Ocado: Fail fast, Learn Faster

Myrna Macgregor, Data Science Festival Oktoberfest 2024

Ocado’s Myrna Macgregor shared their philosophy of “Fail Fast, Learn Faster”, focusing on their approach to e-commerce, supply chain management, fulfilment, and delivery. Here’s my personal takeaways from the talk:

What makes Ocado E-commerce Innovative?

  • Extensive Product Range: Ocado boasts an inventory of 50,000 products, catering to a diverse customer base, and handling supply and logistics for multiple supermarkets.
  • Market positioning: Ocado think of themselves primarily as a technology company, which helps guide their thinking and innovation.
  • Personalised Storefronts: Each user enjoys a personalised storefront, enhancing their shopping experience through tailored recommendations that cater to individual preferences.
  • Machine Learning for Ordering: The company utilises machine learning to optimise ordering processes, aiming to reduce wastage and improve inventory management.

Encouraging Experimentation through Data Bets

Ocado created an initiative they dubbed “Data Bets” to encourage innovation in a product-driven way.

Data Bets allows teams to quickly and cheaply validate data-driven ideas. Data Bets operate outside of the typical business-as-usual (BAU) work, giving the data team the freedom to explore new concepts.

The process is structured thus:

  • Formulate a hypothesis: a concept or process that the team think will have a good payoff, but might be outside the scope of the team’s usual work. In this case it might take the form of a data product, but a the concept could be applied more generally to product work
  • Gather resource: Assign a dedicated resource for sponsoring and leading each experimental initiative. This person acts as product manager and evangelist (more on that in a mo). The rest of the team (developers, designers, whoever you need) can be drawn from anywhere else in the organisation and use their “10% time” (time set aside each week for new ideas and side projects) to complete the work.
  • The team prototypes the product, and the dedicated resource steps into their evangelist role: touting ideas within the organisation to raise visibility and encourage usage
  • They maintain a structured review cycle to assess the progress and outcomes of experiments, ensuring that learning is continuous and integrated into future strategies.

Experimentation with Gen-AI

Ocado used this framework to explore the use of a Generative AI assistant using RAG (retrieval-augmented generation) to enhance internal document querying.

However, the initial results did not meet expectations (hallucinations, inaccuracies etc.), leading to the decision to discontinue the project. Instead, they pivoted to utilising an external knowledge base, but the experiment helped to verify that a solution was needed, and that the 3rd party supplier was doing a better job than Ocado themselves could deliver given the available resources.