Skip the chatbot-on-the-homepage reflex. The best first AI projects are boring, internal and easy to measure.
Founders ask us the same question every month: where should we use AI? The honest answer is usually not where they expect. The highest-return projects are rarely customer-facing chatbots. They're the repetitive, internal tasks that quietly eat hours every week.
Good first projects
- Answering staff questions from your own documents, policies and past tickets.
- Drafting first replies to common support or sales enquiries for a person to approve.
- Turning messy inputs (emails, PDFs, forms) into structured records in your CRM or database.
- Summarising long threads, calls or reports into the three things someone needs to act on.
What these have in common: the data already exists, a person stays in the loop, and you can measure the result in time saved or errors avoided.
Projects to be careful with
- Anything that gives advice customers will act on without a human check: medical, legal, financial.
- Features that only work in a demo with hand-picked examples.
- AI added because a competitor has it, with no clear job to do.
How we run an AI pilot
We pick one workflow, collect real examples, and agree on what “good” looks like before we write code. Then we build a small version connected to your actual data, with logging so every answer can be reviewed. After two to four weeks of real use, the numbers tell you whether to expand it, change it or stop.
The best AI feature is the one your team stops noticing because it just works.
If you have a process that's slow, repetitive or easy to get wrong, that's usually the place to start. We're happy to look at it with you.
Related services: AI and automation.