Retail data leaders in today’s high-velocity retail landscape face a paradox: more data, yet more complexity. As retailers race to meet rising consumer expectations while navigating fragmented systems and evolving technologies, the challenge isn’t just capturing insights — it’s making them actionable, scalable, and trustworthy.

Kristin Foster, SVP of Data Science and AI at 84.51˚  (data science arm of American retail leader Kroger) sits at the heart of that transformation. At the core of Foster’s leadership is a focus on infrastructure that turns raw data into real value. Whether it’s eliminating data siloes, operationalizing AI models, or applying GenAI to improve everything from HR workflows to product search relevance, her approach is rooted in aligning technology with real-world outcomes.

While many AI initiatives get stuck in perpetual pilot mode, Foster’s team relies on its “AI Factory” to accelerate the path to production — a robust system of governance, reusable tooling, and cross-functional collaboration.

In this interview, she breaks down the pain points data leaders are grappling with today — from scaling AI responsibly to choosing between real-time and batch pipelines. She also reveals how 84.51˚  is exploring agentic AI, tapping into GenAI and small language models for business acceleration, and designing retail experiences that make life easier for customers and associates alike.

Edited Excerpts 

Q: What are some of the biggest pain points retail data leaders face today? How have the challenges evolved?

Retail data leaders today are navigating a landscape that’s evolving faster than ever across technology, operations, and culture. These rapid changes require a new level of coordination, speed and strategic thinking. Some of the top pain points data leaders continue to have are:

• Data fragmentation: Many retailers continue to have data trapped in siloes, making it difficult to create holistic insights or connected experiences. We’re focused on unifying and building a centralized data strategy that enables streamlined access across siloes and supports smarter decisions.
• Data for AI: AI and analytics need high-quality, governed, and accessible data — not just for innovation, but for daily operations. We’re helping teams adopt the standards, tools and technologies needed for this.
• Scaled AI: A lot of AI opportunities still get stuck in POCs and not built for scale. These siloed efforts can lead to duplicated work, inconsistent standards and missed opportunities to drive real impact. Our AI Factory platform model is intended to help close that gap. It’s our enterprise approach to make AI cheaper, faster, smarter and better through shared infrastructure, reusable components and accelerators, and governance for teams to make an impact faster.

Q: How do you operationalize data science in products — i.e., going from a prototype in Python or R to a production-ready application used by clients or Kroger internally?

It is both a strategic and technical journey, starting with identifying the problem and scaling only what delivers measurable value.

It starts with identifying high-impact opportunities through quarterly innovation sessions with scientists, engineers, and business stakeholders. We use a funnel approach to quickly test and iterate ideas based on feasibility, business value and build/buy decisions.

Once a concept is validated, we form a cross-functional team — including product, engineering, science and business — to carry it forward. Key success factors include ROI clarity, strong collaboration, and infrastructure readiness.

Technically, most data scientists aren’t production engineers, so we support them with internal platforms that simplify deployment. These include CI/CD tools, serverless compute, curated data products, and MLOps features like model registries and monitoring.

Our Kroger AI Gateway provides accelerators and reusable patterns so teams can easily integrate approved models into enterprise systems with governance and observability baked in. This entire framework helps us scale responsibly and effectively.

Q: How do your products ensure freshness and accuracy of insights? Are you working with real-time or batch data pipelines?

Well, we start with the problem before the pipeline. Ensuring freshness and accuracy of insights means first understanding what we are solving for and then the data, platform, and science needed to solve it. Sometimes that calls for real-time data because it’s tied to a real-time operational decision, while batch data feeds are a better fit for other times.

It’s not all or nothing – which is why critical thinking and problem-framing are among the most important skills for our technologists. For instance, Dynamic Batching is a real-time solution that helps our associates pick online orders more quickly and efficiently. It processes new orders in seconds, optimizing routes based on store layout and item location saving our associates time and enabling faster pick-up time for our customers. That requires real-time data feed.

On the flip side, our assortment sciences do not require real-time operations. It helps us determine the right mix of products for each store by analyzing customer needs within that store – location specific preferences, seasonal trends, new product innovation, and emerging market dynamics. In this case, real-time data is less critical than understanding broader patterns and customer needs. 

Our ability to deliver accurate, timely insights depends on more than just data pipelines, but also to a deep understanding of business context. That’s how we ensure our science is solving the right problems, in the right ways, for the people we serve. 

Q: How has GenAI helped the company improve its product and service offerings? Are you exploring large language models (LLMs) or multimodal AI?

We’re being intentional about how we apply GenAI across the business to help our associates work smarter and always keeping responsible AI and security at the core. We’re using GenAI to help our teams better organize and clean up product information, making it easier to surface the right details for both customers and internal teams.

Those details range from improving product descriptions and grouping items correctly to ensuring seasonal or local relevance, which is especially important as we respond to local events or regional preferences.

We’re also focused on making information more accessible for our associates. Proprietary LLMs and other models help us build tools that let teams query complex documentation, policies or historical insights without needing to know exactly where or how to look. It’s about reducing the friction to get answers, so teams can move faster and make more informed decisions. A good example of this is an HR assistant that helps store managers and leaders execute more streamlined admin tasks for efficiency, freeing up their time to provide support to customers in store.

In addition to LLMs, we’re investing in small language models (SLMs) that are tuned to Kroger-specific language and tasks. These are especially useful for things like in-store tools, product classification or automating high-volume, repeatable work — and they allow us to scale more cost-effectively, while keeping performance aligned to our unique business context

While we’re excited about what GenAI can do, we’re taking a measured approach – making sure every use case goes through our responsible AI process, with clear privacy, security and governance controls.

Q: What are your thoughts on agentic AI? Have you developed any use cases at 84.51˚?

We’re still early, but agentic AI has strong potential for scaling intelligence across the enterprise. We are actively exploring its applications for analytics, innovation and productivity. In analytics, agents can automate time-consuming tasks like data sourcing, cleaning, applying statistical methods, and drafting initial insights. That helps speed up the insight cycle.

We’re also building Agent Barney, an agentic framework for Kroger Manufacturing. It incorporates market trends and customer insights to identify new product opportunities and streamline development.

All agents are built within our AI Factory framework — with existing governance, delivery and AI Gateway integration — so they’re scalable, transparent and secure.

Q: How does 84.51˚ ensure that consumer data is used ethically, especially as AI capabilities become more powerful? Are there any guiding principles you follow to build trustworthy AI in a retail context?   

We don’t want to use AI just to use AI – we want to use it when it will make a meaningful and positive difference. And in doing that we take Responsible AI (RAI) very seriously. RAI is not a checkbox, instead it is something that should be embedded into how we build, deploy and scale AI.

It’s part of our values and process. We developed RAI core principles around transparency, accountability, compliance, security, privacy, safety and reliability, for example, as well as a “how to AI” process that provides teams clarity into the steps that need to be followed.

We have an AI Governance council – comprising technology, legal, privacy, security, HR and business partners – who help govern our AI solutions – both built and purchased from third parties. The Kroger Privacy Office plays a critical role in this council, promoting privacy and compliance from the start.

We are committed to transparency and openness. We disclose the use of AI and make sure users are aware of the risks. Building and maintaining customer and associate trust is critical, and trust is earned through transparent and responsible practices.

RAI is not a one-time effort – it is a continuous commitment. It must be prioritized as within our systems, processes, upskilling, values, and culture.

Q: With access to Kroger’s vast retail data, what are some of the most exciting insights or trends you’re seeing in consumer behavior?

What excites me most isn’t just the data, but how it can be used to make everyday life easier for our customers. We all want that experience to be faster, simpler and more relevant. So, when we look at consumer behavior, we’re asking: How can we help people save time, save money, and reduce stress?

When I am grocery shopping, I want help planning meals, finding the right products at the best value, and getting groceries when and how I want. Usually delivered right to my front door! I want fewer steps, fewer decisions, and more inspiration. That’s the lens I think about when approaching how to leverage our data: How do we make shopping feel less like a chore and more like a seamless service?

So, while the data helps us understand patterns and preferences, my focus is always on turning those insights into meaningful improvements — like relevant coupons, smarter search, better product availability, and more intuitive eCommerce tools. Because at the end of the day, it’s not about the data — it’s about making life a little easier for busy people.

Q: You’ve led AI initiatives across industries. How has your leadership style evolved in your role at 84.51˚?

My core values (empathy, curiosity, accountability) have always grounded me. But leading in retail data science has taught me a few new things.

AI moves fast. I’ve had to get comfortable with failing fast, iterating quickly and creating space for continuous learning. It’s about progress over perfection.

I’ve also leaned into curiosity and critical thinking. The best talent today solves ambiguous, messy problems. I look for people who are adaptive learners, strong collaborators, and great question-askers — and I try to model that myself.

Q: Where do you see the biggest opportunities for AI and data science in retail over the next 3–5 years?

A few that excite me. First, truly operationalizing AI — embedding it into daily processes, not just using it for experimentation. AI should be part of how we work and live.

Second, moving from point solutions to connected intelligence across the retail ecosystem. AI should power the full journey — planning, purchasing, fulfillment.

Third, personal AI-powered agents. I see a near future where an AI helps me meal-plan, build baskets, find deals, and auto-reorder staples. Shopping becomes proactive, relevant, and time-saving.

As this happens, transparency and trust become more important than ever. Consumers need to know when AI is used, how data is handled, and that systems are safe and fair. That’s why Responsible AI must be built in from the start.