← All posts
Automation & AIAI at work

AI Automation Tools for Business: What's Worth Using

Most of the useful AI is already inside software you pay for. The trick is knowing which job to give it.

In a typical week an office manager might get two emails selling an AI assistant, a webinar invite about “agentic automation” and a note from the accounting software saying it now has AI built in. Meanwhile someone on staff has been pasting customer emails into a free chatbot to write replies, and nobody decided whether that was a good idea.

The tools are real, and some of them save a lot of time. But the useful question is narrower than “should we use AI”: which repeated job in this office involves reading or writing something, and what’s the safest way to hand that part over?

What makes a tool an AI automation tool

Ordinary automation follows rules you write. When an email arrives with “invoice” in the subject line, save the attachment to the accounting folder. It works perfectly until a vendor sends an invoice with the subject “Payment due”, and then it quietly does nothing.

An AI step reads the content instead of matching a word. It can look at that email, see that the attachment is an invoice, and pull out the vendor, the date and the total, even though the subject line says something else. That’s the whole difference. AI is good at the messy input step: reading documents, sorting email, summarizing a long thread, drafting a first version of a reply. The rest of the process (who approves, where it’s filed, which system gets updated) is still best handled by plain rules, because rules do exactly the same thing every time.

Four kinds of tool you’ll be offered

AI inside software you already have

Microsoft, accounting packages, CRMs and help desks are all adding AI features to their existing products. Microsoft’s AI Builder, for example, adds prebuilt and custom AI models to Power Automate and Power Apps, and Microsoft’s own documentation gives automating document processing as an example of what it’s for (Microsoft Learn). This is often the cheapest place to start, because the data is already there and the sign-ins are already managed.

Workflow platforms with AI steps

Tools like Make and Microsoft Power Automate connect your apps and run a sequence of steps when something happens: a form is submitted, an email arrives, a file lands in a folder. Most now let you drop an AI step into the middle, such as “classify this email” or “summarize this ticket”, with ordinary steps either side of it.

Document capture

Document management systems such as DocuWare read invoices, forms and letters as they come in, recognize the important fields and file the document without anyone retyping it. This kind of AI has been around longer than the chatbots and is usually the most predictable, because the job is narrow: find these six fields on this kind of page.

Chat assistants and AI workspaces

ChatGPT, Claude and their competitors are general-purpose: write, summarize, analyze, explain. They’re useful to almost everyone in an office. The catch is the account. A personal account is outside your company’s control, and you can’t see what goes into it or switch it off when someone leaves. A company workspace with managed sign-in is a different thing, and the difference matters more than which model it runs. We cover that in more detail in Is It Safe to Put Company Data Into ChatGPT?

Where AI helps and where a rule is better

The job Better with AI Better with a plain rule
Sorting incoming email Working out what an email is about when senders word things differently Routing everything from one address to one folder
Invoices Reading vendor, date and total from many different layouts Sending every invoice over a set amount to the owner for approval
Customer replies Drafting a first version for a person to check Sending the same confirmation every time an order is placed
Reports Summarizing what changed and why Building the same table every Monday morning

A good automation usually mixes both. AI reads the invoice, a rule checks that the total matches the purchase order, and a person approves anything that doesn’t match.

AI is good at the messy input step: reading documents, sorting email, summarizing a long thread, drafting a first version of a reply.

Six questions before you buy any AI tool

  1. Where does our data go? Which company processes it, in which country, and is it kept after the request?
  2. Is our data used to train the vendor’s models? The answer should be a plain no for business data, in writing.
  3. Does it connect to what we already use? A clever tool that needs someone to copy and paste into it just moves the manual work.
  4. Who owns the account? If it’s tied to one employee’s personal email, it leaves when they do.
  5. What happens when it’s wrong? It will be, sometimes. Can a person check its work before anything is sent or paid?
  6. Who looks after it? Every automation needs someone to notice when it stops working, usually after a software update.

If a vendor can’t answer the first two clearly, that’s your answer.

Do the math on one job

Pick one repeated job and put numbers on it. These are example numbers; use your own. Say someone spends 20 minutes every morning sorting a shared inbox and forwarding emails to the right person. Over about 250 working days, that’s roughly 83 hours a year, or more than two working weeks. If an AI step sorts the obvious 80 percent and leaves the unclear ones for a person, most of that time comes back.

That calculation also tells you when not to bother. A job that takes ten minutes once a month isn’t worth building anything for, AI or otherwise.

When to hold off

AI tools are a poor fit when every answer has to be exactly right and nobody will check it (payroll figures, legal deadlines, anything sent to a regulator), when the input is already structured and a simple rule would do, or when the process itself is a mess. Software following a muddled process makes the same mistakes, only sooner. Fix the steps first, then decide which ones a machine should do.

It’s also worth writing down a short AI use policy before tools spread further. What to Put in an AI Use Policy has a plain template.

Where we come in

We build automation for small and mid-size businesses using Make, Microsoft Power Automate and DocuWare, and add AI steps where they earn their place, with a person approving anything that matters. We also run a private AI workspace for teams who want staff off personal accounts. If you’d like to see how AI changes a whole process rather than a single task, read How AI Is Changing Business Process Automation, or tell us the one job that eats your team’s mornings and we’ll tell you whether it’s worth automating.

More on Automation & AI

Let's build your foundation.

Tell us where your business is today and we'll help map the next step.