When adopting new enterprise technologies, companies traditionally used small, controlled pilot projects to test and validate new ideas and innovations. Artificial intelligence follows this familiar trajectory, with companies establishing what amount to AI workshops—carefully crafted but ultimately siloed initiatives.
While these pioneering efforts demonstrate AI’s potential, they also reveal the limitations of treating each AI implementation as an artisanal endeavor. At 84.51°, we discovered that replacing our workshop mentality with an “AI factory” model — where core capabilities, services, and infrastructure are developed for reuse — accelerates development and changes how we extract value from our AI investments.

The initial innovation: Dynamic batching

The journey began with developing a routing and batching optimization system for in-store order fulfillment at Kroger. This system, powered by advanced science capable of making 200,000 decisions per second, optimizes the distance traveled when collecting items for customer orders.

The results were up to a 10% reduction in distance traveled across nearly 2,500 stores, and the ability to fulfill orders in fewer than two hours, even at the busiest locations.

While the business challenge was different, the team recognized an opportunity to reuse the same development process and underlying architecture — accelerating delivery by building on what worked before.

Enter VROOM

With this realization, the team developed VROOM (Vehicle Routing and Order Optimization Model), applying the same approach to a different but similar challenge: Optimizing distribution center-to-store delivery routes.

VROOM uses advanced analytics to evaluate thousands of possible route options in seconds, creating routes that minimize miles driven and costs, as well as streamline transportation. The transition from in-store routing to vehicle routing demonstrated the power of reusable AI capabilities and shared support services:

• While the original dynamic batching solution took six to nine months to develop, the adaptation for vehicle routing was piloted in half the time.
• Actual results significantly outperformed expectations, delivering even greater efficiency in distance traveled (up to 8% reduction in distance traveled).
• The benefits extended beyond efficiency to increased flexibility in the supply chain and carbon footprint reduction.
• Additionally, the underlying capabilities were built in a way that they can be portable to future applications across the enterprise.

The shift to an AI factory mindset

This success story illustrates the shift from an “AI workshop” mentality — where each solution is crafted individually — to an “AI factory” approach focused on building reusable capabilities and building off the success of previous endeavors. Key aspects of this transformation include:

1. Modular design: Rather than building narrowly focused solutions, teams are encouraged to develop capabilities with reuse in mind from the start.
2. Pattern recognition: Identifying common underlying problems across different business challenges that can be solved with similar algorithmic approaches.
3. Efficient scaling: Reusing code, data pipelines, and engineering patterns lets us spin up new POCs in significantly less time — giving teams faster feedback loops and higher chances of success.
4. Standardized operations: With shared deployment and operational frameworks, we reduce the overhead of standing up new solutions. Teams can plug into centralized monitoring, governance, and support models, freeing up resources to focus on innovation.
5. Jump-starting the next use case: Previous successes serve as storytelling blueprints — clear examples that help business partners quickly understand the value and build confidence in new ideas. This shortens the time to buy-in and helps accelerate pilot approvals.

Looking ahead in the age of AI

The success of this approach validated the strategy of building shared, reusable capabilities to reduce the cost of building and maintaining AI solutions. As organizations continue to scale their AI initiatives, the ability to identify and leverage reusable components will become increasingly crucial for maintaining efficiency, compliance, and driving economic returns.

The transformation from “workshop” to “factory” is about more than doing things faster — it is about fundamentally changing how we approach AI development to create sustainable, scalable solutions that can evolve with business needs.

As head of data science and AI, I lead a team of 250+ data scientists focused on researching, developing, and deploying customer-first science in grocery retail, insights, customer loyalty and media. By partnering with the sales, product, and engineering teams, my team creates products and solutions with AI/ML science at the point of decision to drive informed business decisions for The Kroger Co. and its consumer packaged goods partners.

About the Author:

As SVP of Data Science and AI at 84.51˚ Kristin Foster leads a team of more than 250 data scientists focused on researching, developing and deploying customer-first science in grocery retail, insights, customer loyalty and media. By partnering with the sales, product, and engineering teams, her team creates products and solutions with AI/ ML science at the point of decision to drive informed business decisions for The Kroger Co. and its consumer packaged goods partners.