How Airlines Unlock More AI Value When Agents Coordinate Decisions Across Every Department

Airlines get the most from AI when agents can coordinate decisions across departments. Many carriers still automate tasks one team at a time, with operations, commercial, and customer service each running separate systems. That approach saves time within each team but misses the decisions that affect all of them at once. Coordinating those decisions requires shared data and clear limits on what each agent is allowed to access.

Christian Spannbauer is Vice President of Aviation Commerce at Lufthansa Industry Solutions, the Lufthansa Group’s IT consulting and services company, where he advises airline technology and business leaders on structuring their AI investments. He has spent more than two decades in enterprise technology leadership, including CTO and CIO roles across aviation and retail. Across the industry, airlines now expect AI to deliver efficiency gains in software engineering and to bring agentic automation into core business processes.

“Airlines find automation potential inside their silos. So the first thing they do is automate, rather than optimize,” Spannbauer said. Starting within one team is a sensible way to learn, but connecting those efforts takes more work. In SITA’s Air Transport IT Insights survey, 49% of airlines named data integration and consistency as the main barrier to scaling AI. The larger returns come once departmental systems start sharing decisions.

• Trade-offs by design: Airlines need to decide what an agent program should improve before they automate anything. Agents often face choices where no single option wins on every measure, so the system needs clear rules for which goals come first. “There’s always a trade-off from an operational viewpoint. The decision can cost you more, but the benefit can be higher on the customer satisfaction side, and that needs to be clear in this framework,” Spannbauer said. Margins make those choices matter. IATA projects net profit of about $4.50 per passenger in 2026, half of last year’s figure. Each goal should tie to a measurable business result that leaders can track. Operations, commercial, and customer teams have to set those priorities together and revisit them as conditions change, because a decision that helps one team can add costs for another.

As airlines operate under strict safety and regulatory rules, the agents working inside their systems have to follow them too. As more companies move agents from proof of concept into production, governance has become a top priority for enterprise technology leaders.

• Need-to-access permissions: Many organizations give an AI agent the same login as an existing employee. The agent can then reach every system that employee can. “You have to think about what the required access is, not the maximum access rights you can grant them, because agents act at machine speed. If they have the access, they will use it,” Spannbauer said. Some companies now run hundreds or thousands of agents without knowing what each one is allowed to do. Before deploying agents, teams should decide which systems each one needs and when a human has to review its work. That plan should also cover security and compliance rules. Setting those rules up front makes it easier to add more agents later.

• Accountability beyond IT: The CIO and engineering teams build the tools that monitor and control agents. Responsibility for what an agent does belongs to the business team it works for and to that team’s managers. That raises practical questions, such as whether each agent should serve one department or several, and who is responsible when it serves several. Employees working next to agents need support too. BCG advises companies to put about 70% of their AI transformation effort into people and processes, so training and change management should start on day one. “The repeatable tasks can be handed over to the agent, but it still needs human judgment, it needs expertise, it needs the human in the loop,” Spannbauer said.

• Oversight at volume: Keeping a human in the loop gets harder as AI produces more work. In software engineering, AI now writes code faster than engineers can review it. “There are senior engineers able to review, but considering the amount of code generated, this is not doable anymore. You have to find other ways,” Spannbauer said. Teams can ask the AI to explain how it reached a result and what assumptions it made, then test those assumptions. That lets reviewers judge whether the work is sound without reading every line. AI can also track running systems and alert the right people when something looks off, so problems surface early.

Large AI transformation programs carry a lot of risk when the technology changes this quickly. Small pilots give airlines a cheaper way to learn. Each pilot starts with the business outcome it should deliver, and the ones that hit their targets get scaled. Keeping each investment small also prevents companies from locking into one vendor or model before they know what works.

New models with new capabilities come out every few days, and no one knows whether progress will speed up or slow down. Airlines that plan for that uncertainty can adopt better tools as they arrive without redoing their work. “You need the flexibility to adapt, because what may be a limitation today could already be solved by default tomorrow,” Spannbauer said.

Most airlines are still building the shared data and clear ownership that agents need to work across the business. The ones that finish that groundwork now will be ready to expand their use of agents as the technology matures. “We are still at the beginning in many cases, but the potential is huge,” he said.