Skip to main content
Coforge
✕

The Invisible ‘Eye’ that is Redefining Retail : The Ghost in the Aisle

article banner

Abstract

Retail is on the brink of its most disruptive transformation since the barcode rewired global commerce. According to Gartner, more than half of large retailers are expected to embed AI-driven decision intelligence into core operations within the next few years, while IDC projects that retail AI investments will accelerate at double-digit annual growth rates. Yet the stakes are higher than innovation alone: in an industry where net margins hover in the low single digits, even a 1% failure in pricing precision, inventory alignment, or demand sensing can quietly destroy hundreds of millions in enterprise value. The next era of retail will not be defined by better dashboards, but by invisible systems that observe, reason, and act autonomously.

This paper examines the rise of the “Ghost in the Aisle”, the convergence of computer vision, generative AI, and agentic systems that together form an invisible operating layer across stores and supply chains. It explores how retail is shifting from predictive analytics to Agentic AI, systems that do not merely identify or forecast issues but autonomously resolve them in real time. Through the lens of industry leaders such as Walmart, Amazon, and Tesco, it analyzes how data is being transformed into a strategic advantage at an unprecedented scale. At the same time, it confronts the growing tension between technological acceleration and regulatory scrutiny, examining how retailers can architect intelligence without compromising privacy or consumer trust amid pervasive observation.

Introduction

When barcodes were introduced in the 1970s, they quietly rewired the economics of retail. What began as a faster checkout mechanism fundamentally transformed pricing agility, inventory visibility, and supply chain coordination. Prices moved from static stickers on products to dynamic signals on shelves. Data became structured. Decisions became measurable. Retail evolved from manual commerce into information-driven commerce.

Today, a far more profound shift is unfolding. Artificial Intelligence is giving rise to what can be described as the “Perception Store”, a physical and digital environment that observes, interprets, and responds to customer behavior in real time. Powered by computer vision as its “eye,” generative AI as its “brain,” and agentic systems as its autonomous “hand,” this invisible layer, the “Ghost in the Aisle”, transforms stores from passive spaces into intelligent ecosystems. Retail is no longer simply recording transactions; it is perceiving intent, simulating outcomes, and acting autonomously at the edge of every aisle.

Part I: The Landscape  

1.1 Experimentation to Necessity

In the past, AI in retail was mostly pilot programs like chatbots or demand algorithms. But recent data shows that the use of AI in retail has moved from an “innovation lab” approach to becoming the default “operating system”.

Private investment in AI exceeded $130 billion in 2024. More than 80% of this investment has been in the supply chain systems, while only 20% has been in customer-facing technologies. This is reflected in how these markets are projected to grow, with supply chain AI tech growing by more than twice the rate of customer-facing AI in retail.

Data reveals that while customer-facing AI (chatbots) receives the marketing buzz, the backend optimisation is where the capital is flowing. Retailers have recognised that in a low-margin environment, the ability to shave basis points off logistics costs via autonomous decision-making is more valuable than a better product recommendation engine.

1.2 Three Keys of Transformation

The technological architecture driving this value can be categorised into three key areas.

Generative AI (GenAI):

GenAI is no longer just about writing marketing copy. It has moved into fields like “intent understanding”. Retailers are using GenAI to power semantic search algorithms that let customers shop by problem statement (e.g., "how to fix a light bulb") rather than searching for a product like “light bulb wiring”. Some reports say that semantic search technology has driven an 830% increase in traffic to retailer sites compared to 2024.

Computer Vision:

Computer vision digitises physical reality. For example, consider a customer who walks into a store and spends a few minutes in an aisle. Computer vision systems can transform the video feed into structured data - like “Entry at 10:02 AM, dwell time at t-shirts section 15 seconds, product interaction positive”.

Agentic AI:

Unlike passive AI models that wait for a prompt, AI Agents possess agency. They perceive their environment, set sub-goals, and execute actions to achieve a high-level objective. For example, an inventory agent might notice a predicted snowstorm, autonomously decide to increase stock of shovels, negotiate a rush shipping rate with a logistics provider, and update the pricing on electronic shelf labels - all without human intervention.

Part II: The Agentic Supply Chain

Covid saw global supply chains in shock. The fragility of “just-in-time” procurement was deeply exposed. The sector is now moving towards "Cognitive Supply Chains". These chains balance efficiency with resilience. The "Ghost" in the warehouse is an autonomous agent that continuously simulates scenarios to prevent disruptions.

The deployment of Agentic AI in supply chain management represents a shift from descriptive analytics (what happened) and predictive analytics (what will happen) to prescriptive and autonomous action (make it happen).

Procurement agents are growing in importance and reach. From autonomous negotiation to autonomous remediation, agents are capable of “chain-of-thought” reasoning to handle procurement. For example, agents can scan for changes in global tariffs, calculate the cost of holding inventory vs paying tariffs and make a decision based on whichever works out cheaper.

This capability extends to daily operations. Microsoft and Blue Yonder have recently demonstrated agents that prevent stockouts by analysing sales spikes in real-time. The agent automatically generates replenishment orders, ensuring optimal stock levels always.

To enable these agents, retailers are building "digital twins which are virtual replicas of their entire supply chain network. From raw material to the store shelf. By integrating data from IoT sensors, procurement portals, and logistics providers into a unified data layer, the digital twin allows the AI to simulate millions of scenarios.

Part III: The "Invisible Eye" – In-Store Analytics and Surveillance

Digitisation of the physical store is the most visible manifestation of the “Ghost in the Aisle”. Retailers are installing sensors across their stores to capture data from customer movement.

3.1 Heat Mapping

Retailers are deploying sophisticated heat mapping technologies to visualise the customer journey.

  • Mechanism: These systems utilize a combination of overhead video analytics and WiFi/Bluetooth signal triangulation. As a shopper moves through the store, their smartphone (and their physical body) leaves a digital trail.
  • Visualisation: The output is a colour-coded map. "Red Zones" indicate areas of high traffic and high dwell time, the "hot" spots where customers linger. "Blue Zones" indicate dead space.
  • Operational Utility: This data allows for scientific store layout optimisation.
    • Placement Strategy: If a high-margin product is sitting in a "blue zone," the data dictates it must be moved.
    • Queue Management: Real-time heatmaps detect congestion at checkout lines. If the "heat" exceeds a threshold, the system alerts managers to open more registers.
  • The Privacy Trade-off: Companies like Placer.ai and FootfallCam utilise anonymised data to track customers. They often hash MAC addresses to decouple the movement from the person’s identity. However, the granularity of the tracking - knowing that a specific person with a MAC address of XYZ lingered in the pharmacy aisle for 20 minutes before moving to the candy aisle, raises significant privacy questions regarding inferential analytics

3.2 Electronic Shelf Labels (ESL) and Dynamic Pricing

The paper price tags are being replaced by Electronic Shelf Labels (ESLs).

  • Scale: The global ESL market is projected to grow at a CAGR of 12% to 16%, reaching $3 billion by 2030. Walmart has committed to rolling out ESLs to 2,300 stores by 2026, a capital investment that signals the end of manual pricing
  • The Controversy: ESLs enable "dynamic pricing" – the ability to change the price of goods in real-time based on demand, time of day, or expiration dates.
    • Retailers argue this enables efficiency by reducing labour costs to change tags and waste reduction by discounting perishable items
    • Consumer/Legislator Fear: Concern about "surge pricing" in grocery stores. Several people have highlighted the risk that ESLs could facilitate price gouging. For example, water could cost more on a sunny day, and umbrellas might become costlier on a rainy day.
  • The Synchronisation of Shelves: The true power of ESLs lies in their connection to the broader AI brain. An AI agent can scrape a competitor's website, detect a price drop, calculate the margin impact, and update the physical price tag in thousands of stores within seconds.

3.3 The Biometric Frontier: Facial Recognition and Privacy

The use of facial recognition technology (FRT) represents the sharpest edge of the surveillance wedge. It is here that the "Ghost" becomes personal.

  • The Rite Aid Precedent: In a landmark enforcement action, the Federal Trade Commission (FTC) banned Rite Aid from using facial recognition for five years. The retailer had deployed an AI system to identify potential shoplifters, but the system was flawed. It disproportionately flagged women and people of colour, leading to the harassment of innocent consumers. The FTC's order required Rite Aid to delete not just the data, but the algorithms themselves - a penalty known as "algorithmic disgorgement."
  • The Shift to Behavioural Analytics: Stung by privacy backlash and lawsuits (such as those under Illinois' BIPA law), retailers are shifting from identity tracking to behavioural tracking. Amazon's updated "Just Walk Out" technology serves as a prime example. Instead of identifying who you are (facial recognition), the system uses computer vision to identify what you do (picking up an item). It analyses the interaction between the hand and the shelf, using multi-modal models to handle complex scenarios like obscured views or "fumbled" returns, without necessarily needing to know the shopper's name.

Part IV: The Giants - Case Studies

To understand the practical application of these technologies, we must examine the strategies of the industry's leaders. Each has adopted a distinct AI persona.

4.1 Walmart

Walmart's AI strategy is defined by scale and unification. They are not building disparate tools; they are building a global operating system.

Walmart is investing in integrating its data sources across multiple markets, including the US, Mexico, Canada, and Chile. This allows their AI models to train on large volumes of data. They are also investing in a self-healing inventory platform that automatically reroutes goods around disruptions.

Walmart is working aggressively on capturing the “search intent” of customers. They developed

“Wallaby,” which is a large language model that specialises in retail. They are also investing in “Sparky”, a GenAI assistant that can fill physical and digital carts by just listening to users speak in natural language.

4.2 Amazon

Amazon’s biggest AI investment is in the JWO (Just Walk Out) 2.0 platform. JWO 1.0, while highly successful, was also very expensive due to complex sensor fusion. Amazon has pivoted to multimodal foundation models. By analysing sensor and camera inputs simultaneously (rather than sequentially as in 1.0), JWO 2.0 achieves higher accuracy at a lower cost.

Amazon’s Rufus assistant is also being developed at pace, integrating LLMs directly into the Amazon mobile app. Rufus is trained on every aspect of every item in Amazon’s catalogue.

Amazon is also pushing “Dash Cart,” a smart cart that uses computer vision to identify items as they are placed into the cart, thereby making the checkout process very seamless.

4.3 Tesco: The Personalised Loyalty Engine

Tesco demonstrates the power of hyper-personalisation driven by long-term data accumulation. Tesco’s AI makes more than 190 million distinct decisions when rewarding and spending loyalty points. Gamification of loyalty points based on customers’ buying patterns not only increases consumption but also improves customer stickiness.

Tesco views its data layers (transactional, loyalty, and sensor data) as a competitive moat. By feeding this data into AI models, they can predict customer needs with high precision, moving beyond generic "coupons" to personalised "offers”.

In Summary

By 2030, the retail environment will be fundamentally unrecognisable from the model that dominated the 20th century. The "Ghost in the Aisle" will no longer be an intruder but the store’s operating system.

By 2030, the retail environment will be fundamentally unrecognisable from the model that dominated the 20th century. The "Ghost in the Aisle" will no longer be an intruder but the store’s operating system.

We can forecast four distinct shifts:

  • Physical stores will be like web browsers: they will have the same level of analytics, tracking, and dynamic adaptability as websites.
  • The End of the Checkout: The concept of "checking out" will dissolve. Payment will become a background process, continuous and invisible.
  • The Agentic Supply Chain: The back-end will run largely on autopilot. Human intervention will be reserved for strategic relationships and crisis management, while AI agents handle the orchestration of global logistics in silence
  • Privacy as a Premium: Privacy-first shopping experiences will be a premium, paid model, while the ghost will follow regular shopping in the aisle.

The invisible eye is watching, the ghost is calculating, and the way we shop is changing forever. The retailer of the future is part merchant, part media company, and part intelligence agency. The success of this transition will depend not just on the code, but on the trust the retailer can maintain with the consumer in an age of total observation.

This transformation demands more than technology selection. It requires deep domain understanding of retail economics, operational complexity, and regulatory realities.

Coforge partners with global retailers to design and implement AI-native retail architectures, from cognitive supply chains and computer vision–enabled stores to GenAI-powered search, pricing intelligence, and autonomous orchestration layers. By combining retail domain expertise, AI engineering capability, and scalable cloud integration, Coforge enables retailers to move from experimentation to enterprise-wide intelligence.

Retailers who architect this shift deliberately will gain margin resilience, operational agility, and a differentiated customer experience.

Those who hesitate risk watching the “Ghost” power their competitors instead.

Sources:

About the Author

Vikrant Karnik
Ram Mamidanna

Ram is Sr. Vice President, Engineering for Coforge. Based in London, Ram leads solutions for Europe and the UK. As a hands-on enterprise architect, Ram works in the intersection of business and technology to help our clients drive their large-scale AI-led transformations.