Most conversations about AI start in the wrong place. They start with the tool: “Should we get Copilot?”, “Can ChatGPT do our admin?”, “What about agents?” The better starting point is your own week. Where do capable people spend hours on work that follows the same pattern every time, with only a little judgement sprinkled on top?

That is where automation, with or without AI, earns its keep. In a business of around 30 people, there are usually three to five such workflows. Find them, fix one properly, and you will learn more than any product demo can teach you.

The sweet spot: rules plus a little judgement

Pure rules-based work (copy this field to that field, every Friday) has been automatable for years. Pure judgement work (should we take on this client?) belongs with people. What has changed is the middle ground: tasks that are mostly routine but need someone to read, interpret or summarise something along the way.

Modern AI models are good at reading messy input, such as an email, a PDF or a scanned invoice, and turning it into structured information that ordinary automation can then act on. That is the combination that pays off.

Here are the workflows we see most often in South African SMEs.

Supplier invoices into the accounting system

An accounting practice in Centurion receives supplier invoices by email in a dozen formats. Someone opens each one, reads the supplier, date, amounts and VAT, and keys it into the accounting package. AI can read the invoice, extract the fields, match the supplier and prepare a draft entry. A person checks the draft and approves it. The typing disappears; the checking remains.

Onboarding and offboarding in Microsoft 365

When someone joins, they need an account, the right licence, the right groups, a mailbox, Teams channels, a laptop and MFA set up. When someone leaves, all of that must be reversed promptly, including mail forwarding and file ownership. This is a checklist that people forget steps in, and a forgotten offboarding step is a security problem. A structured, automated workflow triggered by a simple form makes it consistent every time. (If you are planning AI tools on top of Microsoft 365, our Copilot readiness checklist covers the groundwork.)

Routing enquiries

The general inbox at an engineering firm in Pretoria East receives quote requests, supplier queries, CVs, invoices and the occasional complaint. AI can read each message, classify it, pull out the key details and route it to the right person with a short summary. Nobody has to triage the shared mailbox at 07:30 any more.

Compiling recurring reports

Monthly management packs, project status updates and client reports often involve pulling numbers from three systems, pasting them into a template and writing a few paragraphs of commentary. Automation can gather the data; AI can draft the commentary for a manager to edit. The manager’s time moves from assembling to reviewing.

Drafting documents from templates

Engagement letters, proposals, standard contracts and compliance letters follow templates with client-specific details. AI can produce a first draft from the template and the client file. A qualified person still reviews and signs off, but starts from a solid first draft rather than a blank page.

How to spot a good candidate

You do not need a consultant’s framework. Three questions will rank your options:

How often does it happen? Daily or weekly tasks beat quarterly ones.
How many minutes does each instance take? Multiply by frequency to get the real time cost per month.
What does an error cost? A wrong VAT figure, a missed offboarding step or a misrouted complaint each has a price, in money, risk or reputation.

Frequency multiplied by minutes gives you the time you could save. The error cost tells you how much the consistency is worth on top. A task that happens forty times a month, takes ten minutes each time and occasionally goes wrong is a far better candidate than an impressive-sounding project that runs twice a year.

A useful test: if you can write the steps down on one page, and a new staff member could follow them after a week of training, it is probably a good candidate for automation.

Where automation doesn’t pay off

Being honest about this saves money and frustration.

Rare tasks. If something happens four times a year, the effort to build, test and maintain an automation will rarely be repaid. A good checklist is usually enough.
High-stakes decisions with no review. AI models can be confidently wrong. They can misread a figure, invent a detail or miss context a person would catch. Anything with legal, financial or safety consequences must have a qualified person checking the output before it takes effect.
Messy or inconsistent data. If your client list lives in three spreadsheets with different spellings, automation will faithfully reproduce the mess at speed. Clean the data first, or pick a different workflow.
Processes nobody agrees on. If three people do the same task three different ways, automating it simply picks one of those ways by accident. Agree the process first.
Work that is really about relationships. A client phoning with a problem wants a person. Use automation to give that person the context faster, not to replace the conversation.

Human in the loop is the default, not a compromise

The safest and most productive pattern for an SME is simple: the automation does the reading, sorting and drafting; a person confirms before anything leaves the building or changes a record.

This is not a sign that the technology is immature. It is how you build trust in it. Over time, as you see which suggestions are reliably right, you may decide that some low-risk steps can run without a check. That decision should be made deliberately, based on evidence from your own business, not assumed on day one.

We follow the same rule in our own operations. You can read how we use AI inside our helpdesk, including the things we deliberately do not let it do.

Start with one workflow and measure it

The businesses that get real value from automation tend to follow the same pattern:

Pick one workflow using the frequency, minutes and error-cost test above.
Write down how it works today, including the exceptions people handle without thinking.
Measure the current state. How long does it take? How often does something go wrong? Even a rough tally over two weeks is useful.
Build the automation with a review step, using the tools you already have where possible, such as Microsoft 365, your accounting package and your helpdesk.
Run it alongside the old way briefly, so you can compare results.
Measure again after a month. Did it save the time you expected? Did errors go down? What did people actually think of it?
Only then choose the next workflow.

One working automation that your team trusts is worth more than five half-finished pilots.

Don’t forget the foundation

Automation touches your email, files, accounting system and user accounts. If those are poorly secured, automation simply gives problems a faster route. Before connecting AI tools to business data, make sure the basics are in place: MFA on every account, sensible access permissions, tested backups and some documentation of how things are set up. It is also worth agreeing what staff may and may not put into AI tools; our guide to POPIA and AI tools is a good starting point.

That is why we approach automation as a managed-IT provider first. The same Discovery, Stabilise, Document, Improve and Operate sequence we use when we take over a client’s IT applies here: secure the essentials, then improve in a structured way.

Where to begin

If you suspect your team is spending too many hours on repetitive work but are not sure which workflow to tackle first, a short outside look often helps. Our free Automation & Risk Check is a 30-minute review of where automation would genuinely pay off in your business, plus a look at the basics: backups, security and documentation. You receive a one-page report with what we found. There is no obligation, and you are welcome to act on it yourselves.

Where AI automation actually pays off in a 30-person business