Amanpal Dhupar, Head of Retail at Tredence – Interview Series

Amanpal Dhupar, Head of Retail at Tredence is an experienced retail analytics and AI leader with over a decade of expertise in designing and developing data-driven solutions that deliver actionable insights for enterprise decision-makers. Throughout his career, he has led strategic analytics transformations for senior executives across major retailers, built AI product roadmaps to drive measurable business KPIs, and scaled analytics teams from infancy to large-scale operations—demonstrating both technical depth and leadership versatility.

Tredence is a data science and AI solutions firm focused on helping enterprises unlock business value through advanced analytics, machine learning, and AI-powered decision-making. The company partners with global brands—especially in retail and consumer goods—to solve complex challenges across merchandising, supply chain, pricing, customer experience, and go-to-market operations, translating insights into real-world impact and helping clients modernize their analytics and intelligence capabilities.

Retailers often run dozens of AI pilots, yet very few move into full-scale deployment. What are the most common organizational mistakes that prevent AI from turning into measurable business results?

A recent MIT Solan study found that 95% of AI pilots fail to achieve full-scale deployment. The reality? Pilots are easy, but production is hard. At Tredence, we have identified four specific organizational reasons driving this gap.

First is the failure to understand the end user workflow. Retailers often insert AI into existing broken processes rather than asking how the workflow itself should be reimagined with AI at the center.

Second is the lack of a platform approach to Agentic AI. Instead of treating agents as one-off experiments, organizations need to streamline the entire lifecycle—from agent design and development to deployment, monitoring, and governance—across the enterprise.

Third is a weak data foundation. It is easy to build a pilot on a clean flat file, but scaling requires a robust, real-time foundation where accurate data is continuously accessible to AI models.

Finally, we see a friction between IT push vs. business pull. Success only happens when business leaders see AI as a value-add tied to measurable impact, rather than a distraction pushed by IT. At Tredence, our focus has always been on the ‘last mile,’ where we bridge this gap between insight generation and value realization.

Tredence works with many of the world’s largest retailers, supporting trillions in revenue. Based on what you’re seeing across the industry, what separates the retailers who scale AI successfully from those who remain stuck in experimentation?

At Tredence, supporting trillions in retail revenue has given us a front-row seat to a clear industry divide: retailers that treat AI as a series of disparate experiments versus those that build an industrialized ‘AI factory.’ The primary differentiator lies in a commitment to Agentic AI Platform foundations. The most successful organizations stop building from scratch and instead invest in a robust ecosystem characterized by reusable component libraries, standard design templates, and pre-built agent patterns aligned to specific retail use cases. When you layer mature LLMOps, full-stack observability, and embedded responsible AI (RAI) guardrails on top of this foundation, the impact is transformative—we typically see speed-to-value for new use cases improve by 80% because the heavy architectural lifting is already done.

However, a platform is only as good as the context it consumes, which brings us to the data foundation. Scaling requires more than just raw access to data; it demands a rich semantic layer where strong metadata and unified data models allow the AI to actually ‘reason’ about the business rather than just process inputs. Finally, the true leaders recognize that this is not just a technology overhaul but a cultural one. They bridge the ‘last mile’ by moving beyond simple automation to human-agent teaming, re-engineering workflows so that associates and merchants trust and collaborate with their digital counterparts, turning algorithmic potential into measurable business reality.

More than 70 percent of retail promotions still fail to break even. How can AI meaningfully improve promotion planning, measurement, and real-time optimization?

The 70% failure rate persists because retailers often rely on ‘rear-view mirror’ analytics that confuse total sales with incremental lift—essentially subsidizing faithful shoppers who would have purchased anyway. To break this cycle, we need to shift from descriptive reporting to a more predictive approach. In the planning phase, we use Causal AI to simulate outcomes and establish ‘true baselines,’ identifying exactly what would have sold without the promotion. This allows retailers to stop paying for organic demand and target only net-new volume.

For measurement, AI solves the ‘portfolio puzzle’ by quantifying halo effects and cannibalization. Human merchants often plan in silos, but AI provides a category-wide view, ensuring that a promotion on one SKU isn’t just stealing margin from another. This holistic measurement helps retailers understand if they are growing the category pie or just slicing it differently.

Finally, for real-time optimization, the industry is moving toward AI Agents that monitor campaigns ‘in-flight.’ Instead of waiting for a post-mortem analysis weeks after the event, these agents autonomously recommend course corrections—like adjusting digital ad spend or swapping offers—to rescue the P&L before the promotion ends. This approach shifts the focus from simply clearing inventory to engineering profitable growth.

Forecasting errors and out-of-stocks continue to cause major revenue losses. What makes AI-driven merchandising and supply chain systems more effective than traditional forecasting approaches?

The first shift is in forecasting, where AI moves us from relying solely on internal history to ingesting external data—like local weather, social events, and economic indicators. When the forecast captures this outside context, the accuracy gains don’t just improve sales number; they cascade downstream, optimizing inventory management, capacity planning, labor schedules, and warehouse operations to align with true demand.

The second shift is in Out-of-Stocks (OOS), which most retailers still fail to measure accurately. AI fixes this by detecting anomalies in sales patterns—identifying ‘Phantom Inventory’ where the system thinks an item is in stock, but sales have stopped—and automatically triggering cycle counts to correct the record. Beyond the data, we are seeing the rise of computer vision to physically flag shelf gaps in real-time and track inventory in backrooms, ensuring the product is not just ‘in the building’ but available for the customer to buy.

Agentic commerce is becoming a major theme in retail innovation. How do reasoning-based AI agents meaningfully change product discovery and conversion compared to today’s search-driven shopping experience?

In today’s search-driven shopping, consumers still do most of the heavy lifting. They have to know what to look for, compare options, and make sense of endless results. Reasoning-based agents disrupt this by dynamically generating ‘synthetic aisles’—custom collections that aggregate multi-category products based on a specific intent. For example, instead of searching separately for five items, a shopper with a ‘healthy morning’ mission is presented with a cohesive, temporary aisle featuring everything from high-protein cereal to blenders, instantly collapsing the discovery funnel from minutes to seconds.

On the conversion side, these agents act less like search engines and more like ‘shopping concierges.’ They don’t just list options; they actively build baskets based on open-ended needs. If a customer asks for a ‘dinner plan for four under $50,’ the agent reasons through inventory, price, and dietary constraints to suggest a complete bundle. This reasoning capability closes the ‘confidence gap’—by articulating why a specific product fits the user’s lifestyle or goal, the agent reduces decision paralysis and drives higher conversion rates compared to a silent grid of product thumbnails.

Finally, we are seeing this extend into hyper-personalized content. Rather than showing everyone the same homepage banner, Agentic AI can generate dynamic landing pages and visuals that mirror the customer’s current shopping mission. However, for this to scale, retailers are finding they need to ground these agents in a Unified Data Model with strict brand and safety governance, ensuring that the AI’s ‘creativity’ never hallucinates products or violates brand voice.

Many retailers struggle with outdated data architectures. How should enterprises modernize their data foundations so AI models can deliver trustworthy and explainable recommendations?

The biggest barrier to AI success is not the models but the ‘data swamp’ beneath them. To modernize, retailers must stop simply collecting data to building a unified semantic layer. This means implementing a standard ‘Data Model’ where business logic (like exactly how ‘Net Margin’ or ‘churn’ is calculated) is defined once and is universally accessible, rather than being hidden in fragmented SQL scripts across the organization.

Second, enterprises need to move to a ‘data product’ mindset. Instead of treating data as an IT byproduct, successful retailers treat it as a product with defined ownership, SLAs, and rigorous quality monitoring (data observability). When you combine this clean, governed ‘golden record’ with rich metadata, you unlock explainability. The AI doesn’t just spit out a black-box recommendation; it can trace its logic back through the semantic layer.

Collaboration between retailers and CPG companies has historically relied on fragmented data and inconsistent metrics. How do unified data models and shared AI platforms unlock stronger category performance for both sides?

So far, retailers and CPGs have looked at the same customer through different lenses, each using their own data and incentives. Unified data models change this by creating a single version of truth across the value chain, be it shelf performance or shopper behavior.

When both sides work off the same AI platform, they can jointly identify what’s driving growth or leakage at a category level. It could be anything- pricing, promotion, assortment, or inventory gaps. This shifts conversations from “my data vs. yours” to “our shared opportunity.”

The result is smarter decisions, faster experimentation, and ultimately, higher category growth that benefits both retailers and brands.

As retail media networks mature, what role will AI play in improving targeting, measurement, and closed-loop attribution while maintaining consumer trust?

AI will transform four key areas as retail media networks mature.

First, in targeting, the industry is evolving from static audience segments to predictive intent. By analyzing real-time signals—like browsing velocity or basket composition—to identify the precise moment of a shopper’s need, AIe ensures we show the right ads when it matters most rather than just targeting a broad demographic label.

Second, for measurement, the gold standard is shifting from simple Return on Ad Spend (ROAS) to incremental ROAS (iROAS). By leveraging Causal AI, we can measure the real impact of the media spend by identifying shoppers who only converted because of the ad versus those who would have that happened organically.

Third, operational efficiency is becoming critical, particularly in creative operations. To support hyper-personalization, retailers are using Generative AI not just for ideation but to scale production. This allows teams to automatically generate thousands of dynamic, channel-specific asset variations in minutes rather than weeks, solving the bottleneck of ‘content velocity’.

Finally, maintaining trust relies on the widespread adoption of data clean rooms. These environments allow retailers and brands to securely match their datasets for closed-loop attribution guaranteeing that sensitive Personally Identifiable Information (PII) ever leaving their respective firewalls.

Looking ahead, what capabilities will define the next generation of AI-powered retailers, and what should leaders start building today to stay competitive over the next five years?

The next era of retail will be defined by the shift from ‘digital transformation’ to ‘agentic transformation.’ We are moving to a future of ‘autonomous orchestration,’ where networks of AI agents collaborate to run complex processes—like a supply chain agent automatically telling a marketing agent to pause a promotion because a shipment is delayed.

To prepare for this, leaders must start building three things today.

First is a unified data model. Agents cannot collaborate if they don’t speak the same language; your data foundation must evolve from a storage repository to a semantic ‘nervous system.’

Second is a governance framework for agencts. You need to define the ‘rules of engagement’—what an AI is allowed to do autonomously versus what requires human approval—before you scale.

Finally, the days of static dashboards providing ‘rear-view’ analytics are numbered. We are moving toward conversational analytics that provide instant, personalized insights. These interfaces go far beyond reporting ‘what happened’; they leverage agentic AI to reason through complex ‘why’ questions and deliver prescriptive recommendations on exactly ‘what to do next,’ effectively closing the gap between insight and action.

Thank you for the great interview, readers who wish to learn more should visit Tredence.