The ACA Corner: October insights from Vikt
The ACA Corner presented in partnership with Official Media Partner PAX International is a column dedicated to voicing expertise directly from ACA members. ACA represents more than 65 percent of the worldwide airline catering market. ACA invites all members to share their views in future editions of the ACA Corner by contacting Mercedes Dieguez.
Supplier's perspective: Manuel Huebschmann, Chief Solution Advisor, Vikt, and ACA Associate Member
Do not buy the pitch. Buy the proof. How airline and caterer decision makers should evaluate AI solutions.
With the summer peak behind us, many airlines and caterers are reviewing their operations and exploring new IT solutions, and AI tools are increasingly on the shortlist. Yet they are often evaluated just like previous IT solutions, even though AI isn't configured like traditional software. It learns from actual operational data, so it is probabilistic, not deterministic. A sales demo, however, runs on data cleaned up for the occasion, so it proves little about real performance.
In my experience with operational AI solutions, the following questions have emerged to separate a capable partner from a convincing presentation.
- Domain fit. Does the solution provider understand airline and catering operational complexity, or is this a generic model with an aviation label attached? Ask what problems they have solved.
- Consistency. Are its outputs consistent and explainable? Airline and catering systems have no tolerance for hallucination, and an operations manager who can't understand why a forecast changed won't trust it.
- Resilience. Can it cope with inconsistent operational data, from manual packing and trolley loading to aircraft swaps and hand-entered crew sales data?
- Adaptability. Can it adjust to new business rules, routes or catering processes without a costly model tuning change request? If every change becomes a project, the savings quickly disappear into the invoice.
These questions are best answered in a data simulation or pilot with actual operations data, not a provider's demo set. If time permits, a one-month A/B pilot across comparable routes, with and without AI, would add additional certainty of the solution fit.
Once live, require monthly ROI transparency, fees against measured impact, not just a business case written before go-live. With the right provider, the learning curve, which is much more dynamic with AI solutions, slopes upwards, not downwards.
Proof also often reveals more than model performance. In our most recent pilot with a European airline, the waste reports used as its baseline did not capture all of the loading: they showed a waste rate of around 35 percent, while the real figure was around 45 percent. The data simulation showed that 25 percent was achievable, yet a reduction of 20 percentage points rather than 10 changed the urgency entirely. The airline set out to test an AI model and came away understanding its own operation better.
As this season's reviews and tenders get under way, I encourage you to write the proof into your evaluation criteria for AI solutions. The providers worth choosing will welcome it.
Manuel Huebschmann
Chief Solution Advisor
Vikt
and ACA Associate Member
Manuel Huebschmann, Chief Solution Advisor, Vikt, and ACA Associate Member
