What can AI do for a small firm right now?
You've used ChatGPT. What changed is that AI can now work for you the way you would, 24/7. How it works, and which tasks to automate first.
You've probably used ChatGPT to tidy up an email, or draft a reply that you then rewrote because it didn't sound like you. Useful enough to keep using it, not enough to justify the "buzz" everyone is talking about with AI. Meanwhile you're reading about firms where AI answers the phone, acts as personal assistant or manages their customer services.
The gap between those two things is smaller than it looks.
How an AI assistant is different from ChatGPT
ChatGPT waits for you to bring up a question. You bring the context, you paste the background, you read the answer, you send it yourself. Close the laptop and the work stops.
Two things changed over the past 18 months. AI models got reliable enough to run a multi-step job without wandering off halfway through. And you can now connect them to your tools, your inbox, your calendar, your WhatsApp.
This means you now have an extra pair of hands that's constantly available, never gets tired, and takes the repetitive work off your desk so you can focus on what matters for your business.
How AI answers a customer enquiry for you
You type your question, AI answers, and the work stops there.
The AI works for you while you are at dinner, answering new enquiries in your voice.
Let's take the example of dealing with new enquiries. First the AI learns your answers: what you charge and what's included, the questions you ask before you can quote, what you say when the answer is "it depends", the enquiries you turn away. Second, it connects to every channel your enquiries arrive through: the inbox, the website form, WhatsApp. Third, when a new enquiry comes in the AI reads it, replies the way you would, asks your usual qualifying questions, and can book a call, all while you are doing something else.
Which tasks should a small business automate first?
Look back at your week and audit your repetitive activities. Mark anything you have done more than twice: the first reply to a new enquiry, the reminder about the missing document, the quote follow-up, the answer to "how much and how long".
List out 5 to 10 activities. Next to each, write why you did it. Below are some of the candidates I've seen in small businesses.
Tap a row for what it takes.
The same 3 questions every time, then a link to book. Needs your standard answers written down once. The lowest-effort job on the list and the one that decides whether the enquiry stays with you.
A checklist, a timer, and a polite nudge on the right channel. Needs a list of what a complete file looks like. Firms lose applications to documents nobody chased.
Day 3, day 7, day 14, in your words, stopping the moment they reply. Needs nothing you do not already have. Most small firms send zero follow-ups because the day fills up.
Cheap if you already track status somewhere, a project if the status lives in your head. Silence is the complaint clients actually make, well ahead of price.
Needs one date per client and a calendar. Brings back last year's clients without anyone remembering to, which is why it usually pays for the rest.
Small on its own, and it rides free once the enquiry desk exists. Worth adding, never worth starting with.
These turn on your read of a person. The most a system should do here is spot one in seconds and put it in front of you. Anything that drafts a reply to this is trading a client for 10 saved minutes.
How to train AI on the way you work
Once you've picked one job to start with, run it with the AI watching. Talk through a real case: what you look for first, what you ask before you can quote, what you charge, what you say when the honest answer is "it depends", which enquiries you turn away.
Then let AI try one and correct it the way you'd correct a new hire in week one. Wrong tone. Wrong order. Missed the thing that mattered. Say so, and make it write the rule down.
None of the work to set it up is technical. You're getting 10 or 15 pages of knowledge out of your head, which is the same reason nobody in your office can cover your inbox properly. The technology has been ready longer than the documentation has.
You can get to the drafting version yourself in a couple of evenings, with any of the main assistants and a notes file. It prepares, you press send. Getting to the version that answers on its own takes the same written answers plus testing on your ugliest cases and a clear line for when it hands back to you. That last part is the one people skip, and it's why so many attempts end at "we tried AI and it wasn't good enough".
Is it too late for a small business to start with AI?
If you're reading this because it feels like everyone else has moved and you haven't, you're not behind. Most firms are still at the chatbot stage. The few ahead of them started by writing down how they work, one process at a time.
If you're keen to see how it would work for managing your new enquiries, the demo answers potential customers in your own words, across all channels, 24/7.