Industry News · CURA Team · 2026-11-10
Airport AI needs a plan beyond prediction
According to The Korea Times, Incheon International Airport uses AI to predict baggage system failures. CURA’s take: the next question is what teams do with that warning.
According to The Korea Times, Incheon International Airport uses AI to predict baggage system failures. The supplied summary does not explain how the system works, what happens after a prediction or what results it has achieved.
That limits the conclusions we should draw about this particular deployment. But it offers a useful starting point for business owners and operations leaders considering predictive AI: do not stop the planning at the prediction.
CURA’s view is that the useful question is not simply whether software can flag a potential problem. It is whether your team has a clear, controlled way to assess that warning and decide what to do next.
Why it matters
For an operations leader, we would treat prediction as one part of a wider workflow, not as the finished product. Before approving a project, ask who receives a warning, what information they need, who can authorise action and how the outcome will be recorded.
Consider a hypothetical maintenance alert. Should it create a task for review, appear in an existing work queue or prompt an immediate escalation? Who decides whether an inspection is justified? What should happen if the person responsible is unavailable? These are design questions for your own operation, not reported details of Incheon’s system.
Our recommendation is to make those decisions before choosing the interface or specifying the automation. Start with the people accountable for the work and build around their responsibilities. Keep human approval wherever your operating procedures require it.
This is also where Custom Software as a Service (CSaaS) fits naturally. CURA’s approach is to build software around how a business works, with AI and automation inside it, then host, support and improve it on a monthly subscription. For a predictive workflow, we would apply that approach to the whole response process rather than treating the AI output as a standalone tool.
What to do next
1. Choose one operational problem. Define the failure or disruption you want to anticipate. Keep the initial scope narrow enough that the team can describe what a useful warning would look like and what action it would justify.
2. Check the information available. Review the records you would use, their consistency and who is responsible for maintaining them. Ask a prospective supplier to explain what those records can support, including the limitations, before agreeing to performance expectations.
3. Map the response. Write down the route from warning to review, decision, action and closure. Assign an owner at each stage. Include a fallback for missed alerts and a way to challenge a recommendation without losing the record.
4. Define success before a pilot. Choose measures that reflect the intended operational benefit. Depending on your objective, you might track warning usefulness, response time, unnecessary inspections or disruption avoided. Agree how you will assess these measures rather than assuming every alert represents value.
5. Plan for ongoing review. Set a review point with the people doing the work. Examine missed problems, unhelpful warnings and awkward handovers. Use that feedback to decide whether to adjust, expand or stop the pilot.
For teams exploring AI in aviation operations, our advice is straightforward: start with a specific decision and its follow-through. Treat prediction as the beginning of the workflow, not the end.