Where do AI features fit in an ERP product?
For an ERP vendor, AI features fit best where users already make repeated judgments: reorder suggestions, schedule changes, document classification, anomaly flags. Each needs a place in an existing workflow and a record of who accepted it.
A chat panel over ERP data is one option. Suggestions placed inside the screens where decisions are made reach users at the moment of the decision.
Table of Contents
Summarize with:
Key facts
- Good candidates are decisions users already make repeatedly.
- Each suggestion should show the data it is based on.
- Acceptance, change or rejection should be recorded.
- A manual path has to keep working.
- With developers, Redbrix defines supported tasks, evaluates errors on representative cases and agrees where review or a manual fallback is required.
- A chat panel as the only AI feature leaves the structured workflows untouched.
Context
Why the question comes up
ERP buyers increasingly ask vendors about AI during evaluation, and vendors look for features that show value without risking the records the business depends on. ERP data is structured and audited, so a feature that writes into it carries more risk than one that only answers questions.
AI feature scope
Redbrix defines the scope of AI functions together with developers, based on what can be performed reliably in the product.
Where AI fits in an ERP
| Area | Example of a repeated judgment |
|---|---|
| Purchasing and inventory | Reorder suggestions |
| Planning and scheduling | Schedule changes |
| Documents | Classification, data extraction from incoming documents |
| Quality and finance | Anomaly flags |
| Sales | Prefilling quotes from customer drawings or requests |
How it shows up in the product
- Suggestions inside the screen where the decision is made
- The data each suggestion is based on, shown with it
- A record of acceptance, change or rejection
- A manual path that keeps working
Options and trade-offs
- A chat panel is quick to add, but users reach it only when they leave their task. Suggestions inside the screens take more design and engineering work and appear at the moment of the decision.
Common mistakes
- Adding a chat panel as the only AI feature
- Suggestions without sources in regulated processes
- Storing AI output as if a user had typed it
- Removing the manual path once the AI feature ships
Where Redbrix has worked on this