AI automation for small companies: where to actually start

Most AI projects in smaller companies fail for the same reason: they get chosen for how impressive they look, not for the hours they give back. A company of ten to two hundred people does not need a grand programme. It needs to stop doing three or four repeated tasks that someone still does by hand every week. Here are the ones that work, and the full approach is in AI automation.

Start with the boring things

Boring means repeated, and repeated means it can be automated with little risk. If the same person does a task every Monday following the same steps, it is a candidate. If it is different each time and requires judgment, it is not, however tedious it may feel.

The other reason to start there is that you can check the result. A repeated task has a known output: you know what last Monday's report looked like, so you can see whether the automatic one matches. With creative tasks nobody can say whether the output is right, and without that reference there is no way to tell if the project is working.

Starting small also protects your team's goodwill. If the first attempt works and gives someone their Monday morning back, the second one gets requested without you asking. If the first is enormous and fails, there will be no second, even when the problem was the scope rather than the technology.

Five things that work today

None of these is futuristic. Every one of them is a task someone in your company is doing by hand right now, probably without complaining, because it has always been done that way.

• Answering repetitive customer emails. Twenty percent of incoming questions are the same five questions. The draft can be prepared automatically and a person reviews it before it goes out.
• Assembling the Monday report. If someone copies numbers from three places into a spreadsheet every week, that part can run on its own and leave the person the interpretation.
• Loading data from invoices and delivery notes. Reading the document, pulling the fields, and putting them into the system is among the first things worth taking off someone's plate.
• Searching inside the company's own documents. Contracts, price lists, procedures, minutes: ask in plain language and get the answer with the exact document it came from. We cover it in internal assistants over private documents.
• Connecting all of that to the CRM, so what gets answered and what gets invoiced is recorded without anyone typing it twice. That is what AI agents for business do.

Notice that none of these replaces a whole person. Each one removes the mechanical part of someone's job, which is usually the part they like least anyway. That matters for how you present it internally. If the team understands the goal is removing repetitive work, they cooperate and point at the tasks that weigh the most. If they suspect the goal is cutting headcount, they stop telling you where the problems are, and that information is exactly what the project needs to succeed.

How to prioritise

The rule is short: start with the task that eats the most hours, not the one that looks best. That is usually something unglamorous in admin or customer service, not the conversational assistant that would look good in a board deck. The boring task returns time from the first week and does not need to convince anybody.

Before starting, measure. For one week, have each person note how long the repeated tasks take. Nothing sophisticated is needed: a shared sheet does it. That number is what will tell you whether the project worked, and without it you will have an opinion instead of an answer. If the task ends up needing something custom-built, that is software development and should be treated as such. Measuring almost always produces a surprise: the task everyone pointed at as the heaviest turns out to take two hours a month, while one nobody mentioned eats a full morning every week. That is why you measure before deciding rather than after, once budget is already committed to the wrong task.

What not to automate yet

• Anything sent outside the company without a person reading it. One email to a client with an invented figure costs more than all the hours saved.
• Decisions about people: hiring, firing, performance reviews. Those need judgment and accountability with a name attached.
• Numbers that end up in accounts, taxes, or a contract. The draft can be prepared automatically, but the signature stays human.
• Processes that change every month. Automating something unstable means rebuilding it every month: wait until it settles.
• Anything you cannot explain in two sentences today. If you cannot describe the steps, it is not ready to be automated.

If nobody can say how many hours a task costs today, nobody will be able to say whether automating it helped.

How to begin

Pick one task, measure what it costs today, and give it four weeks. That tells you whether this path is yours, and if it is not you have spent a month rather than a year. We work this way on our own products: A2GROUP is a product studio with twelve apps published on the App Store, among them AI Investing, and how we do it is described in inside A2GROUP. If your company is based in the Principality, see also AI automation in Andorra, and if you want to talk through your case, write to us.