The unsexy work that makes everything else possible — and the discipline that AI makes non-negotiable.
Every BI project eventually hits the same wall: two teams with two different numbers for the same metric, each convinced theirs is right. That’s a governance failure, not a data failure.
I’ve built a data dictionary, a certified data program, a reporting repository, and a full Tableau audit across an organisation with seven internal stakeholder groups. None of it was glamorous. All of it was necessary.
And then AI arrived. The moment you start feeding LLMs operational data, every governance gap becomes a liability. This pillar covers how to build the foundation — before you need it.
“LLMs don’t replace BI fundamentals — they make them non-negotiable. Garbage in, garbage out gets a lot more expensive at scale.”
How to build a data dictionary that actually gets used — tied to your reporting layer, owned by the people who generate the data, and maintained through a governance cycle.
Master data management for ops teams — deduplication, entity resolution, and the reporting standards that prevent seven teams from producing seven different answers to the same question.
The audit process for an existing Tableau environment — how to identify what’s being used, what’s stale, and what’s actively misleading. Plus the cadence that prevents the graveyard from growing back.
The VOC programme at Tyco/JCI was a governance exercise as much as a process one. We built a structured 18-category tagging taxonomy, embedded it into the ticketing workflow, and produced monthly intelligence reports for product and leadership. Three product roadmap changes were driven directly by VOC data within the first year — possible only because the data was clean, consistent, and governed from the point of capture.