We use AI in client projects where it removes routine, not where it sounds impressive. The test is simple: it has to save more hours than it costs to run, and a person has to stay accountable for anything a customer sees.
Where it earns its place
- Documents: reading invoices, contracts and forms into your systems instead of retyping them.
- First-line replies: drafting answers to repeated questions for a human to approve and send.
- Search across your own knowledge — tickets, wikis, past projects — so answers stop depending on who is in the office.
- Integrations: moving data between systems that were never designed to talk, with the rules written down.
- Content support in marketing: drafts and variants, always edited by the person whose name goes on them.
How we keep it honest
- You are told which model runs where, and what it costs per month.
- Company data is not used to train anyone else’s model.
- Anything customer-facing is reviewed by a person before it goes out.
- If a task is cheaper as a plain script than as a model, we write the script.
We are equally willing to say when AI is the wrong tool. Most automation wins in a small company are still ordinary integration work.