In the retail industry, margin pressures are structural. Companies that are moving forward are making faster, more accurate decisions across merchandising, workforce, and supply chains, and doing so consistently across thousands of locations. Problems faced by most large retailers: Is their organization built to scale AI quickly enough to matter? Albertsons Companies is one of the largest food and drug retailers in the United States, operating approximately 2,300 stores and generating revenues of $80 billion. Sunil Gopinath leads the company’s data and AI globally and also runs Albertsons Companies India, the company’s largest technology and AI hub. His mission is to build the foundation of AI and data to transform leading retailers into data-driven companies at speed and scale.
The beliefs that pervaded our conversation were direct. Stop tolerating fragmentation. Companies that combine their AI ambitions with strong corporate fundamentals will win. Everyone else is doing expensive experiments.
Underpinning this strategy is the Databricks platform, which Albertsons uses across data engineering, ML, governance, and analytics. This shared foundation makes the “one platform” mandate a reality, giving all teams the same starting line rather than a different set of tools.
Building AI muscle: Why centralization is non-negotiable
Ally McGu: How did you move your organization from fragmented AI experiments owned by business units to a centralized AI core team and operating model?
Sunil Gopinath: We stopped tolerating fragmentation and made solid architectural decisions. One team, one platform, one operating model. We’ve organized around four big bets in AI: customer experience, merchandising intelligence, workforce, and supply chain. These have allowed us to focus strategically. Execution power is enhanced with a centralized AI core.
The logic was simple and clear. There was a clear organizational need for common horizontal components such as governance, security, and a central repository for reusable models. With a dedicated team focused on these building blocks, application teams no longer have to worry about hygiene or infrastructure. They can fully focus on making their business better, more predictable, and more viable.
We also have a company-wide governance committee that brings together senior stakeholders and leaders to establish common and acceptable standards for AI and AI governance. It’s collective decision-making at the leadership level. That’s why it persists.
Franchise model for AI at scale
Allie: What was the strategy for building shared standards, central platforms, and reusable accelerators to drive efficiencies across Albertsons while considering local innovation and use cases?
Sunil: The best way to think about it is as a franchise model. Common infrastructure, standards and governance are at the heart. Execute locally and innovate at the edge.
We built reusable accelerators: ingestion pipelines and templates. Feature store pattern. Model monitoring. Performance observability. and governance wrapper. Any team can use these to work 10x faster. The key to this platform is that it doesn’t limit innovation. That accelerates.
Our philosophy is that we must balance innovation with trust and governance from both our employees and customers. Therefore, the criteria are not arbitrary. These reflect what businesses, merchants, and customers need to actually trust AI to work.
Talents that combine in changing situations
Allie: How are you rethinking the skills and leadership required to run this central AI core, and how are you ensuring your platform effectively empowers non-technical teams?
Sunil: Our approach works in three layers: machine learning to predict, genAI to respond, and agent AI to act. All of this is built into the way employees work.
As for our technical team, we have moved to AI augmented engineering. In 9 months, we accepted 1.38 million lines of AI-generated code and over 90% of our engineers used our AI tools. We’ve fundamentally changed the speed at which we build and ship, and now we’re even bigger.
We built low-code dashboards, prompt libraries, and conversational agent generation for non-technical teams. We have a proprietary agent AI platform that allows non-technical teams to drag and drop agents. If you don’t feel comfortable doing that, just have the conversation, “Build an agent to monitor these KPIs,” and it will happen. The goal for both is the same. Spend less time searching for answers and more time making decisions.
Especially when it comes to talent questions, we’re not just looking for technical ability or familiarity with the latest AI tools. We value and hire for an attitude of learning, experimenting, and innovating. Tools will continue to evolve at a record pace. But if those cultural traits are ingrained, people will pick them up and run with them.
top discipline
Allie: Who on the executive team is ultimately responsible for the success of the enterprise AI core, and how have the KPIs changed?
Sunil: Ownership is at the top. For us, AI is a business strategy. Our metrics reflect market-wide reuse rates, time to adoption, responsible AI compliance, and most importantly, business outcomes related to improving AI. If an initiative doesn’t work, it won’t scale. This discipline needs to be enforced from the top and is what makes AI a real advantage, not just an expensive experiment.
lastly
Sunil does not explain the gradual evolution towards centralization. He describes an intentional effort with one team, one platform, and one operating model: strategic bets to focus work and reusable accelerators to increase velocity.
Merchandising intelligence is one of four strategic priorities for AI, a big bet Albertsons is making as part of a broader company-wide transformation, and what a centralized model looks like when faced with real-world business problems. The platform is built on Databricks and includes Genie in the interaction layer. Sellers can ask complex questions in plain language and receive disciplined and trusted answers without having to write a query or submit a ticket. Databricks provides data engineering, ML, and analytics foundations underneath.
For executives struggling with how to move AI from the realm of experimentation to enterprise capabilities, Albertsons’ franchise model provides a useful framework for governing the center, liberating the edge, and allowing all teams to build on what’s already proven.
Download Databricks State of AI Agents to benchmark your investments and create a roadmap for embedding AI across your organization and products.
