Start with a task, not a subscription
“We should use AI” is too broad to act on. “Our operations manager spends every morning turning customer emails into order records” is a process you can examine. Write down what arrives, what the team does with it, and what a correct result looks like.
For a first project, choose work that happens often enough to test but is small enough to stop without disrupting the business. Summarising internal notes or preparing a draft is easier to supervise than allowing software to send a binding quote.
Make a shortlist from your actual week
| Task | Useful first test | Human checkpoint |
|---|---|---|
| Incoming enquiries | Suggest a category and a draft reply | Approve before sending |
| Meeting notes | Extract decisions and action items | Check names, owners, and dates |
| Research | Group relevant material with source links | Read the source before using a claim |
| Data transfer | Check whether ordinary rules are enough | Validate required fields |
These are potential use cases, not a promise that AI will improve each one. A simple form, template, or existing integration may solve the problem more reliably.
Measure the work before changing it
Record the number of items processed, time spent per item, and common corrections over a representative period. Include time spent chasing missing information. A process that looks slow may actually be waiting on a decision that AI cannot make.
For example, 30 enquiries taking six minutes each represent three hours of work. If an assisted process still needs two minutes of review per enquiry, the theoretical saving is two hours before setup and exception handling. This is an illustrative calculation, not a forecast for your business.
Run a small, reversible pilot
- Use representative examples, with sensitive data removed where possible.
- Describe what the system may do and what requires approval.
- Compare its output with the team’s normal work.
- Track review time, errors, and cases it cannot handle.
- Decide whether to expand, change the approach, or stop.
Check the provider’s data handling and your organisation’s rules before uploading customer information. Keep the original workflow available while testing.
What to bring to an AI consultation
Bring one process, the tools involved, a few representative examples, and a rough estimate of the time it takes. The useful outcome is a decision about your next step, not a shopping list of AI products.
One process. A useful next step.
Bring the workflow you want to improve. We’ll discuss where AI can help, what needs checking, and what to do next.
Book a consultation — $59 ↗