Co esponding au ho : Samuel Addo
Copy igh © 2025 Au ho (s) e ain he copy igh o his a icle. This a icle is published unde he e ms o he C ea i e Commons A ibu ion License 4.0.
Causal AI o s a egic business planning: unco e ing la en d i e s o long- e m
o ganiza ional pe o mance and esilience
Samuel Addo*
Depa men o Ma hema ics and Philosophy, Wes e n Illinois Uni e si y, USA.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
Publica ion his o y: Recei ed on 29 Ma ch 2025; e ised on 04 May 2025; accep ed on 07 May 2025
A icle DOI: h ps://doi.o g/10.30574/wja .2025.26.2.1738
Abs ac
In he e a o digi al ans o ma ion and da a ubiqui y, o ganiza ions a e inc easingly shi ing om desc ip i e and
p edic i e analy ics owa d causal AI o in o m long- e m s a egic planning. While adi ional machine lea ning models
excel a ecognizing co ela ions and o ecas ing ou comes, hey o en ail o e eal he unde lying causes ha d i e
pe o mance. This limi a ion becomes pa icula ly c i ical when businesses mus make high-s akes decisions in ol ing
esou ce alloca ion, policy implemen a ion, o cus ome engagemen , whe e unde s anding he impac o in e en ions
is essen ial. Causal AI o e s a powe ul amewo k ha goes beyond p edic ion o unco e la en d i e s o
o ganiza ional beha io , enabling decision-make s o simula e, es , and op imize s a egic ac ions wi h scien i ic igo .
This pape p o ides a comp ehensi e explo a ion o how causal AI enhances s a egic business planning. I begins wi h
a mac o-le el iew o he limi a ions o co ela ion-based analy ics in ola ile en i onmen s and ansi ions in o he
ounda ions o causal in e ence—including s uc u al causal models, coun e ac uals, and do-calculus. The discussion
hen na ows o he p ac ical applica ion o causal machine lea ning algo i hms such as causal o es s, upli modeling,
and Bayesian ne wo ks. These models help iden i y he e ogeneous ea men e ec s, op imize ma ke ing and
ope a ional in e en ions, and p o ide obus insigh s unde unce ain y. By embedding causal logic in o en e p ise
analy ics pla o ms and business in elligence dashboa ds, o ganiza ions gain ac ionable cla i y on "wha wo ks" and
"why"— ans o ming da a in o a p oac i e ool o g ow h, inno a ion, and esilience. The pape concludes by ou lining
implemen a ion pa hways and go e nance conside a ions o ensu e esponsible and scalable adop ion o causal AI
ac oss sec o s.
Keywo ds: Causal AI; S a egic Business Planning; S uc u al Causal Models; Coun e ac ual In e ence; En e p ise
Analy ics; O ganiza ional Resilience
1. In oduc ion
1.1. Backg ound and Mo i a ion
In he pas decade, en e p ises ha e unde gone a p o ound ans o ma ion owa d da a-d i en decision-making,
le e aging digi al oo p in s, eal- ime analy ics, and au oma ed pipelines o guide business ope a ions. F om ma ke ing
o inance, o ganiza ions now ely on s uc u ed and uns uc u ed da a o unco e insigh s, o ecas ends, and
op imize wo k lows [1]. This ansi ion has been accele a ed by ad ancemen s in a i icial in elligence (AI), machine
lea ning (ML), and scalable da a in as uc u e, making i possible o model complex beha io s ac oss as da ase s.
Howe e , a c i ical limi a ion o con en ional app oaches lies in hei emphasis on p edic ion o e explana ion. Mos
machine lea ning models a e designed o op imize accu acy me ics, such as classi ica ion accu acy o mean squa ed
e o , ye hey o e limi ed insigh in o he causal ela ionships unde pinning business phenomena [2]. Fo example, a
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
896
model may p edic which use s a e likely o chu n bu canno de e mine whe he sending an o e will ac ually educe
chu n isk. This lack o causali y impedes s a egic planning, especially when deploying in e en ions ha equi e
cla i y on cause-and-e ec ela ionships.
In an en i onmen ma ked by unce ain y—economic shi s, consume beha io ola ili y, egula o y dis up ion—
o ganiza ions canno ely on eac i e, co ela ion-d i en ac ics. Ins ead, hey equi e p oac i e, in e en ion- eady
amewo ks ha no only p edic ou comes bu explain how hose ou comes a ise and unde wha condi ions hey can
be al e ed [3]. Causal in e ence, pa icula ly when in eg a ed wi h deep lea ning ools such as con olu ional neu al
ne wo ks (CNNs), enables his nex s age o analy ic ma u i y. By unco e ing he la en d i e s o long- e m
pe o mance and esilience, causal AI posi ions i sel as a c i ical capabili y o en e p ise s a egy in he digi al age.
1.2. Resea ch P oblem and Objec i es
The co e esea ch challenge add essed in his a icle lies in he dis inc ion be ween co ela ional signals and causal
d i e s o business pe o mance. While p edic i e analy ics has been widely adop ed o asks such as cus ome
segmen a ion, sales o ecas ing, and anomaly de ec ion, i o en lacks he s uc u al ounda ion o guide in e en ions
o policy changes. Businesses equen ly mis ake s a is ical co ela ion o causa ion, leading o ine ec i e esou ce
alloca ion, misaligned s a egies, and subop imal ou comes [4].
Fo example, i cus ome engagemen is co ela ed wi h highe spending, a business may w ongly assume ha
inc easing engagemen will cause g ea e spend. Wi hou causal analysis, such assump ions can back i e, leading o
cos ly ini ia i es ha do no deli e he expec ed impac . This gap becomes especially signi ican in s a egic planning,
whe e decisions mus be jus i ied no only by wha has happened bu by wha would happen unde di e en ac ions o
condi ions.
The objec i e o his pape is o explo e how causal AI models, suppo ed by ML echniques like CNNs, can be in eg a ed
in o s a egic decision amewo ks o unco e he unde lying mechanisms o long- e m o ganiza ional ou comes.
Speci ically, we aim o:
• In oduce s uc u al and coun e ac ual app oaches o causali y;
• Demons a e a me hodological amewo k using beha io al da a, CNNs, and causal modeling ools; and
• E alua e hei s a egic alue h ough in e p e abili y, p ecision a ge ing, and simula ion o in e en ions [5].
• In doing so, his esea ch b idges a c i ical gap be ween machine lea ning e iciency and execu i e-le el
decision-making, p o iding o ganiza ions wi h a oolki o esilien , e idence-based s a egy o mula ion.
1.3. A icle S uc u e and Con ibu ion
This a icle is s uc u ed o o e bo h a heo e ical and p ac ical oadmap o in eg a ing causal AI in o s a egic
business planning. I begins by si ua ing he ise o da a-cen ic s a egy and e iews he limi a ions o p edic i e-only
models in managing unce ain y. I hen in oduces ounda ional concep s in causal in e ence—including s uc u al
causal models, coun e ac ual easoning, and he use o he "do" ope a o o es ima e in e en ion e ec s [6]. The
dis inc ion be ween obse a ion and manipula ion is emphasized, se ing he s age o applied causal me hodologies.
The me hodology sec ion ou lines a comp ehensi e pipeline combining CNN-based beha io al ea u e ex ac ion wi h
ea men e ec es ima ion using upli modeling and causal o es s. A mul i-s age expe imen al design is de ailed,
in ol ing eal-wo ld beha io al da a, ea men assignmen , and coun e ac ual simula ion o p edic long- e m
ou comes o s a egic decisions. Implemen a ion de ails—such as da a no maliza ion, model a chi ec u e, and ou come
me ics—a e ho oughly add essed o suppo eplicabili y.
Subsequen sec ions ocus on esul s analysis, including in e p e a ion o causal es ima es, he e ogeneous ea men
e ec s, and KPI impac by cus ome segmen . Business in elligence in eg a ion is also discussed, highligh ing how causal
insigh s can be embedded in o dashboa ds o c oss- unc ional access and execu ion [7].
The key con ibu ion o his a icle lies in demons a ing a seamless in eg a ion o machine lea ning and causal
in e ence, showing how his syn hesis enables decision-make s o iden i y no jus wha is likely o happen, bu wha
can be changed and how. This hyb id pa adigm mo es beyond p edic ion o s a egic in luence, o e ing ac ionable,
anspa en , and esilien insigh s o en e p ise leade s.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
897
2. Li e a u e e iew
2.1. T adi ional P edic i e Analy ics and Limi a ions
T adi ional p edic i e analy ics has long been he ounda ion o da a-d i en decision-making in en e p ises, o e ing
powe ul ools o o ecas ou comes such as cus ome chu n, sales ends, o p oduc demand. These models, including
linea eg ession, logis ic eg ession, decision ees, and ensemble me hods like andom o es s and g adien boos ing,
iden i y s a is ical pa e ns and ela ionships be ween inpu a iables ( ea u es) and ou comes ( a ge s) [6]. In e ail,
o example, p edic i e models a e used o o ecas which cus ome s a e likely o make epea pu chases, enabling
imely ma ke ing in e en ions.
Despi e hei e ec i eness in o ecas ing, hese models ha e a c i ical limi a ion: hey canno in e causali y. P edic i e
algo i hms a e op imized o accu acy, no unde s anding. They ely on co ela ions, meaning hey may cap u e
spu ious ela ionships ha a ise due o con ounding a iables o coinciden al ends. A model migh p edic high
pu chase likelihood based on ce ain web beha io s, bu i does no de e mine whe he hose beha io s cause he
pu chases o simply accompany hem [7].
This limi a ion becomes pa icula ly p oblema ic when o ganiza ions a emp o ac on model p edic ions. Decision-
make s equi e answe s o “wha -i ” ques ions, such as whe he sending a discoun will change a cus ome ’s beha io
o i ealloca ing esou ces o one channel will imp o e con e sion a es. P edic i e analy ics canno add ess such
ques ions, as i lacks he s uc u al amewo k o dis inguish in e en ion e ec s om obse a ional pa e ns [8].
In high-s akes en i onmen s—such as inancial se ices, heal hca e, and public policy— elying solely on p edic i e
models isks deploying s a egies based on misleading signals. As a esul , he e is a g owing consensus ha decision
sys ems mus go beyond co ela ion and emb ace amewo ks capable o quan i ying causal ela ionships o suppo
obus , e idence-based ac ion.
2.2. O e iew o Causal In e ence in Business Con ex s
Causal in e ence e e s o he p ocess o iden i ying and es ima ing he e ec s o one a iable on ano he , especially in
he con ex o delibe a e in e en ions. Unlike p edic i e modeling, which ocuses on co ela ions, causal in e ence aims
o de e mine whe he changing one ac o (e.g., launching a new p icing s a egy) will lead o a speci ic ou come (e.g.,
inc eased e enue) [9]. This dis inc ion is cen al o s a egic business planning, whe e leade s a e less conce ned wi h
pa e ns and mo e in e es ed in le e s o in luence.
In business con ex s, causal in e ence is ypically applied o e alua e ea men s, such as ma ke ing campaigns, policy
changes, o ope a ional adjus men s. These ea men s can be analyzed using a ious causal amewo ks, including
andomized con olled ials (RCTs), obse a ional s udy adjus men s, and coun e ac ual modeling. RCTs, conside ed
he gold s anda d, andomly assign ea men s o con ol o con ounding a iables and ensu e unbiased es ima es o
ea men e ec s. Howe e , RCTs a e o en imp ac ical in la ge-scale comme cial se ings due o cos , e hical, o
logis ical cons ain s [10].
In esponse, companies a e adop ing quasi-expe imen al echniques such as p opensi y sco e ma ching, di e ence-in-
di e ences, and eg ession discon inui y designs. These app oaches use obse a ional da a o app oxima e causal
conclusions whe e andomiza ion is no easible. Fo ins ance, an online e aile may compa e use s who saw a
p omo ion o hose who didn’ , adjus ing o p io pu chase beha io o es ima e he campaign’s impac [11].
Causal in e ence o e s businesses a obus mechanism o simula e al e na e decisions, op imize in e en ions, and
p io i ize ac ions based on expec ed ou comes, no jus his o ical ends. I ans o ms da a analysis om a desc ip i e
unc ion in o a s a egic o ecas ing engine capable o d i ing measu able change.
2.3. Machine Lea ning and Causali y: Eme ging In eg a ion
The con e gence o machine lea ning and causal in e ence is gi ing ise o a powe ul new ield known as Causal
Machine Lea ning (Causal ML). This in eg a ion enables o ganiza ions o scale causal es ima ion using he lexibili y
and complexi y-handling capabili ies o mode n ML algo i hms. Unlike adi ional s a is ical me hods, causal ML is
designed o manage high-dimensional da a, nonlinea in e ac ions, and indi idual-le el ea men he e ogenei y—all
c i ical o accu a e in e en ion modeling in en e p ise se ings [12].
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
898
One widely adop ed app oach is upli modeling, also known as ue-li modeling. Unlike con en ional classi ica ion
models ha p edic ou comes (e.g., con e sion), upli models p edic he inc emen al e ec o a ea men on an
indi idual’s beha io . This means hey es ima e no whe he a cus ome will buy, bu whe he hey will buy because
hey ecei ed a p omo ion. Upli modeling segmen s use s in o ou ca ego ies: Pe suadables (posi i e ea men
e ec ), Su e Things (would ac ega dless), Los Causes (una ec ed), and Do No Dis u b (nega i ely a ec ed) [13]. This
a ge ing p ecision d ama ically imp o es ma ke ing ROI and educes cus ome a igue.
Ano he b eak h ough in causal ML is Causal Fo es s, an adap a ion o andom o es s ha es ima es Condi ional
A e age T ea men E ec s (CATEs). Causal o es s pa i ion he ea u e space o unco e he e ogenei y in ea men
esponse, allowing o g anula , segmen -le el op imiza ion [14]. Fo example, a bank may use causal o es s o
de e mine which c edi isk segmen s bene i mos om inancial li e acy in e en ions, in o ming pe sonalized
s a egies.
Deep lea ning models, pa icula ly con olu ional neu al ne wo ks (CNNs), a e also en e ing he causal in e ence space.
While CNNs a e p ima ily known o image and pa e n ecogni ion, hey can be epu posed o ex ac s uc u ed
ea u es om high- equency beha io al sequences—such as clicks eams o ime-s amped in e ac ions [15]. These
ea u es se e as ich inpu s in o causal models, enabling ine-g ained segmen a ion and ea men e ec p edic ion.
Fo ins ance, CNNs can de ec nuanced empo al pa e ns in use beha io ha signal esponsi eness o e en ion
s a egies.
As causal ML con inues o e ol e, i b idges he gap be ween explainabili y and scalabili y. I empowe s en e p ises no
only o p edic wha will happen, bu o s a egically in luence ou comes h ough a ge ed, e idence-based
in e en ions— u ning passi e analy ics in o p oac i e le e s o o ganiza ional pe o mance.
Figu e 1 Compa a i e diag am o p edic i e s. causal AI pipelines
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
899
3. Theo e ical amewo k
3.1. S uc u al Causal Models and Do-Calculus
S uc u al Causal Models (SCMs) o m he backbone o mode n causal in e ence, o e ing a ma hema ical and g aphical
amewo k o ep esen cause-and-e ec ela ionships in complex sys ems. Unlike pu ely s a is ical models ha ocus
on co ela ion, SCMs explici ly de ine how a iables in luence each o he h ough s uc u al equa ions and causal
diag ams [11]. These diag ams a e ypically ep esen ed as Di ec ed Acyclic G aphs (DAGs), whe e each node deno es
a a iable, and edges ep esen causal pa hways. This s uc u e allows analys s o isualize and es assump ions abou
con ounding a iables, media o s, and collide s, which a e o en in isible in s anda d p edic i e modeling.
One o he mos powe ul aspec s o SCMs is hei abili y o dis inguish be ween obse a ion and in e en ion. This is
ope a ionalized using do-calculus, a o mal sys em in oduced by Judea Pea l ha enables easoning abou
in e en ional dis ibu ions such as P(Y | do(X))— he p obabili y o an ou come Y when a a iable X is o cibly se ,
simula ing a eal-wo ld in e en ion [12]. Do-calculus includes a se o ules ha help de i e pos -in e en ion
dis ibu ions om obse a ional da a unde speci ic assump ions encoded in he DAG.
Fo ins ance, in a ma ke ing con ex , SCMs can be used o de e mine whe he sending a discoun o e (X) causes an
inc ease in con e sion (Y), o whe he bo h a e in luenced by an unobse ed a iable like cus ome loyal y. By
iden i ying and adjus ing o hese con ounde s, businesses can make c edible claims abou causal impac .
SCMs hus p o ide he heo e ical ounda ion o causal AI in s a egic planning. They ans o m business analy ics om
a e ospec i e exe cise o a o wa d-looking simula ion engine, capable o e alua ing hypo he ical scena ios and
planning in e en ions ha a e likely o yield measu able and in en ional e ec s [13].
3.2. Coun e ac ual Reasoning and Business Simula ion
Coun e ac ual easoning is a key componen o causal analysis ha seeks o answe “wha i ” ques ions—wha would
ha e happened i a di e en ac ion o decision had been aken. This is dis inc om obse a ional o e en in e en ional
easoning, as i in ol es cons uc ing a hypo he ical al e na i e eali y based on he same unde lying da a s uc u e
[14]. In business con ex s, coun e ac uals allow decision-make s o simula e and compa e ou comes unde di e en
s a egic scena ios, guiding esou ce alloca ion and policy design.
Technically, coun e ac uals a e modeled wi hin he amewo k o S uc u al Causal Models. The p ocess begins by
obse ing a eal-wo ld ins ance—e.g., a cus ome who ecei ed a e en ion o e and enewed hei subsc ip ion. Using
he model, a coun e ac ual es ima e is p oduced o in e whe he he cus ome would ha e enewed wi hou he o e .
This allows businesses o es ima e indi idual ea men e ec s (ITEs) and segmen cus ome s by esponsi eness,
in o ming mo e e icien in e en ion s a egies [15].
Business simula ion using coun e ac uals enables en e p ises o conduc i ual expe imen s a scale. Ra he han
elying solely on A/B es ing, which can be cos ly and ime-consuming, i ms can simula e a ious ope a ional s a egies
and e alua e hei p ojec ed ou comes. Fo example, a logis ics company may model how educing deli e y ime impac s
cus ome e en ion ac oss di e en egions, con olling o seasonal a ia ion and p oduc ca ego y [16].
Coun e ac ual easoning also suppo s ai ness audi ing and impac assessmen . O ganiza ions can e alua e whe he
in e en ions ha e dispa a e e ec s ac oss demog aphic g oups, helping ensu e e hical and inclusi e decision-making.
As such, i o ms he compu a ional co e o causal AI’s abili y o simula e, adap , and op imize business s a egies in
unce ain en i onmen s.
3.3. F amewo k o Causal AI in S a egic Planning
The in eg a ion o causal AI in o s a egic planning in ol es a sys ema ic amewo k ha connec s in e en ions o long-
e m business ou comes. This amewo k begins wi h p oblem o maliza ion, whe e decision-make s de ine he
ea men (e.g., p icing change), he ou come (e.g., p o i ma gin), and ele an con ounde s (e.g., p oduc seasonali y).
A S uc u al Causal Model is hen cons uc ed o encode domain knowledge and assump ions in he o m o a causal
g aph [17].
Nex , ea u e ep esen a ions a e ex ac ed om his o ical beha io al o ansac ional da a. I complex pa e ns exis —
such as in ime-se ies logs o use in e ac ion lows—deep lea ning models like CNNs can be used o gene a e compac
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
900
and exp essi e ea u e embeddings. These embeddings a e passed in o causal ML models such as upli models o causal
o es s o es ima e he e ogeneous ea men e ec s.
The amewo k culmina es in a decision simula ion laye , whe e planne s can e alua e “wha -i ” scena ios and op imize
in e en ions. By using coun e ac uals and do-calculus, he long- e m ipple e ec s o s a egies—such as ma ke ing
pe sonaliza ion, p oduc bundling, o policy e o m—can be assessed ac oss mul iple cus ome segmen s [18].
This end- o-end pipeline allows s a egic eams o mo e beyond p edic ion and in o causal easoning, enabling
e idence-based, adap i e planning ha di ec ly ies analy ic insigh s o o ganiza ional pe o mance and esilience.
4. Me hodology
4.1. Resea ch Design O e iew
This s udy employs a mixed-me hod esea ch design, in eg a ing empi ical beha io al da a wi h machine lea ning and
causal in e ence modeling o unco e la en d i e s o s a egic business ou comes. The app oach blends quan i a i e
da a analysis, including he use o Con olu ional Neu al Ne wo ks (CNNs) o ea u e ex ac ion, wi h causal modeling
echniques such as upli modeling and causal o es s o simula e in e en ion e ec s ac oss di e se cus ome segmen s
[15]. This combina ion allows o he iden i ica ion o no only high-le el pa e ns bu also causal ela ionships ha
in o m ac ionable s a egic decisions.
The esea ch is s uc u ed a ound h ee co e s ages: da a p epa a ion and ea u e enginee ing, causal model aining
and e alua ion, and simula ion o s a egic scena ios. Beha io al da a is collec ed om a eal-wo ld e-comme ce
pla o m o e a mul i-pe iod ime ame, cap u ing use s’ in e ac ions wi h p oduc s, campaigns, and digi al
expe iences. The model de elopmen p ocess begins by encoding hese beha io s in o s uc u ed inpu s using domain-
speci ic and da a-d i en ea u e cons uc ion me hods, including Recency-F equency-Mone a y (RFM) and session
unnel analysis [16].
Once ans o med, he beha io al ea u es a e p ocessed h ough CNN laye s o cap u e empo al and spa ial
dependencies, p oducing dense embeddings ha ep esen each use 's in e ac ion signa u e. These embeddings a e
hen ed in o causal ML models, ained o es ima e Condi ional A e age T ea men E ec s (CATEs) o speci ic
in e en ions such as p omo ional emails o loyal y p og am exposu e [17].
The inal s age in ol es he applica ion o coun e ac ual easoning o simula e "wha -i " business scena ios unde
a ying le els o s a egic in e en ion. This end- o-end pipeline— om da a cap u e o simula ion—p o ides a obus
amewo k o e alua ing he long- e m impac o s a egic ac ions in a dynamic en i onmen . The mixed-me hod
design ensu es bo h s a is ical igo and eal-wo ld ele ance, enabling o ganiza ions o blend p edic i e p ecision wi h
e idence-based s a egic o esigh .
4.2. Da ase and Business Con ex
The da ase used in his s udy o igina es om a mid-sized e-comme ce e aile ope a ing in he consume elec onics
and li es yle ca ego y. I encompasses a 12-mon h pe iod o use in e ac ion logs, ansac ional eco ds, and ma ke ing
campaign da a. The goal is o unde s and he impac o a ious s a egic in e en ions—such as a ge ed discoun s,
pe sonalized ecommenda ions, and loyal y ewa ds—on key business ou comes including e enue pe use , e en ion,
and engagemen longe i y [18].
The da ase includes o e 150,000 unique use IDs, each linked o a ime-s amped sequence o ac ions, including page
isi s, ca addi ions, p oduc iews, and pu chases. Each use session is segmen ed ch onologically, allowing o he
cap u e o eal- ime beha io al sequences. In addi ion o beha io al logs, he da a includes me ada a such as de ice
ype, b owse , e e al sou ce, and campaign exposu e, p o iding con ex o use in e ac ions ac oss ma ke ing and
ansac ional ouchpoin s.
T ansac ional his o y is de ailed, ea u ing o de amoun s, imes amps, i em ca ego ies, and discoun u iliza ion.
Campaign da a includes in o ma ion on email opens, click- h ough a es, and con e sion om speci ic p omo ional
lows. These componen s allow o he de ini ion o ea men s, such as eceip o a p omo ional code o display o a
ecommended bundle, and measu emen o ou comes, such as subsequen spending o epea isi s [19].
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
901
This da ase is well-sui ed o causal in e ence due o he p esence o na u ally occu ing ea men -con ol pai s, non-
andom exposu e a ia ions, and ich empo al g anula i y. The business con ex —whe e equen mic o-decisions can
agg ega e in o long- e m cus ome alue—p o ides an ideal en i onmen o es he e ec i eness o causal AI o
s a egic planning and alida e i s po en ial o d i e sus ainable business pe o mance.
4.3. Fea u e Enginee ing
Fea u e enginee ing is a c i ical s ep in ansla ing aw beha io al logs in o meaning ul inpu s o causal and p edic i e
modeling. In his s udy, we employ bo h domain-speci ic echniques and au oma ed ea u e ex ac ion using CNNs o
de i e a obus beha io al ea u e se . The ounda ion o he ea u e space is he RFM amewo k—cap u ing ecency
( ime since las pu chase), equency (numbe o pu chases o e ime), and mone a y alue ( o al spend). These me ics
o e a compac ep esen a ion o cus ome alue and li ecycle s age [20].
Addi ional ea u es a e cons uc ed o cap u e session-le el dynamics, including a e age session du a ion, click dep h,
bounce a e, and ime-o -day ac i i y. Use s a e u he cha ac e ized h ough con e sion unnel analysis, iden i ying
d op-o poin s and mo emen ac oss key s ages such as p oduc iew → add- o-ca → checkou → pu chase. These
unnel-based ea u es e lec in en and beha io al ic ion, o e ing key signals o segmen a ion and a ge ing [21].
De i ed ea u es also include p omo ion esponsi eness, calcula ed as he di e ence in engagemen o e enue
be ween ea ed and un ea ed pe iods. This a iable is use ul o aining upli models, whe e he goal is o isola e he
inc emen al impac o in e en ions. Fu he mo e, con en di e si y, ime since las in e ac ion, and de ice-swi ch
equency a e encoded o cap u e digi al beha io ichness.
To suppo deep lea ning-based modeling, aw ime-se ies da a om use sessions is p ocessed h ough a CNN, using
sliding windows o e ain empo al o de and encode sequen ial dependencies. The con olu ional laye s lea n
hie a chical in e ac ion pa e ns ha a e no easily cap u ed h ough s a ic ea u es, allowing he model o dis inguish
be ween ou ine and anomalous beha io [22].
The inal ea u e ma ix in eg a es manually enginee ed and CNN-de i ed ea u es, ensu ing ha bo h expe
knowledge and da a-d i en ep esen a ions in o m he causal modeling pipeline. This dual-laye ea u e s a egy
suppo s obus es ima ion o ea men e ec s and enables ine-g ained s a egic simula ions ac oss beha io al
a che ypes.
4.4. Causal Modeling Pipeline wi h CNN/ML
The p oposed causal modeling pipeline in eg a es deep lea ning-based ea u e embedding, machine lea ning-d i en
ea men e ec es ima ion, and coun e ac ual simula ion o e alua e s a egic in e en ions. This pipeline is designed
o quan i y he impac o speci ic business ac ions—such as p omo ions o use in e ace (UX) changes—on key
pe o mance ou comes. The i e-s ep amewo k ou lined below enables scalable, da a-d i en s a egic planning
g ounded in causal in e ence p inciples.
4.4.1. S ep 1: Fea u e Embedding Using CNN
The i s s ep in ol es p ocessing aw, ime-s amped beha io al da a using Con olu ional Neu al Ne wo ks (CNNs).
Use in e ac ions—such as page isi s, sea ch que ies, and ca modi ica ions—a e s uc u ed in o sequen ial ime-
se ies windows. These windows a e con e ed in o mul i-dimensional ma ices whe e ows ep esen ime in e als
and columns ep esen encoded e en s (e.g., e en ype, p oduc ca ego y, engagemen sco e). CNNs a e applied o hese
ma ices o ex ac hie a chical spa ial- empo al ea u es ha cap u e pa e ns like epea ed isi s be o e pu chase o
in e ac ions wi h high- alue p oduc s [19]. This app oach allows o au oma ic lea ning o beha io al embeddings ha
e lec no only equency bu also sequence and in ensi y o ac ions.
4.4.2. S ep 2: T ea men De ini ion
Once use beha io has been embedded, he nex s ep is o de ine he ea men a iable— he speci ic business
in e en ion whose causal impac we aim o es ima e. In he con ex o his s udy, ea men s include eceip o
p omo ional emails, exposu e o ecommended bundles, and in e ace edesigns (e.g., pe sonalized landing pages).
These ea men s a e eco ded as bina y o ca ego ical indica o s, whe e ea ed uni s ecei ed he in e en ion while
con ol uni s did no . C ucially, ea men s mus a y na u ally ac oss he popula ion and be ime-s amped o enable
empo al alignmen wi h ou comes [20].
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
902
4.4.3. S ep 3: Ou come Me ics
To es ima e ea men e ec s, clea ly de ined ou come a iables a e needed. Fo his pipeline, we conside mul iple
ou comes aligned wi h long- e m business goals, including:
Re enue pe use o e a 30-day window pos - ea men
Engagemen longe i y, measu ed by he numbe o sessions o ac i e days
Con e sion a e, indica ing he p opo ion o ea ed use s who comple ed a pu chase. These ou comes a e selec ed
o hei ele ance o bo h inancial pe o mance and cus ome e en ion. Each ou come is lagged app op ia ely om
he ea men o e lec ealis ic causali y and a oid simul anei y bias [21].
4.4.4. S ep 4: Upli Modeling and CATE Es ima ion
Wi h ea u es, ea men s, and ou comes de ined, he pipeline mo es o upli modeling, which es ima es he Condi ional
A e age T ea men E ec (CATE) o indi idual use s. Upli models di e om adi ional classi ie s by p edic ing he
di e ence in ou come likelihood be ween ea ed and un ea ed condi ions. We employ Causal Fo es s—an ex ension
o Random Fo es s buil o pa i ion he ea u e space while p ese ing ea men -con ol in eg i y [22]. Causal Fo es s
p o ide no only a e age ea men e ec s bu also he e ogeneous e ec s ac oss use segmen s.
Al e na i ely, a T-Lea ne amewo k may be used, which ains wo sepa a e models—one on ea ed use s and
ano he on un ea ed— o lea n he ou come unc ion o each g oup. The indi idual ea men e ec is hen es ima ed
as he di e ence in p edic ed ou comes be ween he wo models o each use [23]. These models a e e alua ed using
me ics such as Qini coe icien , upli a op decile, and ou -o -sample ea men e ec calib a ion.
4.4.5. S ep 5: Simula ion o In e en ions ia Coun e ac ual Es ima ion
The inal s ep in ol es simula ing s a egic decisions by le e aging coun e ac ual es ima es. Fo each use , he pipeline
gene a es wo po en ial ou comes: one assuming he ea men occu ed, and one assuming i did no . The di e ence
ep esen s he indi idual ea men e ec , which can be agg ega ed ac oss segmen s o es ima e expec ed business
impac unde a ying le els o in e en ion co e age [24].
Fo example, simula ion can es ima e he o al e enue gain i only he op 30% o use s ( anked by p edic ed upli )
ecei e a p omo ion. These simula ions in o m budge alloca ion, campaign a ge ing, and s a egic p io i iza ion.
Fu he mo e, by unning simula ions unde mul iple hypo he ical ea men s, he pipeline can suppo policy
op imiza ion, helping leade s choose he mos impac ul in e en ion o each segmen [25].
In summa y, his end- o-end pipeline— om CNN-based embedding o coun e ac ual simula ion—p o ides a scalable,
in e p e able, and ac ionable causal in e ence amewo k. I empowe s o ganiza ions o mo e beyond p edic ion and
owa d e idence-based s a egic expe imen a ion, ensu ing ha e e y decision is aligned wi h measu able, long- e m
pe o mance ou comes.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
903
Figu e 2 Flow diag am o he pipeline in eg a ing CNN, ea men modeling, and causal in e ence
5. Implemen a ion and expe imen al se up
5.1. Da a P ep ocessing and CNN T aining
E ec i e da a p ep ocessing is ounda ional o any deep lea ning pipeline, especially when using Con olu ional Neu al
Ne wo ks (CNNs) o ex ac beha io al ea u es om use session logs. The p ep ocessing phase ensu es ha inpu
da a is s uc u ed, no malized, and encoded in a way ha cap u es bo h empo al pa e ns and e en seman ics.
The i s s ep in ol es segmen ing aw beha io al da a using sliding windows. Each use session is b oken in o ixed-
size in e als— o example, 15-minu e o 1-hou chunks—whe e each window cap u es a sequence o e en s such as
clicks, page isi s, and ca in e ac ions. Sliding windows help p ese e ch onological dependencies while enabling he
model o lea n pa e ns in use beha io o e ime. Each window is ep esen ed as a 2D ma ix, whe e ows co espond
o ime in e als and columns o e en ypes, nume ical encodings, o in ensi y sco es [23].
Nex , all beha io al ea u es wi hin each session a e subjec ed o no maliza ion. Con inuous a iables, such as ime
spen on a p oduc page o ca alue, a e scaled using min-max no maliza ion o i wi hin he [0,1] ange. This helps
p e en skewed lea ning due o dispa a e ea u e magni udes. Ca ego ical a iables, such as de ice ype o e e al
channel, a e one-ho encoded o p ese e ca ego ical dis inc ions while main aining compa ibili y wi h CNN il e s [24].
Once he p ep ocessing is comple e, hese ma ices a e inpu in o a 1D o 2D CNN a chi ec u e, depending on he
g anula i y o he empo al s uc u e. The ne wo k ypically includes con olu ional laye s o pa e n de ec ion, pooling
laye s o educe dimensionali y, and ully connec ed laye s o embedding gene a ion. D opou and ba ch no maliza ion
a e inco po a ed o minimize o e i ing and s abilize aining. The ou pu embeddings se e as condensed beha io al
signa u es ha e lec empo al engagemen pa e ns and in e ac ion dep h [25].
These embeddings a e used as inpu s o he subsequen upli modeling phase, ensu ing ha he causal models ope a e
on obus , empo ally-awa e ep esen a ions o use beha io .
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
910
Thi d, go e nance mechanisms mus ensu e anspa ency, ai ness, and epea abili y in causal es ima es. This includes
documen a ion o ea men de ini ions, e sioning o causal g aphs, and egula audi s o model d i and bias. E hical
o e sigh is pa icula ly c ucial when in e en ions a ec use expe ience, access, o p icing.
By embedding hese elemen s, en e p ises can align Causal AI wi h hei s a egic p io i ies and compliance s anda ds,
accele a ing he ansi ion om eac i e o p oac i e decision-making.
8.3. Di ec ions o Fu u e Resea ch
Fu u e esea ch can explo e he applica ion o Causal AI ac oss b oade o ganiza ional domains. In En i onmen al,
Social, and Go e nance (ESG) con ex s, causal modeling can be used o assess he long- e m impac o sus ainabili y
ini ia i es, egula o y compliance, o communi y in es men s, enabling i ms o quan i y non- inancial e u ns and align
wi h esponsible business goals.
In Human Resou ces (HR), causal in e ence could guide hi ing p ac ices, di e si y in e en ions, and employee
de elopmen p og ams, ensu ing ha HR policies esul in measu able imp o emen s in pe o mance, engagemen , and
e en ion.
Finally, in supply chain esilience, Causal AI can help iden i y weak links and simula e dis up ion scena ios, guiding
in es men in edundancy, agili y, o local sou cing s a egies. By modeling he ipple e ec s o logis ics decisions,
o ganiza ions can imp o e obus ness wi hou incu ing excessi e cos .
These ex ensions a i m ha causal easoning is no limi ed o ma ke ing o inance bu ep esen s a scalable analy ical
lens o managing complexi y and achie ing sus ainable impac ac oss he en e p ise.
Re e ences
[1] Ghazi KM, Salem IE, Da H, Elbaz AM. Le e aging s a egic leade ship o boos ing ope a ional esilience in ho els:
he ole o c isis esponse s a egies and e- eadiness. In e na ional Jou nal o Con empo a y Hospi ali y
Managemen . 2024 Sep 2;36(10):3300-23.
[2] Alghamdi OA, Agag G. Boos ing inno a ion pe o mance h ough big da a analy ics powe ed by a i icial
in elligence use: an empi ical explo a ion o he ole o s a egic agili y and ma ke u bulence. Sus ainabili y.
2023 Sep 27;15(19):14296.
[3] Noah GU. In e disciplina y s a egies o in eg a ing o al heal h in na ional immune and in lamma o y disease
con ol p og ams. In J Compu Appl Technol Res. 2022;11(12):483-498. doi:10.7753/IJCATR1112.1016.
[4] Le BT, Liu X. D i ing Social Sus ainabili y and Resilience o SMEs Amids Tu bulence: The Role o In o ma ion
Technology and Dynamic Capabili ies.
[5] González-Mohíno M, Dona e MJ, Muñoz-Fe nández GA, Cabeza-Ramí ez LJ. Robo ic digi aliza ion and business
success: he cen al ole o us and leade ship in ope a ional e iciency. A hyb id app oach using PLS-SEM and
sQCA. IEEE Access. 2024 Dec 2.
[6] Chukwunweike Joseph, Salaudeen Habeeb Dolapo. Ad anced Compu a ional Me hods o Op imizing Mechanical
Sys ems in Mode n Enginee ing Managemen P ac ices. In e na ional Jou nal o Resea ch Publica ion and
Re iews. 2025 Ma ;6(3):8533-8548. A ailable om: h ps://ij p .com/uploads/V6ISSUE3/IJRPR40901.pd
[7] Olan F, A akpogun EO, Suklan J, Nakpodia F, Damij N, Jayawick ama U. A i icial in elligence and knowledge
sha ing: Con ibu ing ac o s o o ganiza ional pe o mance. Jou nal o Business Resea ch. 2022 Jun 1;145:605-
15.
[8] Nei oukh S, Emeagwali OL, Aljuhmani HY. A i icial in elligence capabili y and o ganiza ional pe o mance:
un a eling he media ing mechanisms o decision-making p ocesses. Managemen Decision. 2024 Jun 13.
[9] Emi-Johnson Oluwabukola, Fasanya Oluwa unmibi, Adeniyi Ayodele. P edic i e c op p o ec ion using machine
lea ning: A scalable amewo k o U.S. Ag icul u e. In J Sci Res A ch. 2024;15(01):670-688. A ailable om:
h ps://doi.o g/10.30574/ijs a.2024.12.2.1536
[10] Co ales-Es ada AM, Gómez-San os LL, Be nal-To es CA, Rod iguez-López JE, Sando al-Reyes J. Dynamic
capabili ies con ibu ion o o ganiza ional pe o mance: he media o ole o business con inui y o dina y
capabili ies. Cogen Business & Managemen . 2025 Dec 12;12(1):2493315.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
911
[11] Kababii o Lillian. Ha nessing A i icial In elligence o Real-Time Compliance in he U.S. Oil & Gas Sec o :
Enhancing Tax Accu acy, Cu bing E asion, and Unlocking Re enue G ow h h ough In elligen Au oma ion.
In e na ional Jou nal o Compu e Applica ions Technology and Resea ch. 2025;14(05):55–70.
doi:10.7753/IJCATR1405.1006.
[12] Depino-Besada NI, Sa al A, León-Ma eos F, Llach J. Explo ing he ela ionships be ween digi al ans o ma ion,
o ganiza ional slack and business pe o mance: a con igu a ional app oach. Business P ocess Managemen
Jou nal. 2025 Ma 14;31(2):416-42.
[13] Emi-Johnson Oluwabukola, Nk umah Kwame, Folasole Ade ayo, Amusa Tope Kolade. Op imizing machine
lea ning o imbalanced classi ica ion: Applica ions in U.S. heal hca e, inance, and secu i y. In J Eng Technol Res
Manag. 2023 No ;7(11):89. A ailable om: h ps://doi.o g/10.5281/zenodo.15188490
[14] Va go J, Se ille E. C isis s a egic planning o SMEs: inding he sil e lining. In e na ional Jou nal o p oduc ion
esea ch. 2011 Sep 15;49(18):5619-35.
[15] O e o-Bla M, Palacios-Ma qués D, Leal-Rod iguez AL, Fe a is A. Beyond digi al ans o ma ion: a mul i-mixed
me hods s udy on big da a analy ics capabili ies and inno a ion in enhancing o ganiza ional pe o mance.
Re iew o Manage ial Science. 2025 Feb;19(2):649-85.
[16] Chukwunweike J, Lawal OA, A ogundade JB, Alade B. Na iga ing e hical challenges o explainable AI in
au onomous sys ems. In e na ional Jou nal o Science and Resea ch A chi e. 2024;13(1):1807–19.
doi:10.30574/ijs a.2024.13.1.1872. A ailable om: h ps://doi.o g/10.30574/ijs a.2024.13.1.1872.
[17] Balaji K. Explo ing he D i e s and E ec s on Supply Chain Resilience and Pe o mance in an Eme ging Ma ke
Using A i icial In elligence. In2023 3 d In e na ional Con e ence on Sma Gene a ion Compu ing,
Communica ion and Ne wo king (SMART GENCON) 2023 Dec 29 (pp. 1-6). IEEE.
[18] Olayinka OH. Big da a in eg a ion and eal- ime analy ics o enhancing ope a ional e iciency and ma ke
esponsi eness. In J Sci Res A ch. 2021;4(1):280–96. A ailable om:
h ps://doi.o g/10.30574/ijs a.2021.4.1.0179
[19] Khan S, Mehmood S, Khan SU. Na iga ing inno a ion in he age o AI: how gene a i e AI and inno a ion in luence
o ganiza ional pe o mance in he manu ac u ing sec o . Jou nal o Manu ac u ing Technology Managemen .
2025 Ma 20;36(3):597-620.
[20] Abdulsalam A, Okechukwu M, Oluko un K, Onagun Q. Analysis o bio-enhance s o pH and iscosi y con ol in
d illing luid sys ems. In . J. Res. Inno . Appl. Sci.(IJRIAS). 2020(I).
[21] Olayinka OH. Da a d i en cus ome segmen a ion and pe sonaliza ion s a egies in mode n business in elligence
amewo ks. Wo ld Jou nal o Ad anced Resea ch and Re iews. 2021;12(3):711–726. doi:
h ps://doi.o g/10.30574/wja .2021.12.3.0658.
[22] Jou abchi Ami khizi P, Ped ammeh S, Pakzad S, Shahhoseini A. Gene a i e A i icial In elligence in Adap i e
Social Manu ac u ing: A Pa hway o Achie ing Indus y 5.0 Sus ainabili y Goals. P ocesses. 2025 Ap
12;13(4):1174.
[23] Diyaolu Ch is ianah Omolola. Mul i-agen AI Sys ems o Adap i e, Cul u ally-Conco dan Ca e Rou ing in
Pos pa um Dep ession Ac oss Medicaid-Dependen Popula ions. In J Ad Res Publ Re [In e ne ]. 2025
May;2(5):1–20. A ailable om: h ps://ija p .com/uploads/V2ISSUE5/IJARPR0601.pd
[24] Olayinka OH. E hical implica ions and go e nance o AI models in business analy ics and da a science
applica ions. In e na ional Jou nal o Enginee ing Technology Resea ch & Managemen . 2022 No ;6(11). doi:
h ps://doi.o g/10.5281/zenodo.15095979.
[25] Riipa MB, Begum N, H iday MS, Haque SA. Role o da a analy ics in enhancing business decision-making and
ope a ional e iciency. In e na ional Jou nal o Communica ion Ne wo ks and In o ma ion Secu i y.
2025;17(2):400-12.
[26] Malik P. Le e aging Causal Models o C a AI S a egy. In e na ional Jou nal o Simula ion--Sys ems, Science &
Technology. 2024 Ma 1;25(1).
[27] Fi ko -No is E, Koche a N. Le e aging AI o s a egic o esigh : Un eiling u u e ho izons. InImp o ing and
Enhancing Scena io Planning 2025 Feb 13 (pp. 286-305). Edwa d Elga Publishing.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 895-912
912
[28] Ca ayannis EG, Dumi escu R, Falkowski T, Papamichail G, Zo a NR. Enhancing SME Resilience h ough A i icial
In elligence and S a egic Fo esigh : A F amewo k o Sus ainable Compe i i eness. Technology in Socie y. 2025
Feb 6:102835.
[29] AlSaidi H, C ow he D. S a egic Fo esigh and Business Analy ics: A Sys ema ic Explo a ion o Media ed Impac s
on O ganisa ional Resilience. InSocial Responsibili y, Technology and AI 2024 No 18 (pp. 49-68). Eme ald
Publishing Limi ed.
[30] Ga ido-Mo eno A, Ma ín-Rojas R, Ga cía-Mo ales VJ. The key ole o inno a ion and o ganiza ional esilience in
imp o ing business pe o mance: A mixed-me hods app oach. In e na ional Jou nal o In o ma ion
Managemen . 2024 Aug 1;77:102777.
[31] Olayinka OH. Causal in e ence and coun e ac ual easoning in high-dimensional da a analy ics o obus
decision in elligence. In J Eng Technol Res Manag. 2024.
[32] Lin J, Wu S, Luo X. How does big da a analy ics capabili y a ec i m pe o mance? Un eiling he ole o
o ganisa ional esilience and en i onmen al dynamism. Eu opean Jou nal o In o ma ion Sys ems. 2024 Jul 8:1-
27.
[33] DENG J. Resea ch on he impac o AI echnology adop ion on o ganiza ional esilience o se ice en e p ises.
[34] Masiale i M, Talaei-Khoei A, Yang AT. Re ealing he ole o explainable AI: How does upda ing AI applica ions
gene a e agili y-d i en pe o mance?. In e na ional Jou nal o In o ma ion Managemen . 2024 Aug 1;77:102779.
[35] Wang D, Cao X. A i icial in elligence ad e si y e en , in e -o ganisa ional us , and i m esilience: he
mode a ing e ec o esponsible inno a ion. Technology Analysis & S a egic Managemen . 2024 Aug 10:1-4.
[36] Roux M, Chowdhu y S, Kuma Dey P, Vann Ya oson E, Pe ei a V, Abadie A. Small and medium-sized en e p ises
as echnology inno a ion in e media ies in sus ainable business ecosys em: in e play be ween AI adop ion, low
ca bon managemen and esilience. Annals o Ope a ions Resea ch. 2023 Dec 29:1-50.
[37] Ode E, Awolowo IF, Nana R, Olawoyin FS. Social capi al and A i icial in elligence eadiness: he media ing ole o
cybe esilience and alue cons uc ion o SMEs in esou ce-cons ained en i onmen s. In o ma ion Sys ems
F on ie s. 2025 Ap 30:1-21.
[38] Zhang J, Long J, Von Schaewen AM. How does digi al ans o ma ion imp o e o ganiza ional esilience?— indings
om PLS-SEM and sQCA. Sus ainabili y. 2021 Oc 18;13(20):11487.
[39] de Sousa Jabbou AB, La an H, Jabbou CJ, Seles BM. Does applying a ci cula business model lead o
o ganiza ional esilience? Media ing e ec s o indus y 4.0 and cus ome s in eg a ion. Technological Fo ecas ing
and Social Change. 2023 Sep 1;194:122672.
[40] Ju idha abongse J, Imjai N, Pan a uk S, Su bak i LP, Auji apongpan S. Explo ing he e ec o managemen con ol
sys ems on dynamic capabili ies and sus ainabili y pe o mance: The ole o open inno a ion s a egy amids
COVID-19. Jou nal o Open Inno a ion: Technology, Ma ke , and Complexi y. 2024 Ma 1;10(1):100224.
[41] Chen Y. Examining he dynamics o en ep eneu ial knowledge and i m pe o mance: a longi udinal s udy o
s a -ups in eme ging ma ke s. Jou nal o he Knowledge Economy. 2024 May 23:1-33.
