What we know
Tools we've used properly, not just watched a demo of.
Anyone can list logos. Here's the same list with an honest opinion attached to each one, because knowing when not to use something is the useful part.
AI
OpenAI
OpenAI's models are useful when there's a real job to do: summarising, drafting, classifying, tidying messy input. They're less useful when they're added to something purely so it can be described as AI.
What it's good at
- Turning messy free text into structured, usable data
- Drafting and summarising so a person can edit rather than start cold
- Classifying enquiries and routing them sensibly
- Fast to prototype, so you find out quickly whether it helps
Where it gets awkward
- It will answer confidently when it's wrong
- Costs are easy to underestimate once volume picks up
- Sending sensitive data anywhere needs a proper decision, not a default
- Model changes can shift behaviour under a working feature
Our experience with it
We build AI features with a human in the loop where it matters, clear boundaries on what the model is allowed to decide, and a fallback for when it gets it wrong.
Things we've built or connected
Enquiry summaries so a person starts the call already informed
Classification that routes messages to the right place
Draft-first tools where a human always approves before anything sends
Usually in the same conversation