Automating your business with AI: where to start

Automating your business with AI starts with a single repetitive task. You measure the time it takes, you define the rule, then you compare the result before and after.

The tools and the next tasks come later. The first goal: check that an automation saves you time without shifting the work onto a new round of checks.

What the problem looks like

An email comes in with a quote request. Someone reads the attachments, copies the quantities into a spreadsheet, works out the price and prepares the reply.

If that person is busy elsewhere, the file waits. The client doesn’t see the data entry or the checks. All they see is how long the reply takes.

The same work turns up in other tasks:

• Chasing an invoice after checking that it is still unpaid.
• Copying an incoming request into the sales tracking file.
• Putting a report together with figures from several tools.

The problem comes from how often they happen. Each task ties up a qualified person, when part of the work simply consists of moving or checking a piece of information.

An absence sometimes reveals another problem: nobody knows where the file stands. The procedure exists in one person’s head, with no record the rest of the team can use.

Add the interruptions: leaving one file to deal with a reminder, then picking up the thread of the first job. That time is hard to estimate from memory. A week of observation gives a more useful basis than a hunch.

You don’t need to map the whole business to get started. Pick a frequent, predictable task whose result you can observe. It will be your first trial to decide whether the method is worth extending.

What doing nothing costs

Start from the time actually spent. Over a representative week, note the number of requests and the minutes spent on each one. Include the data entry, the checks and the usual corrections.

The formula: number of requests × time per request ÷ 60 × working weeks × fully loaded hourly cost. The result estimates the cost of the time tied up over a year.

The table below is a simulation. The assumptions of 47 weeks and €35 an hour are the basis I use in my own estimates. Replace each value with your own, especially if your business is seasonal.

This amount is not a promised saving. Part of the time will still go on exceptions and sign-offs. You also have to count the set-up, the tools and the upkeep of the automation.

To decide, compare the cost of the project with the time you can really get back. If the task rarely comes up, or if each file calls for a different judgement, the automation can cost more than it brings in.

Before you start, also note who will pick up the exceptions. Ten requests handled automatically are not progress if the eleventh stays stuck without anyone noticing. The time spent monitoring is part of the comparison.

Do the same exercise on two or three tasks, keeping them separate. You’ll see which one deserves the first trial. Avoid adding up hours that overlap, for example a check already counted in handling the quote.

The first investment can simply be documenting the procedure: where the request arrives, which information is often missing, who signs off and what result has to be recorded. That work stays useful even if you finally decide to keep the task manual.

Keep a second measure alongside time: how often the work has to be redone. If the draft is prepared faster but needs three extra corrections, the apparent gain can vanish. Look at the whole journey, from the request received to the file that is actually finished.

Time freed up doesn’t automatically cut a cost. Someone can spend it replying faster, looking after more clients or catching up on a backlog. Decide what you want to do with it: that concrete result is what lets you judge whether the project is worth it.

Finally, set the success criterion before the trial: fewer minutes per file, less re-keying or a shorter response time. Compare weeks with a similar volume so you don’t mistake a drop in activity for progress.

What we put in place

For a quote request, the journey can fit into four steps. Each one needs a clear input, an expected result and a stop rule if a piece of information is missing.

AI can help read an email or prepare a draft. A fixed rule is often enough to copy a field, apply a known price or trigger a reminder on a set date. Not every step needs AI.

I use n8n, among other tools, to connect these steps. But the choice of tool comes after the rule: a vague instruction stays vague once it is automated.

What breaks, and how to avoid it

Two pitfalls come up most often.

These two pitfalls apply to any automation, even the simplest. If you go further with an agent that chains several actions without supervision at each step, I explain what an AI agent really does for a business in a separate article.

The risks also change in scale at that stage: I have listed them in an article on the risks to know before deploying an AI agent.

Where to start if you do it yourself

1. Pick a task. Take the most repetitive and predictable one in your week. Name the person who will be able to check its result.
2. Measure the starting point. Over a representative week, note the frequency, the time taken and the exceptions. Keep this measure for the comparison.
3. Write the rule. Describe the trigger, the steps, the information needed and the moment a person takes back control.
4. Test under supervision. Plan a trial period, for example two to three weeks, with every output signed off. How long it takes depends on the number of cases to observe.
5. Compare before extending. Count the time spent checking and correcting. Then decide whether to keep the rule, simplify it or stop it.

A rule you can’t explain in a few lines still needs scoping. Break it down into a smaller task before building anything.

Pick a no-code tool for this first trial rather than custom development. A tool like n8n, whose installation I have already covered in a practical guide, lets you test a first rule without writing any code.

Accept that some cases will keep coming back to you. Handling the exceptions is still human work. The first trial should make day-to-day work simpler, with a result you can check and a recovery procedure everyone understands.

Frequently asked questions

Do I need a developer to automate a task in my business?

Not for a well-defined first task. Most of today’s tools can be run without writing code. What takes experience is the scoping: choosing the right task, writing the rule and deciding where to put the human check. A developer becomes useful when the rule touches several systems at once.

How long before I see a result?

Allow two to three weeks to scope a task, write the rule and run it under supervision. The time saved becomes visible once it runs without systematic rework, which assumes you measured the time spent beforehand, otherwise there is nothing to compare.

Which task should I start with?

The most repetitive and predictable one, not the most visible or the most complex. A task that comes up several times a week, with few special cases, is quick to scope and gives a measurable result. A rare task, or one full of exceptions, costs more to scope than the time it saves.

Can an automation make mistakes?

Yes, like any automated system. That is exactly why the human check stays a step of the journey, not an option you remove once things run smoothly. An error caught before sending costs one read-through. An error sent straight to a client costs a great deal more.

Do I need to automate several tasks at once for it to be worth it?

No, and it is actually a bad idea to start that way. A single well-scoped task, measured before and after, gives you a method you can then repeat on the next one. Trying to automate everything at once multiplies the rules to write and the checkpoints to monitor.