Automation
AI where it changes something
The question is not whether you can use AI, but where it measurably beats what you do today.
Area
Automation
Typical for
Documents, enquiries, research
Settle first
Which data the system may see
What it is about
Language models are good at ordering, summarising and classifying unstructured text. They are bad at calculating reliably or guaranteeing facts.
So we build them in where a person checks the result anyway, or where a mistake is cheap.
Where it usually pays off
- Pre-sorting incoming documents and suggesting field values
- Classifying enquiries and routing them to the right place
- Summarising long texts for a decision
- Search across your own documents, with a link to the source
What has to be settled first
- Which data the model may see and which it may not
- Where the model runs and who the processor is
- How a person checks the result before it takes effect
- How you measure whether it actually got better
Related