Large Language Models (LLMs) are Artificial Intelligence algorithms that use massive data sets to summarize, generate and reason about new content. LLMs are built on a set of neural networks based on the transformer architecture. Each transformer consists of encoders and decoders that can understand text sequence and the relationship between words and phrases in it.
The generative AI technologies that have been enabled by LLMs have transformed how organizations serve their customers, how workers perform their jobs and how users perform daily tasks when searching for information and leveraging intelligent systems. To build an LLM we need to define the objective of the model and whether it will be chatbot, code generator or a summarizer.
Building an LLM
Building an LLM requires the curation of vast datasets that enable the model to gain a deep understanding of the language, vocabulary and context around the model’s objective. These datasets can span terabytes of data and can be grown even further depending on the model’s objectives [1].
Data collection and processing
Once the model objectives are defined, data can be gathered from sources on the internet, books and academic literature, social media and public and private databases. The data is then curated to remove any low-quality, duplicate or irrelevant content. It is also important to ensure that all ethics, copyright and bias issues are addressed since those areas can become of major concern as the model develops and begins to produce the results and predictions it is designed to do.
Selecting the model architecture
Model selection involves selecting the neural network design that is best suited for the LLM goals and objectives. The type of architecture to select depends on the tasks the LLM must support, whether it is generation, translation or summarization.
Perceptrons and feed-forward networks: Perceptrons are the most basic neural networks consisting of only input and output layers, with no hidden layers. Perceptrons are most suitable for solving linear problems with binary output. Feed-forward networks may include one or more layers hidden between the input and output. They introduce non-linearity, allowing for more complex relationships in data [2].
Recurrent Neural Networks (RNNs): RNNs are neural networks that process information sequentially by maintaining a hidden state that is updated as each element in the sequence is processed. RNNs are limited in capturing dependencies between elements within long sequences due to the vanishing gradient problem, where the influence of distant input makes it difficult to train the model as their signal becomes weaker.
Transformers: Transformers apply global self-attention that allows each token to refer to any other token, regardless of distance. Additionally, by taking advantage of parallelization, transformers introduce features such as scalability, language understanding, deep reasoning and fluent text generation to LLMs that were never possible with RNNs. It is recommended to start with a robust architecture such as transformers as this will maximize performance and training efficiency.
Figure 1. Example of an LLM architecture [3].
Implementing the model
Implementing the model requires using a deep learning framework such as TensorFlow or PyTorch to design and assemble the model’s core architecture [4]. The key steps in implementing the model are:
Defining the model architecture such as transformers and specifying the key parameters including the number of heads and layers.
Implementing the model by building the encoder and decoder layers, the attention mechanisms, feed-forward networks and normalizing the layers.
Designing input/output mechanisms that enable tokenized text input and output layers for predicted tokens.
Using modular design and optimizing resource allocation to scale training for large datasets.
Training the model
Model training is a multi-phase process requiring extensive data and computational resources. These phases include:
Self-supervised learning where the model is fed massive amounts of data, so that it can be trained in language understanding and predicting missing words in a sequence.
Supervised learning where the model is trained to understand prompts and instructions allowing it to generalize, interact and follow detailed requests.
Reinforcement learning with Human Feedback (RLHF) involves learning with human input to ensure that output matches human expectations and desired behaviour. This also ensures that the model avoids bias and harmful responses and that the output is helpful and accurate.
Fine tuning and customization
Customization techniques include full model fine-tuning where all weights in the model are adjusted to focus on task-specific data. It is also possible to fine-tune parameters and engineer prompts to focus on smaller modules, saving resources and enabling easier deployment.
Training a pre-trained model based on domain-specific datasets allows the model to specialize on target tasks. This is easier and less resource-intensive than training a model from scratch since it leverages the base knowledge already learned by the model.
Model deployment
Deploying the LLM makes it available for real-world use, enabling users to interact with it. The model is deployed on local servers or cloud platforms using APIs to allow other applications or systems to interface with it. The model is scaled across multiple GPUs to handle the growing usage and improve performance [5]. The model is continually monitored, updated and maintained to ensure it remains current and accurate.
Ethical and legal considerations
The ethical and legal considerations are important in the development and deployment of LLMs. It is important that the LLM is unbiased and that it avoids propagating unfair and discriminatory outputs. This extends to discriminatory and harmful content which can be mitigated through reinforcement learning with human feedback (RLHF).
Training data may contain sensitive and private information, and the larger the datasets used to train the model the greater the privacy risks they involve. It is essential that privacy laws are adhered to and followed to ensure the models can continue to evolve and develop while preventing unintended memorization or leakage of private information.
Copyright and intellectual property must also be protected by ensuring that the proper licenses are obtained. Regular risk and compliance assessments and proper governance and oversight over the model life cycle can help mitigate ethical and legal issues.
Conclusion
Developing and deploying an LLM in 2025 requires a combination technical, analytical and soft skills. Strong programming skills in Python, R and Java are critical to AI development. A deep understanding of machine learning and LLM architectures including an understanding of the foundational mathematical concepts underlying them, are also critical. It is also important to have a good understanding of hardware architectures including CPUs, GPUs, TPUs and NPUs to ensure that the tasks performed by the LLM are deployed on the most suitable hardware to ensure efficiency, scalability and cost-effectiveness.
Other skills related to data management, problem-solving, critical thinking, communication and collaboration, and ethics and responsible AI are also essential in ensuring the models remain useful and sustainable.
While LLMs and their revolutionary transformer technology continue to impress us with new milestones, their foundations are deeply rooted in decades of research conducted in neural networks at countless institutions, and through the work of countless researchers.
Introduction
Large Language Models (LLMs) have gained momentum over the past five years as their use proliferated in a variety of applications, from chat-based language processing to code generation. Thanks to the transformer architecture, these LLMs possess superior abilities to capture the relationships within a sequence of text input, regardless of where in the input those relationships exist.
Transformers were first introduced in a 2017 landmark paper titled “Attention is all you need” [1]. The paper introduced a new approach to language processing that applied the concept of self-attention to process entire input sequences in parallel. Prior to transformers, neural architectures handled data sequentially, maintaining awareness of the input through hidden states that were recurrently updated with each step passing its output as input into the next.
LLMs are only an evolution of decades old artificial intelligence technology that can be traced back to the mid 20th century. While the breakthroughs of the past five years in LLMs have been propelled by the introduction of transformers, their foundations were established and developed over decades of research in Artificial Intelligence.
The History of LLMs
The foundations of Large Language Models (LLMs) can be traced back to experiments with neural networks conducted in the 1950s and 1960s. In the 1950s, researchers at IBM and Georgetown University investigated ways to enable computers to perform natural language processing (NLP). The goal of this experiment was to create a system that allowed translation from Russian to English. The first example of a chatbot was conceived in the 1960s with “Eliza”, designed by MIT’s Joseph Weizenbaum, and it established the foundations for research into natural language processing.
NLPs relied on simple models like the Perceptron, which were simple feed-forward networks without any recurrence features. Perceptrons were first introduced by Frank Rosenblatt in 1958. They were a single-layer neural network, based on an algorithm that classified input into two possible categories, and tweaked its predictions over millions of iterations to improve accuracy [3]. In the 1980s, the introduction of Recurrent Neural Networks (RNNs) improved on perceptrons by handling data sequentially while maintaining feedback loops in each step, further improving learning capabilities. RNNs were better able to understand and generate sequences through memory and recurrence, something perceptrons could not do [4]. Modern LLMs improved further on RNNs by enabling parallel rather than sequential computing.
In 1997, Long Short-Term Memory (LSTM) networks introduced deeper and more complex neural networks that could handle greater amounts of data. Fast forward to 2019, a team of researchers at Google introduced the Bidirectional Encoder Representations from Transformers (BERT) model. BERT’s innovation was its bidirectionality which allowed the output and input to take each other’s context into account. This allowed the pre-trained BERT to be fine-tuned with just one additional output layer to create state-of-the-art models for a range of tasks [5].
From 2019 onwards, the size and capabilities of LLMs grew exponentially. By the time OpenAI released ChatGPT in November 2022, the size of its GPT models was growing in staggering amounts, until it reached an estimated 1.8 trillion parameters in GPT-4. These parameters include learned model weights that control how input tokens are transformed layer by layer, as discussed later in this article. ChatGPT allows non-technical users to prompt the LLM and receive a response quickly. The more the user interacts with the model, the better the context it can build, thus allowing it to maintain a conversational type of interaction with the user.
The LLM race was on. All the key industry players began releasing their own versions of LLMs to compete with OpenAI. To respond to ChatGPT, Google released Bard, while Meta introduced LLaMA (Large Language Model Meta AI). Microsoft had partnered with OpenAI in 2019 and built a version of its Bing search engine powered by ChatGPT. DataBricks also released its own open-source LLM, named “Dolly”
Understanding Recurrent Neural Networks (RNNs)
The inherent characteristic of Recurrent Neural Networks (RNNs) is their memory or hidden state. RNNs process input sequentially, token by token, with each step considering the current input token and the current hidden state to calculate a new hidden state. The hidden state acts as a running summary of the information seen so far in the sequence. RNNs understand a sequence by processing them word by word, while keeping a running summary of the words seen so far [6].
The recurrent structure of RNNs means that they perform the same computation at each step, with their internal state changing based on the input sequence. Therefore, if we have an input sequence x = (x1, x2, …, xt), the RNN updates its hidden state ht at time step t using the current input xt and the previous hidden state ht-1. This can be formulated as:
Where:
ht is the new hidden state at time step t
ht-1 is the hidden state from the previous time step
xt is the input vector at time step t
Whh and Wxh are the shared weight matrices across all time steps for hidden-to-hidden and input-to-hidden connections, respectively.
bh is a bias vector
tanh is a common activation function (hyperbolic tangent), introducing non-linearity.
This process can be visualized by breaking out the RNN timeline to show a snapshot of the system at each step in time, and how the hidden state is updated and passed from one step to the next.
Figure 1. An RNN through time. The RNN cell processes input and the hidden state from time t-1 to produce the hidden state that is used as input together with token input from time t.
Limitations of RNNs
The sequential nature of RNNs limits their ability to process tasks that contain long sequences. This may be one of the main limitations that gave rise to the development of the transformer architecture and the need to process sequences much faster and in parallel. The following are some of the main limitations of RNNs.
Limitations in modeling long-range dependencies: RNNs are limited in capturing dependencies between elements within long sequences. This limitation is due primarily to the vanishing gradient problem. As gradients (error signals) occur during later steps in time, it becomes increasingly more difficult for the signal to flow backwards to earlier time steps and adjust the weights. This is because the longer the sequence the fainter the signal becomes. As the signal becomes weaker, by the time it reaches the relevant steps it becomes increasingly more difficult for the network to learn and trace back the relationship between those earlier inputs and the later outputs.
Sequential processing: RNNs process sequences token by token, in order. Hidden states must also be processed sequentially such that to obtain the hidden state at time t, the RNN must use the hidden state from t-1 as input. Modern hardware like GPUs and TPUs are well equipped to work with parallel computation. RNNs are unable to make use of this hardware due to their sequential processing, which leads to longer training times compared to parallel architectures.
Fixed size of hidden states: In the sequence-to-sequence model, the encoder must process and compress the entire input sequence into a single fixed size vector. This vector is then passed to the RNN decoder which is used to generate the output for the next hidden state. The compression of potentially long and complex input sequences into fixed-size vectors can be challenging. It also makes it difficult to retrain the network on all the input details that may have been compressed, and thus potentially causing some important information required for training to be missed.
How Transformers Replaced Recurrence
The limitations of RNNs in optimizing learning over large sequences and their sequential processing gave rise to the transformer architecture. Instead of sequential processing, transformers introduced self-attention, enabling the network to learn from any point in the sequence, regardless of distance.
Self-attention in transformers is analogous to the way humans process long sequences of text. When we attempt to translate text or process complex sequences, we do not read the entire text and then attempt to translate or understand it. Instead, we tend to go back and review parts of the text that we determine are most relevant to our understanding of it, so that we can generate the output we are trying to produce. In other words, we pay attention to the most relevant parts of the input that will help us generate the output. Transformers apply global self-attention that allows each token to refer to any other token, regardless of distance. Additionally, by taking advantage of parallelization, transformers introduce features such as scalability, language understanding, deep reasoning and fluent text generation to LLMs that were never possible with RNNs.
How Transformers Pay Attention
Self-attention enables the model to refer to any token in the input sequence and capture any complex relationships and dependencies within the data [7]. Self-attention is computed through the following steps.
1. Query, Key, Value (Q,K,V) matrices
Query (Q): Represents what current information is required or must be focused on. This can be described by asking the question “What information is the most relevant right now?”
Key (K): Keys act as identifiers for each input element. They are compared against the input sequence to determine relevance. This is analogous to asking the question “does this input token match the information I am looking for?”
Value (V): Values are also associated with each input token and represent the content or the meaning of that token. Values are weighted and summed to produce the context vector, which can be described as “this is the information I have”.
The model performs a lookup for each Query across all Keys. The degree to which a Query matches a Key determines the weight assigned to the corresponding Value. The model then calculates a weight or an attention score that determines how much attention a token should receive when generating predictions. The attention scores are used to calculate a weighted sum of all the Value vectors from the input sequence. The result is a vector containing the weighted sum of all the weighted value vectors.
Figure 2. Calculating the weighted sum vector. Attention scores are multiplied by the Value (input) vectors which are then summed to produce the weighted sum vector.
We have already discussed how traditional RNNs struggle to retain information for distant input due to the sequential nature of their hidden state. Attention, on the other hand, allows the model to consider the weights of all inputs and by summing them up. The resulting vector incorporates information from all inputs with the proper weights assigned to them. This allows the model to have a context of all input, while focusing on the most relevant information in the sequence, regardless of their distance.
2. Multi-head attention
When we consider a sentence, we do not consider it one word at a time. Instead, we look at each specific word in the sentence and consider whether it is the subject or the object. We also consider the overall grammar to make sense of the sentence and what it is trying to convey.
The same analogy applies when calculating attention. Instead of performing a single attention calculation for the Q, K, V vectors, multiple calculations are performed each on a single attentionhead, such that each head looks at a different pattern or relationship in the sequence. This is the concept of multi-head attention which allows the parallel processing of the Q, K, V vectors. It allows the model to look at different patterns or relationships within the sequence.
3. Masked multi-head self-attention
Masking ensures that the head focuses only on the tokens received so far when generating output, without looking ahead into the input sequence to generate the next token.
Attention score: The dot product of the Q and K matrices is used to determine the alignment of each Query with each Key, producing a square matrix reflecting the relationship between all input tokens.
Masking: A mask is applied to the resulting attention matrix to positions the model is not allowed to access, thus preventing it from ‘peeking’ when predicting the next token.
Softmax: After masking, the attention score is converted into probability using the Softmax function. The Softmax function applies a probability distribution to a vector whose size matches the vocabulary of the model, called logits. For example, if the model has a vocabulary of 50,000 words, the output logits vector will have a dimension of 50,000. Each element in the logits vector corresponds to a score for one specific token in the vocabulary. The Softmax function takes the logits vector as input, and outputs a probability vector that represents the model’s predicted probability distribution over the entire vocabulary of the model for the current position in the output sequence.
When it calculates attention for the Q, K, V vectors, the model does not recalculate attention for the same original Q, K, V vectors. Rather, it learns separate linear projections for each head. If we have h attention heads, then each head i learns the projection matrices
Each head performs the scaled Dot-Product Attention calculation using its projected Qi, Ki, Vi:
Where dk is the dimension of the Ki vectors within each head. Each projection of Qi, Ki, Vi allows a head to focus on and learn from a different representation of the original input. By running these calculations in parallel the model can learn about different types of relationships within the sequence.
4. Output and concatenation
The final step is to concatenate the output from all attention heads and apply a linear projection, using a learned weight matrix, to the concatenated output. The concatenated output is fed into another linear feedforward layer, where it is normalized back to a constant size to preserve the original meaning of the input before it is passed deeper into additional layers in the network [8].
Conclusion
There is no doubt that transformers have revolutionized the way LLMs have been deployed and applied in a variety of applications, including chatbots, content creation, agents and code completion. By relying on large and ever-growing volume of parameters, and an architecture that is designed for scalability and parallel computing, we are only beginning to discover the breadth of applications transformers can have.
As the challenges facing LLMs continue to be overcome, such as the ethical and environmental concerns, we can expect them to continue to become more efficient, more powerful and ultimately more intelligent. While LLMs and their revolutionary transformer technology continue to impress us with new milestones, their foundations are deeply rooted in decades of research conducted in neural networks at countless institutions, and through the work of countless researchers.
References
[1] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin. “Attention Is All You Need.” In Proceedings of the 31st Conference on Neural Information Processing Systems (NeurIPS 2017). arXiv:1706.03762 cs.CL, 2017
The retail landscape is being quickly transformed by Agentic AI programs that are driving a competitive race to lead in autonomy, speed and personalized customer experiences. In 2025, retailers cannot afford not to move quickly and aggressively in implementing agentic AI in all business functions or they risk being left behind, or worse, forced to exit.
Introduction
AI agents are redefining retail and are evolving into autonomous assistants that plan, recommend and take action. One of the most prominent examples of this shift is Walmart’s “Sparky”, a conversational AI shopping assistant in the mobile app that can understand customers shopping needs, suggest relevant products, answer questions and provide recommendations based on preferences [1]. Walmart is betting big on AI to drive its e-commerce growth and is aiming for online sales to account for 50% of its total sales [2].
Amazon, another retail giant, is using AI on a different scale by creating a harmonious ecosystem of AI and Machine Learning (ML) models across the different functional areas of the business. For example, demand forecasting is accomplished using models that leverage sales history, social media, economic trends and weather to predict demand more accurately. Machine learning (ML) algorithms use data across the supply chain to optimize stock levels and replenishment strategies to ensure alignment with predicted demand. Amazon is also using AI to automate inventory management and using AI-driven robots to manage the movement of good within warehouses. Other AI models optimize delivery routes in real-time using inputs like traffic conditions and weather among other factors [3].
Retailers that make use of AI and ML will ensure they maintain a competitive edge, and those that do not, risk being left behind or forced to exit. Amazon’s example of creating an ecosystem that uses the output from one AI model as input into another ensures that the business continues to add efficiencies and boost future profitability. Across the U.S., retailers are investing heavily in AI agents, with 83% of companies claiming AI is a top priority in business plans [4].
These statistics bring about an interesting question: what if every customer and every employee had their own AI agent, helping find products and optimize their shopping experience, or helping with labor-intensive tasks? AI agents are evolving from pilot projects to front-line and business critical applications, enabling businesses to gain a competitive edge and attract customers with better online shopping experiences.
What Are AI Agents?
In the context of AI, “agentic” refers to autonomous systems capable of making decisions and acting independently. AI agents are a more advanced form of AI that can make decisions and take actions with little or no human intervention. Agentic AI can combine multiple interconnected AI agents that are continuously learning, reasoning and acting proactively. Businesses can customize AI agents to meet their needs, given the flexibility and adaptability of AI agents for a wide range of industries and applications [5][6].
The key features of agentic AI include:
Autonomy: the ability to work autonomously to analyze data and solve problems in real-time with little human intervention.
Collaboration: the ability of multiple AI agents to work together leveraging Large Language Models (LLMs) and complex reasoning to solve complex business problems.
Learning and adaptation: dynamically evolving by interacting with its environment, and refining strategies, based on feedback and real-time data.
Reasoning and proactivity: identifying issues and forecast trends to make decisions such as reordering inventory or resolving customer complaints.
The adoption of Agentic AI in 2025 is gaining momentum as businesses aim to move from insight to action at greater speed and efficiency. Agentic AI solves the problem of scarce human resources needed to deal with the growing volume, complexity and inter-dependence of data. By moving at the speed of machine computation, agentic AI allows businesses to be more agile in real-time, act on business-critical insights more quickly, and scale more rapidly.
The competitive edge introduced by agentic AI is driving its rapid adoption, and it is due to the following factors [7][8][9]:
Speed: Businesses must move and react to customer needs, supply chain factors and market conditions at unprecedented speeds in 2025. It is no longer sufficient to use traditional AI that still relies on human intervention. Agentic AI is not only able to forecast problems and issues, but it can also act and execute upon them. Agentic AI can forecast and resolve customer issues before they even occur, and it can react to supply chain disruptions by forecasting and acting upon them, before they happen.
Reduce reliance on humans, not replace them: Agentic AI does not aim to replace humans and take away jobs, but rather to augment them. It acts as a co-worker that enhances productivity by focusing on analysis of repetitive data-intensive processes, creating forecasts that enable faster decision-making, and enabling employees to focus on business strategy and the creative, innovative decisions that will allow the business to continue to grow. Agentic AI allows businesses to increase performance while cutting costs without the need for increased human intervention.
Cost reduction and improved ROI: Agentic AI is also unlocking vast opportunities for cost reduction, through quick evaluation of data, testing strategies and adjusting operations in real-time. By automating repetitive and data-intensive processes, AI agents reduce the dependence on manual labor, minimize errors that translate to rework and add cost-effectiveness and efficiency that in turn result in higher ROI.
Enhanced customer experience: AI agents are capable of contextual understanding, proactive assistance and continuous learning. This allows them to boost customer satisfaction and loyalty by offering instant, real-time assistance and answers to customers queries while reducing wait times and improving resolutions rates.
Business must adapt or die: Agentic AI allows businesses to remain at the forefront of their market by learning and adapting in real-time. In 2025, customers expect instant and personalized service. It is becoming easier for businesses to integrate agentic AI into their various systems, especially with the introduction of Model Context Protocol (MCP) integration framework enabling intelligent agents to interact with external systems in a standardized, secure, and contextual way. User-friendly applications allow businesses to easily connect and deploy AI agents via a visual workflow builder without coding. Business have the opportunity to adapt by leveraging the technologies and capabilities available to them today to implement agentic AI.
The following table illustrates how AI is being implemented across various areas within Retail.
Functional Areas
Applications
Examples
Customer experience
• Personalize services, answer questions, and process orders • Offer product and project guidance • Smart kiosks assist with product search, availability, and location • AI delivers instant answers, recommendations, and smoother shopping
• Walmart’s “Sparky” suggests products and summarizes reviews Lowe’s AI assistant offers DIY and product support via app H&M’s chatbot recommends outfits, boosting satisfaction by 30% [10]
Inventory Management
• Streamline store ops and inventory management AI robots track stock and automate restocking • Smart shelves auto-detect low stock and reorder • Forecast demand using sales and market data • AI schedules staff based on foot traffic forecasts Video analytics detect theft and safety issues
• Zara’s AI cut stockouts by 20% and excess by 15% by using data from sales, customer behavior and market trends to forecast demand • Walmart uses robots for real-time shelf scanning • Home Depot AI helps staff quickly access data and gain necessary information to help customers
Supply Chain
• Adjust orders and routing using sales, weather, and trend data • Track shipments, suppliers, and logistics for full supply chain visibility • Improve forecasting to optimize supply chain operations • Cut costs by aligning forecasts with supply chain efficiency
• Kroger’s AI forecasting cut food waste by 10% and improved inventory accuracy by 25% • Unilever’s AI use reduced supply chain costs by 15% and improved delivery times by 10% • Walmart also achieved major gains through AI-driven supply chain improvements
Marketing
• Agentic AI manages end-to-end customer journeys across commerce, content, loyalty, and service [11] • AI interprets real-time journey data to adapt marketing strategies • Retailers use AI insights to keep campaigns fast, relevant, and effective • AI analyzes feedback to spot improvements and cut manual tasks
• Nike uses AI to predict purchases and personalize marketing, boosting engagement by 20% and driving sales • Coca-Cola uses predictive analytics to shift budget to high-performing channels, increasing Instagram spend by 20% and sales by 15%.
Table 1: Retail Examples Where AI Is Already Driving Impact
What Executives Should Do To Drive The Agentic AI Shift
AI agents are changing how organizations can deliver value to their customers, improve customer experience and manage risks. Executives are becoming increasingly aware that agentic AI is not just an automation tool, but rather a new way to drive deep business innovation and, if harnessed correctly, a way to maintain a competitive advantage.
Executives must lead the shift in the organization towards agentic AI by aligning governance and priorities to support IT and data investments required. To facilitate this shift to agentic AI the CEO must focus on [12][13]:
Investing in labor and technical infrastructure: this is accomplished by removing the barriers across the various systems in the organization to enable AI agents to operate across the various functional areas. In addition, upskilling and retraining the workforce is required to learn how to work with the new technologies introduced by agentic AI.
Lead the organizational shift: establish the goals and intended values of using agentic AI in the organization, and how it is to be used as a partner in creating value. The goal should not be simply to focus on optimizing headcount and reducing costs, it is about leading the organization into the future of retail.
Highlight key projects: by spearheading key and high-value projects in areas of the organization such as supply chain management, operations and customer service, the CEO can help build momentum and rally resources. They can also demonstrate the value of agentic AI by tracking key KPIs.
Oversee risk, compliance, and ethics: it is essential for the CEO to oversee all regulatory, privacy, transparency and risk issues related to the adoption of agentic AI. This is crucial in allowing the organization to proceed with confidence in implementing the various technical and IT infrastructures needed, and to realize the value and gains from agentic AI quickly and efficiently.
It is important to note that organizations that can quickly adopt and adapt to agentic AI will gain the competitive edge. The value proposition for executives in adopting this technology can be summarized in the following key elements:
Business transformation through automation and productivity: Agentic AI goes beyond the range of capabilities offered by Gen AI and can handle complex workflows through autonomous decision-making. This allows staff to work alongside AI agents and use its output while focusing on strategic and high-value tasks that boost workers productivity and allow them to use their time efficiently.
Gaining a competitive edge: AI agents work continuously adapting to real-time issues, learning and making decisions quickly. This enhances customer experience, boosts innovation and resilience against market changes.
Boost ROI and increase revenues: Studies have shown that agentic AI contributes up to 18% improvement in customer satisfaction, employee productivity, and market share, with $3.50 in return for every $1 invested realized over a 14-month payback period [14]. This is driven primarily by redirecting human resources from focusing on repetitive low-value tasks to more strategic and high-value ones.
Enable rapid scaling and agility: AI agents can help lead the transformation of the organization to be more forward-looking and competitive, by driving business transformation, upskilling the workforce and enabling the rapid scaling and adaptation of business models.
Implementation Priorities: How to Get Started
The diagram below illustrates the interconnected functional areas and visually describes how they intersect with Inventory Management in an omnichannel retail environment. The data that flows between each area is what is used in AI models to enhance decision making. The interconnected data that flows between functions feed AI models which generate insights needed to optimize inventory, fulfillment, and customer responsiveness.
Figure 1: Inventory Management across Functional areas in Retail
The table below outlines key functional areas, the associated data points, and how AI is applied to enhance operational efficiency.
Inventory Layer
Key Data Points
AI Usage to Improve Efficiency
Factory / Seller*
• Proforma Invoice • Commercial Invoice • Packing List
• Predict lead times and invoice anomalies • Detect supply risk patterns
Shipper
• Advanced Shipping Notice (ASN)
• Predict shipment delays • Optimize dock scheduling at warehouse
Warehouse
• Putaway Status • Inventory Quantity & Location • SKU Detail • Cycle Count Accuracy • Labor Handling Time
Table 2: How Data Enables AI to Improve Inventory Across the Supply Chain *Assumes FOB Incoterms
Implementing Agentic AI follows a multi-phased approach that integrates technology, business and culture. This approach can be iterative and repeated as necessary depending on the complexity and scope of the processes being automated [15].
Readiness ➡ Design ➡ Pilot ➡ Scale
Assessing readiness
Assessing readiness involves evaluating and auditing workflows, data infrastructures and IT capabilities to ensure compatibility with the agentic AI needs. These include ensuring that AI model outputs will be compatible with the organization’s future audit needs and that IT infrastructures can support the AI models data requirements.
It is also important to evaluate the company’s culture and assess the adaptability and openness to automation. This is a good opportunity to address any resistance and skill gaps through education and training to ensure that teams see the value agentic AI will add to their work.
The readiness phase is also a good opportunity to identify high-impact business use cases that can be used to pilot the implementation of agentic AI processes, and scale as necessary to the rest of the organizations, as these processes are further developed and defined.
Design
The design phase is important in defining objectives and scope, ensuring leadership buy-in and that data systems are properly integrated to meet the needs of the agentic AI models.
Defining scope and objectives involves setting clear and measurable business goals and aligning AI initiatives with the overall company strategy. This is best achieved by identifying key business processes and applications that could provide the highest impact, show the best ROI and serve as the benchmark for future projects and applications.
Securing leadership and cross-functional team buy-in is also critical in ensuring that AI models are fully adopted into the various business processes, and that communicated ROIs are realized to their fullest potential. This is achieved by securing sponsorship at the executive level, and assembling multi-disciplinary teams from IT, data science and engineering, operations and compliance. It is essential that clear, attainable and measurable ROIs are clearly communicated to ensure that teams work collectively towards achieving the defined goals and objectives.
Mapping data and systems integration ensures that agentic AI systems have easy and real-time access to data across various silos including CRM, EPR and other cloud applications. This is essential in allowing agentic AI models to ingest all data required for the algorithms and produce accurate and timely outputs to guide their decisions. It is essential that close attention is paid to upgrading the security of all systems as they are integrated to ensure that no vulnerabilities are introduced as part of this process.
Pilot
Deploy the AI models in a contained environment that allows collecting live data for training. This is a great opportunity to train, fine-tune and iterate on the agents to ensure they produce accurate output, ROIs are met and compliance is achieved. Correct errors in the models and the algorithms, monitor output and behavior, and document outcomes.
Scale
Scale the phased approach across additional business functions and processes while increasing integration across the various AI agents as they are scaled. Continue to retrain agents and monitor their performance and output, paying close attention to monitoring and updating the risks and adding controls as necessary. It is also essential to continue to train and upskill employees to enable them to collaborate productively with agents.
Risks, Realities, and Responsible Scaling
Agentic AI is projected to automate up to 15% of day-to-day enterprise decisions by 2028, and potentially resolve 80% of standard customer service issues [16]. However, this also introduces a large risk surface, especially for critical systems.
Increased cyber-attack and security risks – agentic AI systems are designed to act autonomously across multiple systems with access to various data silos across the organization. This creates a multitude of entry points and vulnerabilities for traditional cyber threats such as data leaks and hijacking. More novel and emergent threats can also be introduced such as “agent hijacking”, which allows malicious software to control agent behavior, directing it to perform unauthorized actions and access to data, and potentially collaborate with other agents through interactions that are difficult to detect and monitor.
Loss of control & unintended outcomes – by reducing human involvement and interactions, agentic AI increases the risk for agents to make incorrect, inappropriate or harmful decisions. This is especially true for LLMs that can misinterpret data and context and lead to unintended outcomes on a potentially massive scale.
Compliance, privacy and operational risks – agentic AI consumes and acts upon large amounts of sensitive data. Without proper oversight this opens the organization to risks of breaching privacy laws. It can also be difficult for large organizations to trace agentic AI decision making, thus making it difficult to audit, correct errors and perform disaster recovery.
In 2025, most enterprises are implementing and running agentic AI pilots, especially in functions like customer service and supply chain management. However, enterprises have yet to achieve true end-to-end adoption of agentic AI across their various business functions. To achieve this requires strong cross-functional alignment and adoption of agentic AI, something that is rare and hard to achieve.
Agentic AI has also been able to deliver value and efficiencies in domain-specific areas such as customer service and logistics, but it has yet to reliably deliver the same value for mission-critical business functions. There are still reliability challenges to overcome for agentic AI in these domain-agnostic areas.
As the market became flooded with a multitude of vendors and start-ups hoping to capitalize on the acceleration of AI technologies, the tools and frameworks offered for agentic AI have become fragmented and difficult to standardize. The pace of demand for these tools continues to far outstrip the pace at which these tools are offered.
What Kind of Retailer Will You Be?
The retail landscape is being quickly transformed by Agentic AI programs that are driving a competitive race to lead in autonomy, speed and personalized customer experiences. In 2025, retailers cannot afford not to move quickly and aggressively in implementing agentic AI in all business functions or they risk being left behind, or worse, forced to exit.
To be on track or ahead of the agentic AI trend in 2025, retailers must already be piloting or implementing it in one or more domains that were identified to have high ROI. Businesses can identify one or more functions such as customer support, supply chain and inventory management or marketing automation, where agentic AI can be strategically deployed to realize high ROIs.
IT infrastructures and systems must also be revamped through APIs and data pipelines that allow for seamless integration of AI agents across various data silos across POS, supply chain and CRM platforms. While these actions are taking place, it is critical for retailers to ensure proper governance and frameworks are put in place to manage agentic AI risks, ethics and compliance. This can be done through maintaining proper audit trails, real-time monitoring of AI agents output and decision-making, and clear disaster recovery plans.
It is also critical for retailers to ensure that employees are continuously educated, trained and upskilled in collaborating with and using AI agents. Maximizing ROIs does not rely entirely on the performance of AI agents. It also requires that employees learn and understand how to use AI agents to gain strategic insights that allow to focus on creative and impactful decisions.
Retailers can also establish agentic AI centers of excellence to ensure proper governance and compliance, manage risks and lead strategies for responsible scaling of agentic AI at the enterprise level. Training and upskilling of employees to collaborate with Agentic AI is also critical. These actions can also be further strengthened through the formation of vendor partnerships to collaborate with AI solutions providers that allow for rapid deployment capabilities and quicker realization of ROIs. Retailers can also participate is industry consortiums to benchmark, share knowledge and establish standards and risk mitigation strategies.
In this case study, we examine a team that has recognized the potential of Machine Learning (ML) and Artificial Intelligence (AI) to refine and enhance a longstanding methodology for forecasting product order volumes. By leveraging AI and ML, the team can achieve more precise ordering based on those forecasts, while also gaining the capability to monitor market prices and receive insights on how to adjust orders to minimize costs. Historically, this team has relied on Excel for manual and meticulous user input, maintaining continuous communication among members and adhering to a process honed over several decades. While this approach has been effective, they have now realized that integrating AI and ML can significantly enhance their workflow by handling larger datasets at a faster pace and generating profound insights aimed at maximizing efficiency, reducing costs, and driving business growth.
Role of the Product Designer
The Product Designer in this role helped initiate a significant project aimed at transforming a long-standing, Excel-based process into a web-based application. This endeavor required not only a user-centered design approach but also an understanding of how AI and ML can be applied effectively to realize the project goals and meet the users needs.
To achieve this, the Product Designer addressed the following considerations:
What processes were being followed?
How did users use Excel for data entry and forecasting?
Which identified business processes needed to be maintained, which could be enhanced with AI, and which could be replaced entirely?
Where could AI and ML introduce efficiencies and savings, and provide valuable insights?
Were the identified AI efficiencies and insights aligned with user expectations?
How would the design help users provide inputs easily, and act upon AI-generated outputs and insights?
How would the design help users enhance their work efficiency and enable them to focus on more strategic and human-centric tasks?
At this stage, and before any of the AI models and algorithms were developed, the Product Designer assumed a strategic role in defining the foundational framework for the data the AI models would use and the outputs and insights they would generate for user consumption and action.
By adopting a product owner’s mindset and taking on a strategic role in determining the user requirements as they pertained to the AI models, the Product Designer shaped the product through the following actions:
Identifying key stakeholders through close collaboration with the project manager
Gaining an understanding of the current business process and the value the project proposes to achieve
Developing interview scripts designed to deepen understanding of:
The current business process and how Excel is used to generate forecasts
The ideal product vision and how it aligns with the user needs and expectations, particularly as it applies to the use of AI and ML
How the product could add value to users’ day-to-day work
Scheduling and facilitating interviews with stakeholders.
Maintaining an openness to any additional insights gained during the interviews such as including additional stakeholders, and exploring other areas of the business as necessary.
Collecting and synthesizing feedback from the stakeholder interviews.
Establishing a framework based on the feedback analysis for user personas, user journeys and user flows.
These activities enabled the Product Designer to acquire deep insight into the current business process, the stakeholders and users and their roles, and most importantly, insights into the AI and ML user needs and what outputs would be expected .
Research & Discovery
Using the insights gained from stakeholder interviews, the Product Designer was able to:
Identify user personas based on the roles and users discussed during the interviews.
Conduct additional workshops to further refine the identified user personas.
Discover new opportunities for additional user personas not previously identified and develop them further.
Develop user journeys for key user personas or those with the most critical needs.
Create user flows that outline the application’s essential features, associated screens, inputs, and outputs.
Refine user flows, enhance identified features, and ascertain any missing components and data.
Develop an information architecture (IA) aimed at providing easy and intuitive navigation that prioritized productivity and ease of navigation between various sections of the application.
The outcome of this process was a well-defined product roadmap and vision that provided a clear framework for technical teams. Data scientists, engineers, back-end and front-end developers could use this roadmap, along with the user needs and technical requirements identified, to begin designing and developing AI models. This structured approach ensured that the AI models were not only functional but also optimized to enhance the user experience.
Designing with Clarity and Logic
The research and discovery phase provided essential insights, enabling the Product Designer to conceptualize how the application would meet users’ needs and allow the AI models to generate the necessary outcomes. The user flow diagrams established an information architecture that formed the basis for the features the application offered and the overall user experience. With the information architecture now established, wireframes and mockups were created to facilitate discussions with users regarding the design direction.
Wireframes and mockups enabled stakeholders and end users to understand how input is provided into the AI models and what the output would look like.
This stage was crucial in the product design process as it helped establish the foundation for robust AI models that would directly meet the users’ needs.
The wireframes and mockups were refined based on feedback from users and stakeholders through recurring reviews and workshops.
Wireframes, mockups, and user flows helped the Product Designer build a prototype that:
Assisted data scientists, engineers, front-end, and back-end developers in visualizing user input and the generated output in the application.
Assisted data scientists and engineers in understanding the requirements for data input and output processing, and design algorithms to meet these requirements.
Illustrated the detailed interactions necessary to enable users to calibrate their inputs into AI models.
Facilitated collaboration between back-end and front-end developers with data scientists and engineers to design and build APIs that support the flow of data and insights from AI models.
Allowed running usability testing sessions with end users to validate the design and iterate based on the feedback received.
The development of a prototype represented a significant milestone and underscored the strategic role of the Product Designer played in guiding stakeholders and users through discovery sessions, workshops and design reviews. The Product Designer’s efforts in understanding data input and output requirements, and visualizing them clearly as part of a prototype, enabled technical teams to design AI models and algorithms that met those requirement. This hybrid mindset adopted by the Product Designer, functioning as an intermediary between business strategy and technical execution, was pivotal in fostering collaboration among product, engineering, and data teams, ensuring a clear understanding of the product vision and roadmap.
To drive the success of this AI-enabled product, the Product Designer delivered a strategic and structured design process that included:
User personas, journeys, flows, and an information architecture that defined core behaviours and ensured the experience aligned with user needs.
Interactive prototypes, refined through multiple rounds of usability testing and stakeholder input.
Detailed interactions clearly outlining user inputs and the data required to support model performance.
Insight-driven visualizations, which shaped how AI outputs were presented and guided model design.
An end-to-end product roadmap, mapping the full product vision while enabling the extraction of an MVP and a plan for iterative, future releases.
Lessons Learned
In this case study, the product designer started their work by identifying users’ needs and pain points related to an existing business process. Stakeholders and users wanted to explore how AI and ML could introduce savings and efficiencies into their business.
The designer’s responsibilities included identifying and analyzing the problem, and integrating ML and AI solutions through extensive collaboration with stakeholders, product managers, engineers, and data teams. This collaboration was crucial for the designer to articulate the product vision using a user-centered approach while also providing comprehensive insights into the data engineering efforts needed to optimize AI model outcomes.
Designing and developing AI and ML models for a product is a time and resource-intensive process. Therefore, it is essential for organizations to ensure these resources and efforts are invested in a manner that maximizes benefits and potential gains. In this case study, the Product Designer’s role was vital in establishing the product vision and roadmap, and in helping the various project teams understand and acheive this vision.
The rapid integration of Artificial Intelligence (AI) and Machine Learning (ML) capabilities into products and applications necessitates a shift in the role of Product Designers.
AI offers significant opportunities for organizations to address complex business problems. Users are now able to provide input into detailed data models that can process extensive datasets and generate insights through what-if scenarios and simulations.
AI-driven scenarios and simulations often rely on substantial input data and calibration tailored to user needs. As output complexity increases, it becomes essential for Product Designers to be at the forefront, understanding these sophisticated models and shaping designs that present the results in an intuitive, user-friendly way.
Consequently, Product Designers must move beyond translating requirements into mock-ups and they must instead lead the vision for how human-interactive design can refine data input, guide calibration, and surface outputs that make AI models more actionable and aligned with business objectives. In this article, I’ll explore the following key areas and what’s needed to succeed:
Adopt a Product Owner’s Mindset
Design with Clarity and Logic
Lead with a Hybrid Mindset
Growing into the Role
Adopt a Product Owner’s Mindset
The Product Designer’s role has evolved to extend far beyond interface design, especially in AI product development. By adopting a Product Owner’s mindset, designers become key contributors to defining and delivering value. They shape business-aligned product strategies that build user trust in AI outputs, accelerate decision-making, and guide teams in aligning technical execution with business goals.
This mindset grounds Product Designers in structured, outcome-oriented thinking. It enables them to ask the right questions and lead cross-functional collaboration with clarity:
Who are the core and secondary user groups, and what are their needs?
What real problems does this product need to solve?
How do users currently generate insights, and where can AI improve the process?
What does the ideal product vision look like, and how can it be prototyped?
What defines the MVP, and how can it evolve into the ideal-state solution?
How will we validate both versions with end users before development?
With this approach, the Product Designer leads the development of a comprehensive product framework that informs decisions across the lifecycle, from early discovery through MVP delivery to long-term iteration. They help align teams around a shared vision, providing structure, clarity, and strategic direction that ensures product decisions are rooted in business value and user impact.
Designing With Clarity and Logic
The role of the Product Designer has become a critical part of AI product development, ensuring the product aligns with both business and user objectives while delivering the intended outcomes. Increasingly, data scientists, engineers, and developers rely on insights and outcomes from discovery work led by Product Designers in collaboration with stakeholders and end users.
By deeply understanding user personas, journeys, and workflows, the Product Designer brings clarity to complex systems. Their work translates business intent and user behavior into logical structures that guide product development and model behavior. This structured approach lays the foundation for AI-powered experiences by turning vision and research into clear, actionable design requirements.
Through this work, the Product Designer can:
Provide detailed, logically structured definitions of features and requirements.
Deliver a comprehensive understanding of the product vision from end to end.
Clearly map how AI models should present results and insights.
Clarify business rules and data relationships within a unified design logic.
Address stakeholder needs with precise, high-fidelity prototypes.
Supply robust data validation requirements based on business-aligned designs.
By designing with clarity and logic, the Product Designer empowers cross-functional teams to move forward with confidence, ensuring that every design decision is grounded in purpose, informed by data, and aligned with user expectations.
Lead with a Hybrid Mindset
As AI and data-driven products reshape the landscape, the Product Designer’s role is evolving. Today’s most effective designers lead with a hybrid mindset, one that combines user empathy, business strategy, and technical fluency.
This mindset is not about owning a product backlog, but about thinking and communicating like a Product Owner or Business Analyst. It’s about reducing ambiguity for technical teams, earning stakeholder trust, and helping cross-functional collaborators understand how design decisions tie to business outcomes.
When Product Designers operate with this hybrid mindset, they:
Translate complex user journeys into actionable design decisions.
Align design efforts with broader business objectives.
Help define clear roadmaps for MVPs and ideal-state experiences.
Communicate stakeholder needs clearly through prototypes and interactions.
Build team confidence in the product’s value and impact.
Serve as strategic partners, connecting vision with execution.
By integrating the language of business and technology into the design process, Product Designers become trusted leaders, not just creative contributors. They provide the connective tissue that links user needs, stakeholder priorities, and technical realities into cohesive, AI-powered product experiences.
Growing Into the Role
A Product Designer can cultivate the following competencies to attain a high level of strategic thinking and leadership:
Focus on core business problems and user needs.
Dive deep with data scientists and data engineers and collaborate on design AI models that meet user needs.
Create value from AI models by clearly visualizing insights and provide clear data calibration.
Turn insights into fast, testable prototypes and iterate.
Collaborate across teams to shape the product framework.
Define the ideal state and guide releases from MVP to full launch.
It is essential that Product Designers work closely with Business Analysts and Product Owners to shape clear, roadmap-aligned backlogs that reflect both user intent and business priorities. These collaborative skills are essential for defining intuitive user interactions within complex, AI-enabled applications. Mastering this intersection of design, strategy, and systems thinking elevates the Product Designer from contributor to strategic leader, capable of influencing both product direction and delivery.
Final Thoughts
Design is no longer just about aesthetics or interaction, it’s about enabling users to extract clarity and insight from complexity, particularly in AI-driven environments. The most effective Product Designers operate as strategists, analysts, and owners of the product experience, guiding teams through ambiguity to unlock value. Whether defining a user flow or shaping a new feature, ask: Does this design simply function, or does it help solve a deeper problem? When it does the latter, you’re not just designing, you’re leading. And in today’s AI-powered applications that leadership is what shapes truly impactful products.
