Data + AI
Data first, then context, then AI. A person signs off.
The position
Take the screens away from an enterprise system and what remains is data. This work starts there.
The position is deliberately narrow. Enterprise data is first organised into a context layer of entities, relationships and attributes. Language models and agents are then connected on top. A person, not another AI, signs off the result.
The purpose is narrow too: decisions for management, and analysis that can be acted on directly.
The four layers
Read from the bottom up. Each layer is built on the one beneath it.
-
1 Data
What your ERP and core systems already record.
-
2 Context layer
Entities, relationships and attributes: what the data means in your business.
-
3 Language model
Reads and reasons over the context instead of raw tables.
-
4 Agents
Carry out defined tasks using what the model understands.
-
A person signs off
The last check on any result is made by a named person in management.
Four rules
Data before models
No model is connected until the context layer exists. The ERP or core system you already run is the anchor.
Effectiveness before efficiency
The first aim is a better decision. Gains in speed come later, once there is enough data to support them.
A person signs off
One AI may check another. The last check above both is always a person in management.
Built in-house
The people who build and run it are your own in-house team, managed under Technology Leadership Services.
Where it usually starts
Identifying high-value customers early, and predicting which customers are about to leave.
What it does not cover
Routine workflow automation. If a fixed rule can do the job, a rule should do it, and we do not call that AI.
Where this came from
In 2018 Chris turned a large durian plantation into a data-driven operation.
A plantation management system, sensors in the field and big data analytics went in. It was an early lesson in turning operational data into decisions, and it is why the data comes first.