What AI can automate in your business (and what it can't)
On this page
- Automating doesn't mean replacing anyone
- What AI adds to a traditional automation
- How to tell whether a task can be automated
- What can be automated, department by department
- Administration
- Customer service
- Sales and marketing
- Management
- Production and quality
- What not to automate (or not without a person)
- What makes the difference
- Where to start in your company
- What I take away
Automating doesn't mean replacing anyone#
When a company hears "automating with AI", it usually pictures one of two things: a robot doing a person's job, or an expensive fad nobody quite knows what to do with. The reality is somewhere else, and much more ordinary: getting rid of the repetitive tasks that eat up hours every week and that nobody wants to do.
Copying the details of an invoice into the accounting software. Answering the same customer question for the fifth time. Putting together Monday's report from three spreadsheets. None of that adds value in itself, but it takes time and attention, and that's exactly where mistakes creep in.
This post isn't a tutorial. It's a map: what an ordinary company can automate today, department by department, what's best left to a person, and how to spot where to start in yours.
What AI adds to a traditional automation#
Automations have been around for decades: "when an order comes in, send a confirmation email". They run on fixed rules, which is why they only work when information always arrives the same way, in the same format and in the same place.
AI adds something rules can't do: read and understand. It can read an email written the way each customer writes, an invoice with each supplier's layout or a voice note, and pull out what matters. With that, many tasks that used to be done by hand because "every case is different" can now be automated.
The key is the last step in the figure: a good automation doesn't remove the person, it puts them where they add value. They review what matters, decide what has consequences and leave the mechanical part to the machine.
AI makes it possible to automate tasks that used to need someone to read and understand; the person stops copying and starts reviewing and deciding.
How to tell whether a task can be automated#
Before thinking about tools, look at the task. These four signs usually point to a good candidate:
- It repeats. Every day or every week, always with similar steps.
- It moves information from one place to another. From email to a spreadsheet, from a PDF to the invoicing software, from a form to the CRM.
- It deals with text, documents or data. Emails, invoices, delivery notes, orders, queries, reports.
- It can be reviewed. If the machine gets it wrong now and then, someone can spot it before it causes a problem.
If a task ticks all four, it can almost certainly be automated, fully or in part.
What can be automated, department by department#
These are common ideas, not a closed list. Not all of them fit every company, and in each case the details matter more than the idea.
Administration#
- Reading invoices and delivery notes and passing their details (supplier, date, amounts, taxes) to accounting or the ERP, flagging only what doesn't add up.
- Sorting incoming email: orders, invoices, queries, complaints, with each one reaching the right person.
- Generating documents from data: certificates, standard contracts, reports or any document that's filled in by hand today with information that already lives somewhere else.
Customer service#
- Answering frequent questions on the website or WhatsApp, with answers based on your documents (opening hours, prices, terms) and not on what the AI "thinks" it knows.
- Sorting and prioritising incidents, so the urgent ones don't wait behind the trivial ones.
- Summarising the conversation before handing it to a person, so they don't have to read all of it.
Sales and marketing#
- Drafting proposals from the notes of a meeting.
- Following up on contacts that were left without an answer.
- Adapting the same content for the website, social media and the newsletter, keeping the brand's voice.
Management#
- Automatic reports: Monday's summary of sales, outstanding payments or production data, without stitching spreadsheets together by hand.
- Meeting summaries with the decisions and who owns each task.
- Asking the company's documents: "what does the contract with this supplier say about deadlines?". When the documents are sensitive, this can be done with an AI that runs on your own servers, without sending anything to third parties.
Production and quality#
- Recording production data without typing it in, and flagging when a value is out of the ordinary.
- Keeping quality and traceability paperwork up to date, which tends to be as necessary as it is tedious.
What not to automate (or not without a person)#
Knowing what not to automate matters as much as all of the above. These are the cases where you should slow down:
- Decisions about people. Screening CVs or assessing employees with AI isn't just another automation. The EU AI Act treats as high-risk the systems intended for recruitment, including those used "to analyse and filter job applications", and those that evaluate workers' performance. There are specific exceptions, for example when the system only performs a narrow procedural task, but never when it profiles people. The obligations it sets, both for whoever develops the system and for the company that uses it, will apply from 2 December 2027. It's worth bearing in mind already. I covered it in more detail in the post on the AI Act.
- Answers with legal, tax or medical consequences that go out without anyone reviewing them. AI can prepare the draft; signing it is a person's job.
- Sensitive data in public tools. Customer, health or employee data shouldn't end up in just any AI service without knowing what it does with it.
- Processes nobody is clear about. If everyone does the task differently, automating it just makes the mess faster. First sort out the process, then automate it.
Building an automation that works on demo day is easy. The hard part is keeping it working when an invoice arrives in an odd format, when a connection fails or when the AI gets it wrong, and making sure someone finds out in time. That's the difference between something that saves hours and something that quietly creates problems.
What makes the difference#
The tools for automating are within anyone's reach, and that's good news. But the result depends less on the tool than on everything around it:
- Understanding the process before touching it. Often what saves the most time isn't the AI, but removing a step that wasn't needed.
- Fitting in with what you already use. Your email, your spreadsheets, your invoicing software. An automation that forces you to change everything doesn't get used.
- Security and privacy from the start: which data leaves the company, where it goes and who can see it.
- Error handling: what happens when something fails and how the person in charge finds out.
- Maintenance: suppliers change their invoices, tools get updated and processes evolve.
Where to start in your company#
To find your first automation, answer these questions thinking about an ordinary week:
- Which task do you do every week that nobody wants to do?
- Where is information copied by hand from one place to another?
- Which customer question gets answered over and over again?
- Which mistake keeps coming back, and at which step does it usually happen?
- Which report or document is prepared the same way every time?
The best first automation is usually small, frequent and easy to review. It's the one that shows soonest whether it pays off, and the one that lets the team trust the next ones.
What I take away#
- Automating with AI removes repetitive tasks, not people. The person goes from copying to reviewing and deciding.
- AI adds the ability to read and understand, and that opens the door to tasks that used to be done by hand.
- Not everything should be automated. Decisions about people, anything with legal consequences and sensitive data need care and review.
- Start with something small and frequent, and check whether it pays off before going for the big things.
I've spent years building automations, agents and chatbots for businesses, from generating official PDF documents with a QR code out of a spreadsheet to assistants that answer using the company's own documents. And I always start the same way: with an assessment to understand the process properly before touching anything.
References
- Anexo III: Sistemas de IA de alto riesgo a que se refiere el artículo 6, apartado 2AI Act Service Desk (Comisión Europea) · ai-act-service-desk.ec.europa.eu
- Artículo 6: Reglas de clasificación de los sistemas de IA de alto riesgoAI Act Service Desk (Comisión Europea) · ai-act-service-desk.ec.europa.eu
- Artículo 113: Entrada en vigor y aplicaciónAI Act Service Desk (Comisión Europea) · ai-act-service-desk.ec.europa.eu
Share
Need help integrating AI into your business?
I help you identify real automation opportunities and implement tailored solutions. No smoke, no unnecessary jargon.
Let's talk→