How to bring AI into your business in 30 days without stopping the business

How to bring AI into your business in 30 days, week by week: assessment, one process, build, real-world testing, training and measurement.

Bringing AI into your business sounds like a long, expensive project with a risk of leaving the business half-running while it happens. In Mallorca, where many businesses live off a six-month season and cannot afford a week of chaos, that idea holds a lot of people back. The reality we see in the implementations we carry out is different: a first process automated with AI goes live in around thirty days, without changing software, without stopping anyone and with the team working as usual in the meantime. This article describes that plan week by week, as we apply it, and ends with the list of mistakes to avoid.

Week 1: assessment and choosing a single process

In the first week nothing is built. We observe. We spend a few days seeing how the team really works: which emails come in, what gets copied from one place to another by hand, what gets forgotten, where time is lost every day. From that comes a list of candidate processes with an estimate of the weekly hours each one consumes.

Then just one is chosen, using these criteria:

  • It is repetitive and follows rules that can be written down on a sheet of paper.
  • It consumes hours every week for someone on the team.
  • If it fails, the damage is small and quickly detected.
  • It has a clear owner who will be able to supervise it.

Common examples on the island: sorting and replying to booking emails, following up on quotes sent, filing receipts and supplier invoices, preparing the daily appointments summary. The full assessment, with the prioritised list and the chosen process, has a fixed price of €1,500.

Week 2: building with the tools you already have

A frequent mistake is to think that bringing in AI requires changing your management software, your email or your calendar. It does not. The agent connects to what your business already uses and works from there. We build on open-source agent platforms (Hermes Agent, OpenClaw) that connect to email, the calendar, shared folders and most management packages, and we choose the model (Claude, GPT, Gemini or Grok) according to the task.

During this week:

  • The rules of the process are written in plain language, with the owner of the process present. Those rules are the agent's instructions and anyone on the team can read them.
  • Permissions are defined: which inbox it accesses, which folders it may touch, what it may never do under any circumstances.
  • It is decided what goes out automatically (template messages) and what requires approval (everything else, at the start).
  • Data hosting is configured on servers in the European Union.

The team carries on working exactly as before. It sees nothing yet.

Week 3: testing with real data and draft mode

In the third week the agent starts working, but in draft mode: it processes real emails, prepares real replies, files real documents, and a person reviews each action before it takes effect. This is the phase where the cases nobody had mentioned come to light: the supplier who sends invoices in an odd format, the German customer who writes in English with questions in three languages, the booking that arrives through two channels at once.

Each correction becomes a new rule. By the end of the week the agent usually gets the vast majority of cases right and the corrections are matters of nuance.

The team is also trained in this week, and the training is short because the agent does the heavy lifting: they learn to review drafts, to correct, to flag when something odd appears and to read the action log. Half a morning is usually enough. It is important to present it as what it is: a tool that takes away the tasks they like least, supervised by them.

Week 4: go-live and measurement

In the fourth week the agent moves to normal operation. Tasks where it has not failed during testing can become automatic if the owner of the process so decides; the rest stays in draft mode for as long as needed. There is no rush to remove human approval. Many businesses keep it permanently on everything that reaches customers, and that is a perfectly reasonable decision.

Three things are measured from day one:

  • Hours of manual work the process consumed before and consumes now.
  • Errors or omissions detected (quotes without follow-up, lost receipts, unanswered emails).
  • Customer response time, where applicable.

With that data, at the end of the month you know whether the process has been worth it and what the next candidate is. An automation of this kind costs between €3,000 and €6,000, and a managed agent, with supervision and continuous improvements, from €500 per month.

What to avoid during those 30 days

What we have seen go wrong, in our own implementations and in others':

  • Starting with three processes at once. Attention is scattered and none of them turns out well.
  • Choosing the company's most important process as the first one. Start with one that hurts but does not kill.
  • Changing management software at the same time. One change at a time.
  • Switching on automatic sending on day one "to see what happens".
  • Not naming an owner. An agent without a supervisor is an orphaned process.
  • Doing it in the middle of high season if it can be avoided. October to March is the natural time for many businesses on the island.
  • Hiding it from the team until it is done. They find out anyway and take it worse.

Frequently asked questions

Do we really not have to stop anything during the implementation?

No. For the first three weeks the team works exactly as always and the agent is built and tested in parallel, in draft mode. What changes in the fourth week is that certain tasks arrive done or prepared. If at any point the agent stops, the process carries on being done by hand as before.

What if halfway through we realise the chosen process was not the right one?

It is usually caught in the assessment, but if it happens we change it. The written rules and the connections to the tools are largely reused, so the cost of the change is a matter of days. That is why we start with a small process rather than the biggest one.

And after the 30 days?

The next process on the assessment list is chosen and the cycle is repeated, shorter this time because the foundation is in place. With a managed agent, those extensions and the daily supervision are done by us, and your business receives a monthly report on what the agent has done and what it proposes to improve.

If you want to know what the first process to automate in your business would be and how this plan would fit your calendar, the assessment is the first week of the thirty days. Request an assessment

If this fits your business

We implement it for you

We work at the leading edge of artificial intelligence and agents: we build on the most advanced agent platforms of the moment and we do it at your business, in Mallorca, with published prices and a person from your team approving anything that goes out. We start with an initial assessment and a written plan.