John Lewis Partnership:
From MLOps Foundations to Exploring Agentic AI

John Lewis Partnership’s (JLP) AI transformation began with a bold ambition: Productionize 12 machine learning models in 12 months. With support from Deloitte and Dataiku, JLP built a scalable MLOps foundation that enabled faster deployment, consistent governance, and reusable workflows across their organization.

This strategic initiative unlocked JLP’s ability to operationalize AI at scale, transitioning from ad hoc experimentation to a robust, repeatable process. As a result, JLP has grown its AI footprint to 18 production-grade models powering use cases from forecasting and pricing to labor planning and checkout optimization, delivering over £40 million in business value.

Modernizing Store Operations With AI, One Loaf at a Time

Before implementing Dataiku, in-store partners at JLP were expected to manually download spreadsheets, interpret them in Excel, and estimate what to bake and when. The process was manual, time-consuming, and inconsistent.

One of the models born from JLP’s AI maturity was the Bakery Optimization Tool (BOT). Using historical sales data, the data science team built a model that generates store-specific baking recommendations, offering visual, real-time forecasts that are fast and easy to act on. BOT quickly scaled across the store network, improving forecast accuracy, boosting sales, and reducing waste, demonstrating the tangible, operational value of AI at the shelf level.

 Since implementing Dataiku, model operationalization has become so much easier. Previously, it took us months to put models into production. Now we've got that down to just a couple of weeks. Barry Hostead Director of Data Management & Intelligent Platforms, John Lewis Partnership

Scaling Enterprise AI With Trust and Speed

The bakery forecasting model was just one example. JLP’s AI footprint now covers use cases like online checkout optimization, product availability forecasting, promotion planning, returns reduction, delivery slot availability, and labor scheduling.

Since launching its MLOps foundation, JLP has deployed 18 production-grade models using Dataiku, The Universal AI Platform™. Model deployment timelines dropped from months to just five weeks, and iteration cycles that once took 10 to 12 weeks now run in just one to two.

 Dataiku is an AI platform that enables us to manage the whole data science lifecycle from data prep all the way through to model training and build. Barry Hostead Director of Data Management & Intelligent Platforms, John Lewis Partnership

Laying the Foundation: MLOps With Deloitte

JLP’s AI journey began with a focused program: a one-year initiative with Deloitte and Dataiku to get 12 models into production in 12 months. At the time, promising models were stuck in notebooks, pipelines lacked consistency, and getting to production was a major hurdle.

Deloitte brought the engineering expertise to design a scalable MLOps framework tailored to JLP’s needs. The program also included targeted upskilling for JLP’s data science team on MLOps best practices and how to effectively use Dataiku, ensuring long-term ownership. Dataiku provided the platform to operationalize it with reusable templates, a visual interface, and built-in governance. That foundation accelerated delivery, but more importantly, it gave JLP full ownership of its AI roadmap.

Dataiku  gave us an environment where we could experiment and test our hypothesis, govern our data, and govern the models that we were going to put through to production and actually go live and deliver value. Richard Cortés AIOps Product Manager, John Lewis Partnership

Built to Integrate, Built to Scale

Furthermore, the JLP tech stack is deeply integrated. Data is stored in Snowflake, visualized in Tableau, and orchestrated through Dataiku, which serves as the connective layer across all AI workflows. Dataiku’s drag-and-drop interface, pre-built components, and reusable templates enable both senior data scientists and analysts to contribute without relying on engineering overhead.

Governance-First Enterprise AI

As JLP scales its AI practice, governance is built into every layer. The team uses Dataiku to enforce standards around model documentation, version control, explainability, and ethical oversight. These guardrails aren’t just best practices, they’re essential for building and maintaining trust at scale.

Before Dataiku, there was no real way of realizing the value our use cases could bring. Richard Cortés AIOps Product Manager, John Lewis Partnership

Looking Ahead: From MLOps to Agentic AI, Done Responsibly

With production models in place and a proven delivery framework, JLP is now focused on the next phase of AI maturity: exploring how GenAI and agentic AI can augment shared decision-making and streamline operational workflows. These new capabilities are being developed with business impact in mind: improving productivity, accelerating time-to-value, and reinforcing transparency at scale.

We found Dataiku incredibly easy to implement across our environment. The speed at which we work, the effortless way we can put a model from experimentation into production, and the value that we've realized in such a short amount of time has basically paid for itself. Richard Cortés AIOps Product Manager, John Lewis Partnership