Which business processes to automate (and how to tell which ones are ready)

Think about the last process you wanted to automate. Now answer this: would you describe it the same way as the person who carries it out every day?

It almost never matches. And that gap, which looks like a detail, is what separates an automation that works for three years from one that has to be taken apart within three weeks.

Before choosing a tool there’s a question that decides the outcome: which of your processes is genuinely ready. That’s what this is about.

What processes can be automated in a business

A process is a good candidate for business process automation when it happens many times, follows rules that can be written down and works with information that’s already in some system. The further it strays from those three traits, the more expensive it is to automate and the less it holds up over time.

In almost any SME, the candidates are the same:

• Invoicing and accounting. Reading invoices, extracting the data, reconciling it and getting it ready for the ERP. It’s the most common case because the volume is high and the format, ugly as it is, turns out to be predictable.
• Moving information between systems. Someone copies from an email into an Excel sheet, and from the Excel sheet into the CRM. Work that needs no judgement, and where most hours are lost without anyone counting them.
• First-line customer service. Sorting what comes in, answering the repetitive questions, escalating the rest. With an important caveat: answering repetitive queries, yes; replacing the conversation you’re hired for, no.
• Recurring reports. Monday’s report, the month-end close, the table someone builds by hand that always pulls the same data from the same places.
• Onboarding and offboarding. A new customer, a new employee, a new supplier: a sequence nobody remembers in full and that always stalls at the same point.
• Sales follow-up. Logging what happened, flagging what’s due, preparing the repetitive paperwork for a proposal.
• Document management. Naming, classifying and filing. Boring, constant, and the place where later nobody can find anything.

That list is the starting point, not the answer. The answer depends on how those processes work in your business, and that’s where almost every project goes off track.

The process you think you have isn’t the one you have

In any business, three versions of the same process coexist.

The one you think you have. The one you’d describe in a meeting. Clean, in five steps, with a clear owner for each.

The one that’s written down. The one in the manual, the onboarding pack, the quality procedure. It’s usually an old version of the previous one, and nobody has opened it since it was written.

The one that really happens. The one that includes the Tuesday exception, the customer who gets treated differently because they’ve been with you for twelve years, the decision someone makes based on a rule that isn’t written down anywhere, and the parallel Excel sheet kept on the side because the official system doesn’t cover their case.

Automating without looking at the third one means building on the first. The system goes live, and a week later the Tuesday exception turns up.

Automating a broken process doesn’t fix it: it multiplies it

This is the part that tends to surprise people.

When a manual process fails, it fails slowly and with witnesses. Someone copies a figure wrongly, someone else spots it when checking, it gets corrected. The error meets friction, and the friction acts as a brake.

When you automate that same process, you remove the friction. The error stops happening once a day and starts happening two hundred times without anyone looking. And because the result comes out fast and looks good, it takes weeks to be spotted.

Automation doesn’t judge the quality of what it runs. It just runs it faster, more often and without complaining. If what you’ve given it is badly thought out, you’ve bought a machine for repeating the error.

That’s why the order matters. It isn’t a methodological preference: it’s the difference between investing and spending.

How to know which processes to automate: the prior assessment

A process assessment is the work of understanding how a task really works before building anything: who does it, with what information, how many times, how much it costs and where it breaks. What comes out of it isn’t a technical proposal, but an informed decision about what to automate, what to fix first and what not to touch.

It isn’t an audit or a hundred-page report to justify fees.

It’s the answer to a very specific question: where is the money you’re losing, and which part of it could a system handle better than a person?

With and without an assessment: the difference in practice

Both paths start with the same sentence, “we want to automate this”, and end up in very different places.

The right-hand column isn’t slower overall. It’s slower to get going and much faster from the second month onwards.

The four questions that decide whether a process is ready

You don’t need any more. If a process answers them clearly, it can be automated. If any of them is left hanging, that’s where your groundwork is.

1. How many times does it happen and how much time does it cost each time? Without volume there’s no return. A perfect process that happens twice a month isn’t worth the effort of automating, however irritating it is.
2. Is the information where it needs to be? If the key piece of data lives in someone’s head, in an email from eight months ago or in a crooked scanned PDF, that’s the real problem. AI can’t guess what isn’t there.
3. Do the decisions follow a rule or depend on someone’s judgement? If they depend on judgement, that judgement has to be written down first. It’s the step nobody wants to take, and the one that creates the most value on its own: it turns knowledge that lived in one person’s head into something the business owns.
4. What happens when it goes wrong? A process without an exception plan isn’t ready. Well-built systems aren’t the ones that never fail: they’re the ones that fail visibly and recoverably.

If you answer all four quickly, you have a candidate. If you hesitate on two, you have work ahead of you.

Five signs you’re automating too early

• Nobody can tell you how much the process costs today. Without a starting point there’s no way to prove the improvement. And if you can’t prove it, the investment becomes an act of faith.
• Everyone on the team explains it differently. It isn’t poor communication. It’s that several processes are running at once under the same name.
• The conversation starts with the tool. “We want to put AI into customer service” describes a technology, not a problem. The useful question is what your team does every day that it shouldn’t be doing.
• The process changed recently and is still bedding in. Automating something unstable freezes a provisional version. Wait until it stops moving.
• The urgency comes from outside, not from within. A competitor announced it, it was on LinkedIn, someone asked about it at the board meeting. None of those is an operational reason.

None of these signs means “don’t automate”. They mean “not yet, and here’s what’s missing”.

What changes when the order is right

At JGM Consultores, a tax advisory firm, the candidate process was handling requests from the Spanish Tax Agency: case files that consumed entire working days and arrived in bursts, without warning.

The groundwork didn’t start with which tool to use. It started by measuring two things: how much time a case file really took from start to finish, and which parts were professional judgement and which were moving information from one place to another.

With that on the table, the result was measurable:

• From 24 hours to 2 hours per case file.
• From €480 to €10 in operating cost per process.

Look at what made those numbers possible. It wasn’t the AI model: it was having first separated out which part of the work was a professional’s judgement (which stays with the professional) and which part was shuffling data around, which is what got automated.

Without that prior cut, the project would have tried to automate the whole case file. And it would have failed precisely in the part that matters.

What you shouldn’t automate yet

Some processes don’t take well to automation, and it’s worth saying so before charging to try.

What rarely happens. If it comes up three times a year, document it properly and keep doing it by hand.

What calls for professional judgement with responsibility behind it. A system can prepare a good part of a professional opinion. The signature still belongs to whoever answers for it.

What doesn’t have written rules yet. If the judgement isn’t written down, the first deliverable isn’t an automation: it’s writing down the judgement. Which is valuable in itself, because the business stops depending on that person being available.

The relationship with the customer when it is the product. Automating the administrative follow-up, yes. Automating the conversation you’re hired for, no.

Saying this out loud costs work in the short term and saves it in the medium term. A project that’s stopped in time works out cheaper than one that’s delivered and never used.

How to automate business processes: choosing your first one this week

You don’t need to hire anyone to get started. You need two hours and a bit of honesty.

1. Pick a process that irritates you. Irritation is a decent indicator: it usually marks where there’s volume and friction at the same time.
2. Sit down with the person who does it, not the person who supervises it. And ask them to show you while doing it, not to explain it. The difference between what’s said and what’s done is exactly what you’re looking for.
3. Note every point where someone decides something. Those are the knots. Each one is a rule that has to be written down, or a decision that stays with a person.
4. Put a number on it. Times per week, minutes each time, hourly cost of whoever does it. Even if it’s rough. A weak number is infinitely better than none.
5. Decide what’s surplus. Often the conclusion isn’t “automate this”, but “this step shouldn’t exist”. Removing a step always works out cheaper than automating it.

If by the end you have the real process mapped out, the decision points marked and a figure on top, you already have more than most businesses buying AI right now.

Frequently asked questions

What processes can be automated with AI?

The ones that combine high volume with poorly structured information: reading invoices or emails and extracting data, sorting what comes in, drafting repetitive documents, reconciling information between systems. The difference from traditional automation is that AI tolerates changes in the input format. What doesn’t change is the underlying requirement: there has to be a rule or a written criterion behind it.

What are the four types of automation?

The most widely used classification distinguishes four levels: automating individual tasks (a macro, an email rule), automating complete processes from start to finish, robotic process automation or RPA (a program that mimics a person’s clicks on systems that have no integration) and intelligent automation, which brings in AI to interpret information and decide within a set of rules. In practice the label matters little: what decides the outcome is whether the process was ready, not which box you put it in.

Where do I start automating my business?

With a process, not a tool. Pick one that happens many times, measure it as it is today and separate which part is professional judgement and which part is moving information around. Starting with the tool is what produces projects that get abandoned: a solution is bought and then something is found to apply it to.

How long does it take to analyse a process before automating it?

For a specific process, between a few days and two weeks depending on the volume of information and the team’s availability. What drags it out is almost never the analysis: it’s gathering scattered data or reconstructing criteria that were never written down.

Can I automate without analysing anything if the process is very simple?

Yes, and sometimes that’s the right call. If the process always happens the same way, has no exceptions and the information is already structured, it can be sorted out in one conversation. The problem comes when that simplicity is assumed without being checked.

What’s the difference between automating and using AI agents?

Automating means running a defined workflow: always the same steps, in the same order. Using AI agents means bringing in systems that can interpret information, decide the next step within a set of rules and coordinate several tasks. The prior analysis determines which of the two fits, because putting agents on a process that only needed an automation means paying over the odds.

In short

Automating is the easy part. Deciding what to automate, with what information and how far is where the project is won or lost.

The right order isn’t a consultant’s formality. It’s what makes sure that, two years from now, the system is still working and you can justify what it cost with numbers.

If you want to start there, describe a process in your business in the free AI assessment and we’ll tell you what we see: whether there’s a case, what kind of solution fits and what would be needed first. No commitment, and if we don’t see a real fit, we’ll tell you.

And if you’d rather have the process analysed, designed and implemented from start to finish, that’s how we work in AI consultancy.