8.

Methodlogy
The world is becoming increasingly interconnected and compact, while the global
supply chain (SC) is growing progressively intricate. As an example, Apple relies on
43 suppliers scattered across 6 continents to assemble and manufacture its iconic
iPhone. Even aseemingly simple product like a tennis ball travels through 11
countries across 4 continents during its manufacturing process (Petrova, 2018;
Carrington, 2016). Moreover, in light of emerging global challenges such as the
COVID-19 pandemic, political tensions, and unforeseen weather events, Supply
Chain Management (SCM) has become riskier than before.
This section presents the findings structured around nine key themes or use cases
aligned with the four SCOR stages. Each theme corresponds to a major AI use case
in supply chain management, derived from nine benefit groups: PLAN (demand
forecasting, inventory optimization, supply risk mitigation), SOURCE (procurement
and supplier selection), MAKE (product quality assurance, smart warehouse
management, predictive maintenance), and DELIVER (route optimization, dynamic
pricing, last-mile delivery, and customer service). For each group, the section
discusses relevant AI techniques, practical use cases, industry examples, and the
associated benefits.
A. Demand Forecasting
Demand forecasting or demand planning holds paramount importance in SCM as it
serves as a foundation for various other SC processes, including procurement,
inventory planning, manufacturing, and distribution planning.
Technology: In their comprehensive literature review on AI/ML techniques for
demand planning in SCM from 2005 to 2019, Seyedan & Mafakheri (2020) found
that the most commonly used method is NN (Neural Networks),SVM (Support Vector
Machines) and Decision Tree methods. Similarly, Carbonneau et al. (2008)
conducted a study on ML in SC demand planning and identified RNN (Recurrent
Neural Networks) as the best predictor model.
Real-WorldExample
An exemplary case is observed in the German e-retailer Otto, which utilizes ML and
deep learning to analyze a vast dataset of 3 billion transactions and approximately
200 variables, including weather, historical sales, and web searches. Based solely
on AI recommendations, without human supervisors' involvement, Otto purchases
around 200,000 items monthly to proactively prepare for future demand and
maintain a high level of delivery efficiency. This approach liberates human
managers from the decision-making process, enabling them to focus on more
strategic issues. It is worth noting that Otto's AI-driven demand forecast for a
specific month is reported to be up to 90% accurate (The Economist, 2017;
Wolfgang, 2021).
Also in Germany, the drugstore chain DM (Drogerie Markt) harnesses the power of
AI to anticipate weekly demand by utilizing input variables, such as SKU (Stock
Keeping Unit)-level data, over a span of 6 months. By using a training dataset from
the past 2.5 years, DM achieves highly accurate demand forecasts, leading to cost
reductions and increased customer satisfaction (Wenzel et al., 2019).
In 2016, Nike from the United States faced a dip in profits due to faulty demand
prediction. Learning from this experience, the sportswear giant later acquired
Celect, an AI data predictive analytics firm, three years later (Thomas, 2019).
Meanwhile, Accenture collaborated with General Electric on TALERIS, a predictive
analytics tool tailored for smart service operations in the airline industry.
In 2012, US ecommerce giant Amazon introduced the concept of anticipatory
delivery and even filed a patent for this approach in August that year. It was
approved in December of the following year. Utilizing customer data, such as search
history, purchase history, and shopping cart contents, Amazon anticipated when a
customer would buy a particular product and shipped it to them before they even
placed the order (Lomas, 2014). Amazon also efficiently calculated the destination
hub for a group of future buyers near their addresses using complex algorithms,
enabling seamless and anticipatory delivery.
American logistics company UPS (United Parcel Service) has also successfully
implemented ML for demand forecasting during holiday peaks and last-minute
demand spikes.
California-based telecom maker Infinera employs Intrigo Systems to predict delivery
dates of components using ML. By analyzing past logistics provider performance
and production lead times, Infinera gains valuable insights into delivery schedules
(Korolov, 2018).
B. Inventory Optimization
Inventory optimization or stock level optimization is the process of determining the
optimal inventory levels to strike a balance between localized and centralized
storage. Overreliance on centralized inventorying poses risks of stockouts, while
excessive localized storage can be costly. Since the annual holding stocks cost
might take up to 35% of a product value (Timme and Williams-Timme, 2003),
inventory optimization could serve as a very effective way to substantially lower
costs. Marr (2017) highlighted that AI-driven inventory optimization models take
into account factors such as demand variability, lead times, and supply chain
constraints. to avoid both overstocking and understocking, businesses track sales
and store inventory in real-time.
Technology: In this domain, GA (Genetic Algorithms) and ANN (Artificial Neural
Networks) are commonly suggested to effectively manage inventory levels (Paul
and Azeem, 2011; Sandra and Bernd, 2021; Praveen et al., 2019). Priore et al.
(2019) adopted ML, specifically inductive learning to determine the best inventory
replenishment policy, achieving an accuracy rate of 88% and reducing operation
costs. Sharma and Singh (2021) also applied ML, specifically Random Forest and
XGBoost, to predict demand and achieve optimum stock levels in what they call
“intelligent warehouse stocking”.
Real-World According to McKinsey (2018a), ML can reduce lost sales from
stockouts by up to 65% and inventory reductions of 20 to 50%. For example,
Microsoft successfully reduced its inventory by a significant US$200 million by
leveraging AI and IoT (Internet of Things) capabilities. Lennox was able to cut data
collection time by an impressive 90%, while the Komplett group managed to
increase SC efficiency by nearly 30% (Ozdogru, 2020). In the UK, the National Grid
employs Google's DeepMind technology to anticipate the optimal level of demand
versus supply, considering various input variables such as weather conditions and
other relevant factors (Yao, 2018). Additionally, the AI system implemented by
German e-retailer Otto has proved highly beneficial and led to a remarkable 20%
reduction in surplus inventory and significantly decreased product returns by over 2
million items per year (The Economist, 2017; Wolfgang, 2021). In the same vein,
Walmart uses AI-based demand forecasting algorithms to adjust inventory levels
based on changing customer preferences and market dynamics. This allows the
retail giant to optimize inventory levels, reduce excess inventory and improve shelf
availability. Companies like Nike also leverage AI to predict upcoming styles and
fashion trends, better preparing them for future inventory supply to meet customer
demands effectively.
C. Supply Risk Mitigation
AI proves to be a valuable tool in effectively minimizing various supply risks, such as bottlenecks,
bullwhip effects, calamities, and other potential disruptions. Predicting the estimated arrival time
of shipments becomes crucial for companies to notify customers and loaders in advance or adjust
cross-docking operations as needed. In scenarios where shipments may be held overnight at
customs due to bureaucracy or missing documents, AI's predictive abilities become vital in
avoiding potential delays that could incur significant costs in JIT mismatches. AI can proactively
suggest solutions in real-time to minimize these supply risks.

Technology Various AI techniques applied in this domain include ANN/SVM which is used to
quantify various resilient strategies for risk mitigation (Rajesh, 2020), and FL (Fuzzy Logic) to
reduce the bullwhip effect as proposed by Balan et al. (2007). Mogre et al. (2016) suggested a
DSS (Decision Support System) using Decision Trees to prescribe SC risk mitigation and
strategies when the risks materialize

Real World: Watson, an AI-enabled tool developed by US giant chipmaker IBM, is equipped with
the capability to track social media trends, predicting potential occurrences like political riots,
work strikes, financial climate, and weather patterns. This analytical prowess allows Watson to
effectively monitor SC risks and offer solutions ranked by priority, helping businesses make
informed decisions and mitigate potential disruptions. Additionally, IBM's AI-powered Supply
Chain Insights further aids clients in streamlining their decision-making processes by reducing
risks and disruptions in SC operations. Germany-headquartered logistics giant DHL (Dalsey
Hillblom Lynn) has developed AI-powered Resilience360, a cloud-based supply chain risk
mitigation platform, that utilizes ML and NLP to analyze unstructured texts from over 300,000
online sources and sentiments from a staggering 8 million posts.

Utilizing ML, DHL leverages predictive capabilities to forecast delays in air freight
transit with remarkable precision. By accurately predicting the average daily transit
time for a particular lane seven days in advance, DHL enables proactive planning
for freight forwarders, streamlining logistics operations and ensuring timely
deliveries (DHL, 2018). Another notable AI solution in the industry is US-based Flex's
AI software, Flex Pulse, which excels in predicting SC risks such as work strikes,
calamities, and other potential disruptions. With an impressive clientele including
Microsoft, Apple, and Ford Motor, Flex Pulse significantly contributes to enhancing
risk management strategies. During the COVID-19 pandemic, companies like US
pharmaceutical firm Johnson & Johnson too used AI-driven analytics to assess and
mitigate SC disruptions.
D. Procurement, supplier selection
AI offers significant advantages in the realm of procurement and supplier selection.
One of its key benefits is the ability to efficiently identify the most optimal suppliers
and cost-effective options. Through comprehensive analysis of suppliers' risk
profiles, track records, and feedback, AI assists in evaluating and ranking potential
suppliers and transporters.
Technology: Sharma et al., (2022) found that the most popular AI techniques in
selecting suppliers are a combination of fuzzy set theory and MCDM (Multicriteria
Decision- Making) models. Other systems include ANN and genetic algorithms (Riahi
et al., 2021) while other scholars combined ANN with additional approaches like
Data Envelopment Analysis (DEA) and Analytic Network Process (ANP) for supplier
selection (Kuo et al., 2010).
Besides ANN, other ML techniques most commonly used in supplier selection are
Decision Trees and SVM and Q learning (Tirkolaee et al., 2021.) FL (Fuzzy Logic) is
also suggested for supplier selection (Carrera and Mayorga, 2008) for new product
development.
Real-World: For supplier selection, Scoutbee from Germany developed an AI-
powered system to help companies find the best suppliers globally by analyzing
vast amounts of data from a variety of sources. It collects and processes data from
various suppliers, including their capabilities, track records, certifications, financial
stability, and customer feedback and uses NLP to interpret and understand
unstructured data, such as supplier profiles, news articles, and social media posts,
to gain deeper insights into potential suppliers. In 2020 Scoutbee raised $60 million
in Series B funding to expand the system (Musgrove, 2020).
Other examples include AI-driven procurement systems such as SAP Ariba that use
ML algorithms to automatically identify and evaluate potential suppliers based on
historical data, performance metrics, and market intelligence (Yarramalli, 2020;
Singh, 2023).
Organizations such as Japanese carmaker Toyota have implemented AI-driven
supplier assessment systems to identify potential SC risks and disruptions. These
systems facilitate informed supplier selection, negotiation, and risk mitigation
decisions. AI collaboration platforms have changed the way SC partners
communicate and coordinate.
In 2023, Bloomberg highlighted that Walmart has used chatbots for negotiating
optimal pricing and payment terms with several suppliers and vendors through
Pactum AI's chatbot. This intelligent system has the capability to analyze present
supplier proposals against historical data and competitor pricing. With the
assistance of AI chatbots, deals that once took weeks or even months when solely
handled by humans are now closed within days, ultimately bolstering profitability.
E. Product Quality Assurance and Smart Warehouse Management
In manufacturing, image recognition powered by Machine Vision plays a vital role in
identifying and sorting products, detecting defects, scratches, or dents. Cameras
equipped with AI can also conduct product counts and promptly alert human
operators in case of missing items or potential thefts.
Technology: Zhao et al., 2021 used CNN to successfully identify nearly 98% of
defects in the textile industry. Similarly, Unajan et al., (2019) applied Machine Vision
with Deep Learning to accurately identify product scraps with an accuracy rate of
85% while Villalba et al., (2019) developed DNN (Deep Neural Network) to
successfully and automatically classify 98.4% of products in the production of
gravure cylinders. In warehouses, AI-driven programs monitor employee
movements, allowing for accurate predictions of task completion times.
Real-World: IBM's Watson, for instance, demonstrates exceptional accuracy in
detecting cargo defects from real-time images, achieving over 90% accuracy (Binns,
2018). Detected anomalies are relayed to maintenance teams for timely correction.
Watson, fed with images from cameras installed along the tracks, will classify and
identify damages and prescribe suitable repairs for wagons (DHL, 2018). Companies
like GreyOrange utilize AI warehouse services to handle changing inventories and
demand volatilities in real-time, reducing lead times, and ultimately improving
customer satisfaction and order fulfillment. In the field of robotics, Deep Learning
has brought about a transformation by empowering robots to manipulate objects
without reliance on fixed positions. Finnish firm ZenRobotics utilizes ML and
Machine Vision in its robot system to identify different labels and logos on beverage
and food cartons, as well as recyclables. The robots efficiently sort and pick up
these items from conveyor belts at an impressive rate of 4,000 items per hour.
(DHL, 2018). UK supermarket Ocado utilizes AI-enabled robots to transport products
to human workers responsible for packing them into shopping bags for delivery
(Dale, 2018). Subsequently, other robots or cobots (collaborative robots) carry the
bags to trucks driven by humans, destined for customers' homes. According to
McKinsey (2017), such cobots can increase productivity by an impressive 20%.
Similarly, German retailer Metro Cash & Carry leverages ML to differentiate between
stationary and moving pallets on RFID-tagged loading ramps, significantly reducing
false positive readings and correctly classifying over 95.5% of data (Wenzel et al.,
2019). Japanese firm NEC utilizes image recognition technology to efficiently sort
products and determine which items to ship, matching a scheduled list (Yosuke et
al., 2017). Meanwhile, US chipmaker Intel Corp employs ML to expedite the
resolution of design problems for semiconductor makers, increasing efficiency in the
manufacturing process (Burgess, 2018).
Canadian startup TwentyBN has developed an AI technology capable of analyzing
human behavior through video feeds, enabling it to detect incidents such as people
falling or engaging in shoplifting activities. This application enhances operational
safety and helps prevent accidents and thefts in various settings. In 2021, it was
acquired by US mobile tech firm Qualcomm (Freeman, 2021).
Other examples of smart warehousing come from the Chinese e-commerce giant
Alibaba, which operates a warehouse in Huiyan. In this warehouse, 60 robots are
employed to transport items weighing up to half a ton to human workers
responsible for packing. These robots are equipped with laser sensors that enable
them to scan their surroundings, avoiding collisions with objects and personnel.
French company Qopius has developed an AI system capable of identifying various
elements such as price tags, logos, brands, and labels. Additionally, it can recognize
when a shelf is out of stock, helping improve inventory management and store
operations (DHL, 2018). In the realm of agriculture, US machinery maker John Deere
employs IBM Watson at its production factory in Mannheim, Germany. Deep
Learning-powered cameras are used to detect product faults, and Watson provides
explanations to human operators on methods to fix the issues. Workers can even
interact with Watson, asking it questions related to the faults. The smart system
then automatically orders the correct parts and suggests an optimal time for
maintenance, enhancing production efficiency. (DHL, 2018).
F. Predictive Maintenance
Predictive maintenance, also known as prognostic maintenance, is a proactive
approach that involves predicting when warehouse machines are likely to break
down and recommending timely repairs and maintenance. This strategy
significantly reduces downtime and repair costs by detecting microscopic cracks,
scratches, and defects in warehouse facilities and production lines Predictive
maintenance relies. on predictive analytics, which collects data from sensors to
forecast machine breakdowns and duration, ultimately minimizing downtimes and
preventing unscheduled repairs.
Technology Predictive maintenance mainly relies on ML applications (Kumar and
Hati, 2021), with Deep Learning specifically being applied for predictive fault
detection in induction motors (Luo et al., 2019). Other techniques to predict
machine failures are RL (Encapera et al., 2021) and Decision Trees (Kaparthi and
Bumblauskas, 2020). In a report titled "Smartening up with Artificial Intelligence"
published in 2018, global consulting firm McKinsey revealed that predictive AI can
lead to a 10% reduction in maintenance costs and up to a 20% decrease in annual
downtime (McKinsey, 2018a). These AI-driven advancements in predictive
maintenance technology contribute to improved warehouse efficiency, cost savings,
& enhanced safety by addressing issues before they escalate and disrupt
operations.
Real-world: Molęda (2023) discussed how companies such as US-based MNC
General Electric use AI algorithms to predict equipment failures and optimize
maintenance schedules, thereby reducing downtime and maintenance costs. Other
real-world examples include Swedish carmaker Volvo which has developed a truck
type installed with multiple sensors inside to identify when and where maintenance
is to be carried out. (DHL, 2016).
 G. Route Optimization
ML and graph theory play a crucial role in determining the most efficient routes for
transportation. Leveraging real-time data from in-transit shipments, AI can
accurately predict the departure, potential delays, and expected arrival times of
cargo.
Technology: GA have been successfully applied for routing optimization in SC,
specifically vehicle routing and scheduling (Chen et al., 1998, Park 2001); delivery
and pickup (Jung and Haghani, 2000); bus network optimization (Bielli et al., 2002);
minimum spanning tree (Zhou and Gen, 1999). Chambers (2001) in fact found out
that GA is among the most popular AI methods used to solve network transportation
issues. Hill and Böse (2016) discovered that applying an ANN-enabled DSS to design
the optimal routing of trucks could lead to a decrease in waiting times and
workloads while Michalewicz and Fogel (2004) found that the famous TSP (traveling
salesman problem) could be solved using FL.
Real-world The impact of AI on logistics and route optimization is evident in cases
such as US transport giant FedEx, which uses AI algorithms to optimize delivery
routes based on real-time traffic data, weather conditions, and delivery schedules.
Loske (2021) highlighted that AI-enabled logistics solutions reduce transportation
costs, shorten delivery times, and increase overall customer satisfaction.
UPS leverages ML to anticipate weather turbulence and recommends alternative
routes accordingly. Through UPS' On-Road Integrated Optimization and Navigation
system, the company calculates the most efficient path for road deliveries. By
utilizing ML algorithms, UPS can dynamically adjust delivery routes based on real-
time weather data, ensuring timely and reliable deliveries even in the face of
challenging weather conditions. This predictive and prescriptive approach enables
UPS to optimize its operations, minimize disruptions, and enhance overall customer
satisfaction. In other instances, DHL and Chinese smartphone maker Huawei
collaborated to develop a sensor system installed inside mailboxes. These sensors
relay information to the nearest or "most optimal" DHL drivers, allowing for more
efficient pickups and reducing drivers' waiting times by an impressive 50% (Alim &
Kesen, 2020). These advancements in AI-driven technologies are revolutionizing
transportation and logistics, leading to enhanced efficiency and sustainability.
H. Dynamic Pricing
Dynamic pricing, unlike fixed pricing, allows for flexibility, ensuring that prices
accurately reflect the current market conditions, optimizing profits. It is a powerful
AI use case that involves continuously adjusting prices based on various factors,
such as demand, seasonality, social media trends, time of the day (time-based
pricing), weather conditions, competitors' prices, and individual shopping habits.
TechnologyLeung et al. (2019) proposed applying FL to enable dynamic pricing
that resulted in an increase in quotes acceptance rate by 78% and a 60% reduction
in the average planning time of offers. AI enables personalized pricing by analyzing
customer data to tailor prices based on individual preferences and purchase history.
Not limited to service providers, commodity firms also leverage AI-prescribed
dynamic pricing through ESL (Electronic Shelf Labelling), allowing them to instantly
adjust product prices based on AI recommendations.
Real-world Amazon uses AI algorithms to recommend personalized prices to
customers, improving customer experience and increasing sales (Chen et al., 2016)
while American Delta Airlines leverages AI to analyze real-time data such as
demand, competitor prices, and historical sales to dynamically adjust ticket prices.
This ensures that products are competitively priced while maximizing returns. E-
commerce platforms like eBay also use AI to adjust prices in real time based on
factors such as customer behavior, competitor pricing, and product availability. This
approach allows sellers to quickly adapt to market conditions and optimize their
revenue potential (Hann, 2006). Dynamic pricing has been considered for the
tourism sector (Guizzardi et al., 2021) with the hotel industry applying AI-based
revenue management systems to optimize pricing for hotel rooms, event spaces,
and other offerings. For example, US-headquartered hotel chain Marriott uses AI
algorithms to adjust prices based on factors such as occupancy, seasonality, and
local events (Kardaras et al., 2013).
I. Last Mile Delivery, Customer Service
AI applications have become pervasive throughout the SC, spanning from the
procurement stage to the final delivery. Even LMD (last-mile delivery) and customer
service are benefiting from AI technologies.
Technology Automatic delivery utilizes a combination of AI methods namely GA
and ANN for route optimization; CNN, Object Recognition, Image Segmentation…for
Machine Vision to recognize objects; and NLP for communication (Chen et al., 1998;
Park, 2001; Jung and Haghani, 2000; Bielli et al., 2002; Zhou and Gen, 1999;
Chambers, 2001; Hill and Böse 2016).
To enhance customer satisfaction, businesses utilize NLP-assisted chatbots like
ChatGPT that possess the ability to comprehend human languages and nuances,
providing instant responses to customer queries 24/7. NLP employs statistical
methods to analyze word phrases and and patterns in speech or text, even
understanding jargon in different contexts and evolving as these jargons change
over time. NLP are ubiquitously used in AI chatbots to answer customer queries and
execute direct sales (Zdravković et al., 2022; Toorajipour et al., 2021).
Another innovative conversational AI, "Jenny," developed by Israel's [Link],
interacts with customers via SMS or Facebook Messenger to gather instructions on a
convenient pickup time and location, as well as confirmations. Jenny also keeps
delivery drivers updated on any last-minute changes and collects feedback from
parcel recipients.
Real-world Google has made remarkable strides through its Wing unit, which has
successfully completed 330,000 drone deliveries. Walmart has also entered the AI-
driven delivery arena, partnering with Zipline for 600,000 air deliveries, primarily
focused on medical supplies in Africa. Meanwhile, Amazon has ambitious plans for
10,000 drone deliveries in 2023, despite having executed only 100 so far (Tarasov,
2023).
UPS has also collaborated with Zipline to deliver medical supplies via drones in
Rwanda. On the ground, AI-powered delivery solutions are transforming the food
industry. Google teamed up with Chipotle industry. Google teamed up with Chipotle
to deliver burritos at Virginia Tech, while US pizza restaurant chain Domino's Pizza
partnered with Flirtey for pizza delivery in New Zealand. Additionally, Domino's has
joined forces with Nuro, a California-based firm, to deliver pizzas in Houston, Texas,
using Nuro's autonomous cars (Wood, 2021).
In the food industry, US-based restaurant KiwiBot utilizes AI-powered robots to
deliver meals, harnessing behavioral neural networks to optimize the delivery
process and improve customer service.
Starbucks also uses sentiment analysis to understand customer reactions to new
menu items and promotions, allowing them to tailor their offerings based on
feedback. Since personalization increases customer engagement and conversion
(Bleier et al., 2018), Amazon's recommendation engine uses AI algorithms to
analyze browsing and purchase history to suggest relevant products.
In the realm of brick-and-mortar stores, Amazon Go has introduced an AI-driven
shopping experience. Shoppers check in at the store, place products in their carts,
and can leave without the need to pay at the checkout counter. The store's system
of cameras detects the products chosen by the shopper, debits their accounts, and
emails the receipts accordingly, revolutionizing the traditional shopping process.
In 2017, DHL Parcel introduced a voice-based service to track LMD using Amazon's
Alexa. Through an Echo speaker, customers can inquire about their product's
whereabouts and expected delivery time. By simply speaking their tracking number,
the Echo speaker provides them with the latest shipment information.
AI in Supply Chain: Techniques, Applications, Real-World Cases and
Benefits under SCOR Framework
9. Global Practices & Case Studies
ECOMMERCE INDUSTRY
 Case studies of AI in Amazon’s supply chain
In today's rapidly evolving digital landscape, AI has become an essential technology
for business success, and Amazon stands as a leader in leveraging its capabilities.
The company has continuously pushed the boundaries of innovation, particularly in
the realm of AI-driven solutions.
Machine learning in supply chain optimization
Amazon applies four key types of machine learning in its supply chain.
In demand forecasting, supervised learning utilizes labeled data to build predictive
models, enabling accurate forecasts. Techniques such as Long Short-Term Memory
networks (LSTM), XGBoost, and ARIMA are employed to predict market demand,
optimize inventory levels, and improve route planning. These advanced algorithms
help Amazon enhance supply chain efficiency and responsiveness to market
fluctuations.
Automated warehouse management
The efficiency of Amazon's supply chain operations is largely driven by its
automated warehouse management system. As early as 2012, Amazon invested
$775 million to integrate robotic automation into its logistics network. Today, the
company operates over 200,000 robots, forming an extensive and highly
sophisticated supply chain ecosystem [2].
Smart logistics and delivery
Amazon's integration of artificial intelligence extends beyond demand forecasting
and warehouse automation—it also plays a pivotal role in logistics and delivery. The
company has leveraged cutting-edge technologies such as autonomous drones and
self-driving delivery vehicles to enhance efficiency and redefine modern logistics.
Current challenges AI implementation in Amazon’s supply chain faces two major
technological obstacles. First, AI algorithms are complex. Global market data is vast
and highly volatile, influenced by economic markets, policies, force majeure events,
and competitive environments. Therefore, even with the most advanced algorithms,
it is difficult to eliminate errors. Moreover, an increase in order volumes may lead to
algorithm bottlenecks, causing delays when calculating optimal routes. Additionally,
there are unstable factors with automated machines. Amazon's supply chain heavily
relies on technologies such as automated robots and smart delivery systems.
However, these technologies have inherent instability. During peak periods, as
order volumes rise, automated delivery vehicles might experience delays, leading
to traffic congestion. There is also a possibility of power outages during transit, and
system errors in the automated delivery process could pose traffic safety risks.
Furthermore, drones also present safety hazards and are affected by weather
conditions
DOI: 10.54254/2754-1169/2025. TheApplication ofArtificial Intelligence in
Supply Chain Management:ACase Study ofAmazon
AUTOMOBILE INDUSTRY
 BMW – Smart Logistics and AI-Driven Quality Control in
Manufacturing
In BMW’s high-precision manufacturing environment, maintaining quality across
thousands of vehicle components and complex assembly workflows is both mission-
critical and resource-intensive. Manual quality checks were time-consuming and
vulnerable to inconsistencies, while real-time visibility into component movement
and status across production sites was limited.
To address these pain points, BMW implemented AI-driven solutions within its
logistics and quality assurance functions. Computer vision systems were introduced
on assembly lines to inspect critical components-such as body panels, welding
seams, and paint finishes-in real time. These systems could detect micro-defects
invisible to the human eye and immediately alert technicians or trigger corrective
actions. Simultaneously, AI-powered tracking systems monitored the flow of
materials and components, predicting delays and optimizing parts delivery to each
station.
The deployment resulted in a measurable drop in defect rates and a smoother,
more synchronized production flow. With real-time anomaly detection and logistics
optimization, BMW was able to accelerate manufacturing cycles, reduce waste, and
ensure consistent product quality-all while enhancing traceability and transparency
across its global production footprint.
[Link]
TECHNOLOGY INDUSTRY
 Google
Google has been one of the early adopters of machine learning in their processes
and has ever been a contributor to the technology by developing and maintaining
ML libraries, APIs, datasets, and so on. Outlining Google’s affinity to machine
learning, Sundar Pichai, who currently heads Google and Alphabet, remarked in
2015 that “Machine learning is a core, transformative way by which we are
rethinking how we are doing everything” [21].
Gmail: Gmail is the most widely used e-mail service worldwide. Owing to a large
number of users, it is also prone to several security threats. E-mail inboxes can also
include malicious attachments and spam messages which can escalate to be a
security threat. Behind the scenes, Google tries to block these attachments before
reaching our inbox. The deep learning-based document scanners work in
association with existing artificial intelligence (AI) and machine learning (ML)
mechanisms in Gmail to enhance the document detection capabilities. The scanner
is composed of a TensorFlow deep learning model developed in TFX (TensorFlow
Extended) and document analysers designated for each type of file
(.docx, .pdf, .exe, etc.). These documents analysers parse the content, detect
macros and perform feature extraction. Multiple scanners running in parallel to the
exiting document detection system helps Gmail’s decision engine to classify
attachments and block malicious documents [22].
Smart Reply and Smart Compose introduced by Gmail also rely on machine learning
and are great tools to save time while drafting e-mails [23]. They suggest the
appropriate sentences while composing a new e-mail or replying to a received e-
mail, with the support of a large, predefined training set. E-mail filtering is another
area where machine learning finds applications. Statistical techniques such as a
Naïve Bayes classifier could be implemented to filter spam messages. Other
algorithms like logistic regression and neural networks could also be used for the
same [24].
Google Maps: Google Maps is another popular service from the house of Google. It
helps commuters and tourists by predicting traffic and suggesting an appropriate
route for them to reach their destinations. It acts as a platform that provides
business listings, driving directions, public transport details as well as real-time
traffic information.
Google, in association with DeepMind, the artificial intelligence subsidiary of
Alphabet Inc., has developed techniques to improve traffic prediction capabilities. It
applies the concept of Graph Neural Networks (GNN) to lower the cases of
inaccurate expected time of arrivals (ETAs) reported. The prediction system is a
combination of a GNN model and a route analyser. The route analyser models
‘super segments’ from the traffic information and geographical data available. The
GNN model then predicts the time to travel for each super segment [26]. The
predictive traffic model can identify the routes which are highly likely to experience
congestion or heavy traffic and suggests alternative routes accordingly.
Google Photos: Computers do not visualize photographs and videos as humans
do. For a computer, a picture is pixel and colour information that may be
interpreted as shapes. Even though the computer’s perception of a photo is not
similar to that of human beings, it can be trained to identify certain patterns,
objects, colours, or shapes from images. A computer can be trained to identify
patterns or the prominent colours in an image that makes up the picture of
structures such as a highway or an object such as a traffic sign. This is the same
technology that permits users to organize images in Google Photos based on
similarity and retrieve them later with a simple keyword search [27]
Google Assistant: A trailblazer among virtual assistant applications, Google
Assistant is a voice assistant service provided by Google and can accept commands
from users to search on the web or control other devices. It was introduced as the
successor to Google Now. Google Assistant allows users to control connected
devices and smart homes. AI plays a pivotal role in the working of Google Assistant.
Natural language processing (NLP), another application of machine learning,
enables it to understand what the user is speaking [30]. Natural Language
Processing is that branch of AI which deals with developing systems and machines
to understand human language and also translate from one language to another
(such as Google Translate). Google Assistant is the confluence of multiple domains,
namely linguistics, computational linguistics, and machine learning. Google
Assistant analyses and processes the words spoken to it and understands the
meaning and context. Based on this understanding, appropriate action is executed
or responses are framed.
Application of Machine Learning in Google Services- A Case Study
[Link]

FASHION INDUSTRY
Zara is Using AI
1. Revolutionizing Inventory Management with AI-Powered Analytics
Challenge
Zara, a global leader in fast fashion, excels at swiftly adapting to changing trends
while maintaining an efficient inventory system. The brand encountered notable
difficulties in managing inventory effectively across its extensive store and
distribution network. Traditionally, inventory decisions were made based on
historical sales data and manual assessments, often resulting in overstocking,
understocking, and wastage.
In the highly competitive retail sector, overstocking results in higher storage costs
and unsold products, while understocking can lead to missed sales opportunities
due to item unavailability. Additionally, Zara’s commitment to a responsive supply
chain made it crucial to predict demand accurately and replenish stock in near real-
time. The need to balance efficiency, responsiveness, and profitability highlighted
the limitations of traditional inventory management practices.

Solution
a. AI-Driven Demand Forecasting: Zara implemented AI-powered analytics to
address these challenges to transform its inventory management processes. AI
tools analyze a wide range of data, such as past sales, current purchasing patterns,
weather forecasts, and social media trends, to accurately anticipate demand. For
example, AI can detect emerging trends in real time, enabling Zara to adjust
production and distribution strategies to align with shifting consumer needs.
b. Dynamic Stock Allocation: Zara’s AI tools dynamically allocate inventory to stores
based on predicted demand and local trends. These systems monitor stock levels
continuously and recommend replenishment schedules to ensure optimal stock
availability without overburdening store inventories. For example, if AI detects a
spike in demand for summer dresses in a particular region due to rising
temperatures, it reallocates inventory to efficiently meet the surge in demand.
c. Warehouse Optimization: The company’s warehouses now leverage AI to enhance
inventory organization and movement. Advanced AI systems enhance storage
efficiency and streamline order fulfillment processes, reducing operational time and
workforce requirements. Furthermore, AI predicts the frequency of restocking
specific items, ensuring that popular products are always available while minimizing
the storage of less in-demand items.
d. Minimizing Waste: Zara also uses AI to identify slow-moving inventory and
recommend strategies to reduce excess stock. For instance, items with declining
demand might be highlighted for promotions or markdowns to clear inventory and
prevent waste. It not only enhances profitability but aligns with Zara’s sustainability
goals.
Result
Zara has significantly improved its operational efficiency and responsiveness by
integrating AI-powered analytics into its inventory management processes.
Accurate demand forecasting has reduced stockouts, improving customer
satisfaction and boosting sales. Dynamic stock allocation has optimized store
inventory levels, minimizing overstocking and understocking issues. Warehouse
optimization has streamlined Zara’s supply chain, reducing costs and enhancing
speed. Furthermore, the company’s ability to cut down on waste reflects its
dedication to sustainability and enhances its public image. With AI-driven inventory
management, Zara continues to lead the fast fashion industry, setting benchmarks
for agility and efficiency.
[Link]
wsiqTeslaUsingAICaseStudy