09:00 – 10:30 The AI Landscape for Manufacturing

AI spans four broad disciplines: perception, prediction, language and reasoning, and autonomous decision-making. Most manufacturing applications in production today live in the first two categories, while most AI headlines live in the third and fourth. Understanding that distinction changes how you evaluate risk, reliability and investment.

• Learning paradigms compared: supervised, unsupervised, reinforcement and generative AI, focused on where each works best
• Why AI projects fail, and how setup and framing drive success more than the technology
• A structured model for matching business problems to the right technique

11:00 – 12:15 Workshop: Mapping Your Operations to AI Opportunity

Working individually and then in small groups, delegates identify three to five persistent, measurable problems from their own facilities and run each through a scoring rubric covering data availability, problem structure and feasibility. Every problem is placed into one of three categories: ready to scope, conditions not yet right, or wrong tool. A full group debrief surfaces assumptions about data quality and operational tolerance. Each participant leaves with a personal opportunity map and a problem selected for development in the Day 2 business case workshop.

• Uses a real scoring rubric applied to problems from your own facility
• Includes a “wrong tool” category where process or maintenance is more appropriate than AI
• Highlights how differing views between similar operations reveal key assumptions

13:15 – 14:30 Case Study: Predictive Maintenance

An end-to-end walk-through of a predictive maintenance build, from raw sensor data to a model running in production. The session works through the actual data, the problems encountered, and the decisions made under pressure.

• Data quality challenges: sensor drift, label scarcity, and model decay over time
• Evaluation trade-offs executives must understand and own
• Success drivers, near-failures, and what would be done differently

14:30 – 15:30 Data: Your Actual Competitive Moat

Proprietary sensor, process and quality data is more valuable than any algorithm. This session explains data quality, labelling, freshness and governance as business assets, not IT concerns, and introduces a data readiness assessment executives should require before approving any AI project.

• Why the algorithm is rarely the constraint, and why your operational data is harder to replicate
• Data quality, labelling and freshness explained in terms of business risk
• A structured data readiness checklist to apply before approving any AI project

15:45 – 17:00 Challenge: Vendor Interrogation

Delegates review three fictional but realistic AI vendor pitches and apply the questions that

separate credible vendors from slide-deck AI. Every question in this session has exposed a real vendor in a real procurement process.

• The five core questions: training data, baseline comparison, and what happens when the process changes
• How to read vendor evaluation metrics and which numbers to demand when missing
• Common signs of weak vendor proposals and the questions that expose them

09:00 – 10:00 Computer Vision and Quality Control: Where It Works and Where It Fails

Applications covered include defect detection, dimensional inspection and assembly verification. Real confusion matrices are used to explain what false positive rate means on a production line, and why lighting, camera placement and training data diversity matter more than model choice.

• Reading confusion matrices as a business document, including the cost of false positives on a production line
• Why deployment conditions, not model architecture, determine success or failure
• Common failure modes in production vision systems and how to design against them

10:00 – 11:00 Case Study: Demand Forecasting and Supply Chain

A live teardown of a real, anonymised forecasting model output. Delegates interpret the feature importance together, then examine what happens when a demand shock hits a model trained on stable cycles, and how practitioners respond.

• Reading feature importance: what the model learned and if it transfers to future conditions
• Distribution shift: how models fail during demand shocks and key warning signs
• Practical mitigation approaches without full retraining

11:15 – 12:15 Generative AI in Manufacturing: Real Uses, Real Limits

Honest coverage of where large language models and multimodal models add genuine value in manufacturing, and where they introduce unacceptable risk. Includes a live demonstration of a model confidently producing a wrong answer from a maintenance manual.

• Practical applications: maintenance documentation, report generation, and knowledge capture
• Multimodal models for drawing interpretation and gaps between demo performance and production reliability
• Hallucination risk in safety-critical settings and structuring effective human oversight

13:15 – 14:30 Workshop: Build Your AI Business Case

Using a structured one-page template, each executive drafts a business case for their highest priority opportunity identified on Day 1. Followed by peer review with practitioner feedback.

• Template covers problem framing, data, success metrics, baseline comparison, and go/no-go criteria
• Risk factors and how to quantify them for finance and board-level decisions
• Peer review using the same criteria a practitioner applies before committing resources

14:30 – 15:30 Governing AI: What Executives Must Own

Covers the governance responsibilities that belong to leadership, not the data team.

• Model monitoring, drift detection and audit trails: what to require, verify, and who is accountable
• Human-in-the-loop design: when to mandate sign-off and prevent automation bias
• KPIs to track AI performance over time and criteria for shutting deployments down

15:45 – 17:00 Closing Workshop: Your 90-Day AI Readiness Plan

Each participant leaves with three concrete, personally owned actions. Followed by a facilitated discussion on organisational blockers.

• One data audit: identify the highest-value data asset not currently fit for use
• One pilot: turn the business case into a concrete next step with a named owner
• One capability to build or buy: assess what your organisation currently lacks