*Co esponding au ho : Aleksand Buinõi.
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 Liscense 4.0.
De elopmen and implemen a ion o a p edic i e analy ical model o op imizing
in en o y managemen in he B2C sec o in highly compe i i e online ma ke s
Aleksand Buinõi *
Bachelo o Compu e Sys ems, Depa men o Compu e Enginee ing, Facul y o In o ma ion Technology, Tallinn
Uni e si y o Technology, Tallinn, Ha ju Coun y, Es onia.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1701-1707
Publica ion his o y: Recei ed on 10 July 2025; e ised on 20 Augus 2025; accep ed on 22 Augus 2025
A icle DOI: h ps://doi.o g/10.30574/wja .2025.27.2.2984
Abs ac
In he con ex o apid g ow h in sales olumes and in ensi ying compe i ion on global B2C ma ke places, e ec i e
in en o y managemen plays a decisi e ole in ensu ing p o i abili y and long- e m business sus ainabili y. This s udy
p oposes a concep ual p edic i e analy ical model aimed a op imizing s ock managemen in companies ope a ing on
highly compe i i e online pla o ms such as Amazon. The objec i e is o de elop a hyb id demand- o ecas ing sys em
ha combines classical ime-se ies me hods (SARIMA) wi h mode n g adien boos ing algo i hms in o de o ensu e
adap abili y o apidly changing ma ke condi ions, accoun o seasonal luc ua ions, long- e m ends, and he
in luence o ex e nal ac o s. The me hodological basis comp ises a c i ical e iew and syn hesis o key publica ions
om ecen yea s, as well as he use o p op ie a y da a on he company Skysales L d. o demons a e he e ec i eness
o he app oach. The esul s ob ained indica e an inc ease in he accu acy o consume demand o ecas s, which leads
o a educ ion in excess in en o y, a dec ease in los sales, and accele a ed capi al u no e . The scien i ic no el y o he
wo k lies in he o ma ion o a hyb id model a chi ec u e speci ically adap ed o he needs o small and medium-sized
en e p ises in he B2C e-comme ce sec o wi h a b oad and dynamically upda ed asso men . This a icle will be use ul
bo h o academic esea che s in he ield o supply chain managemen and da a analys s, and o p ac i ione s—
execu i es and e-comme ce manage s—seeking o imp o e he ope a ional e iciency o hei en e p ises.
Keywo ds: In en o y Managemen ; P edic i e Analy ics; B2C; E-Comme ce; Online Ma ke s; Machine Lea ning;
Demand Fo ecas ing; Op imiza ion; G adien Boos ing; SARIMA
1. In oduc ion
The business- o-consume segmen o elec onic comme ce is unde going an unp eceden ed phase o expansion.
Acco ding o expe es ima es, he global olume o online ade will each app oxima ely 4.32 illion USD by 2025 [1].
In an en i onmen o in ensi ied compe i ion, pa icula ly on la ge ma ke places such as Amazon, which suppo s i s
Eu opean selle s h ough a ne wo k o mo e han 250 ul illmen and so a ion cen e s, as well as ai deli e y and
deli e y cen e s, in en o y u no e and in en o y accoun ing accu acy become key ac o s o compe i i eness [2].
Ine icien in en o y managemen leads ei he o excessi e accumula ion o goods — o e s ocking, which eezes
wo king capi al and inc eases s o age cos s wi h he isk o obsolescence, o o sho ages o high-demand i ems, which
no only educes sales in he sho e m bu also unde mines consume loyal y, p omp ing swi ching o al e na i e
selle s.
T adi ional me hods o in en o y le el planning ha ely on a e age sales me ics om pas pe iods and expe
judgmen p o e insu icien ly adap i e in he apidly changing en i onmen o online e ail, whe e demand noise is
high, asso men li e cycles a e sho ened, and consume ends a e con inually ans o ming. The ele ance o he s udy
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1701-1707
1702
is de e mined by he need o shi om eac i e o p oac i e decision-making ools o p ocu emen and wa ehouse
logis ics. P edic i e analy ics, based on he applica ion o s a is ical algo i hms and machine lea ning me hods o
iden i y pa e ns in his o ical da a and o cons uc models o u u e demand, is conside ed one o he mos p omising
solu ions. Al hough among la ge e aile s he di usion o AI echnologies has al eady eached a signi ican le el —
acco ding o NVIDIA, 89% o esponden s epo ed using a i icial in elligence o p edic i e analy ics [3] — in he SME
segmen he adop ion o such sys ems emains cons ained by limi a ions in human expe ise, compu ing esou ces,
and ailo ed so wa e solu ions.
The esea ch gap mani es s i sel in a sca ci y o comp ehensi e and p ac ice-o ien ed p edic i e models capable o
accoun ing o he speci ics o compe i i e p essu e and demand ins abili y on SME B2C pla o ms. Exis ing s udies
gene ally ocus ei he on compa a i e analyses o indi idual ime-se ies algo i hms o on p ac ical cases o la ge
companies wi h well-es ablished supply chains.
● The aim o he s udy is o de elop a hyb id demand o ecas ing sys em ha combines classical ime-se ies analysis
me hods (SARIMA) wi h mode n g adien boos ing algo i hms o ensu e adap abili y o apidly changing ma ke
condi ions and o accoun o seasonal luc ua ions, long- e m ends, and he in luence o ex e nal ac o s.
● The scien i ic no el y o he wo k lies in shaping a hyb id model a chi ec u e speci ically adap ed o he needs o
small and medium-sized en e p ises in he B2C e-comme ce sec o wi h ex ensi e and dynamically upda ed
asso men s.
● The hypo hesis is ha he applica ion o he hyb id model will imp o e in en o y managemen pe o mance
indica o s such as o ecas accu acy, in en o y u no e , and p o i abili y compa ed wi h classical app oaches and
single-algo i hm solu ions.
2. Ma e ials and me hods
In he esea ch on in en o y managemen op imiza ion in he B2C sec o in highly compe i i e online ma ke s, se e al
esea ch di ec ions a e dis inguished. Fi s o all, analys s pay a en ion o assessing he scale and g ow h a es o e-
comme ce, which guide de elope s o p edic i e models. Acco ding o S a is a, he global e-comme ce ma ke shows
s eady annual g ow h, which c ea es addi ional p essu e on in en o y planning sys ems amid inc easing compe i ion
[1]. Simila conclusions a e con ained in he Amazon Eu ope S a is ics epo , which emphasizes he impo ance o
accu a e demand o ecas ing o Eu opean selle s on he Amazon pla o m [2]. A he same ime, a 2025 NVIDIA s udy
shows ha 90% o e aile s ha e ei he al eady implemen ed o a e es ing AI-based solu ions in hei supply chains,
which indica es an ac i e ansi ion om heo y o p ac ice in he use o p edic i e analy ics in in en o y managemen
[3]. The Mo do In elligence epo o ecas s u he expansion o he B2C e-comme ce ma ke h ough 2030 while
main aining high g ow h dynamics, which inc eases he need o lexible and scalable in en o y managemen models
[10].
Alongside empi ical ma ke s udies, he li e a u e p esen s sys ema ic e iews and in eg a i e wo ks ha p o ide a
c i ical assessmen o exis ing app oaches o demand o ecas ing. Thus, Chowdhu y A. R., Paul R., Rozony F. Z. [4]
conduc an analysis o models o e ail e-comme ce, classi ying hem by algo i hmic amilies and e alua ing
pe o mance in compa a i e expe imen s. Yuso Z. B. [12] ocuses on he ole o machine lea ning in op imizing
in en o y managemen and no es ha despi e he g owing popula i y o deep neu al ne wo ks, many p ac ical solu ions
s ill ely on classical s a is ical me hods due o hei anspa ency and ease o implemen a ion.
T adi ional s a is ical me hods and hyb id models occupy a dis inc place among o ecas ing ools. Çe in B., Taşdemi Ç.
[5] demons a e he success ul applica ion o an op imized SARIMA model o sales o ecas ing, showing ha e en in a
apidly changing en i onmen i is possible o achie e high accu acy wi h p ope uning o seasonal and end
componen s. Lin Y. e al. [8] p opose a hyb id CEEMDAN-LSTM a chi ec u e o inancial ime se ies, combining
p elimina y signal decomposi ion wi h ecu en neu al ne wo ks, which makes i possible o accu a ely cap u e
nonlinea dependencies and da a ola ili y. Te ada L., El Khaili M., Ouajji H. [6] ex end his di ec ion by compa ing
se e al deep lea ning a chi ec u es o demand o ecas ing in SCM 4.0 and no e he ad an ages o CNN- and LSTM-
based app oaches in he au oma ic ex ac ion o ea u es om his o ical ime se ies.
Fu he de elopmen is obse ed in ein o cemen lea ning-based me hods and hyb id AI a chi ec u es. Bou e R. N. e
al. [9] p esen a oadmap o applying deep ein o cemen lea ning (DRL) o he in en o y managemen p oblem,
desc ibing he key elemen s o he en i onmen , ewa d unc ion, and lea ning algo i hms. De Moo B. J., Gijsb ech s J.,
Bou e R. N. [11] p opose a ewa d shaping mechanism o pe ishable goods managemen asks ha accele a es DRL
agen con e gence and inc eases obus ness o changes in demand dynamics.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1701-1707
1703
A sepa a e body o esea ch ocuses on ex ending he unc ions o adi ional logis ics h ough ML-based op imiza ion.
Pasupule i V. e al. [7] analyze me hods o inc easing supply chain lexibili y and esilience, p oposing op imiza ion
echniques based on andom o es s and g adien boos ing o ou e planning and in en o y le el con ol wi h
conside a ion o en i onmen al and economic ac o s.
Despi e ex ensi e a en ion o a ious algo i hmic amilies, con adic ions a e obse ed in he li e a u e. On he one
hand, wo ks [4, 12] emphasize he limi ed p ac ical ad an ages o deep models due o hei black-box na u e and high
compu a ional cos , whe eas [6, 8] indica e hei supe io i y in o ecas ing accu acy. Simila ly, in he discussion o DRL
me hods he e a e bo h en husias ic o ecas s o adical imp o emen s in in en o y con ol [9] and skep icism
ega ding hei ma u i y and he need o subs an ial aining da a [11]. A he same ime, he li e a u e pays insu icien
a en ion o he in eg a ion o p edic i e models wi h p icing s a egies, he impac o seasonal and ma ke ing
p omo ions on o ecas accu acy, and issues o in e p e abili y o complex AI solu ions o end use s and manage s. In
addi ion, he e is a se e e lack o esea ch de o ed o he adap abili y o models o changes in compe i ion in online
ma ke s and o mul i-objec i e op imiza ion ha simul aneously accoun s o deli e y speed, s o age cos , and
en i onmen al ac o s.
3. Resul s and Discussion
Based on he esul s o he conduc ed s udy o he exis ing li e a u e and he ope a ional cha ac e is ics o small and
medium-sized en e p ises in he B2C e-comme ce segmen , a hyb id p edic i e analy ical model (GPAM) has been
de eloped, he s uc u al diag am o which is p esen ed in Figu e 1. The a chi ec u e o he p oposed model includes
ou in e ela ed componen s: an in o ma ion collec ion and p ep ocessing module, a o ecas ing module, an o de
placemen op imiza ion module, and a moni o ing module wi h subsequen eedback.
Figu e 1 A chi ec u e o he hyb id p edic i e analy ical model (GPAM) (compiled by he au ho based on [4, 8, 11])
As can be seen om Figu e 1, he i s block is he p ocedu e o da a collec ion and p ep ocessing. In his case,
in eg a ion o he e ogeneous in o ma ion sou ces is implemen ed as he ounda ion o he o ecas ing sys em. The
in e nal da ase s include he sales his o y o each SKU, p ice in o ma ion, calcula ed cos p ice, and cu en wa ehouse
in en o ies. Fo demons a ion pu poses, he au ho ’s da a on Skysales L d., ope a ing on Eu opean Amazon
ma ke places, a e used. The ex e nal componen is ob ained ia ma ke place APIs and includes he empo al dynamics
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1701-1707
1704
o Bes Selle s Rank, compe i o s’ p ices, he numbe o e iews, and pa ame e s o ad e ising campaigns (PPC). In
addi ion, mac oeconomic indica o s and holiday calenda s, which exe a subs an ial in luence on consume beha io ,
a e aken in o accoun . A he p ep ocessing s age, he da ase s a e cleansed o ou lie s and duplica e eco ds, missing
alues a e impu ed, and new ea u es a e enginee ed ( ea u e enginee ing), o example, mo ing a e ages o sales
olumes o p ice me ics.
Nex , he second block is he demand o ecas ing s age. A hyb id a chi ec u e is used o cons uc he o ecas . Fi s , he
SARIMA (Seasonal ARIMA) model es ima es he baseline le el o sales, iden i ying ends and seasonal luc ua ions
solely om he ime se ies o he sales his o y o he speci ic p oduc [5]. Then, a g adien boos ing algo i hm (Ligh GBM
o XGBoos ) adjus s he ini ial o ecas by in eg a ing he in luence o ex e nal de e minan s: p ice changes, BSR,
compe i o ac i i y, and he conduc o p omo ional campaigns [6, 7]. The inal o ecas is c ea ed by combining he
ou pu s o bo h models, anging om simple weigh ed a e aging o s acking, in which he esul s o SARIMA and
boos ing se e as inpu ea u es o a me a-model ha op imizes hei combina ion.
The nex block is o de op imiza ion. Tha is, he ob ained o ecas s se e as he s a ing poin o p ocu emen
decisions. Fi s , he sa e y s ock is calcula ed on he basis o he s a is ics o o ecas e o s (s anda d de ia ion), which
makes i possible o mi iga e demand unce ain y. Second, he eo de poin is de e mined by he o mula:
Reo de Poin = Demand o ecas × Lead ime + Sa e y s ock (1).
Finally, he economic o de quan i y (EOQ) is calcula ed, adap ed o he speci ics o ma ke place FBA models wi h
conside a ion o ime- a ying s o age and o de p ocessing cos s. Fo p ac ical demons a ion, hese p inciples we e
applied o an anonymized Skysales L d. da ase o he pe iod om 01.06.2022 o 31.05.2024 using he example o a
single SKU in he consume elec onics segmen .
Table 1 Compa ison o he accu acy o o ecas ing models o SKU-12345 (compiled by he au ho based on [1-3])
Me ic / Model
T adi ional Me hod (A e age
Sales o 3 Mon hs)
SARIMA
G adien
Boos ing
GPAM
(Hyb id)
MAPE (Mean Absolu e
Pe cen age E o )
38.5%
24.2%
19.8%
15.1%
RMSE (Roo Mean Squa e
E o )
125.1
88.7
75.3
61.9
As can be seen om he da a in Table 1, he p oposed hyb id model (GPAM) p o ides he highes o ecas ing accu acy,
educing he mean absolu e pe cen age e o (MAPE) o 15.1 %, which exceeds he esul s o bo h he classical app oach
and he indi idual cons i uen componen s o he model. This educ ion in he o ecas ing e o di ec ly co ela es wi h
inc eased economic e iciency and educed cos s. The cha shown in Figu e 2 p esen s a compa ison o he ac ual sales
olume wi h he p edic ions ob ained using a ious models. A isual analysis indica es ha he hyb id model (GPAM)
mo e adequa ely ep oduces seasonal peaks (in pa icula , du ing No embe –Decembe ) and accu a ely cap u es
sho - e m demand luc ua ions d i en by p omo ional campaigns.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1701-1707
1705
Figu e 2 Compa ison o demand o ecas s o SKU-12345(compiled by he au ho based on [1-3])
The implemen a ion o GPAM enables a ansi ion om eac i e o p oac i e in en o y managemen . Figu e 3 shows a
concep ual diag am o he model’s impac on key pe o mance indica o s (KPI).
Figu e 3 Impac o GPAM on key indica o s o in en o y managemen (compiled by he au ho based on [9, 10, 12])
I should be emphasized ha he e ec i eness o implemen ing he p oposed model is de e mined no only by he
accu acy o he algo i hms used, bu also by he quali y o he inpu da a and he ma u i y o exis ing business p ocesses.
Fo small and medium-sized en e p ises such as Skysales L d., whe e he ounde s a e di ec ly in ol ed in ope a ions
and possess su icien IT skills ( hey de elop in e nal analy ical ools in Py hon o assess ma ke oppo uni ies), he
en y ba ie is signi ican ly lowe . The au ho ’s expe ience indica es ha au oma ing he p ocessing o la ge da a
olumes ( o example, e i ying 10 000 p oduc i ems akes abou 2 hou s ins ead o 250 hou s wi h manual
e i ica ion) makes i possible o op imize asso men managemen and enables apid scaling o he business. The GPAM
unde de elopmen o malizes his app oach and ex ends i by in oducing ad anced o ecas ing me hods.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1701-1707
1706
The p ima y ad an age o he chosen me hod is i s high scalabili y. A g adien boos ing model can e icien ly p ocess
housands o SKUs, while i s aining and applica ion do no equi e compu a ional esou ces as powe ul as hose
needed o complex neu al ne wo ks, which makes i pa icula ly a ac i e o SMEs. The eedback loop embedded in
he a chi ec u e (Block 4) ensu es egula e aining on up- o-da e da a and he eby main ains he ele ance o
o ecas s in a cons an ly changing ma ke .
A he same ime, he model has i s limi a ions. I s ope abili y is la gely de e mined by he a ailabili y and comple eness
o ex e nal da a: no all ma ke places o e open and comp ehensi e API in e aces. Howe e , by p o iding a apid and
accu a e assessmen o he cu en s a e o he ma ke , he p oposed sys em enables companies o p omp ly adap hei
s a egies in esponse o sudden changes.
4. Conclusion
In he cou se o he s udy, a hyb id p edic i e analy ical model (GPAM) was de eloped and heo e ically subs an ia ed,
aimed a op imizing in en o y managemen in B2C companies ope a ing in highly compe i i e online ma ke s. The
p oposed GPAM combines he ad an ages o s a is ical ime se ies analysis using he SARIMA me hodology and
ensemble machine lea ning me hods (G adien Boos ing), which makes i possible o achie e signi ican ly highe
demand o ecas ing accu acy by simul aneously accoun ing o bo h in e nal seasonal pa e ns and he in luence o
di e se ex e nal ac o s.
The analysis esul s demons a ed ha in eg a ing a ious o ecas ing models is an e ec i e s a egy o imp o ing
bo h he accu acy and he obus ness o p edic ions in he ola ile e-comme ce ma ke . The implemen a ion o GPAM
ans o ms in en o y managemen om a eac i e o a p oac i e mode, which di ec ly con ibu es o educing
wa ehousing cos s, dec easing los sales, accele a ing capi al u no e , and ul ima ely inc easing p o i abili y le els.
The GPAM a chi ec u e in eg a es modules o da a collec ion and p ep ocessing, o ecas ing, o de olume
op imiza ion, and moni o ing o key pe o mance indica o s, o ming a comp ehensi e solu ion adap able o small and
medium-sized en e p ises wi h basic IT compe encies. The au ho ’s hypo hesis ha he use o a hyb id app oach can
signi ican ly imp o e key in en o y managemen pe o mance indica o s was con i med.
The p ac ical signi icance o he wo k lies in he ac ha he de eloped model p o ides SMEs in he e-comme ce sec o
wi h a speci ic and accessible ool o enhancing compe i i eness. In u he esea ch, i is ad isable o conduc GPAM
ials unde eal p oduc ion condi ions, as well as o expand i s unc ionali y h ough he in eg a ion o na u al language
p ocessing (NLP) algo i hms o analyzing cus ome e iews and news backg ound in o de o u he imp o e o ecas
accu acy.
Compliance wi h e hical s anda ds
Disclosu e o con lic o in e es
No con lic o in e es o be disclosed.
Re e ences
[1] eComme ce - Wo ldwide [Elec onic esou ce]. - Access mode:
h ps://www.s a is a.com/ou look/emo/ecomme ce/wo ldwide (da e o access: 19.06.2025).
[2] Amazon Eu ope S a is ics You Need o Succeed as an Amazon Eu ope Selle in 2023 [Elec onic esou ce]. -
Access mode: h ps://ma ginbusiness.com/amazon-eu ope-s a is ics/ (da e o access: 22.06.2025).
[3] NVIDIA. 9 Ou o 10 Re aile s Now Adop ing o Pilo ing AI, La es NVIDIA Su ey Finds. 2025 [Elec onic
esou ce]. - Access mode: h ps://blogs.n idia.com/blog/ai-in- e ail-cpg-su ey-2025/ (da e o access:
23.06.2025).
[4] Chowdhu y A. R., Paul R., Rozony F. Z. A sys ema ic e iew o demand o ecas ing models o e ail e-comme ce
enhancing accu acy in in en o y and deli e y planning //In e na ional Jou nal o Scien i ic In e disciplina y
Resea ch. – 2025. – Vol. 6 (1). – pp. 01-27. h ps://doi.o g/10.63125/mbb w637
[5] Çe in B., Taşdemi Ç. Fo ecas ing elec ic ehicle sales using op imized SARIMA model: A wo-yea p edic i e
analysis //Ve i Bilimi. – 2024. – Vol. 7 (2). – pp. 41-51.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1701-1707
1707
[6] Te ada L., El Khaili M., Ouajji H. Demand o ecas ing model using deep lea ning me hods o supply chain
managemen 4.0 //In e na ional Jou nal o Ad anced Compu e Science and Applica ions. – 2022. – Vol. 13 (5).
- pp. 704-711.
[7] Pasupule i V. e al. Enhancing supply chain agili y and sus ainabili y h ough machine lea ning: Op imiza ion
echniques o logis ics and in en o y managemen //Logis ics. – 2024. – Vol. 8 (3). – pp. 1-16.
h ps://doi.o g/10.3390/logis ics8030073
[8] Lin Y. e al. Fo ecas ing s ock index p ice using he CEEMDAN-LSTM model //The No h Ame ican Jou nal o
Economics and Finance. – 2021. – Vol. 57. h ps://doi.o g/10.1016/j.naje .2021.101421
[9] Bou e R. N. e al. Deep ein o cemen lea ning o in en o y con ol: A oadmap //Eu opean jou nal o
ope a ional esea ch. – 2022. – Vol. 298 (2). – pp. 401-412. h ps://doi.o g/10.1016/j.ejo .2021.07.016
[10] Mo do In elligence. B2C E-Comme ce Ma ke Size, Sha e and T ends Resea ch Repo , 2030[Elec onic
esou ce]. - Access mode: h ps://www.mo do in elligence.com/indus y- epo s/b2c-ecomme ce-ma ke
(da e o access: 05.07.2025).
[11] De Moo B. J., Gijsb ech s J., Bou e R. N. Rewa d shaping o imp o e he pe o mance o deep ein o cemen
lea ning in pe ishable in en o y managemen //Eu opean Jou nal o Ope a ional Resea ch. – 2022. – Vol. 301
(2). – pp. 535-545. h ps://doi.o g/10.1016/j.ejo .2021.10.045
[12] Yuso Z. B. Analyzing he ole o p edic i e analy ics and machine lea ning echniques in op imizing in en o y
managemen and demand o ecas ing o e-comme ce //In e na ional Jou nal o Applied Machine Lea ning. –
2024. – Vol. 4 (11). – pp. 16-31.
