Every week we speak to business owners who know AI could help them but can't see where to begin. The tools change monthly, the vocabulary is confusing, and the examples online are either toy demos or enterprise programmes with enterprise budgets.
The good news is that the first step is usually smaller and more practical than people expect.
Start with the work, not the tools
Before choosing a platform or a model, write down where your team's time actually goes. Ask each person to list the tasks they repeat every week, how long each one takes, and how much judgement it needs.
You're looking for work that is:
- Frequent. It happens daily or weekly, not once a quarter.
- Rule-based or close to it. Most cases follow a pattern, even if a few need a human.
- Spread across tools. Someone is copying information from email into a spreadsheet, or from a form into a CRM.
- Low risk to get slightly wrong. A draft that a person reviews is a safer first project than a payment that goes out automatically.
Lead follow-up, invoice processing, weekly reporting, onboarding checklists and answering common customer questions almost always show up on this list.
Score your shortlist
Once you have ten or fifteen candidates, score each one on two simple scales: how much time or money it would save, and how hard it would be to automate. Hard usually means messy data, lots of exceptions, or many systems involved.
The best first projects sit in the high-value, low-effort corner. Pick two or three. Resist the urge to start with the most impressive idea on the list.
Decide what kind of automation it needs
Not every process needs AI. Many are better served by a plain workflow that moves data from one place to another on a trigger. AI earns its place when the work involves reading, writing or deciding: extracting details from a PDF, classifying an enquiry, drafting a reply, or summarising a long thread.
A useful rule of thumb:
- If the steps are always the same, use a workflow.
- If one step needs judgement on unstructured text, add an AI step inside the workflow.
- If the system needs to choose which steps to take to reach a goal, consider an agent, with clear limits on what it can do alone.
Keep people in the loop
The fastest way to lose trust in automation is to let it act without oversight on day one. Build in review points: approvals for anything that leaves the business, a daily summary of what ran, and an easy way for staff to flag mistakes. As confidence grows, you can remove checkpoints that no longer add value.
Measure before and after
Record how long the process takes today and how often it goes wrong. After launch, measure the same things. This turns "AI is helping" into a number you can take to the rest of the business, and it tells you which project to do next.
The short version
Find the repetitive work. Pick the few processes with the best return. Use the simplest tool that does the job. Keep a person in the loop until the system has earned trust. Then do it again.
If you'd like help finding your first three, book a discovery call and we'll work through it with you.
