From virtual models to real-world performance: exploring digital twins and simulations

Online grocery is uniquely hard to get right. Basket sizes vary, orders span multiple temperature zones, and customers frequently edit their orders, often including time-sensitive fresh items.

Equally, customers have come to expect a best-in-class service: minimal (preferably zero) substitutions, ultra-fresh food, on-time delivery, and competitive pricing. 

Online grocery retailers have to consider these demands to win in their markets and develop a competitive advantage at the same time.

This combination of complexity, expectations and cost pressure makes online grocery one of the most demanding domains in retail logistics. It’s also what makes it an ideal environment for advanced robotics, automation, and increasingly, simulations (the use of virtual models to replicate and test how real-world systems behave, without disrupting live operations).

A short history of simulations at Ocado Group

In 2002, when Ocado launched our first fulfilment centre in Hatfield, almost every process innovation was tested directly in the real world. Some concepts, like a suspended monorail system (designed to move totes across the warehouse), were quickly abandoned when they proved inefficient. Others, like an early zone-pick model (where human pickers stayed in fixed zones and items were brought to them) became foundational to Ocado’s fulfilment design.

In 2009, Ocado built our first simulation model of the Hatfield zone-pick system to evaluate investment options and how to increase capacity. That early model mimicked both the physical infrastructure and the control logic: a critical step toward the simulation-first approach our business has today.

Developing our simulation methodology took time to establish. However, more and more successful use-cases and projects established the model as a trusted tool. As our confidence in it increased, we started to experiment with connecting the actual control software to our simulation model.

A simulation first approach

The Hive was the first Ocado system developed using a simulation-first approach, where modelling and virtual experimentation informed hardware design, layout, and control software from day one.

We used simulations to determine the specifications for everything from Bots to peripheral machines and grid layout.

They also enabled us to iteratively develop real algorithms and software, and ultimately optimise our processes across the board.

‍

‍

Simulations support optimisation

The systems we build make thousands of interdependent decisions. Each one of those micro-decisions can affect performance, cost, throughput, or adaptability in unexpected ways.

Of course, in a real-world fulfilment centre, it’s virtually impossible to test every scenario. But in a simulation we can use virtual models to evaluate trade-offs, stress-test for edge cases (rare or unusual situations the system might encounter), and explore how small design changes might cascade through an entire system and mitigate for those eventualities. 

This allows us to optimise for multiple goals at once, whether that’s minimising capital and operational costs, maximising item throughput, adapting to partner-specific constraints, or simply improving our overall service-level performance.

Simulations let us explore trade-offs and optimise across a complex set of priorities, including:

• Efficiency: minimising capital and operational cost to achieve a target throughput of items shipped
• Adaptability: adapting to a diverse range of contexts and sites Flexibility: customising prioritisation of trade offs for partner needs

The result: faster learning, lower risk, and more confident decision-making before new ideas are deployed into a live environment.

Uncertainty and variability

Real-world operations have many variables. Customer demand fluctuates. Staffing levels change. Inbound stock doesn’t always arrive on time. Small delays or unexpected events can have big knock-on effects.

Simulations give Ocado the tools to understand how our systems behave in the real world, and help us to answer "what if?" questions, without hunches or guesswork.

For instance, building our first Ocado Smart Platform (OSP) Customer Fulfilment Centre (CFC) marked a significant shift away from our established conveyor-based systems. Simulations were crucial in supporting this choice. They helped us test and expose the limitations of conveyor-based designs, such as single points of failure, and explore the potential of a modular system that could boost productivity and resilience. The insights gave us confidence to invest in a new model that would go on to define the OSP approach.

Production Replay validation

As real sites started to come online, we invested in proprietary technology to help us build and maintain credibility in our simulations.

Production Replay Simulation is an Ocado tool which loads the previous day’s data into a simulation: real orders, inbound stock arrivals, staffing levels and more to recreate how the system actually ran: a digital twin.

By constantly feeding operational data back into the simulation environment, we can benchmark against real outcomes, build trust in our models, and support faster tactical decision-making.

Expanding simulation across the value chain

Today, simulation at Ocado goes far beyond warehouse optimisation. It’s becoming a shared capability across the organisation: embedded in how we explore, evaluate and optimise decisions in multiple domains:

• A flexible OSP Grid simulation environment: enabling rapid experimentation with warehouse configurations, algorithm development, and early-stage design
• Supply chain simulation: modelling inbound stock flow, inventory profiles and delivery frequency to inform purchasing decisions, optimise delivery schedules and reduce out-of-stock risk
• Warehouse flow simulation: modelling grid congestion, tote routing and processing rates to identify bottlenecks, improve layout and enhance throughput
• Manual picking simulation: testing layouts, staff allocation and task sequencing in ISF and hybrid sites
• Vehicle routing simulation: exploring last-mile delivery plans and constraints, such as shift length and drop density
• Robotics simulation: improving coordination between bots, robotic arms and other automation elements, and enabling early testing of new designs and control strategies through detailed performance modelling.

The future is now

From early design to live operations, simulations help Ocado Group and our partners move faster, adapt smarter, and make better-informed decisions across the value chain. As technology continues to evolve, we explore how it can help us shape the future of online grocery and beyond.