AI automation: build it yourself or outsource it?

Build AI automation yourself or outsource it, that is the question nearly every business owner asks the moment the first plan gets concrete. On one side: someone on your team is handy, the tools cost a few tens of euros a month, so why pay thousands for something you can click together yourself? On the other: you have no time, no idea where to start, and you have heard about companies where the whole thing quietly stalled halfway. Both feelings are right. Below is the trade-off the way we make it with business owners, including the cases where our answer is: do it yourself, then you do not have to pay us.

First: where does the mid-market actually stand?

Before you choose, it helps to know how fast this is really moving. According to Statistics Netherlands (CBS), one in six Dutch companies used AI in 2025. Among companies with 10 to 50 staff it was 27 percent, with 50 to 250 staff 45 percent, and above 250 people 66 percent (CBS, December 2025).

Two things follow from that. One: you are not late. Almost three quarters of companies your size are doing nothing yet. Two: the ones that are doing it tend to be bigger, and they have someone on the payroll who can take this on. You probably do not. Which is exactly why the build-or-outsource question weighs heavier for a mid-sized company than for a corporation with its own IT department.

What building it yourself actually means now

Let us be fair: building it yourself has become a lot easier. There are tools where you click blocks together without programming, when this happens, do that. For a decent range of jobs that is plenty. You can perfectly well handle yourself:

• Routing a form on your site to the right person automatically.
• Getting a short overview in your inbox every Monday morning.
• Sending a reminder when a quote is still open after a week.
• Filing receipts in the right folder automatically.
• Copying new customer details from one list to another.

What these have in common: one or two systems are involved, there is no customer on the other end, and if it stops for a day nothing bad happens. Hiring an agency for this is a waste of your money. We say so in the first conversation.

The sum almost everyone makes

The comparison you make as an owner is usually this: the tool costs 15 to 50 euros a month, and an agency wants thousands for the same thing. Hardly a discussion. Except one item is missing from the do-it-yourself side of the bill: your own hours.

Someone on your team who really digs into this is easily 60 to 80 hours in before the first serious automation works. Figuring out how it fits together, building something, discovering it does not quite do what you meant, starting over, and then removing the errors that only show up once real data runs through it.

At that rate, 60 to 80 hours is roughly 2,850 to 3,800 euros of time that never appears on any invoice. That is not an argument against doing it yourself, those hours can be well spent, because the knowledge stays in house. But call it what it is. Anyone who tells you building it yourself costs almost nothing is valuing your own people at zero.

Four questions that expose the real cost

Building is not the expensive part. What costs you for years is keeping it running. These four questions expose that, and they work for both options:

• Who fixes it when it stops at seven in the morning? If the answer is that one colleague, but they are on holiday, you do not have automation. You have a dependency.
• Who notices when it stops? This is the dangerous one. Automation that fails usually fails silently. No error message arrives, things simply stop coming.
• Who still knows how it works a year from now? Home-built things often live only in the builder's head. Once they leave, nobody dares touch it.
• Who is responsible when it sends something wrong to a customer? If you built it, that is always you. If you outsourced it, there should be an agreement about it.

Why AI projects stall

Failure here is not pessimism. In June 2025 Gartner predicted that over 40 percent of projects involving AI that acts on its own will be cancelled before the end of 2027. The reasons they list: rising costs, unclear business value and insufficient control of the risks.

And it is not just large companies. Research by SAS and IDC among more than 1,600 mid-market leaders (July 2026) found that nearly 70 percent are still stuck in the experimental phase: loose trials, no coherence, no plan to move it into the actual work.

Research agency Dialogic reached the same core conclusion for the Dutch Ministry of Economic Affairs (September 2025): a lack of knowledge and skills, a limited degree of digitalisation and excessive costs hold back AI adoption in smaller companies. What owners mostly need is practical knowledge and guidance. Not another tool.

When you should simply do it yourself

This section costs us revenue, but it belongs here. In these cases outsourcing is money thrown away:

• The process does not touch a customer. If it goes wrong, you notice and nobody else does.
• One or two systems are involved, not five.
• You can explain what has to happen in one sentence. If you need three sentences with except when in them, it is more complicated than it looks.
• You have someone who genuinely gets time for it during working hours, not evenings on top of a full week.
• It can sit still for a week without hurting.
• It happens a few times a week. At a few times a day the sum changes.

One more variant we run into regularly: the software you already have can do it. Many packages have reminders, automatic emails or standard reports built in that nobody ever looked at. Switching that on takes an afternoon and zero euros.

When outsourcing pays for itself

The other way round, there is a point where doing it yourself gets more expensive than it looks. You recognise it by this:

• There is a customer on the other end. A mistake is then not an inconvenience but reputational damage.
• Several systems need to know something about each other, and they do not speak the same language.
• The data is messy: the same customer twice, names spelled differently, half-filled fields. This is the silent project killer. Cleaning up is almost always more work than building.
• It has to keep running when the person who made it is gone.
• You need it in three weeks, not in six months.
• It touches money: invoices, quotes, payments. You do not want to be learning there with real amounts.

The hybrid approach, and the trap in it

Almost every article on this subject ends with: go hybrid. Have the first, hardest automation built and pick up the rest yourself. That advice is sound, but it tends to fail at the same point.

Handover is not an afternoon of explaining. If nothing is written down and one person has been trained, that person is your new weak spot. So ask for something simple: a plain-language description of what happens, where it lives, and what to do when it stops. Not a technical document, something their successor would also understand.

What does it cost, on both sides?

Doing it yourself, per automation:

• The tool: roughly 15 to 50 euros a month, depending on how much you push through it.
• Your own hours: 60 to 80 for the first one, then a few hours a month keeping it alive.
• The mistakes you make along the way. Usually time, sometimes an angry customer.

Outsourcing to us looks like this:

• First conversation: free. We look at whether there is anything worth doing at all.
• Small project (1 to 2 weeks): 300 to 5,000 euros.
• Medium project (3 to 6 weeks): 5,000 to 12,000 euros.
• Large project (6 weeks or more): 12,000 to 25,000 euros.
• Maintenance after that: 100 to 750 euros a month for something small, 750 to 1,500 if your business runs on it daily, 1,500 to 3,000 if it sits everywhere.

See that 300 euros at the bottom? It is not there for show. Sometimes the answer after a conversation is that one small thing needs building and the rest can stay as it is.

Five things to agree if you outsource

Outsourcing fails on agreements, not on technology. Settle this before anything gets built:

• Who owns it if we stop? You have to be able to take it with you. If it is locked to an agency, you are not buying a solution but a subscription you cannot cancel.
• Can you cancel monthly? With us you can, no long-term contracts. Always ask, because that is not standard everywhere.
• Who gets told when it goes down, and how fast is it back up?
• What does it cost per month while running? Not just the build price. Also ask what happens if volume runs higher than expected.
• What does it do when unsure: stop and ask a human, or guess? Anything that goes outside or touches money should stop and ask.

Since 2 August 2026 there is one more thing

Since 2 August 2026, European rules require that people know when they are talking to AI rather than to a person. So if something at your company ends up dealing with customers, that has to be clear to them.

For this decision that matters for one reason: the responsibility sits with you as a company, whether you built it yourself or had it built. Doing it yourself does not shrink the rule, it only means nobody is checking whether you got it right.

How to decide tomorrow

This does not need weeks. Five steps, one morning:

• Write down what goes wrong, not which tool you want. We respond to enquiries too late is a starting point. We want to do something with AI is not.
• Count how often it happens per week. Under five times it is rarely worth a project.
• Work out what one mistake costs. A missed enquiry, a wrong invoice, a customer walking away. That number sets how much it is worth doing properly.
• Ask whether there is a customer on the other end. If so, the bar is higher and learning on real customers is a bad idea.
• Only then look for the cheapest thing that actually solves it. Sometimes that is switching on a setting, sometimes a 30-euro tool, sometimes something built for you.

With those five answers in hand you can decide on your own. Which is the point: you should not need to call us to find out whether you need us.