AI Pricing & Demand Forecasting Today: Insights from a Decade in Applied AI

Episode Summary

In this episode of Pricing Heroes, we are joined by Jayadeep Shitole, previously Principal Data Scientist of Applied AI and Machine Learning at Walmart Global Tech. With over a decade of experience in retail pricing and demand forecasting, Jayadeep has built AI systems for some of the world’s largest retailers, including Walmart and Petco. He shares how AI is transforming retail—from dynamic pricing and predictive demand modeling to real-time decision support and end-to-end strategy.

We explore how AI-based pricing works in practice, what’s changed in recent years, and how to build trust and transparency into next-generation retail pricing systems.

From math to merchandising: A career in applied AI

Jayadeep’s entry into the world of pricing didn’t begin with retail, but with mathematics. After completing degrees in applied math and operations research, he started his career building machine learning models across sectors like finance and automotive. His early work included designing trading algorithms for hedge funds—systems built to spot tiny inefficiencies in fast-moving markets.

But he soon realized he wanted to work on problems that impacted people more directly. “I wanted more impact on everyday life,” he explains. That led him into retail, where he joined a major pet retailer’s newly formed shopping intelligence team. “That’s where I realized how even small, tiny changes in pricing or demand forecasting can have a dramatic effect on both profitability and customer satisfaction.” Since then, he’s spent four years at Walmart Global Tech, leading initiatives in AI pricing, demand forecasting, and decision intelligence—helping shape the future of scalable, automated, but still human-aware pricing systems.

From rules to reinforcement: The evolution of AI pricing technologies

Over the past five years, Jayadeep has seen retail pricing undergo a fundamental transformation. What was once a rigid, rules-based function has evolved into a dynamic, data-driven system that can respond in real time to shifting conditions. But these changes haven’t just been technological—they’ve reshaped how organizations think about pricing as a strategic lever.

He outlines three major shifts:

1. From static to dynamic – “We’ve moved away from simple rule-based pricing, like cost-plus or competitor matching, to far more dynamic data-driven algorithms,” Jayadeep explains. In modern systems, prices can update multiple times a day based on competitor behavior, inventory availability, and local demand. “Especially in e-commerce, this level of granularity is essential. You’re not just setting one price—you’re constantly recalibrating.”
2. Wider accessibility – Thanks to cloud computing and open-source tooling, AI-powered pricing is no longer the domain of tech giants alone. “Even mid-market retailers can now adopt machine learning for pricing,” he says. “It’s no longer just a luxury that’s reserved for some of the largest companies in the world.”
3. Explainability and trust – With the rise of AI came a new concern: can teams trust the system? Jayadeep believes this is one of the most significant evolutions. “Companies want to understand why algorithms recommend a certain price. They need to ensure that pricing models reflect not just profit goals, but also brand image, customer loyalty, and ethical boundaries.” This has led to an explosion of interest in interpretable AI, human-in-the-loop systems, and governance frameworks designed to ensure alignment between models and mission.

Together, these changes signal a new era—one in which pricing is fast, flexible, and increasingly central to the retail value chain. 

Bridging the adoption gap: Why AI pricing still faces roadblocks

Despite the momentum behind AI pricing, Jayadeep is candid about the barriers that still slow adoption. First and foremost: data. “Probably the most costly aspect of adopting AI pricing isn’t the actual tools themselves,” he notes. “It’s organizing the data.” Many retailers continue to struggle with fragmented infrastructure, siloed teams, and inconsistent definitions across systems. Without a strong data pipeline, even the most advanced models will fall short.

But technical readiness is only part of the story. Organizational alignment is just as critical. “Going from manual, rules-based methods to fully automated AI is a huge jump,” he says. “Not everybody is comfortable trusting an algorithm with something as important as pricing.” Trust isn’t built overnight—it requires visibility, education, and collaborative design processes that include the commercial teams who use the insights every day.

And then there’s the business case. While the benefits of AI pricing can be enormous, the upfront investment in talent, tooling, and integration can be a hard sell—especially if leadership hasn’t seen clear ROI examples. Jayadeep recommends starting small. “Pilot a narrow product category, refine your model, and build confidence through results. Once your team understands the system and sees it working, you can expand.”

A practical framework: Four steps for implementing AI-powered pricing

For retailers ready to modernize their pricing capabilities, Jayadeep offers a clear roadmap—an end-to-end pipeline that transforms raw data into optimized price recommendations aligned with business goals. Each stage is an opportunity to build trust, embed domain knowledge, and refine model accuracy.

• Gather data – “Start with what you have,” Jayadeep advises. That means collecting historical sales data, website traffic, competitor prices, external factors like weather or holidays, and anything else that could influence demand. “Don’t underestimate the complexity of this step. You’re trying to build a data foundation that can support real-time decisions.”
• Engineer features – Once data is collected, the next step is to transform it into meaningful signals. “This is where the art meets the science,” he explains. “You’re translating messy real-world inputs into something a model can learn from. And you need to work with business partners here—they know what moves the needle.”
• Train models – With feature sets defined, teams can now train demand prediction models. “We often use XGBoost or deep neural networks,” Jayadeep says. “The goal is to predict how demand responds to different prices—so you’re not just looking at historical averages, but at how elastic demand is under different conditions.”
• Optimize and deploy – Finally, optimization layers take these predictions and generate price recommendations that align with the company’s specific goals—whether that’s profitability, waste reduction, price perception, or a mix. “If you’re selling fresh food, maybe your goal is to reduce spoilage. If you’re managing a high-end brand, it could be preserving price integrity. The model has to reflect your strategy.”

Crucially, Jayadeep emphasizes that this isn’t a linear process. “You’ll constantly loop back—refining features, adjusting models, improving constraints. And you’ll get better over time.”

Pricing and forecasting: A single system with two engines

While pricing and forecasting are often treated as separate workflows, Jayadeep argues they should be deeply interconnected. “The demand of your product depends on the price that you're offering,” he says. “So when you build AI-based forecasting, you're also learning what levers you can pull—and how pricing decisions shift demand.”

Traditional time series models struggled to account for external shocks or behavioral nuance. AI forecasting tools, by contrast, can incorporate dozens of variables: promotions, seasonality, social sentiment, competitor actions, and more. These systems not only adapt faster to market shifts—they provide a richer foundation for dynamic pricing strategies.

Jayadeep offers the example of markdowns in fresh grocery. “If you’re not selling through fast enough, you risk spoilage and waste. But if you lower prices in a smart, targeted way, you can reduce loss, improve turnover, and align with sustainability goals—all at once.” AI allows retailers to embed these competing objectives into a single optimization framework.

From tariffs to DeepSeek: The future of retail decision intelligence

Looking ahead, Jayadeep sees AI moving beyond narrow optimization to broader decision support. Tools that once answered “what price should I set today?” will increasingly be able to simulate complex tradeoffs involving inventory, supply chain constraints, and even macroeconomic policy.

Take tariffs. As Jayadeep explains, “Trade policies can shift overnight—and that has a ripple effect on costs, supply chains, and consumer behavior.” AI models can integrate signals like currency fluctuations, supplier lead times, and tariff rates to help retailers simulate scenarios: Should we raise prices? Absorb costs? Pre-purchase inventory? “You can run what-if simulations across thousands of SKUs in seconds.”

He’s also watching the rise of reasoning models like DeepSeek—next-gen AI systems capable of ingesting unstructured data and generating strategic insights. “These models won’t just optimize—they’ll support planning,” he says. “Imagine combining customer analytics, supply chain metrics, and macro trends into one pricing decision engine. That’s where we’re heading.”

Still, he stresses the role of human oversight. “AI can handle the predictable parts of your business. Humans should focus on the unpredictable—and on setting the goals that matter.”

Recommended resources

For pricing professionals and data leaders eager to keep learning, Jayadeep suggests following engineering blogs from companies leading the way in applied machine learning. “DoorDash, Uber, Netflix, LinkedIn—these teams share what it really takes to build, scale, and manage ML systems in production,” he says. “It’s not just theory. It’s practice.”