Skip to main content

Practical exercise

Instruction and application
In Progress

AutoDeliver life cycle challenge​

In this activity, you will act as AutoDeliver’s ML team to design a life cycle plan for its autonomous delivery models. You’ll identify scalability risks, propose governance practices and decide how and when to retire outdated models safely.

These skills are essential for organisations that run ML systems at scale. By practising life cycle planning, governance design and decommissioning strategies, you’ll learn how to keep models efficient, compliant and aligned with business needs while avoiding disruption.

Practical exercise context​

AutoDeliver is a fast-growing logistics startup operating an autonomous delivery fleet in urban areas. The company relies on ML models to coordinate operations and deliver packages efficiently at scale.

Models in use:

  • Real-time navigation: Directs vehicles through city streets.

  • Traffic prediction: Anticipates congestion patterns to optimise routing.

  • Package routing: Balances delivery priorities, vehicle capacity and time windows.Current challenge:

  • Rapid fleet expansion: Fleet size has tripled in the last year. At peak hours, deliveries are projected to grow from500 per minute to 1,500 per minute over the next 12 months. Current models must generate navigation updates in under200ms to keep vehicles on schedule.

  • Supply chain dependencies: Reliance onthird-party map APIs (licensing limits, rate throttling, outages) andcloud GPUs (pricing spikes, regional shortages) introduces supply chain and governance risks.

  • New model rollout: AV2 navigation model is scheduled for release in the next six months. Older models (V1) remain active in production and are still connected tocustomer-facing apps and partner APIs, creating challenges for dependency management and service continuity.

  • Governance and logging pressures: Regulators now requiretraceability of ML decisions for autonomous vehicles. Past routing failures at AutoDeliver revealed gaps in logging, making incidents difficult to investigate.

  • Retirement and compliance: Regulators also require AutoDeliver to archivemodel versions, training dataset snapshots and inference logs for at least three years to ensure they can conduct safety investigations and compliance audits.Your task: Step into the role of AutoDeliver’s ML operations team. Design alife cycle plan that ensures:

  • Scalability to meet growing demand.

  • Governance through structured deployment and logging practices.

  • Safe retirement of outdated models with compliance and business continuity in mind.

Activity instructions​

Work with your group to complete the following tasks:

Review the context​

Start by reviewing AutoDeliver’s situation. Focus on the scalability pressures,supply chain dependencies and the presence ofoutdated models still in production.

Plan for capacity​

  • Calculate basic resource needs (traffic × latency).
  • Analyse how data variety (e.g. multimodal inputs) and data quality (e.g. schema changes) affect scalability.
  • Propose one scaling strategy (horizontal, vertical or autoscaling).
  • Identify two supply chain risks, and suggest mitigations.

Design governance practices​

  • Draft a structured rollout plan for V2.
  • Specify at least three key logs/metadata to capture.

Develop the retirement protocol​

  • Outline the steps for retiring V1.
  • List two or three artefacts to archive for compliance and explain why.

Regroup and share your insights​

Return to the main session after 20 minutes to discuss key insights and recommendations.

Action item: Activity share out​

  • What is one scalability risk your group identified in AutoDeliver’s system?
  • What is one governance practice you would recommend to strengthen reliability and compliance?
  • How would you decide when and how to retire an outdated modelwithout disrupting service?