Your AI Pilots Won't Move the P&L. Here's What Will

Adam Nahari, Apollo Global Management15:52 · Nov 2026 · 3,317 views
Thumbnail for Your AI Pilots Won't Move the P&L. Here's What Will Watch on YouTube
TL;DR
  1. 1

    Most companies use AI, but only a small share see significant EBIT impact because time savings often do not reach the P&L.

  2. 2

    The largest gains come from redesigning a core workflow around AI, rather than adding a tool to an existing process.

  3. 3

    Successful transformations start with one measurable business problem, have committed owners, involve the business in building the solution, and use kill metrics before the pilot begins.

Summary

Adam Nahari argues that AI transformation is mainly a change-management problem. General-purpose copilots can improve individual productivity, but those gains often get competed away or remain invisible in company margins. Larger financial impact comes from core operational AI embedded in a business process. He uses an airline crew-scheduling example: instead of making a scheduler slightly faster, the company built an optimization system that considered constraints, generated possible crew solutions, ranked them by cost, and supported a redesigned workflow. The engineering was smaller than the work of getting the system adopted. Nahari's six moves are to choose one or two P&L-linked problems, redesign the process end to end, secure CEO commitment, build with business owners and users, give employees hands-on experience with AI, and define time, performance, adoption, and economic gates before starting. The airline moved from finding one answer in 15 minutes to finding the answer in less than a minute.

Key ideas
01:22

AI adoption does not automatically produce EBIT impact

Nahari contrasts widespread AI use with limited financial results. Almost 90% of businesses report using AI in at least one function, while only 6% report significant EBIT impact. Reported time savings add up to about 2% of total US working hours, yet almost none appears in the P&L. He connects this pattern to Robert Solow's 1987 observation about computers and productivity statistics.

02:42

Change management takes more work than engineering

For the AI transformations Nahari has led or watched closely, he uses a mental model of five units of change management for every one unit of engineering. Choosing the right problem, creating buy-in, driving adoption, and measuring impact take most of the effort. He says the engineering share is shrinking as models improve, while failures still tend to happen around adoption and business change.

03:42

Core operational AI has a better chance of retaining margin

Nahari describes three layers of business AI. General productivity tools include copilots and chatbots, but their gains tend to appear in prices rather than margins because competitors can access the same tools. Domain-specific AI improves retrieval and use of enterprise knowledge. Core operational AI is embedded in the processes that make a business work, such as airline crew scheduling. These systems are harder to buy off the shelf, so their gains are more likely to remain in margin.

05:27

Start with one measurable problem tied to the P&L

Nahari's first move is to pick one or two problems instead of 40 and work backwards from the P&L. In the airline example, crew was the second-largest cost after fuel, so the team focused there. A good use case moves a line the CFO recognizes, has an agreed baseline, uses ready infrastructure and company-owned data, has a named user who can champion it, and belongs among the CEO's top priorities.

07:21

Redesign the workflow instead of adding a faster tool

Nahari distinguishes an incremental 10% play from a 10x play. The incremental version would give the existing scheduler a tool that makes the current process somewhat faster. The larger redesign built an optimization engine that considered all constraints, generated possible solutions, ranked them by cost, and produced results in under a minute. The team also consolidated flight, crew, and maintenance data and built a decision-support application.

08:59

Executive ownership and business participation determine adoption

Nahari will not fund an initiative unless the CEO has publicly committed to it. Without that commitment, data requests and staffing decisions repeatedly lose priority. The P&L owner, CFO, and CTO each have distinct responsibilities, and their alignment must be renewed. In the airline, the VP of crew scheduling co-sponsored the work, while schedulers helped design the interface, took part in A/B testing, and persuaded peers to use it.

11:42

People need hands-on AI experience before they can redesign work

Nahari says it is impossible to redesign a workflow with AI if the people involved have never used AI tools. At a workshop with around 30 executives, three out of four had never used an AI tool until that week. He treats experimentation as the minimum requirement for developing intuition about where AI can create value and which use cases are possible.

12:40

Kill metrics should be set before a pilot starts

The airline initiative used three gates. Each stage had a 90-day time gate. A metric gate compared A/B test results with a baseline agreed in advance. An adoption gate required more than 50% of crew schedulers to use the tool. The team also worked with the CFO to check whether the scaled cost, including compute, tokens, and staff time, was significantly lower than the initiative's expected impact.

"Experimentation with the tools is the absolute minimum requirement to drive this type of transformation."12:23
Who should watch
  • You are funding AI pilots but cannot explain how they will affect a CFO-recognized line in the P&L.
  • Your team is improving an existing workflow one tool at a time and needs a concrete case for redesigning it end to end.
  • You need operating rules for ownership, employee adoption, pilot gates, and the cost of scaling an AI system.