. h p s : / / d o i . o g / 1 0 . 1 0 0 7 / 9 7 8 - 3 - 0 3 1 - 3 4
2 4 1 - 7 _ 1
Delias P (2017) A posi i e de iance app oach o elimina e was es in business p ocesses. Ind Manag Da a Sys 117:1323–1339
Di F ancescoma ino C, Donadello I, Ghidini C, Maggi FM, Puu a J (2025) Business p ocess de iance mining wi h sequen ial and
decla a i e pa e ns. Bus In Sys Eng
Eid-Sabbagh R-H, Kunze M, Meye A, Weske M (2012) A pla o m o esea ch on p ocess model collec ions. In: an de Aals W,
Mylopoulos J, Rosemann M, Shaw MJ, Szype ski C, Mendling J, Weidlich M (eds) Business p ocess model and no a ion, ol
125. Sp inge Be lin Heidelbe g, Be lin, Heidelbe g, pp 8–22
Fahland D, Fou nie F, Limonad L, Ska bo sky I, Swe els AJE (2024) How well can la ge language models explain business
p ocesses?
Feue iegel S, Ha mann J, Janiesch C, Zschech P (2024) Gene a i e AI. Bus In Sys Eng 66:111–126
F anzoi S, Ha l S, G isold T, an de Aa H, Mendling J, om B ocke J (2025a) Explaining p ocess dynamics: a p ocess mining
con ex axonomy o sense-making. P ocess Sci. h ps://doi.o g/10.1007/s44311-025-00008-6
F anzoi S, Delwaulle M, Dyong J, Scha ne J, Bu ge M, om B ocke J (2025b) Using la ge Language models o gene a e p ocess
knowledge om en e p ise con en . In: Gdowska K, Gómez-López MT, Rehse J-R (eds) Business p ocess managemen
wo kshops, ol 534. Sp inge Na u e Swi ze land, Cham, pp 247–258
F ied ich F, Mendling J, Puhlmann F (2011) P ocess model gene a ion om na u al Language ex . In: Mou a idis H, Rolland C
(eds) Ad anced in o ma ion sys ems enginee ing, 23 d edn. Sp inge Be lin Heidelbe g, Be lin, Heidelbe g, pp 482–496
Galpe in BL (2012) Explo ing he nomological ne wo k o wo kplace de iance: de eloping and alida ing a measu e o con-
s uc i e de iance. J Appl Soc Psychol 42:2988–3025
G ego S, He ne AR (2013) Posi ioning and p esen ing design science esea ch o maximum impac . MIS Q 37:337–355
Ma in G, Hellen N, Jjingo D, Naka umba-Nabende J (2024) P omp enginee ing in la ge Language models. In: Jacob Ij, Pi a-
mu hu S, Falkowski-Gilski P (eds) Da a in elligence and cogni i e in o ma ics. Sp inge Na u e Singapo e, Singapo e, pp
387–402
G isold T, an de Aa H, F anzoi S, Ha l S, Mendling J, om B ocke J (2024) A Con ex F amewo k o Sense-making o P ocess
Mining Resul s. In: 2024 6 h In e na ional Con e ence on P ocess Mining (ICPM). IEEE, pp 57–64
Ha l M, Zilke S, Weinzie l S (2024) Towa ds au oma ed business p ocess edesign in un ime using gene a i e machine lea ning.
P oceedings o he Eu opean Con e ence on In o ma ion Sys ems (ECIS)
He ne AR, Ma ch ST, Pa k J, Ram S (2004) Design science in in o ma ion sys ems esea ch. MIS Q 28:75–105
He ne AR, Pa sons J, B endel AB, Lukyanenko R, Tie enbeck V, T emblay MC, om B ocke J (2024) T anspa ency in design sci-
ence esea ch. Decis Suppo Sys 182:1–11
Kampik T, Wa mu h C, Rebmann A, Agam R, Egge LNP, Ge be A, Ho a J, Kolk J, He zig P, Decke G, an de Aa H, Poly yanyy
A, Rinde le-Ma S, Webe I, Weidlich M (2024) La ge P ocess Models: A Vision o Business P ocess Managemen in he Age
o Gene a i e AI. KI - Küns liche In elligenz:1–15
Ko J, Comuzzi M (2023) A sys ema ic e iew o anomaly de ec ion o business p ocess e en logs. Bus In Sys Eng 65:441–462
König UM, Linha A, Röglinge M (2019) Why do business p ocesses de ia e? Resul s om a Delphi s udy. Bus Res 12:425–453
Kou ani H, Be i A, Schus e D, an de Aals WMP, an de Aa H, Bo k D, Schmid R, S u m A (2024) P ocess modeling wi h la ge
Language models. En e p ise, Business-P ocess and in o ma ion sys ems modeling, ol 511. Sp inge Na u e Swi ze land,
Cham, pp 229–244
Leopold H, Eid-Sabbagh R-H, Mendling J, Aze edo LG, Baião FA (2013) De ec ion o naming con en ion iola ions in p ocess
models o di e en languages. Decis Suppo Sys 56:310–325
Lo LS (2023) The a and science o p omp enginee ing: a new li e acy in he in o ma ion age. In e ne Re Se Q 27:203–210
Ma in-To al S, Sainz-Palme o G, Dimi iadis Y (2008) De ec ion O Incohe ences In A Technical And No ma i e Documen
Co pus. In: Co dei o J, Filipe J (eds) P oceedings o he Ten h In e na ional Con e ence on En e p ise In o ma ion Sys ems.
SciTeP ess - Science and and Technology Publica ions, pp 282–287
Ma in-To al S, Sainz-Palme o G, Dimi iadis Y (2010) Hyb id App oach o Incohe ence De ec ion Based on Neu o- uzzy Sys ems
and Expe Knowledge. In: Co dei o J, Filipe J (eds) P oceedings o he 12 h In e na ional Con e ence on En e p ise In o -
ma ion Sys ems. SciTeP ess - Science and and Technology Publica ions, pp 408–413
Page 33 o 33Schul e e al. P ocess Science (2025) 2:22
Mcin osh TR, Liu T, Susnjak T, Wa e s P, Halgamuge MN (2024) A easoning and alue alignmen es o assess ad anced GPT
easoning. ACM T ans In e ac In ell Sys 14:1–37
Mendling J, Pen land BT, Recke J (2020) Building a complemen a y agenda o business p ocess managemen and digi al
inno a ion. Eu J In o m Sys 29:208–219
Me ens W, Recke J (2017) Posi i e De iance and Leade ship: An Explo a o y Field S udy. In: Sp ague R, Bui TX (eds) P oceed-
ings o he 50 h Hawaii In e na ional Con e ence on Sys em Sciences (2017). Hawaii In e na ional Con e ence on Sys em
Sciences
Mo ana S, K oenung J, Maedche A, Schach S (2019) Designing p ocess guidance sys ems. JAIS 20:499–535
Nelson E, Kollias G, Das P, Chaudhu y S, Dan S (2024) Needle in he hays ack o memo y based la ge language models. h p s : / /
d o i . o g / 1 0 . 4 8 5 5 0 / a X i . 2 4 0 7 . 0 1 4 3 7
Nwankpa JK, Roumani Y, Da a P (2022) P ocess inno a ion in he digi al age o business: he ole o digi al business in ensi y
and knowledge managemen . JKM 26:1319–1341
Pe e s K, Tuunanen T, Ro henbe ge MA, Cha e jee S (2007) A design science esea ch me hodology o in o ma ion sys ems
esea ch. J Manage In Sys 24:45–77
Poly yanyy A, Smi no S, Weske M (2015) Business p ocess model abs ac ion. In: om B ocke J, Rosemann M (eds) Handbook
on business p ocess managemen 1, ol 1. Sp inge Be lin Heidelbe g, Be lin, Heidelbe g, pp 147–165
Rosemann M, om B ocke J (2015) The six co e elemen s o business p ocess managemen . In: om B ocke J, Rosemann M (eds)
Handbook on business p ocess managemen 1, ol 1. Sp inge Be lin Heidelbe g, Be lin, Heidelbe g, pp 105–122
Rosemann M, Recke J, Flende C (2008) Con ex ualisa ion o business p ocesses. IJBPIM 3:47
Rosemann M, om B ocke J, an Looy A, San o o F (2024) Business p ocess managemen in he age o AI – h ee essen ial d i s.
In Sys E-Bus Manage. h ps://doi.o g/10.1007/s10257-024-00689-9
Sai C, Win e K, Fe nanda E, Rinde le-Ma S (2023) De ec ing de ia ions be ween ex e nal and in e nal egula o y equi emen s
o imp o ed p ocess compliance assessmen . In: Indulska M, Reinha z-Be ge I, Ce ina C, Pas o O (eds) Ad anced in o -
ma ion sys ems enginee ing, ol 13901. Sp inge Na u e Swi ze land, Cham, pp 401–416
Sahoo PK, Da a R, Rahman MM, Sa ka D. Sus ainable en i onmen al echnologies: ecen de elopmen , oppo uni ies, and key
challenges. Applied Sciences. 2024;14(23):10956.
Sain -Dizie P (2018) Mining incohe en equi emen s in echnical speci ica ions: analysis and implemen a ion. Da a Knowl Eng
117:290–306
Sànchez-Fe e es J, an de Aa H, Ca mona J, Pad ó L (2018) Aligning ex ual and model-based p ocess desc ip ions. Da a
Knowl Eng 118:25–40
Schulho S, Ilie M, Balepu N, Kahadze K, Liu A, Si C, Li Y, Gup a A, Han H, Schulho S [Se ien], Dulepe PS, Vidyadha a S, Ki D,
Ag awal S, Pham C, K oiz G, Li F, Tao H, S i as a a A, . . . Resnik P (2024) The p omp epo : a sys ema ic su ey o p omp -
ing echniques. h ps://doi.o g/10.48550/a Xi .2406.06608
Schulho S, Ilie M, Balepu N, Kahadze K, Liu A, Si C, Li Y, Gup a A, Han H, Schulho S [Se ien], Dulepe PS, Vidyadha a S, Ki D,
Ag awal S, Pham C, K oiz G, Li F, Tao H, S i as a a A, . . . Resnik P (2024) The p omp epo : a sys ema ic su ey o p omp -
ing echniques. h ps://doi.o g/10.48550/a Xi .2406.06608
Se iawan MA, Sadiq S (2013) A me hodology o imp o ing business p ocess pe o mance h ough posi i e de iance. In J In
Sys Model Des 4:1–22
Sonnenbe g C, om B ocke J (2012) E alua ions in he science o he A i icial – Reconside ing he Build-E alua e pa e n in
design science esea ch. In: Hu chison D, Kanade T, Ki le J, Kleinbe g JM, Ma e n F, Mi chell JC, Nao M, Nie s asz O,
Pandu Rangan C, S e en B, Sudan M, Te zopoulos D, Tyga D, Va di MY, Weikum G, Pe e s K, Ro henbe ge M, Kuechle B
(eds) Design science esea ch in in o ma ion sys ems. Ad ances in heo y and p ac ice, ol 7286. Sp inge Be lin Heidel-
be g, Be lin, Heidelbe g, pp 381–397
Sun Y, Kan o PB (2006) C oss-e alua ion: a new model o in o ma ion sys em e alua ion. J Am Soc In Sci 57:614–628
Teinemaa I, Dumas M, Maggi FM, Di F ancescoma ino C (2016) P edic i e business p ocess moni o ing wi h s uc u ed and
uns uc u ed da a. In: La Rosa M, Loos P, Pas o O (eds) Business p ocess managemen , ol 9850. Sp inge In e na ional
Publishing, Cham, pp 401–417
Tuunanen T, Win e R, om B ocke J (2024) Dealing wi h complexi y in design science esea ch: a me hodology using design
echelons. MIS Q 48:427–458
an de Aa H, Leopold H, Reije s HA (2017) Compa ing ex ual desc ip ions o p ocess models – he au oma ic de ec ion o
inconsis encies. In Sys 64:447–460
an de Aa H, Ca mona J, Leopold H, Mendling J, Pad ó L (2018) Challenges and oppo uni ies o applying na u al language
p ocessing in business p ocess managemen . In: Bende EM (ed) The 27 h In e na ional Con e ence on Compu a ional
Linguis ics - p oceedings o he con e ence: Augus 20–26, 2018, San a Fe, New Mexico, USA: COLING 2018. Associa ion
o Compu a ional Linguis ics, S oudsbu g, PA, pp 2791–2801
an Dun C, Mode L, K a sch W, Röglinge M (2023) P ocessGAN: suppo ing he c ea ion o business p ocess imp o emen
ideas h ough gene a i e machine lea ning. Decis Suppo Sys 165:113880
Venable J, P ies-Heje J, Baske ille R (2016) FEDS: a amewo k o e alua ion in design science esea ch. Eu J In Sys 25:77–89
Vidgo M, Bachho ne S, Mendling J (2023) La ge Language models o business p ocess managemen . Oppo uni ies and
Challenges
om B ocke J, Win e R, He ne A, Maedche A (2020) Special issue edi o ial –accumula ion and e olu ion o design knowledge
in design science esea ch: a jou ney h ough ime and space. JAIS 21:520–544
Weinzie l S, Zilke S, Dunze S, Ma zne M (2024) Machine lea ning in business p ocess managemen : A sys ema ic li e a u e
e iew. Expe Sys Appl :1–43
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.