AI Basics: What needs to be right before AI makes any sense.
AI tools are powerful. But performance without the right AI basics is not an advantage. It is a multiplier. For everything, including what does not work.
Pilot launched. Tool introduced. Use case presented. And then?
We see this in many companies right now. The playbook is always the same: a pilot is launched, a tool introduced, a use case presented. Everyone is enthusiastic. And then, a few months later, someone asks the question nobody wanted to ask: "What did this actually achieve?"
The honest answer is usually: less than hoped. Not because AI does not work. But because the AI basics were missing.
Three signs that the AI basics are missing.
We see these situations in almost every company that comes to us.
The pilot runs but nothing changes
The AI tool is live. The demo was impressive. But in day-to-day use hardly anyone uses it, results do not land where they are needed, and nobody can say whether it has actually improved anything.
The result: Budget spent, expectations disappointed, scepticism towards the next project grows.
The data is there but nobody trusts it
The model runs on real company data. But the quality was already a problem before AI. Duplicate entries, missing fields, inconsistent formats. The model produces structured output, but the foundation remains shaky.
The result: Decisions based on results that nobody can really verify.
AI runs but ownership is missing
The automation works technically. But who is responsible when it delivers wrong results? Who maintains the model? Who decides when requirements change?
The result: A system that quietly ticks along, nobody touches anymore, and then suddenly fails.
The problem does not disappear through AI adoption. It gets covered up. Briefly.
When a company has problems, data silos, unclear ownership, processes that run on goodwill, those do not disappear through AI. They get covered up. Briefly. Then they come back. Louder.
Because AI systems were built on these foundations. Because the team now produces twice the output, but still on the basis of wrong or missing data. Because the automation now runs the broken process at three times the speed. The problem has not been solved. It has been multiplied.
And then comes phase 2: people keep going enthusiastically. Which is fundamentally good, because it means progress. More use cases, more automation, more output. Until someone from the business side says: "Something is not right here."
The AI is running on autopilot. And automating the wrong things. The tricky part: this cannot be completely avoided. But those who have their AI basics clear notice it earlier. Because most of these errors are not surprises. They are the edge cases the business team already knew about before AI. They were never documented, never defined as exceptions, never fed into the system. And now the AI processes them as normal cases.
The awareness of this needs to be there from the start: AI needs not only a good beginning, but also someone who keeps watching.
What belongs to the AI basics that actually matter?
We sometimes say this directly to our clients: before we talk about AI, we need to talk about your data. About your processes. About who is actually responsible when something goes wrong.
Those who do not clarify the AI basics are building on shaky ground. That sounds uncomfortable. And sometimes it is. Those who come into a conversation with AI enthusiasm do not want to take a step back. But the step back is often the fastest way forward.
Before AI can be used sensibly, a few unspectacular questions need answers: what data do we have, and how clean is it really? Which processes should be automated, and do they run reliably today? What is the goal, and how do we measure whether we have reached it? Who owns the result when the AI is running?
Those who can answer these questions have already got the hardest part behind them.
Three months AI pilot. Then the honest question.
A company has invested three months in an AI pilot. The tool runs, the use case works technically. But when reviewing the results, it turns out the data quality in the source system was already a problem before. The model produced structured output, on the basis of data nobody fully trusts.
We take a step back. Not to the beginning, but to the cause. Data quality, process ownership, clear goals. Then, and only then, AI on top.
- Data quality in source system checked and cleaned
- Process ownership clearly assigned
- Success criteria defined before the next step starts
- AI pilot reset with clean AI basics
- Results this time measurable and traceable
- Team understands what the model does and why
“Every week the basics are not clarified is a week the real problem grows. And at some point the moment comes when you have to go back to basics. Just then with three years of AI debt on top.”
What you usually ask us about AI basics.
AI MAKES SENSE. WHEN THE BASICS ARE RIGHT.
If you want to know whether your company is ready for AI, or where the AI basics are still missing, talk to us. Ten minutes is usually enough for a first assessment.