What happens after the AI demo works? In this focused presentation, Stuart Houston, Data Scientist at CNOOC, examined why many AI proof of concepts generate early excitement but fail to become trusted production capabilities.
Using a live example from an internal control document review process, Stuart explained how AI was applied to identify responsibilities spread across multiple documents, compare overlapping ownership and highlight duplicated or inconsistent wording. The real value was not simply extracting information faster, but making hidden patterns visible across a complex document set.
The presentation then moved into the harder question: not whether the technology works, but whether the business can rely on it. Stuart highlighted the practical issues that often stop pilots from progressing, including output validation, traceability, ownership, source transparency and whether users would be confident enough to act on the information provided.
A key message was that responsible AI cannot be added at the end of a project. It has to be designed in from the start through human validation, clear accountability, source-level traceability and a focus on measurable value. Without those foundations, even a successful pilot can stall before reaching production.
Watch the full presentation to hear Stuart explain why the real challenge is not building AI that works, but building AI-enabled capability that the business is willing and able to trust.
Speaker
Stuart Houston, Data Scientist, CNOCC
