Straight answers on ERP data, reporting and AI.
What actually works when you put analytics and AI on ERP data — written by the people who do it every day.
Finance can’t deal in “probably”
Give an AI chatbot the same sales table twice and you can get two different totals, two different best-performing regions and two different services mixes. The arithmetic is rarely the problem. Nobody told it your rules, so it made its own.
What is a semantic layer?
Most businesses carry several versions of the same number, all defensible, none of them written down. Somebody decides what 'sales' means from memory, one conversation at a time, and the only corner of the business where that job was ever recorded properly is your chart of accounts.
When your reporting depends on one person
Getting a number right takes someone who knows what it's supposed to mean and someone who knows where it lives. In a small team that's usually one person, and often the only one.
Why ERP data breaks the textbook semantic layer
Most writing on semantic layers assumes the modelling problem upstream is already solved. ERP data is where that assumption falls over, and where the reports go wrong without a single word being used loosely.
How to make your ERP data ready for AI
An AI assistant is probably already somewhere in your finance team. Someone is asking Copilot to explain a variance, dropping an ERP export into ChatGPT, or using Claude to turn a spreadsheet into commentary for a board pack.
Nothing in that category yet.