Nobody else could price a custom charter flight upfront, in an app. At Vetifly, we could do it because I spent weeks on the data before writing any logic.
In the app, a customer could build their own helicopter route and see the price before booking. Helicopter booking via an app was still new, and even private jet platforms hadn't touched this bit. Charter booking still worked through calls and emails, with someone checking feasibility, availability, allocation of aircraft type and then pricing. Typical response times ranged from a few hours to two days.
A lot of my time during Covid went into picking the brain of our COO, a helicopter pilot, working out the 'master data' for helicopters, helipads, routes as well as global configs, like what timings needed night-time certifications. Take the range of an aircraft, for example. The app used four different variables to get to this, and would then restrict which helipads could be selected when building the route.
Getting the data right first is something people working with AI are (re)discovering. Today, I can imagine someone pointing an AI at the aircraft manuals and helipad lists, and using the single spec-sheet range confidently, not knowing the other factors that affected the range in practice. That knowledge is likely still not written down. The charter booking specialists know it, pilots know it, but it isn't in the specs.
AI is very good at finding linkages across things that already exist somewhere. Unfortunately, it has no way of knowing what's missing. Someone still has to do the BA work of sitting with the expert and getting things out.
An additional question now gets added though, about who governs the data. At Vetifly, I had put the master data under the same change control as the algorithm, with periodic forced reviews and data completeness requirements for new aircraft or helipads. If any rule live only in an agent prompt, or in an assumption the model has made, nobody owns it and nobody notices when it's wrong.
So the starting point for an AI project isn't the model or the vendor. It needs all three: what data is needed, what data actually exists and who will own each piece of data once it's in the system.
Who will be the first to notice a wrong rule in your AI pilot?
#AIStrategy #AIGovernance #DataGovernance #Algorithms