Most of the value of AI for a small business isn’t a chatbot on the homepage. It is in the repetitive work behind the scenes: sorting enquiries, drafting replies, moving data between tools and turning documents into something useful. Here are the automations we build most, and how to approach them safely.
Start with the boring work
The best candidates for automation are tasks that are:
- Repeated many times a week
- Rule-based, or close to it
- Easy to check
- Annoying enough that people put them off
Automations that work well
Enquiry triage
New form submissions are read, categorised and routed to the right person, with key details pulled out: budget, timeline, service, location. Urgent ones are flagged.
Draft replies
For common questions, a draft reply is prepared in your tone for someone to check and send. The human stays in control; the blank page goes away.
Document processing
Invoices, quotes, forms and PDFs are read and their details entered into your accounting, CRM or spreadsheet automatically.
Meeting and call notes
Recordings or notes turned into a summary, action items and an update in your CRM.
Content support
First drafts of product descriptions, alt text for images and summaries of long documents, always reviewed before publishing.
Website editing by asking
Describe a change to your site to Claude or ChatGPT, review it, and publish. We set this up so changes go through a review step rather than straight to the live site.
Automate the tasks people put off. Keep people on the decisions that matter.
What to watch out for
Keep a human in the loop
Anything that goes to a customer, touches money or makes a decision should be reviewed by a person, at least at first.
Protect your data
Know where your data goes. We use providers with clear business data terms, and avoid sending personal or sensitive information anywhere it isn’t needed.
Start small and measure
Automate one workflow, measure the time saved and the error rate, then decide what to do next.
Plan for mistakes
AI tools can be confidently wrong. Build in checks: confidence thresholds, validation against your data, and an easy way to correct and report errors.
How we build it
We connect AI models to the tools you already use through their APIs, with the workflow running on infrastructure such as Cloudflare Workers. That means no new app for your team to learn: things simply start happening in your inbox, CRM or spreadsheet.