Post

Enterprise AI is exposing a problem many companies have been quietly living with for years: data that works because humans know what it really means. Once machines start making decisions, that hidden context becomes operational risk. #ai #artificialintelligence #agenticai

15/10/26, 7:52 am1 repost4 likes

Dermot Casey

@dermotcasey.bsky.social

I remember a project 20 years ago where we had four conflicting definitions of a "sale" inside the business that had to be argued through before we could build the software.

Tom Raftery

@tomraftery.com

Take “order delivery date”. Warehouse may mean the truck has left. Finance may mean the customer has received the goods and revenue can be recognised. Both can be technically correct. An AI agent still has to know which meaning applies.

Tom Raftery

@tomraftery.com

This is why “clean data” is no longer enough. Accuracy, completeness and consistency matter. But AI also needs semantics, provenance, freshness, ownership and intended use. A perfectly formatted field can still be wrong for the decision being made.

Tom Raftery

@tomraftery.com

McKinsey reported in 2026 that just 7% of companies had fully scaled AI, while more than two-thirds of high-performing companies identified data as their primary obstacle. The bottleneck is shifting from model capability to organisational context.

Tom Raftery

@tomraftery.com

That has practical consequences. An AI agent can spot procurement leakage across thousands of transactions. But if supplier identities, contract terms or definitions are inconsistent, it can also automate that leakage faster than a human team ever could.