How long does AI automation take? Honest timelines

It is almost always the second question. First: what does it cost? Then: how long until it works? How long AI automation takes rarely depends on the technology. It depends on how small you make the first job, how quickly you can get at your own data, and who in your business makes the call. This article gives you realistic timelines per type of project, shows where the time quietly leaks away, and tells you what you can do yourself to speed it up.

The short answer: two weeks to two years

That fifty-fold difference is largely your own choice. A tightly scoped job, answering every enquiry from your contact form within a minute and booking it straight into the diary, is a matter of weeks. 'We are going to introduce AI' is a matter of years, because nobody can say when that is finished. So the question of how long it takes is really the question of how small you dare to start.

Why large companies take years

In the autumn of 2025, Deloitte surveyed 1,854 executives across Europe and the Middle East, the Netherlands included. Most said a typical AI use case takes two to four years to pay for itself. For ordinary technology those same companies expect seven to twelve months. Only 6 per cent saw payback within a year; even among the most successful projects it was just 13 per cent (source: Deloitte, 2025).

Before that frightens you: those figures are about companies with thousands of staff rolling AI out across the whole organisation, alongside a reorganisation and a clean-up of their data. Deloitte itself compares it to the switch from steam to electricity: factories had to redesign their entire production line before the gains showed up. You do not have to redesign your production line. You only have to take away that one piece of retyping.

Realistic timelines per type of job

These are the timelines we work with, with the price alongside so you can weigh it up straight away:

• One tightly scoped task, answering incoming enquiries automatically and putting them in the diary, or reading receipts: 1 to 2 weeks of building, 300 to 5,000 euros.
• A complete process from start to finish, connected to the software you already use: 3 to 6 weeks, 5,000 to 12,000 euros.
• Several processes that feed into each other, or several departments at once: 6 weeks and up, 12,000 to 25,000 euros.

Before that comes one more step: the scan. That review of how your business runs costs 1,500 euros and produces a report saying which process is up first and what it is expected to deliver. How fast that step goes depends mostly on how quickly I can speak to your people and see your numbers, we agree that up front, so you can put it in your own diary.

The clock does not start at the building

Almost everyone counts only the build weeks. But three things have to happen before anything gets built, and that is exactly where the time goes at companies that end up taking months.

• Choosing which process goes first. Not the most annoying process, but the one that comes back often and follows fixed patterns.
• Arranging access to your own systems. Who holds the login details for your software? Does your supplier have to switch something on first? That can mean a week of waiting, and you can start that waiting today.
• Agreeing who decides. One person who cuts through keeps a two-week project at two weeks. A committee that meets on Fridays turns it into three months.

Five things that really drive the timeline

• How tightly the job is scoped. 'Enquiries from the contact form' is tight. 'Sales' is not.
• Whether the work follows patterns. Forty enquiries of the same kind go fast. Forty one-offs do not.
• How quickly you can reach your own data. Is it all in one package, or in four systems that know nothing about each other?
• How many exceptions there are, and whether anyone can list them. Exceptions that only surface once it is live always cost an extra round.
• Who decides, and how fast they answer. In practice this is the biggest factor, and the cheapest one to fix.

Where the time quietly leaks away

These five delays are the ones I see most, and none of them are technical:

• Waiting for access. Your software supplier has to open something up and only replies next week.
• Testing with examples that are too tidy. Everything works perfectly until the first real enquiry arrives with a photo attached and half an address.
• Exceptions nobody mentioned, because they are so normal you no longer see them.
• The people who have to work with it only get asked at the end, and then raise objections you could have collected up front.
• The project that grows halfway through: 'could we also just...'. That is the end of any schedule.

Deloitte names that second point as one of the main reasons returns fail to appear: a trial on made-up test data goes beautifully, and the moment real data goes in the problems start (source: Deloitte, 2025). That is why I would rather test straight away on your real cases from last month, especially the messy ones.

TNO followed four Dutch organisations introducing AI and published the results in 2026. At insurer a.s.r. development was abandoned along the way. At Customs the project was still in development after months. At e-commerce company HelloPrint it did work: AI now handles roughly half of all customer contact there. TNO's main conclusion: the organisations had given too little thought up front to what it does to their people's work (source: TNO, 2026). That is not a soft side issue, it is exactly what stalls a project.

Time to build is not time to pay back

Two different clocks. The first runs from 'we start' to 'it is running'. The second runs from 'it is running' to 'it has paid for itself'. For a scoped job of 4,000 euros that saves five hours a week, that second clock runs in months, not years. For a roll-out across your whole company you are in the territory where Deloitte measures two to four years. That is the real reason to start small: not because big is impossible, but because with small you know within a quarter whether it works.

Count the upkeep too. Your software changes, your prices change, your customers start asking different things. With us maintenance runs from 100 to 750 euros a month for something small, up to 1,500 to 3,000 euros for a setup your business depends on daily. That is not a hidden cost, it is the reason it still works a year from now.

An example schedule for a small first project

Not a promise, but the shape I see most often for a two-week job:

• Day 1: walk through the process together and collect the last fifty real cases.
• Day 2: arrange access and agree what the system may do on its own and what goes to a person.
• Days 3 to 6: building, with a look along the way so you are not in the dark for two weeks.
• Days 7 and 8: running on last month's real cases, without anything going out the door.
• Day 9: let your people look it over and collect the exceptions they see and we do not.
• Day 10: live, on the agreement that everything still passes a person in the first week.

When it takes longer, and when you are better off waiting

Honestly: sometimes starting now is not smart. Rather wait if one of these applies:

• You are about to switch to different software. Do not build on something that disappears in three months.
• The process lives only in one person's head and that person has no time in the coming weeks. Then you are building on guesswork.
• You have just hit a busy period. In the first weeks, automation asks for attention from your people.
• The process happens less than a few times a week. Then the time saved is too small to be worth it, and I would rather say so before you pay.

What you can do to make it go faster

• Collect fifty real examples of the work you want automated up front, exactly as they came in.
• Find out now who manages the login details for your systems and whether your supplier has to open something up.
• Appoint one decision-maker who answers within a day.
• Tell your team what is going to happen before anything runs, not afterwards.
• Keep the first job small, even if you fancy more. The second one goes twice as fast, because the access and the agreements are already in place.