AI Agents – Baha Jabarin M.Sc., M.A.

Understanding Agentic AI: Key Insights for Retail Leaders

Introduction

The term “Agentic AI” is now commonly used in industry conversations, yet its meaning often ranges from simple automation tools to advanced digital workers. Retail leaders typically envision Agentic AI as a capable junior employee able to understand goals, reason, take action across platforms, and learn, setting high expectations for implementation.

This broad perception is close to the research-based definition: systems that pursue goals, understand context, plan, act, and collaborate with other agents. In practice, however, many solutions labeled as agentic simply combine automation, machine learning, language models, and APIs.

In this discussion:

Agentic AI means sophisticated, enterprise-level autonomous systems focused on defined objectives.
Autonomous Workflow Orchestration (AWO) reflects current retail tools: smart workflows still guided by human priorities.

Key questions covered:

What systems are in use today?
Which technologies are mislabeled as agentic?
What advancements are needed in tech, data, and processes to move from AWO to true agentic AI?

What People Think “Agentic AI” Is (And Why That Matters)

Many view an “agent” as more than a rule-based system. They expect it to handle complex tasks, strategize, and act independently. Technically, such agents should:

Understand goals rather than just react to inputs.
Make multi-step plans involving various systems.
Select and sequence tools or APIs appropriately.
Adapt when things go off course.

This distinction affects leadership expectations: if leaders think they’re getting fully capable agents, they may incorrectly assign responsibility. Confusing automation with autonomy can lead to inadequate oversight and accountability gaps. Accurate descriptions of “agentic AI” are crucial, as mislabeling advanced workflow automation may cause governance failures when organizations rely on abilities these systems don’t possess.

What AWO Really Is: Architectural Reality, Not Just Buzz

AWO is an integrated stack supporting autonomous workflows:

The Workflow/RPA layer manages tasks between systems.
Machine learning models assess risk, sort tickets, predict demand, and spot patterns.
LLMs process unstructured text, summarize, draft, and converse.
The integration fabric links retail and supply chain apps with APIs and queues.
Rules and policies set boundaries, manage thresholds, and handle approvals.

Compared to traditional automation, AWO uses machine learning to trigger workflows based on data, rather than fixed rules. LLMs interpret complex inputs, enabling routing by predictions or classifications instead of basic logic. While adaptable, these systems don’t independently pursue high-level goals; they follow designed workflows.

In retail, AWO can validate return requests, resolve delivery issues, and spot shelf gaps from images. Problems occur when model assumptions fail, rules conflict, or policies change. Because workflows drive actions, solutions often require process redesign, underscoring the gap to fully goal-driven, agentic systems.

The Spectrum of Automation and Agentic Behaviour in Retail

The spectrum of automation and agentic behaviour provides leaders with a framework to benchmark their current capabilities and chart a path for future development. Retail organizations typically progress through four distinct stages, each with its own strengths, weaknesses, and operational implications.

The spectrum: Automation → AWO → Narrow Agents → Agentic Ecosystems

Stage 1: Rules Automation

At this stage, automation is driven by macros, scripts, and Robotic Process Automation (RPA) bots. The primary advantage of this approach is its predictability and controllability. However, these systems are inherently brittle; any change in user interface or data format can cause the automation to break, leading to disruption in operations.

Stage 2: Adaptive Workflow Orchestration (AWO)

AWO systems can adapt within established workflows but lack the ability to modify the workflow structure itself. These systems remain workflow-centric but incorporate machine learning (ML) and large language models (LLMs) to make smarter decisions within the flow. The strength of AWO lies in its ability to handle greater variation and reduce manual handoffs. The limitation, however, is that goals are externally defined and the workflow logic is still hard-coded, constraining the system’s ability to respond to new or unexpected challenges.

Stage 3: Narrow Agents

Narrow agents introduce the capacity to make decisions based on trade-offs, not just rigid rules. These domain-specific agents can reason within a tightly defined scope. For example, a pricing agent can select among pre-approved strategies within established guardrails, while a disruption-management agent may propose and sometimes execute remediation steps. At this stage, the distinction between a “smart workflow” and an “agent” begins to blur, as the system starts to optimize rather than merely execute scripted actions.

Stage 4: Agentic Ecosystems

In this most advanced stage, agents operate under high-level goals and possess autonomy in selecting methods. Multiple agents with different roles and perspectives collaborate, sharing goals or negotiating trade-offs such as margin, service level, and inventory risk. These agents are empowered to choose their tools and may even propose new process variants, reflecting a dynamic and adaptive approach to retail operations.

Current State and Key Takeaway

Most retailers today find themselves between Stages 2 and 3, with Adaptive Workflow Orchestration present in several workflows and a few narrow agent-like pilots underway. Despite these advancements, governance, data foundations, and integration patterns remain rooted in traditional workflow-centric models, rather than in structures that support agents capable of initiating or reshaping work.

Importantly, progression through these stages cannot be achieved in a single leap. Each stage introduces new potential failure modes, ranging from simple bot breakdowns to workflows making poor decisions, to agents optimizing for objectives that may not align with organizational goals. Leaders must be deliberate and explicit about which stage they are designing for, ensuring that systems and processes are properly aligned with their intended capabilities.

Practical Examples: Where Automation Excels and Where It Falls Short

Automated Refunds and Returns: The Limits of Autonomy

Automated refund and return processes demonstrate how advanced orchestration systems streamline routine workflows. The standard – or “happy path” – scenario is handled efficiently: the system classifies the return reason, checks applicable policies, processes the refund, and notifies the customer. However, the process becomes more complex when exceptions arise. Critical questions include: Who is responsible for resolving edge cases such as suspected fraud, chronic returners, or policy conflicts? Is the automated system empowered to weigh cost against customer goodwill, or does that authority remain with humans?

Typically, automation is permitted only within a defined risk band. For instance: if the risk score is below a certain threshold (X), the system approves the refund automatically; if the score falls between X and Y, the case is escalated; if above Y, the refund is blocked. This illustrates classic Adaptive Workflow Orchestration (AWO) – the system applies a business’s predetermined risk appetite on a larger scale but does not set or adjust that appetite itself.

Computer Vision in Planogram Checks: From Task Generation to Strategic Action

In another example, computer-vision-powered systems conduct planogram checks, detecting gaps on shelves and prompting the workflow to generate corrective tasks. The deeper, strategic questions are: Can the system reprioritize these tasks based on factors such as sales impact or labour constraints? Is it able to propose alternative merchandising layouts in response to local store behaviour?

At present, the answer is generally no. The system continues to follow a linear process: detect an issue, then raise a task. True agentic behaviour would involve the system analyzing a store’s unique traffic patterns and sales profile, proposing a new display layout, simulating the impact, and rolling out the change as a test.

The Analytical Gap in Current Automation

A common pattern emerges across these scenarios. The “sense” and “act” phases of automation are becoming more intelligent and hands-off. Yet, determining the broader objectives – deciding what trade-offs are acceptable and which “game” to play – remains mostly a human-driven and static process.

This highlights a key analytical gap. While much is said about “autonomous AI,” closer examination reveals that most autonomy is local and tactical, not global and strategic. As a result, Adaptive Workflow Orchestration delivers strong return on investment (ROI) but does not fundamentally transform the underlying operating model.

A More Rigorous Look at Future Agentic Scenarios

Let’s revisit the future supply chain scenario in a more structured way. When an agent spots a disruption, it goes through several processes: monitoring data continuously, maintaining contextual awareness of business-critical variables, and communicating efficiently with other agents to coordinate responses.

The replenishment agent, in turn, considers constraints like supplier lead times and contractual limits, understands service levels and margin goals, and prioritizes options that best fit business objectives.

As more agents are added, covering margins, stores, and customer interactions, the challenges shift from simply integrating systems to ensuring all agents share accurate information, resolve conflicts, and know when to involve humans.

These issues mean automation is not just about upgrading technology. Key concerns include who defines agent goals, how often they’re reviewed, and what oversight exists for agent decisions. As a result, agentic pilots tend to focus on narrow tasks, such as dynamic pricing or local optimization, rather than handling entire supply chains. The primary hurdles relate to governance, data quality, and accountability, not just technical sophistication.

The Leadership Imperative: Why the AWO vs. Agentic AI Distinction Matters

Mischaracterizing Automated Workflow Orchestration (AWO) as fully agentic artificial intelligence can lead to notable repercussions for leadership and organizational effectiveness. When this distinction is not explicitly acknowledged, three primary challenges frequently emerge: architecture drift, risk blind spots, and talent misalignment.

1. Architecture Drift

Integrating agents into a workflow-centric environment without comprehensive planning often results in their function being limited to advanced decision points rather than serving as fundamental system components. Such an approach neglects critical design considerations including shared memory, a unified goal repository, and event-driven architecture, each essential for enabling agents to operate as integral contributors within the broader ecosystem.

2. Risk Blind Spots

The presumption that “the agent knows what it’s doing” may result in inadequate investment in vital safety and governance controls. These include:

Observability: Mechanisms enabling tracing and explanation of agent decisions.
Kill Switches: Capabilities to quickly intervene and suspend agent actions when necessary.
Sandboxes: Controlled environments for safely testing new agent behaviours prior to deployment.

3. Talent Mismatch

Prioritizing recruitment of only prompt engineers overlooks the comprehensive skills required for effective agentic AI implementation. Beyond technical expertise, organizations benefit from engaging:

Professionals skilled in designing robust machine–human workflows.
Individuals capable of defining agent objectives, constraints, and developing meaningful evaluation frameworks.

Retail-Specific Sequencing Challenges

Within the retail sector, misconstruing “buying agents” may result in omitting foundational activities such as:

Data cleansing and standardization for products, locations, and customers.
Streamlining process variants to minimize operational complexity.
Establishing standardized integrations across Order Management Systems (OMS), Warehouse Management Systems (WMS), Enterprise Resource Planning (ERP), and e-commerce platforms.

Neglecting these prerequisites often causes agentic initiatives to stagnate or devolve into isolated, non-scalable solutions. This may foster the erroneous belief that agents are inadequate, when in fact, the organization was insufficiently prepared for adoption.

Importance of Distinguishing AWO from Agentic Ecosystems

Differentiating between AWO and agentic ecosystems is imperative, as it significantly influences leadership approaches and talent requirements. While workflow enhancements primarily necessitate expertise in workflow engineering and machine learning/large language models (ML/LLM), transitioning to agentic systems demands reimagining organizational decision-making structures and recruiting individuals adept at architecting resilient socio-technical systems.

Practical Steps for Leaders: Navigating Agentic AI in Retail

If you are a CIO, COO, or Head of Digital responding to board-level questions about “agentic AI,” the following structured approach outlines what you should focus on over the next 12 to 18 months.

1. Maximize the Value of Automated Workflow Orchestration (AWO)

Identify five to ten high-volume, rules-based processes. Typical examples include returns management, handling order exceptions, vendor queries, and store-level tasks.
Redesign these processes explicitly as AWO, ensuring each has defined inputs, outputs, and key performance indicators (KPIs). Carefully consider where machine learning or large language models (ML/LLMs) can add measurable value.
Implement instrumentation for these flows to track and measure improvements such as reduced cycle times, lower error rates, and customer impact.

2. Develop Targeted Agent Pilot Projects

Deliberately design one or two narrow agent pilot initiatives. Select domains with clear objectives and manageable risks, such as dynamic pricing within set ranges, markdown optimization, or tuning localized assortments.
Allow agents to propose actions within predetermined guardrails. Initially, keep humans in the approval loop, gradually shifting to exception-only review as confidence in the system grows.
Treat these pilots as experiments in operational autonomy, not just as new digital tools. Document and analyze any challenges encountered, including data quality issues, policy conflicts, or trust barriers.

3. Lay the Foundation for “Agent Readiness”

Data: Clearly define what data agents will need to operate cross-functionally across the organization.
Events: Transition from nightly data batches to real-time event streams for key operational signals.
Governance: Establish an “autonomy matrix” to clarify which decisions can be fully automated, which require human review, and which should remain exclusively human-driven for the time being.

By systematically following these three steps, you will be building the necessary infrastructure and capabilities to progress from today’s orchestrated copilot models to tomorrow’s more autonomous agentic ecosystems, without exposing your organization to undue risk or succumbing to industry buzzwords.

Reframing “Progress” in Retail AI

The core message is not that “Agentic AI is years away, so wait,” but rather: “Retail is currently experiencing an AWO phase that offers notable value, and the approach taken to AWO will either position businesses for agentic ecosystems in the future or pose significant challenges later.”

If AWO implementations are opaque, rigid, and confined to singular applications, they limit long-term progress. Conversely, instrumented, integrated, and well-governed AWOs serve as foundational platforms for developing agent-based systems. While the underlying technologies may be similar, the resulting strategic trajectories differ substantially.

For organizational leaders, the critical consideration is not simply whether agents have been adopted, but whether today’s automation strategies are being designed to enable greater autonomy in the future, should that become desirable. Affirmative action in this regard ensures that organizations are leveraging current capabilities to strategically prepare for a transition toward autonomous retail operations.

December 14, 2025December 16, 2025 by bahajabarin Tags: agentic AI, AI Agents, Automated Workflow Orchestration (AWO), Retail Leave a comment

Retail Is Entering Its Agentic AI Era

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	• Predict slotting needs • Detect discrepancies • Optimize workforce allocation
Available Inventory	• On-hand quantity • Committed vs Free inventory • Safety stock levels	• Dynamic Available to Pick (ATP) calc • Reallocation suggestions • Overstock / stockout alerts
Allocation	• Demand forecasts • Store sales velocity • Promotion calendar	• Optimize first allocation • Recommend flow-through allocation
Replenishment	• Sell-through data • Min/max thresholds • Lead times	• Auto-trigger replenishment • Predict out-of-stock risk • Dynamic reorder points
Store Inventory	• Store on-hand inventory • Returns & damages • Shelf vs backroom split	• Optimize replenishment routing • Detect phantom inventory
Customer Order	• SKU ordered • Delivery preference • Fulfillment location	• Predict best node to fulfill (e.g., ship-from-store vs DC) • Reduce split shipments
Fulfillment / Distribution	• Pick time • Delivery time • On-time % • Exception logs	• Route optimization • Predict fulfillment delays • Auto rerouting
Reorder Loop	• Real-time sales data • Inventory velocity • Reorder frequency	• Adaptive reorder intervals • Prevent overstock / stockouts

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.