The useful version of AI automation in a small business is unglamorous. It reads things that arrive, pulls out what matters, prepares the next piece of work, and hands it to a person to check. It doesn’t run the company. Done well, it removes the copying, sorting and first-drafting that fills the gaps between real work.
Jobs it handles well
Each of these has a recognisable input, a step that repeats, and a person who can judge the result.
- Enquiry triage. Emails and form submissions are read, sorted by service and urgency, and matched with a draft reply. Someone on your team edits or sends.
- Documents into systems. Supplier invoices, order forms or delivery notes are read and the fields typed into your accounting tool, ERP or spreadsheet. Anything the system isn’t sure about is flagged instead of guessed.
- Product and listing drafts. Descriptions are drafted from your spec sheets and brand rules. A person checks every claim before it’s published. Google’s own guidance on AI-generated content says to fact-check and review it, including titles and meta descriptions.
- Internal answers. Staff ask “what’s our returns policy for custom orders?” and get an answer drawn from your own documents, with the source shown.
- Recurring summaries. The weekly numbers someone currently copies into a report are collected and summarised, so that person reads the report instead of assembling it.
Jobs to keep away from it
- Decisions with real consequences for people. Hiring, credit, pricing a specific customer, refusing a claim. The GDPR’s Article 22 gives people the right not to be subject to decisions based solely on automated processing when those decisions have legal or similarly significant effects. A person should own them.
- Work nobody can check. If there’s no one who can tell whether the output is right, automation just hides the errors.
- Processes that aren’t defined yet. If three people do the task three different ways, agree one way first.
- Rare tasks. Something done twice a year rarely pays back the time to build and test it.
Score your candidates
List the tasks your team complains about. For each one, write down:
- How often it happens, per week.
- How long it takes each time. Measure it for a week rather than guessing.
- How consistent the input is. Same form every time, or free-text emails in four formats?
- Whether you could write the rules down for someone new.
- What a mistake costs. A typo in an internal note, or a wrong invoice amount?
- Who would review the output.
A fictional worked example. A five-person wholesale distributor receives about 40 order emails a week. Each takes around six minutes to read, check stock and type into the order system. That’s four hours a week. The emails vary, but they always contain product codes and quantities. The rules are clear. A mistake means a wrong delivery, so a person should confirm each order before it’s created. That’s a good first candidate: frequent, consistent enough, and easy to review.
Compare it with “write our monthly newsletter.” It happens once a month, the input is whatever’s new that month, and quality depends on judgement. Worth doing, but not first.
Use real numbers from your own business. The point of the arithmetic is to compare tasks with each other, not to promise a saving.
Check the data before you build
Find out where the inputs live and whether the systems involved can be connected: an inbox, a store, a CRM, an accounting tool. Some have usable APIs; some only export files.
If the workflow sends customer or employee personal data to an AI provider, that provider is processing it on your behalf. Under GDPR Article 28 you need a data processing agreement with them, and you should know where the data is processed and whether it’s retained.
Start narrow
Pick one task. Run the automated version alongside the manual one for a few weeks and compare the outputs. Track two numbers: time spent, and errors caught at review. Switch over only when the review step is mostly confirming, not correcting. Then move to the next task.
If you’re unsure whether a task needs a simple workflow or something more autonomous, AI agents vs chatbots sets out the difference.
That’s the way we scope AI workflows: one contained process, a visible approval step, and a team that can see what the system did. Tell us which task you’re tired of doing by hand and we’ll say honestly whether it’s worth automating.