Engineering Meaning Before Automation: My AI Transformation Philosophy
Most AI initiatives fail before a single model is trained. They fail in the slide deck. They fail because the organisation reaches for automation before it has done the harder, slower, far less glamorous work of engineering meaning. After five years building AI-native enterprise systems, I'm convinced of one thing: semantic clarity always precedes automation. Skip that step and your 75%+ ROI promise becomes another failed POC slide in next quarter's review.
The Semantic Processing Unit
I think of every AI transformation as building a Semantic Processing Unit (SPU) for the business — a layer that translates messy human language, tribal knowledge and disconnected spreadsheets into a structured, machine-readable model. The SPU is not a tool you buy. It's a discipline you adopt. Every enterprise that has compounded value from AI in the last decade — from the Googles to the lean B2B SaaS shops — built this layer first, often without calling it by that name.
The Pathway: Glossary → CDM → LDM → AI
The pathway I run with every client is deliberately boring on the front end and explosive on the back end:
- Business Glossary. Every key term, defined and owned. What does "active customer" mean? "Margin"? "Lead"? You'd be surprised how often three executives in the same room hold three different definitions.
- Conceptual Data Model. The entities and relationships of the business, drawn at the kitchen-table level. No technology decisions yet.
- Logical Data Model. The CDM elaborated with attributes, cardinality, constraints. This is where data governance starts to bite.
- AI Systems. Only now do we wire models, RAG pipelines, agents and automations into a structure that actually means something.
Why this beats the POC trap
At Drinkizz, we didn't have time for the full enterprise dance — we were a Series A startup with no IT team. But we still ran the pathway in compressed form. The result: a No-Code system that 3× production capacity with no added headcount and 100% digital traceability. At F2Tech, the same pathway — executed at enterprise depth — replaced a legacy ERP with an AI-native semantic platform. Revenue went from $10k to $150k per month. ROI hit 75%. Seed round closed at $3M.
Both companies "did AI." Both achieved outsized returns. Neither started by training a model. They started by engineering meaning.
The lesson for every leader
If your AI POC died on the vine, the model wasn't the problem. The data wasn't the problem either. The problem was that nobody agreed on what the words meant. Fix that, and the rest is engineering. Skip it, and you'll keep paying consultants to rebuild the same broken foundation.
Engineer meaning first. Automate second. ROI third. In that order.