World Aviation Festival 2026 | Industry Insights
Artificial intelligence is no longer a distant ambition for aviation. Airlines are experimenting with generative AI, developing intelligent agents and exploring new ways to automate decisions, support employees and improve passenger experiences.
Yet an important question remains: how can airlines move beyond promising pilots and make AI work reliably, responsibly and profitably across their organisations?
At World Aviation Festival 2026, Ink Innovation joined industry leaders from Norwegian, JetBlue, United Airlines and Air Canada for a panel - βDiscussing the challenges of driving successful strategic AI adoption, impactful solutions and evaluating the financial return.β
Moderated by Alex Goman, Director at STS Aero, the session brought together:
- Ivan Karlovic, Director, Data Analytics & AI, Norwegian
- Kumar Abhishek, Director, Data Science, JetBlue
- Kushal Pasari, Global Head of AI/ML, United Airlines
- Tauseef Khan, Manager, Data & AI, BI Platforms, Air Canada
- Oana Savu, Head of Growth, Ink Innovation
The discussion explored the practical realities behind AI adoption: data readiness, technology selection, organisational change, operational applications and the challenge of demonstrating a meaningful return on investment.
Several important themes emerged.
1. Successful AI starts with trusted data, not technology
β2. Not every problem needs generative AI
β3. Moving from a successful pilot to enterprise-wide adoption
4. Measuring AI success means measuring business outcomes
5. Disruption management presents a significant opportunity for AI
6. The next stage of AI adoption is operational transformation
Let's break it down.
1. Successful AI starts with trusted data, not technology
Before airlines can scale AI, they need to establish confidence in the information powering it.
Tauseef Khan emphasised the importance of strong data foundations, clear ownership and consistent governance. If employees spend weeks finding, validating or combining information, introducing AI does not automatically solve the underlying problem.
Drawing on his experience at Air Canada, he discussed how technologies such as Collibra and Snowflake support data governance and analytical capabilities. Equally important is collaboration between business teams and data owners to establish shared definitions and trusted datasets.
After all, the same operational or commercial metric can mean different things to different departments.
Kushal Pasari introduced another important dimension: the institutional knowledge held by airline employees. Much of an organisation's operational intelligence exists in people's experience, established practices and everyday decision-making, rather than in formal databases.
Capturing this knowledge and incorporating employee feedback into AI-supported workflows could help airlines improve the accuracy and usefulness of their systems.
Meanwhile, Kumar Abhishek highlighted the emerging shift from traditional dashboards towards natural-language interactions with data, making information potentially more accessible to employees across an airline.
The discussion highlighted that AI adoption depends not only on new technology but also on making organisational knowledge accessible, reliable and useful.
2. Not every problem needs generative AI
With growing interest in generative and agentic AI, airlines face an increasingly important decision: choosing the right technology for the right problem.
Ivan Karlovic discussed Norwegian's work connecting AI capabilities with its existing data environment, including Snowflake and dbt, alongside initiatives involving external and unstructured information.
This is particularly relevant in aviation, where valuable operational knowledge is distributed across standard operating procedures, spreadsheets, partner systems and other sources that do not necessarily sit within a conventional data warehouse.
He also highlighted Norwegian's work with Google on AI-based weather prediction.
However, greater technical capability does not mean AI should become the default solution for every business challenge.
Generative and agentic systems remain probabilistic. They can produce inaccurate information, fail during intermediate steps or deliver inconsistent results. Established analytics, business intelligence and conventional automation may still be more appropriate for certain tasks.
Regulatory considerations, including GDPR, add another layer to technology selection and implementation.
Oana Savu, Head of Growth at Ink Innovation, brought the discussion back to the business and operational context.
While AI applications in shopping and customer engagement attract significant attention, the operational data layer deserves equal consideration.
For technology providers and airlines alike, the objective should extend beyond demonstrating what AI can predict or automate. The more important question is how it changes an existing workflow, supports employees and improves the resulting service.
AI should be selected because it solves a business problem, not because a business problem can be adapted to accommodate AI.
3. Moving from a successful pilot to enterprise-wide adoption
An AI proof of concept can demonstrate impressive results in a controlled environment. Deploying the same capability consistently across an airline is a different challenge.
Scaling requires reliable infrastructure, governance, clearly defined ownership and employees who understand both the capabilities and limitations of the tools they use.
Tauseef highlighted the importance of progressing from validated information towards certified datasets that can be trusted and reused across business functions.
He also raised a critical consideration for long-term adoption: the economics of generative AI.
Although the cost per token may decrease, increased usage can drive overall expenditure higher. Airlines, therefore, need to understand the full operating cost of their AI applications rather than assuming that falling unit prices automatically translate into better financial performance.
Kumar shared examples of work involving operational agents and bots, including applications designed to identify flights at risk of delay and assess whether potential delays may be controllable.
These capabilities demonstrate the opportunity for AI to support operational decision-making, while also highlighting the work required to progress from identifying a problem to reliably recommending or supporting action.
The growing range of AI products also makes the traditional build-versus-buy decision more complex.
For airlines, the challenge is not simply to find a promising solution. It is to determine whether that solution can be integrated, governed, maintained and scaled within an already complex operational environment.
4. Measuring AI success means measuring business outcomes
How should airlines evaluate AI's financial return?
The panel explored why conventional measures of technology adoption do not always capture the value an AI application creates.
Kushal argued for evaluating operational impact rather than AI activity in isolation.
For example, the success of an AI-powered customer service tool should not be determined solely by how many employees use it. More meaningful indicators include whether it helps agents resolve customer issues faster, navigate legacy systems more effectively or deliver better outcomes.
Ivan raised a related challenge: translating productivity improvements into measurable financial value.
If AI reduces a task from 45 minutes to five, the efficiency gain is evident. But how does that improvement translate into financial performance?
The answer depends partly on what happens to the time saved.
When employees can redirect their attention towards higher-value activities, the business benefit may be substantial, even if it does not immediately appear as a distinct saving on the profit and loss statement.
This reinforces the importance of establishing a clear business case before implementation.
Rather than measuring AI adoption as an objective in itself, airlines need to connect each application to outcomes such as improved operational performance, reduced processing time, better resource utilisation, service quality or lower overall costs.
The complete economics matter too: implementation, infrastructure, ongoing consumption, maintenance and governance must all be considered alongside the value generated.
5. Disruption management presents a significant opportunity for AI
One of the most compelling areas of discussion was the potential for AI to improve irregular operations (IROPS) and disruption recovery.
Airline disruption management involves numerous interconnected decisions, often distributed across operational departments, customer-facing teams and separate technology systems.
Oana Savu highlighted how this fragmentation can make it difficult to manage the complete operational, financial and passenger impact of a disruption.
Even when individual teams optimise their own activities, the overall outcome may remain inefficient if responsibilities and information are disconnected.
This creates an opportunity to rethink how airlines coordinate disruption management.
AI could help connect information, identify emerging risks and support more informed decisions across operational functions. However, achieving meaningful improvements also requires airlines to examine the processes and organisational structures surrounding those decisions.
Who owns the end-to-end disruption outcome? How are decisions coordinated between teams? And how can airlines evaluate the wider financial and customer impact of an operational intervention?
Kumar's example of using AI-supported tools to identify flights at risk of delay illustrates how these capabilities could develop.
Predicting a potential disruption is valuable. Turning that prediction into timely, reliable operational action is where the greater opportunity lies.
For airlines, this means progressing beyond isolated intelligence towards more connected decision-making, supported by appropriate human oversight.
6. The next stage of AI adoption is operational transformation
Across the discussion, a consistent theme emerged: successful AI adoption cannot be treated as a standalone technology initiative.
It requires alignment between data, people, processes and commercial objectives.
Airlines need trusted information before they can depend on intelligent outputs. They need to select technologies according to the problem being solved. They need governance and infrastructure that can support enterprise-wide deployment. And they need a clear understanding of how new capabilities will change the work employees perform.
Perhaps most importantly, they need to define success in operational and financial terms.
From AI experimentation to measurable impact
The conversation at World Aviation Festival 2026 highlighted an important shift in how the aviation industry approaches artificial intelligence.
The question is no longer simply what AI can do.
The discussion reinforced that moving beyond AI pilots requires more than technology. For airlines, the challenge is to integrate AI into everyday operations, establish clear ownership and demonstrate its value through measurable business outcomes.
