Automate the dull work first. Leave the judgement calls to people.

Make the list before you buy anything

For one week, everyone in the business keeps a note of the repetitive tasks they do. Not the interesting work: the copying, chasing, reminding and re-typing. For each one, write down:

• What triggers it: a form, a sale, a date, an email.
• Who does it, how often, and roughly how long it takes.
• What goes wrong when it's late or missed.
• Whether it follows a rule you could write on one line.

Then rank the list. Frequent, slow, rule-based and low risk goes to the top. Rare, quick, judgement-heavy or high risk goes to the bottom, or off the list. The owner's own week is usually the richest source, which is why the owner is the bottleneck and automation tend to come up in the same conversation.

Five jobs worth automating first

1. The first reply to every enquiry. An instant, useful acknowledgement in the channel the customer used, a text back on missed calls, and an alert to whoever owns the enquiry. This usually pays back fastest, because speed often decides who wins the job. How to follow up enquiries before they go cold covers the full sequence.
2. Moving data between systems. Form to CRM, CRM to quote, accepted quote to invoice, order to spreadsheet. If someone is copying the same details into two places, software should be doing it, and the typing mistakes go with it.
3. Reminders and chasing. Appointment reminders, overdue invoices, missing documents, renewal dates. Your accounting software and booking system may already do some of this, so switch those features on before you buy anything new.
4. Review requests. A message after every finished job or delivered order, with a direct link. Reviews stop depending on someone remembering on a good day.
5. The weekly numbers. Enquiries, sales, cash in and overdue invoices, pulled together and sent every Monday, instead of someone building a report on Sunday night.

None of these needs AI. That's the point: the biggest early wins are plain automation, and plain automation is reliable.

Where AI actually helps today

AI is good at reading, sorting, summarising and drafting. It's bad at judgement, and it will state wrong things with total confidence. So use it where a mistake gets caught before it costs anything:

• Drafting replies to enquiries and emails, for a person to check and send.
• Summarising calls, meetings and long email threads into notes on the customer record.
• Sorting and tagging incoming enquiries, documents and support messages so they reach the right person.
• Answering common questions on WhatsApp or the website, from information you've written and approved, with a clear hand-off to a person for anything else.
• First drafts of product descriptions, job adverts, social posts and standard documents.

The pattern is the same each time. AI does the first draft, a person does the last check.

What not to automate yet

• Anything you haven't done by hand. If the process isn't written down and working manually, automating it just makes the mess faster.
• Quotes and pricing for non-standard work. Templates help. The number should still come from a person.
• Complaints and upset customers. An automated reply to an angry customer usually makes things worse.
• Advice with consequences. Medical, legal, financial or safety answers should never come from a chatbot without a qualified person checking.
• Messages sent in your name without review. Until a type of AI-drafted message has been checked many times and proved reliable, a person approves it.
• Decisions about people. Hiring, pay and performance stay human.

Rules that stop automation breaking quietly

Automations rarely fail loudly. A form field gets renamed, a login expires, and a step stops working at 2am without telling anyone. A few rules prevent most of the damage:

• Every automation has a named owner and a one-line description of what it does.
• Failures send an alert to a person, not just a line in an error log.
• Anyone can switch it off, and everyone knows how.
• Test with real examples before it touches customers, and check the first week's output by hand.
• Review the list once a month: what ran, what failed, what nobody uses any more.
• Keep customer data in tools you control and understand. In the UK, that means thinking about UK GDPR before you connect anything. In the US, texting customers needs their consent, and business texting needs carrier registration (A2P 10DLC).

Use what you already pay for

Before buying another subscription, check the software you already have. Most CRMs have built-in workflows. Accounting tools like Xero and QuickBooks can send invoice reminders. Booking systems send appointment reminders. Google Workspace and Microsoft 365 both offer AI assistants, built in or as add-ons. Tools like Zapier or Make fill the gaps between systems. A business with five tools that talk to each other is better off than one with fifteen that don't.

A first month, week by week

1. Week one: make the list and rank it.
2. Week two: automate the first reply and the missed-call text.
3. Week three: connect the form to the CRM, and the CRM to invoicing.
4. Week four: reminders and review requests, then add up the hours saved.

Bring AI in after that, one job at a time, starting with drafts a person approves.

If you'd rather someone else did it

This is the work behind our AI and automation service: find the jobs, automate them in order of hours saved and risk, then keep them running. If there's no CRM yet, that comes first, because automation needs somewhere to put things; see CRM and follow-up. The audit produces the ranked list for your business before anything gets built, and automation runs as one part of the transformation engagement rather than a one-off project. Or tell us what's going on.