Where to start with AI automation (and where not to)

There's a lot of noise around AI. For most businesses the useful question isn't "what can AI do?" but "which hours of our week are being spent on work a system could handle?" Start there and automation becomes a practical project rather than a buzzword.

Automation and AI are not the same thing

Traditional automation follows fixed rules: when a form is submitted, add a row to a spreadsheet and send a confirmation email. It's predictable and cheap. AI adds the ability to handle messy, unstructured input: reading an email and working out what the person wants, summarising a document, or drafting a reply.

The most reliable workflows combine both. Rules handle the structure; AI handles the parts that need judgement; people handle the decisions that matter.

What makes a good first candidate

Look for work that is:

• Repetitive. It happens many times a week in roughly the same way.
• Time-consuming. Someone would notice the hours coming back.
• Well understood. You can explain the steps to a new employee.
• Low risk. A single mistake is cheap, or easy for a person to catch.
• Already digital. The information lives in email, forms, spreadsheets or software, not on paper.

Typical examples include capturing enquiries from your website and WhatsApp into one place, qualifying and routing leads, sending follow-ups, booking appointments, turning documents into structured data, producing weekly reports, and answering common customer questions.

What not to automate first

• Anything you can't yet describe clearly. Automating a messy process just produces mess faster.
• High-stakes decisions. Approving payments or giving legal or medical advice should never run without a human check.
• Processes that happen rarely. The setup time won't pay back.
• Personal moments. Conversations where customers value a real person should stay with a real person.

Fix the process on paper first. Then automate the version that works.

A simple way to begin

• Map the process. Write every step, who does it, which tools are involved and where information gets copied by hand.
• Find the bottleneck. Usually it's a step where someone reads something and retypes it somewhere else.
• Decide where people stay in the loop. For example, AI drafts the reply and a person approves it before it is sent.
• Build small. Automate one workflow end to end, test it with real examples, then expand.
• Measure. Note how long the task took before and after, and how often the system needs correcting.

Doing it safely

AI models can be confidently wrong. Design for that. Give the model clear instructions and only the information it needs, check its output against rules where possible, and route uncertain cases to a person. Every workflow should log what it did and alert someone when it fails, so problems are visible rather than silent.

Be careful with data. Know which tools your customer information passes through, avoid sending more personal data than necessary, and make sure the services you use meet your obligations to customers.

The real payoff

The best automations are boring. They quietly do the same job every day so your team can spend time on work only people can do: talking to customers, making decisions and improving the business. Start with one workflow, prove the value, and build from there.