Why AI Pilots Don't Change the Bottom Line

Adam Nahari of Apollo Global Management explains his approach to AI transformation, using an airline crew-scheduling project as his example.

Start with the business problem

Pick one or two initiatives tied to a financial measure, with an agreed baseline, usable data, a named user champion and executive backing.
Nahari divides business AI into general productivity tools, domain-specific assistants and systems embedded in core operations. He argues that the last category can create more durable financial gains.
His five-to-one ratio of change management to engineering is a mental model from his experience, not a measured law.

When a flight disruption occurs, a scheduler must find a legal crew assignment under time pressure; a first feasible solution need not be the cheapest.
Nahari says the project combined flight, crew and maintenance data with an optimization engine and a decision-support app. It ranked possible solutions by cost in under a minute.
His “10% versus 10x” comparison is illustrative. He does not disclose a measured tenfold financial return or actual savings for the airline.
Crew schedulers helped design the interface, participated in testing and promoted it to peers; the scheduling VP co-sponsored the project.

Secure visible CEO support and a business owner who can prioritize staff, funding and data access; renew that alignment over time.
Give stakeholders hands-on experience with AI so they can identify realistic uses.
Set stop/go criteria before a pilot: Nahari describes 90-day stage gates, baseline-based A/B tests and an adoption threshold above 50% of schedulers.
Have the CFO compare expected impact against costs at scale, including compute, tokens and staff. The talk presents a management framework and one case, not an independent controlled study of its financial effects.

“Picking the right problem, creating buy-in, driving adoption, measuring impact, that’s where the real time and effort goes.”