Rolling out AI in a company that already works happens in five phases: understanding how you work and deciding what not to build, building on the tools you already use, seeing it working with your data, measuring against the "before", and supporting and adjusting. You do not have to stop the business or change programs. If you do not yet know whether your company is ready, take the AI maturity self-assessment first; if what is unclear is where to start, read how to choose your first AI project.
You keep putting it off because you picture it as building work. Knocking down partitions, cutting the water, living among boxes for a few weeks and, in the end, something that is almost what you asked for but not quite. So it is not the moment: you will look at it in autumn, when the season eases off.
Rolling out artificial intelligence in a business that already works is nothing like that. You do not have to stop. You do not have to throw away anything you have already set up. And it does not start with technology: it starts with someone watching how you work on any ordinary Tuesday, nothing special, to understand where your time really goes.
And it is not just for multinationals with their own department. According to the ONTSI indicators of artificial intelligence use in Spain (Red.es), 11.4% of companies with ten or more employees used AI in 2024. And according to the INE ETICCE survey, 13.4% of companies with fewer than ten employees were already using it in the first quarter of 2025. That is the size of the neighbourhood business, not of the office tower.
This article takes the process apart piece by piece. The five real phases, what you will be asked for in each one, what you decide and how you measure whether it has worked. And why nobody serious gives you a delivery date before seeing your case.
What does "rolling out AI" mean, and what does it not mean?
It is worth clearing up first, because the phrase frightens people more than it should.
Rolling out AI is not buying a tool and switching it on. Nor is it replacing your team with a machine. It is building software made to measure for your business (a chatbot that knows your opening hours and your services, a workflow that puts invoices where they belong, an agent that picks up the phone when you cannot) and fitting it into the way of working you already have.
The difference from an off-the-shelf template is exactly that: you do not adapt your business to a generic product; the product is built around how you work. That is why the time saving, when it shows up, is a consequence of having built the right thing, not the starting point. The starting point is understanding what is worth building. And that is where phase one begins.
Phase 1: understand how you really work (and decide what not to build)
This is the phase almost everyone skips, and the one that costs most to skip.
Before touching a single tool, someone has to sit down with you and understand how your business works inside. Not the tidy version you would put in a presentation: the real one. Who answers the phone, what happens when that person is off sick, where each thing gets written down, how many times a day the same question comes up, what falls through the cracks when things get hectic.
Two lists come out of that. One is what does make sense to build: repetitive, predictable processes that eat time without calling for judgement. The other is just as important and hardly anyone shows it: what not to build. Some tasks depend on your eye, on a difficult conversation, on deciding an unusual case. Automating those goes badly and costs a lot. Saying "leave this as it is" is part of the job, not a renunciation.
If a supplier turns up with the solution before asking you these questions, be wary. They are selling what they already had built, not what you need. This preparatory work of organising the process is also what most determines how much everything else will cost and how long it will take. A clear process is automated quickly; one that nobody has written down has to be organised first.
Phase 2: build on the tools you already use
This is where many people expect the worst: migrations, changing programs, learning a new system from scratch. Almost never needed.
The normal thing is to build on top of what you already have. If you serve customers on WhatsApp, the chatbot lives in WhatsApp. If you keep your customers in a spreadsheet or in a particular CRM, the workflow writes there, not somewhere else that nobody will look. If you invoice with a certain program, the automation hands the invoice to that program. The point of integrations and automation is exactly that: connecting what you already use so the pieces talk to each other, not forcing you to start from scratch.
This has an advantage you notice the next day: the team has to learn almost nothing new. They carry on working on the same screens as always; what changes is that now part of the work is done by itself in the background. The less the surface your people touch changes, the less resistance and the fewer mistakes.
If you are thinking of setting it up yourself, without a supplier, with no-code tools such as Make or n8n, that is another route and we cover it in how to start with AI in your business without coding.
Phase 3: showing it to you working, not in a PowerPoint
The moment comes to see what has been built. And here there is a line that separates those who build software from those who sell slides.
What you need to see is the system really working, with data from your business, doing what it will do once it is live. You write the chatbot a trick question and see what it answers. You feed in an odd invoice and see where it puts it. You simulate the call that always gets complicated and see how it handles it. Not a nice video, not a promise: the thing itself, in front of you, responding.
That demonstration serves two purposes. First, so you can see whether it really fits how you work or whether something needs adjusting before it is released. And second, so that you understand what has been built. If you leave that meeting without knowing what the system does, who maintains it and what happens when it fails, something has gone wrong. A supplier who shows you nothing creates dependency, and dependency gets expensive over time.
Phase 4: measure against the "before"
An automation that is not measured is an act of faith. And faith is no way to run a business.
That is why you should have the "before" before switching anything on. If the project is a chatbot, how many repeated enquiries came in per day and who answered them? If it is an invoice workflow, how much time went on entering them by hand? If it is an agent that takes calls, how many went unanswered? That starting figure, the one for your real situation today, is what gives meaning to everything else.
When the system has been running for a few weeks, you look at the same metric again and compare. No headlines and no brochure percentages: your figure against your figure. Sometimes the result is striking and sometimes it is modest, and both are useful information. It may even turn out that a particular process was not worth automating. Knowing it with numbers is better than suspecting it, and adjusting on data is cheaper than arguing over impressions.
Phase 5: support and adjust
The day it goes live is not the end. It is the start of the part almost nobody budgets for.
A chatbot improves as it piles up conversations and you discover questions it did not know how to answer. A workflow has to be touched up when something changes in your business: a new supplier, a service you stop offering, a different time of year. Software is alive, and a system abandoned on day one slowly rusts until it becomes a source of errors instead of a help.
Supporting means being there when the unusual case nobody foresaw appears, fine-tuning what falls short and deciding, with the data in front of you, what to automate next. This phase has no end date on a calendar, because a business does not have one either.
What will you be asked for, and what do you decide?
Throughout the process the split of roles is fairly clear, and it helps to know it in advance.
What you will be asked for is, above all, access and time from whoever knows the process. Access to the tools you already use so they can be connected. A while with the person who really knows how each thing is done (not always you) to explain the real process, with its exceptions. And real examples: the questions that come up most, the invoices as they arrive, the calls as they are. The more faithful the material, the better what is built fits.
What you decide is what matters. You decide the scope: where to start and what to leave for later. You decide what goes live and when, after seeing it working. You decide, looking at the measurements, whether to continue down that path or change it. And you have the right to decide that something is not built. A good supplier gives you the judgement to decide; they do not decide for you and then send the bill.
How the result is measured you have already seen: against your starting point, with the metric agreed at the beginning, not against anybody's average. And what tends to go wrong along the way is in the 7 mistakes when rolling out AI in a small company.
And how long does all this take?
Time to be honest here, even if the honest answer sells less than a round date.
There is no standard timeline, and anyone who promises you one without having seen your case is selling you smoke. "Live in 72 hours", "you'll have it in a week": those phrases are said before knowing what needs building, and that is why they are worthless. The real time depends on specific things in your business. On how many processes you want to cover. On how many systems have to be connected. On how many exceptions each process has. And, above all, on the state of your information: if it is already organised, it connects; if it is scattered across folders, emails and one person's head, it has to be organised first, and that takes time.
A chatbot with a tightly defined scope (the questions you already know by heart and a clear protocol for what it cannot answer) is ready much sooner than an automation that touches four systems, one of them undocumented. Both are "rolling out AI", but they look nothing alike on the calendar.
What can be settled from day one is what will be built, in what order and what you will measure to know whether it works. That conversation happens before signing anything. The exact date comes from seeing your case, not from a price list.
Frequently asked questions
Do I have to change the tools I already use?
Almost never. The usual approach is to build on what you already have: your WhatsApp, your spreadsheet or CRM, your invoicing program. Those pieces are connected so they work together, instead of forcing you to migrate to a new system nobody asked for. If in some specific case your current tool falls very short, you are told and it is decided with you, not as a surprise.
Do I need to know about technology to go through this process?
No. What you need is to know your business, and you already do. The questions in the first phase are about how you work, not about which technology to use. The technical side is the responsibility of whoever builds; your role is to explain the real process, see the system working and decide. If at any point you do not understand what has been built, it is a sign that it has not been explained well, not that something is missing on your side.
What if it turns out in the end that building something is not worth it?
You are told. Some processes depend so much on your judgement, or happen so rarely, that automating them does not pay off. Saying so is also part of the job, and it is better to know before spending on building something that was not going to be useful. A supplier who says yes to everything is not looking after you, they are invoicing you.
How do I know the rollout has worked?
By comparing with your starting point. If before starting you note how much time went on a process, or how many enquiries went unanswered, afterwards you measure the same thing again and compare. Your figure against your figure. That is the only proof that matters, and it does not fit in a catalogue percentage.
Let's talk
You can use everything above to think through your own case by yourself: look at what repeats for you, what depends on your judgement and what state your information is in. With that you already have half of phase one done.
If you would rather we look at it together, tell us what you do and which process is eating your week, and we will tell you whether it makes sense to build something, how we would approach it and what it would take. It may turn out yes and it may turn out no. Tell us what you need and we will give you a frank answer; if it is not for us, we will tell you that too. And if you want to see what we work on first, here are our services.
Xora IntelligenceAI consultancy and custom software in Los Barrios (Cádiz). We write about what we build for businesses in the province. If you have a question about your case, write to us.




