Can't we just build the system ourselves using AI?
More and more people are asking us about this exact thing, and the answer is yes. With ChatGPT, Claude, or Copilot, you can build a simple CRM or Quality management system You know your business best, and we’re not here to stop you. What we can contribute is our own calculation. We’ve been building on Business Online since 2008, and our log records thousands of development tasks. So you know what you’re saying yes to, and what you’re saying no to using AI time for.
The short answer is that you can build the system yourself with AI, whether it’s a CRM or a quality system. The question isn’t whether you can do it, but whether you want the job that comes with it. Operations, access management, privacy, integrations, and an owner who stays. Some companies will do that, and they should build it. Most have more valuable things to spend both people and AI on.
What AI is actually based on one afternoon
Ask ChatGPT, Claude, or Copilot for a system to keep track of customers and projects, and you get something that works. Forms, lists, and a dashboard. It looks good in a leadership meeting. We see the same thing in our sales meetings. The documents are uploaded to Claude, the projects live there, and it works.
It’s smart use of tools, and the companies that work that way are ahead of those that haven’t started. If you’re there now, you’re on the right track.
A prototype and a system the company is working on are nevertheless two different things. The difference is not evident in week one. It becomes clear in month seven.
The recipe, if you want to build the system yourself with AI
If you’re going to build it yourself, this is the job. Please do it; it’s free and it’s right.
First, what has become simple. Logging in, storage, and backup are now handled by cloud platforms like Azure, and AI writes the code faster than any consultant. What remains is what the platforms cannot take responsibility for.
- The data model. The AI writes the code, but it doesn’t know that a discrepancy with you needs to be connected to both the project and the supplier, because the auditor is going to ask questions. Errors here are only discovered when the system is full of real data. Then it’s called a redesign.
- Access management in practice. The platform provides the login. However, some of you still need to decide, test, and maintain who sees payroll data, customer contracts, and personnel matters, every time someone starts or quits.
- Privacy. The system processes personal data from day one. By then, you should have an overview of how the data flows, deletion procedures, and a data processing agreement with each service that touches it. If you inadvertently input any personal data into a private ChatGPT account, you have a discrepancy to handle.
- The integrations. The system is supposed to talk to Outlook, the accounting system, and your documents. This is the biggest part of the job, and it will never be finished, because the systems at the other end change without asking you first.
- Operation and changes. When something goes wrong, someone needs to discover it, understand it, and fix it. AI agents help with more and more of this, but agents also need a boss. Someone needs to decide what they can touch and assess what they delivered. The discovery can be delegated. Responsibility cannot. In quality work, this is tightened up. ISO 9001 requires controlled change, with testing, approval, and traceability. An agent who rectifies the system on their own is seen as a strength. For the auditor, it is a deviation.
- Training and documentation. A system that only one person understands is not a system. There is a risk with logging in.
- Ownership. What happens when the person who built it quits, changes roles, or gets bored? This is the point that determines the majority of construction projects, and it is never mentioned in the prototype.
If you can handle these seven things, you don't just have a system. You have a small software department. Some companies want that. It's a completely legitimate choice, as long as it's a choice.
What happens in month seven?
We meet Norwegian growth companies every week, and many have already moved parts of their work into AI clients like Claude. We haven’t seen any real AI magic yet, and it’s worth being honest about why. This AI wave is still young. The reality of the systems being built in the fall won’t hit until the month of July to spring.
The mechanism, however, is old. Companies have been building their own Excel solutions, Access databases, and SharePoint structures for twenty years, and that history follows the same path every time, no matter how good the tool was. The table below shows how it usually goes.
| Custom-built system, week 1 | Custom-built system, month 7 | |
|---|---|---|
| Data | Five test customers and two projects | Real customer data, contracts and personal data |
| Maintenance | Nothing to maintain | An integration has stopped working after an update on the other end |
| System responsibility | The most eager in the company | The same person, at the peak of full-time work |
| Time consumption | An engaging afternoon | The times no one drives |
The tool gets better with each wave. The work in the right column still comes up. We have written about the SharePoint variant in dThe hidden costs of SharePoint.
What are your AI hours worth?
Here is the calculation we believe you should actually set up, and it’s not about fear. It’s about what your hours are worth.
A Norwegian development consultant typically costs 1,200 to 1,800 kroner per hour, according to Folke's hourly pricing statistics, Therefore, operating hours can be priced. But the interesting cost is another. The hours and the best people who agree to build and maintain their own control system are always taken from something else. The offers. The profession. The customers. You can certainly build it. But is that what you should be doing?
And work with AI and agents, for all that. The stupid thing is to not do it. But AI is an amplifier, and an amplifier works best where you already excel. If you are best at piping or subsea inspection, that is where AI hours are most valuable. A control system is not the place for that. It already exists, from us and from many others.
Soiltech They did just that. The proceeds went to the treatment of drilling waste; the turnover increased from 80 to 275 million kroner, and the system is operated by someone else.
Our own accounting illustrates the point. Many of the development tasks in our log have the same origin. Something changed at the other end. An API was updated, a standard was revised, a requirement came up. Those tasks don’t disappear if the system is self-built. They just move home to you. The difference between self-built and purchased is not whether the maintenance comes, but who takes it on. When Visma or Microsoft changes something, a system vendor fixes it once for all their customers at the same time. That applies to us, and it applies to our competitors.
And the most important thing is that no agent can help you afterwards. The history and traceability of a system are built while the work is being done. Those who weren’t caught can’t be re-filled.
Want to see what this looks like in practice?
PSS Securitas grew from a few guards to over 2,500 employees on the same system, without changing platforms in 11 years. Now they use AI directly against the data they have collected all along the way.
What if the supplier disappears?
The question goes both ways, and it deserves an answer. What if the system provider is closed down, bought up, or doubles its price? Ask the same control questions of any provider, including us. Who owns the data? Can it be exported? What happens to it if the agreement is terminated?
Our answer is that the data is already theirs and can be extracted, and that the documents are in their own Microsoft 365 regardless. But don’t take our word for it. Ask any vendor to show it before you sign.
Build ahead, if this is you
The tool is for under ten users, without customer data or sensitive personal information, Outside the audit framework and with a lifespan of a couple of years, build. Preferably already today. An internal tool that saves one department from plunder is a brilliant AI job; you learn to know your own processes, and you don’t have to ask anyone for permission.
The system that keeps track of customers, quality, and documentation is another category. The auditor and the customers must see it, and it must survive both enthusiasm and turnover.
Four questions are enough to know which category you belong in. Ask them in peace and quiet, without us present.
- Who will maintain this in two years, and does that person know about it?
- Who notices that something is wrong, and how long does it take?
- Who do you call when it’s quiet on a Tuesday in March?
- Would you have let the same colleague own the invoicing alone?
If you answer confidently to all four questions, then build, with our sincere cheers from the sidelines. If you answer no to one of them, you know it now, and it was free.
And if you’d rather use the AI for something more fun
We use AI ourselves in development every day. It has made us faster, but the maintenance, integrations, and responsibility are still there, and it is our job to carry that. Therefore, we have also built MCP support in Business Online, So that ChatGPT, Claude, and Copilot can work directly against your data in a system that someone else runs.
In practice, the workday can look like this with Business Online and an AI client. You come from a customer meeting, Copilot, Claude, or ChatGPT tells you what happened, and asks it to record it. The company, contact, and follow-up are structured in Business Online, with history and traceability, without anyone having to open a form. This is the workflow people envision when they want to build themselves. It already exists.
So, feel free to build. You know best what your company needs. There are also several ready-made systems out there, and ours is one of them.
FAQ
No, not alone. ChatGPT is good at working with the information you give it, but it doesn’t keep structured customer data consistent over time, doesn’t control who sees what, and doesn’t provide audit trails. It works best as a tool on top of a system that is being operated.
Expect ongoing hours as long as the system is operational. With an outsourced developer at Norwegian consultant prices of 1,200 to 1,800 kroner per hour, even a few hours per month becomes a visible annual cost. If the work is done in-house, the price remains the same for the hours taken from other tasks, plus the vulnerability of having the knowledge in one person's head.
Partly. Agents can monitor, detect, and correct more and more, but the agent setup itself is a system that must be owned, and someone has to approve and assess what the agents are doing, especially where an auditor or privacy is involved. We use agents in our own development every day. They make us faster. They do not remove the need for someone to have responsibility.
A prototype shows that something can work. A production system works when fifty people use it simultaneously, when the system at the other end changes, when the auditor asks for traceability, and when the person who built it has quit. The distance between the two is operation, responsibility, and time.
That could be the case, if you have control over where the data flows, data processing agreements with the services that handle it, and deletion and access procedures. The job is solvable, but it must be done and maintained, and the responsibility lies with you alone.
See it finished
We'll show you how Business Online meets ISO requirements, with your processes and documents. 30 minutes, no obligation.