{
  "video": {
    "id": "vUJDyDnqT4c",
    "title": "Your AI Pilots Won't Move the P&L. Here's What Will - Adam Nahari, Apollo",
    "duration": 952,
    "upload_date": null,
    "channel": "AI Engineer",
    "source": "AI Engineer",
    "event_format": ""
  },
  "analysis": {
    "video_id": "vUJDyDnqT4c",
    "title": "Your AI Pilots Won't Move the P&L. Here's What Will - Adam Nahari, Apollo",
    "one_liner": "Getting AI to move the P&L is mostly a change-management problem, not an engineering one: Nahari's rule of thumb is five units of change management for every one unit of engineering. The 6% of companies seeing real EBIT impact redesign whole core workflows end to end, with CEO backing, embedded users and kill metrics set up front.",
    "summary": "Nearly 90% of businesses use AI, but only 6% report significant EBIT impact. Nahari argues this repeats Solow's 1987 productivity paradox. Gains from general-purpose copilots get competed away, while core operational AI built into the processes that define a business keeps its gains in margin. Using a cargo airline's crew-rescheduling project as the running example, he sets out six moves: pick one or two problems, reimagine the process end to end, start with the CEO, build with the business, put tools in everyone's hands, and set kill metrics first with the CFO keeping score.",
    "key_points": [
      "Adoption is fast but P&L impact is small. Over 55% of US adults have used generative AI within 3 years, compared with 20% for the PC and 30% for the internet at the same point. Almost 90% of businesses use AI in at least one function, but only 6% report significant EBIT impact. Reported time savings add up to about 2% of US working hours, and almost none of it shows in the P&L.",
      "Mental model: for every five units of change management there is one unit of engineering. As models improve, the engineering share keeps shrinking. Picking the problem, creating buy-in, driving adoption and measuring impact are where the effort goes and where failures happen.",
      "Three layers of AI in business. (1) General productivity (copilots and chatbots): everyone has access, so gains show up in price, not margin. (2) Domain-specific AI: function-specific copilots and retrieval over business documents. (3) Core operational AI: vertical systems embedded in the business processes. Because these can't be bought off the shelf, their gains stay in margin. Companies with material EBIT impact are 3x as likely to redesign a whole workflow end to end.",
      "Airline example. When an aircraft fails, a scheduler has 15 minutes to re-crew the flight by hand, working through rest rules, qualifications and pilot locations. They often find one legal solution that isn't the cheapest, or none, in which case premium pay is the fallback. Crew is the second-largest cost after fuel.",
      "The '10x play' (as opposed to a 10% tool that makes the scheduler a bit faster) was an optimization agent engine. It takes all the constraints into account, produces every possible solution and ranks them by cost, all in under a minute. It required consolidating flight, crew and maintenance data, building the engine, building a decision-support app and redesigning the scheduler workflow.",
      "Five criteria for a good use case: it moves a line the CFO recognizes; it's measurable against a baseline agreed upfront; the tech and data infrastructure is ready and the company owns the data; a named person (not a department) wants it and will champion it; and it's one of the CEO's top five priorities.",
      "Nahari won't fund an initiative unless the CEO has publicly committed to it as a priority. Without that, you get scheduling conflicts, data requests that take weeks, and pilots that get delayed. Alignment wears off over time and has to be renewed.",
      "Kill gates set before the pilot: a time gate (90 days per stage), a metric gate (A/B testing against a baseline decided in advance) and an adoption gate (over 50% of crew schedulers using the tool). There is also a CFO test that the cost at scale, including compute, tokens and people, is significantly lower than the impact. 'Many pilots die at this stage.'"
    ],
    "takeaways": [
      "Work backwards from the P&L to a North Star metric (e.g. EBITDA) and its drivers. Pick one or two needle-moving initiatives, not 40 pilots.",
      "Redesign the whole workflow around the new technology instead of automating an existing process that may be obsolete, and invest in layer-three core operational AI as well as copilots.",
      "Get a public CEO commitment before putting money at risk. Make the P&L owner a co-sponsor, and embed named end-user champions in design from day one: they met engineering every other day, co-designed the UI, took part in A/B testing and drove adoption among peers. Hire leaders with both AI and domain expertise; here that meant a CIO, a VP of data and AI, and a senior director of data science.",
      "Make sure decision-makers actually use AI tools themselves. In one executive workshop, three of four attendees had never used any AI tool before that week, and you can't redesign a workflow with AI if you've never used it.",
      "Write down kill metrics (time, metric and adoption gates) before starting a pilot, and work with the CFO to check that unit economics at scale, including compute and token costs, are clearly below the impact."
    ],
    "topics": [
      "ai transformation",
      "change management",
      "enterprise ai",
      "roi",
      "workflow redesign",
      "optimization",
      "adoption",
      "kill metrics"
    ],
    "tools": [
      "Apollo"
    ],
    "quotes": [
      {
        "text": "For every five units of change management, there's only one unit of engineering.",
        "at": "02:51",
        "url": "https://www.youtube.com/watch?v=vUJDyDnqT4c&t=171s"
      },
      {
        "text": "If they haven't, I don't fund the initiative.",
        "at": "09:24",
        "url": "https://www.youtube.com/watch?v=vUJDyDnqT4c&t=564s"
      },
      {
        "text": "It's impossible to redesign a workflow with AI if you've never used AI.",
        "at": "12:09",
        "url": "https://www.youtube.com/watch?v=vUJDyDnqT4c&t=729s"
      },
      {
        "text": "the ability to absorb these models is a competitive differentiator,",
        "at": "15:07",
        "url": "https://www.youtube.com/watch?v=vUJDyDnqT4c&t=907s"
      }
    ],
    "words": 2075
  },
  "summary_url": "/#vUJDyDnqT4c",
  "transcript": {
    "html": "/transcripts/vUJDyDnqT4c.html",
    "txt": "/transcripts/vUJDyDnqT4c.txt",
    "vtt": "/transcripts/vUJDyDnqT4c.vtt"
  }
}