Quality data will always surpass aimless AI experiments, and data-driven AI tools with a strong business case can enhance decision-making, writes Luxaviation Group CITO Pieter Steyn.
We are in the middle of an AI gold rush. Every boardroom conversation I enter these days, or even casual conversations with friends, seems to circle back to AI. What people mean is usually machine learning, predictive models, or someone asking what our AI strategy is. I support the potential of AI, but I have also seen the mess that happens when we get the order wrong. Chasing AI without getting the data right first. We have made that mistake in our own research and development team.
In aviation, trust, compliance and safety are non-negotiable. You cannot build AI on shaky ground. Everything starts with data. Clean, structured, governed data. You also have to be transparent about where your data comes from and sensitive about how it is used.
There is plenty of hype about AI models but not enough conversation about what they are built on. To be honest, AI often feels like a marketing gimmick. Under the hood, it is still messy. AI is not magic. It is maths. And if the data is incomplete, inconsistent, or stale, the output is not just wrong, it is dangerous.
When you train as a data scientist or machine learning engineer, you build model after model just to get a feel for what good looks like. Then you need strong statistical knowledge to check whether the answers make sense. We still have our data challenges, but we understand them and we are clear on where we want to be. That is why we are strict about how data enters our systems.
In aviation, there is no room for close enough. Whether we are tracking aircraft performance, managing sustainability metrics, or predicting charter demand, we rely on data to make safe, informed decisions. Take our carbon calculator. If the distance flown, fuel type, or aircraft profile is even slightly off, the result loses all value. Yes, there are margins of error, and no model is perfect, but when decisions and money are on the line, you have to be close.
Finding the data
Our approach has always been simple. Build strong pipelines. Curate the right data layers. Only then introduce models or intelligence. We never experiment on live systems. We only test where the data is ready.
There is a clear difference between analytics and AI. CIOs need to protect that line. Analytics explains what happened and why. It supports daily decisions, real-time monitoring, and business visibility. AI tries to predict what might happen next or suggest what to do.
Before you even touch AI, you can go far with prescriptive analytics and what-if scenarios. AI requires maturity. Analytics sparks the right questions. Statistics help you get to the answers.
True AI is machine learning or deep learning. To build it, you need to find the data, understand it, clean it and keep it clean. Then you need to engineer features, know the business context, and finally ask yourself if it is even worth it.
We once explored predictive models for helicopter maintenance. But when we looked closer, the data we had did not go back far enough or capture what we needed. So we used analytics instead. AI requires strategy, alignment and readiness across the business.
The main difference is risk. With analytics, you act based on known patterns. With AI, the model is guessing. And if your data is shaky, it guesses wrong. Analytics can mislead too, but it is easier to see why. AI is a black box unless you build explainability into it.
Most of our insights still come from analytics. We do not need a neural net to tell us when charter demand peaks. We have flight data going back years. AI only steps in when the scale or complexity goes beyond what humans can handle.
High-quality data creates value every day. AI only creates value if it works. And that is a big if when the data is missing or wrong.
A smart investment
What is high-quality data? It is accurate, complete, consistent, timely and relevant. It is stored in the right format and maintained over time. It has a clear owner.
I have seen projects go nowhere because the data did not exist. The assumptions were off. The integrations broke. On the flip side, projects like our unified reporting layer gave us big wins just by improving visibility.
In our R&D team, we sometimes need to collect data that is not available anywhere. We have to research, verify and build the dataset from scratch. That is exactly what we did with Wayfinder. That process took us three years.
Data is not the boring part. It is a smart investment. It drives better decisions, smoother operations and prepares the ground for everything else.
It does not matter how advanced your model is. If your data is not governed, you are guessing.
AI runs on more than data. It runs on trust. That trust comes from governance. You need data ownership, access control, audit logs, versioning, and validation. In aviation, it is not optional. It is required.
Depending on the jurisdiction, AI tools must be explainable and safe. We cannot just deploy systems that affect human safety without oversight.
Wayfinder was built with governance from day one. Every source is logged. Every calculation is tracked. If an aircraft owner questions a number, we can trace it all the way back.
Maturity curve
That is what builds trust. That is what makes AI safe.
I have made the mistake myself. We have built things that looked good, got people excited, and then failed in production. That is a hard lesson.
When AI goes wrong, the first question is always the same. How did this get through? The answer is usually in the data.
I have seen forecasts fall apart because of a single bad data point. Bad data does not just make you look bad. It erodes trust. Your clients do not care how advanced your technology is. They care whether the numbers are right.
We simulate failure before any AI goes live. We test what happens with missing values, outdated inputs, and corrupt files. If it breaks early, it is not ready. We fix the data first.
There is a maturity curve to data. You start with collection, then move to storage, cleaning, analytics – and only later do you get to data science and AI. Trying to jump ahead never works.
Here is what I recommend to any CIO trying to get real value from AI:
- Start with business outcomes: Be clear on what decisions you want to improve.
- Make your data visible: People cannot use what they cannot find or trust.
- Do not skip governance: Build it in early, as retrofits are painful.
- Get teams talking: Data is not owned by one department so create shared ownership.
And only once you have done all of this. Only once the data is ready. Then you start thinking about AI!
