Blog · AI and automation
Automation follows the rules you give it. AI reads, writes and makes a best guess. They are not the same thing, you will probably need both, and knowing which is which decides what you can trust.
At a glance
In plain words
Automation is a set of rules that runs on its own. When a client signs, create the record. When an invoice is fourteen days overdue, prepare the reminder you wrote for someone to check and send. Its intended behaviour follows the rules, rather than improvising. Bad data, expired access or a failed connection can still stop it or produce a wrong result.
AI, in the sense most people mean today, is a language model such as ChatGPT or Claude. It can read an email and say what it is about, pull the key details out of a document, summarise a long thread or draft a reply. It is useful precisely where rules break down, because the input is messy. It is also capable of being wrong while sounding sure of itself.
Automation and AI, side by side
| Compare | Automation | AI |
|---|---|---|
| What it does | Follows rules you set: when this happens, do that | Reads, sorts, summarises and drafts language |
| Good at | Repeating the same steps exactly, every time | Messy input: emails, documents, free text |
| Example | Creating an invoice when a stage is approved | Sorting incoming emails and drafting a reply |
| When it is wrong | Wrong rules, bad data, lost access, failed connections or software changes | Sometimes, and confidently. It needs a check |
| Running cost | Depends on your tools, usage, monitoring and support | A subscription or a charge per use |
A simple test
If you can write the rule down, use automation. Creating an invoice from recorded work, raising a direct debit, moving a client to the next stage: none of it needs a model. It needs a reliable connection between two systems, with agreed checks for missing data and failed connections.
If the job is reading or writing language, AI may help. Sorting an inbox, extracting details from a form someone filled in their own way, preparing a first draft for a person to approve.
If it needs your expertise or your relationship, it stays with you. That is what your clients pay for.
Much of what I build is not AI. It is plumbing, well laid.
In practice
Automation is not a robot or a chatbot. It is a rule that runs on its own: when this happens, do that, without anyone remembering to do it. Three examples from delivered work, followed by an enquiry-intake pattern you could use. None requires AI.
What clients tell me
The first three examples describe client work or my own service business. The fourth is a possible intake workflow. I add AI where reading or writing language helps, and identify the review and exception steps alongside it.
An ordinary ChatGPT conversation, without connected tools or a configured workflow, does not run your business processes. Connected assistants can take actions, but they still need the right access, triggers, rules and approval steps. A scheduled follow-up needs a defined process, not just a good prompt.
That is why amounts, payments and statuses need agreed rules, accurate records and checks, and why client-facing drafts wait for a person to read them and press send. A rule can also fail, so we test and monitor the whole process.
If a mistake would cost money or trust, a rule does the work and a person checks it.
Better together
The useful systems combine both, each doing what it is good at. Take a client email arriving in a busy inbox:
1. Automation picks it up the moment it arrives and links it to the right client in the CRM.
2. AI reads it, recognises a request for a project update, and prepares a draft from the project status.
3. Rules route it to the person responsible for that client, with a deadline.
4. That person checks the draft, changes what needs changing, and sends it.
5. Automation logs the reply on the client record and sets the next follow-up.
Three of those five steps are plain automation. One is AI. The one that matters most is a person.
AI is a tool in the box, not the box.
If you only use ChatGPT for emails
Many small business owners I meet are in the same place. They have heard a great deal about AI, tried it a little, mostly to write emails, and feel overwhelmed by everything else. That is a perfectly sensible place to be.
Three uses worth trying this week, with no set-up: paste in a long email thread and ask for a summary before you reply; turn your notes from a client call into a follow-up email or the outline of a proposal; turn a messy list into a clean checklist or table. Use a business account, and leave out personal details your clients would not expect you to share.
But the bigger win is often not more AI. It is removing the repetitive steps around your work: the retyping, the chasing, the checking. That is automation, and it is where most of the hours come back.
A word of caution
Do not let an unchecked AI answer set an invoice amount, approve a payment or send a client message. Use agreed business rules and verified records, with people checking exceptions and approving what goes out.
Before adding AI to any step, four questions are worth asking. What happens when it gets something wrong? Who checks, and when? Where does the data go, and on what terms? And can the process carry on without it if the tool changes or goes down?
Automation follows rules you set: when this happens, do that, the same way every time. AI reads and writes language, such as sorting emails or drafting a reply, and can be wrong. Many small business processes need automation first, with AI added only where reading or writing helps, and a person checking anything that reaches a client or touches money.
No. You need to understand your business, and someone needs to understand which parts of it AI can help with. That is a diagnosis, not a technology course.
In the firms I work with, no. It takes over reading, sorting and first drafts, which frees people for the work that needs judgement and relationships. Decisions and client conversations stay with people.
For helping one person write and think, often yes. To run part of a process reliably, AI needs to be placed inside a workflow with clear rules, checks and a person who approves what goes out.
It depends on the tool, the plan and its settings. Use business accounts, read how the provider treats your data, and only send it what the task needs. I set this out for every process where AI is involved.
Keep reading
Your business
Fifteen minutes, no charge. Bring one process and we will sort it into the three columns together.
15 minutes · No charge
Prefer to write? [email protected]