What AI actually does for small businesses, and when you don't need it

A lot of what small businesses are told about AI runs it together with ordinary automation, which is a different thing. And a lot of admin problems can be solved by a setting in software you already pay for. The difference matters, because the three cost different amounts, go wrong in different ways and need different kinds of checking.

This guide helps you sort one task at a time. For each task you repeat, the question is the same: does it need a better setting in the software I already have, a plain rule that runs on its own, or AI that prepares a draft for a person to check?

Three answers to the same question

• A setting you already have. The same message, reminder or filing step, every time, inside one tool. Email templates, inbox rules and invoice reminders belong here.
• Ordinary automation. A clear trigger and the same steps each time, often across two tools that do not talk to each other. No judgement is needed, and none is wanted.
• AI assistance. The input varies and has to be read: a rambling enquiry, rough notes, a long document. AI can summarise it or draft a reply. A person decides what happens next.

Most real workflows mix the three. A rule files the message, AI drafts the reply and a person approves it. The point is to use each where it fits, and not to pay for AI where a rule would do.

Start with the software you already pay for

Before buying anything, look at the settings and help pages of what you use now. Useful admin features are easy to miss.

• Email templates. Gmail lets you save a message that does not change often as a template and reuse it. Google's help page says templates can only be turned on and used from Gmail on a computer. See Google's guide to Gmail templates.
• Inbox rules. Outlook rules can automatically move incoming email to other folders, change its importance or delete it, based on criteria you set. See Microsoft's guide to Outlook rules.
• Invoice reminders. QuickBooks can send email reminders up to 90 days before or after an invoice's due date, only for invoices you have already emailed, and you can edit the wording of each one. See QuickBooks' guide to invoice reminders.

Other accounting, booking and job tools have their own versions of these. If one of them does the job, you have spent nothing extra and there is nothing new to check.

Not needing AI is a perfectly good answer. In the government's AI adoption research, a telephone survey of UK private sector businesses with at least five employees, carried out between February and May 2025, the most common reasons for not adopting AI were a lack of identified need and limited AI skills.

When a plain rule is enough

Ordinary automation joins steps together. Zapier, one of several tools that do this, describes a workflow as a trigger, the event that starts it, followed by actions, such as creating a contact, sending a message or updating a record. See Zapier's explanation.

An illustrative example: when someone fills in the enquiry form on your website, a new row appears in your job list and the person who handles enquiries gets a notification. Nothing in that sequence needs to understand the message. It does the same thing every time, which is exactly what you want from it.

A rule is usually the right choice when:

• You can write the steps down as "when this happens, do that".
• The result should be the same every time.
• A mistake would be easy to spot, such as a missing row or a reminder that did not go out.

Rules still need looking after. A connected app changes, a password expires or a form gains a field, and the rule can quietly stop. Ask for an alert when a step fails, and agree who fixes it.

Where AI adds something a rule cannot

AI earns its place when the input is different each time and someone would otherwise have to read it and write something back. For example:

• Summarising a long customer message and listing what is missing from it.
• Turning rough notes into a tidy record or a first draft of a description.
• Drafting a reply in your usual tone from facts you have already confirmed.

Your existing software may already offer this. Gmail's "Help me write" can generate a draft or refine text you have written, with an eligible Google Workspace or Google AI plan. On a personal Google account it is limited to the US. See Google's guide to drafting with Gemini in Gmail. Outlook has a similar feature, Draft with Copilot, and Microsoft's own instructions are to review the draft and edit it as needed before sending. See Microsoft's guide to drafting with Copilot in Outlook.

AI is the wrong tool for working out a price, a quantity or a date that must be right, and for anything that should give the same answer every time. Keep those in the software built for them, and use a rule or a setting where one will do.

What a person must still check

The National Cyber Security Centre's AI guidance warns that AI can get things wrong and present incorrect statements as facts. A fluent draft is not evidence that it is right.

Customer details raise a data protection question too. The ICO's guidance on generative AI says that when using an AI tool, organisations must check that outputs are accurate and correct anything inaccurate, and that personal data shared with the provider of an AI tool must be shared in line with data protection law. Check how a provider handles your data before you put real customer details into it.

A workable set of checks for a first trial:

• AI prepares drafts. It does not send messages, change prices or book anything.
• The person checking sees the original beside the draft.
• Missing information is flagged, not filled in with something plausible.
• Prices, scope, technical advice and promises stay with a named person.
• There is a manual route if the tool fails or gives an unclear answer.

What it costs, including your time

Whichever answer you reach, ask for the costs separately: setup, subscriptions, any AI usage charges, the connections between tools, ongoing maintenance, and your own time checking the output. The last one is easy to leave out, and it can decide whether the change is worth making.

At Appening Now, an automation sprint, which connects one recurring task across up to two existing tools, starts at £1,250 + VAT, one-off. Platform subscriptions and usage are separate, and so is any AI usage. A bespoke business tool starts at £2,500 + VAT, one-off, for a defined small scope. That is for work that does not fit the tools you have, not a first step. See current pricing and what each option includes.

Test it on one task first

Pick one task you repeat, one person to check the output and a fixed trial period. Write down how the task goes today, then try the change. Each time it runs, record:

• Time spent, including checking and correcting.
• What had to be corrected, especially anything invented.
• Whether the next step happened without chasing.

A quiet week or a handful of examples can make the result look better or worse than it is, so compare like with like. If checking takes as long as doing the task yourself, change the approach or stop. If it works, extend it to the next step in the same workflow.

For a worked example in one sector, AI for trade businesses follows an illustrative enquiry from arrival to a checked reply.

Where to start

Bring one task you repeat and a list of the software you already use. That is enough to work out whether the answer is a setting, a rule or AI. A free AI audit is a conversation about how your business works and where each of the three fits. If you already know the workflow you want, plan your project.