GUIDE · NO. 1
AI that lasts
A guide for industrial companies that want the solution to still work a year from now.
Does it solve the right problem?
Does it hold up legally?
Does it still work in a year?
More and more companies are building their own AI solutions. That is a good thing. The tools have become so accessible that an afternoon is enough for a working demo. But what decides whether the solution creates value is rarely the demo. It is everything around it: that it solves the right problem, that it holds up legally, and that it still works a year from now.
This guide walks through those three things, with a checklist for each. Use it when you build yourselves, or when you evaluate what you have already built.
01 · BUSINESS VALUE
Start with the business value
The technology is a solution, not the starting point. The solutions that survive are the ones that start from a real problem: something repetitive, time critical or something that takes up your most experienced people's time. The same question answered over and over. Answers that take too long while the customer waits. Knowledge that only exists in the heads of a few.
Set a measure before and after. Hours, response time, number of tickets. Without a measure, no one knows whether the solution works, and then it dies quietly.
And expect more than the time to be affected. When answers come faster, quality, customer satisfaction and sales tend to follow.
What takes the most time today?
What does it cost in hours?
How do you know it got better?
CHECKLIST · BUSINESS VALUE
Examples of where the value usually sits: knowledge, technical support, customer service and inside sales, aftermarket.
02 · SECURITY AND COMPLIANCE
Does it hold up legally?
This is where most homegrown builds fail. The rules apply from day one, even for an internal test.
- Data
- Where is it stored and processed? Inside or outside the EU?
- Access
- Who can reach the solution and the data behind it? An internal chat that can answer everything can also leak everything.
- Permissions
- A colleague, a customer and a reseller should not be met by the same answer. Can the solution control who sees what?
- GDPR
- Personal data in questions, documents and answers. Is there a data processing agreement with the supplier?
- EU AI Act
- Transparency requirements, for example that users know they are talking to AI.
- Sources
- AI is happy to answer even when it does not know. Demand sources for the answers, and the ability to trace why an answer came out the way it did.
- Manipulation
- A solution that meets customers or reads incoming email can be tricked into doing the wrong things. Set boundaries for what it is allowed to answer.
Questions in
Answers out, with sources
The data is stored and processed inside the boundary. You know where, and who has access.
CHECKLIST · SECURITY AND COMPLIANCE
03 · FROM DEMO TO OPERATIONS
Does it still work in a year?
A demo is built in a day. Operations is the rest of the solution's life, and that is where most homegrown builds end.
- Ownership
- Who maintains the solution, and does the person responsible for your IT know it exists? When many people build on their own, the overview disappears, but the bill, the security and the troubleshooting still land with IT. What happens when the person who built it leaves?
- Updates
- Products, prices and documents change. How do the changes get into the solution, and how quickly?
- Feedback
- How are wrong answers caught, and how are they corrected? Without feedback the solution gets worse every month.
- The human in the loop
- Where is a human needed to review before an answer goes out? Toward customers the answer is almost always: somewhere.
- Consistency
- If different departments build their own variants, different truths appear. The same question should give the same answer, whether it comes through the web chat, the email or the internal chat. That requires a common foundation to build every solution on.
It is never finished. That is the point.
CHECKLIST · OPERATIONS
GAIDE
You don't have to do it yourselves
Everything in this guide can be solved on your own. Many companies do, and it is a good way to understand what AI can do in your specific business. That knowledge stays with you, whatever you choose afterwards.
Getting help from someone who does this full time is mostly about the cost of learning. We have asked the questions before, seen where solutions tend to die and know what it takes for them to last. It usually turns out faster and cheaper than finding the way on your own.
If you have questions about anything in the guide, we are happy to answer, whether it leads anywhere or not. If you want to know what this would look like for you, we can talk about how you work today, what you have tried and what we have seen at other industrial companies.
Written by us, with help from our AI assistant.