{
 "byKey": {
  "cover": [
   {
    "say": "Welcome back to Business Model Design. In this chapter, we ask how A I actually changes a business model.",
    "item": 0
   },
   {
    "say": "We'll look at cost structure, value propositions, the difference between A I-enhanced and A I-native, and when not to use A I at all.",
    "item": 0
   },
   {
    "say": "Then we'll work through a full numbers case on customer support, so you can see where the profit really goes.",
    "item": 0
   }
  ],
  "Principles: How AI Changes Cost Structure and Value Propositions#1": [
   {
    "say": "Let's start with a small story about a fictional company called Pine Street Translation.",
    "item": 0
   },
   {
    "say": "They have five translators doing English-Spanish document work. The owner has them use A I for a first draft, then proofread it themselves.",
    "item": 1
   },
   {
    "say": "Three months later, three things happened. The same five people can take on more work, and competitors start bidding lower.",
    "item": 1
   },
   {
    "say": "Customers also ask, if A I did the translating, why isn't it cheaper? And once, the A I mistranslated a currency unit in a contract, which the proofreader caught just in time.",
    "item": 1
   },
   {
    "say": "Those are this chapter's three themes: cost structure, value and price expectations, and new quality risks. A I is not magic. It reshuffles a few boxes of the business model.",
    "item": 1
   }
  ],
  "Principles: How AI Changes Cost Structure and Value Propositions#2": [
   {
    "say": "Remember the nine-box business model canvas from Chapter 1? A I rarely changes all nine at once. It usually moves a handful.",
    "item": 0
   },
   {
    "say": "First, the value proposition. It becomes faster, cheaper, more personalized, and sometimes it offers things that were impossible before, like twenty-four-hour replies or instant summaries.",
    "item": 1
   },
   {
    "say": "Second, the cost structure. For repetitive knowledge work, cost moves from payroll to model usage fees, review labor, and upkeep.",
    "item": 2
   },
   {
    "say": "Third, key activities and resources shift from lots of manual operation to organizing data, designing workflows, and monitoring quality.",
    "item": 3
   }
  ],
  "Principles: How AI Changes Cost Structure and Value Propositions#3": [
   {
    "say": "Two more boxes get touched as well.",
    "item": 0
   },
   {
    "say": "Customer relationships and channels. One-to-one interaction becomes affordable, but it can also feel cold to customers, so watch for that.",
    "item": 1
   },
   {
    "say": "And the revenue model. When delivery costs fall, customers expect prices to fall too. Pricing by the hour or by the item may need to change, and Chapter 9 goes deeper on that.",
    "item": 2
   }
  ],
  "Principles: How AI Changes Cost Structure and Value Propositions#4": [
   {
    "say": "Now let's look at cost structure more closely. Traditional software is expensive to build, but one more user adds almost nothing, which gives it beautiful gross margins.",
    "item": 0
   },
   {
    "say": "A I products are different. Each use usually calls a model, which creates an inference cost billed by usage. More usage means higher cost.",
    "item": 1
   },
   {
    "say": "Second, output can be wrong, so you need review or spot checks. This quality cost is often overlooked.",
    "item": 2
   },
   {
    "say": "Third, you must keep maintaining the knowledge base, the prompts, and the evaluation data. That's a fixed upkeep cost.",
    "item": 3
   }
  ],
  "Principles: How AI Changes Cost Structure and Value Propositions#5": [
   {
    "say": "There's one more piece to the cost picture.",
    "item": 0
   },
   {
    "say": "Prices and features vary by vendor and version, so check the official documentation. Vendors can also change them, which makes your costs uncertain.",
    "item": 1
   },
   {
    "say": "So A I swaps part of the payroll for variable inference fees, fixed upkeep, and quality cost. It pays off only if those three add up to less than the old labor cost.",
    "item": 1
   },
   {
    "say": "Don't judge by how cheap a single model call looks.",
    "item": 1
   }
  ],
  "Principles: How AI Changes Cost Structure and Value Propositions#6": [
   {
    "say": "On the value side, A I usually creates four types of value for customers.",
    "item": 0
   },
   {
    "say": "Speed, because things are done in seconds. Cost, through a lower price per unit of service. Availability, because it works at any hour. And personalization, where everyone gets different content.",
    "item": 1
   },
   {
    "say": "But A I also brings new worries. Will it make mistakes? Will my data leak? Who is responsible if something goes wrong?",
    "item": 1
   },
   {
    "say": "A good A I value proposition spells out the result, how reliable the quality is, and what happens when it's wrong. Customers buy results and trust, not buzzwords.",
    "item": 1
   }
  ],
  "Principles: How AI Changes Cost Structure and Value Propositions#7": [
   {
    "say": "Let's be honest about the limits, because A I does not automatically become profit.",
    "item": 0
   },
   {
    "say": "Efficiency is not profit. If competitors are all cutting prices, customers capture the efficiency gain, and you may only keep the profit you already had. The case study shows this with numbers.",
    "item": 1
   },
   {
    "say": "Model capability gets commoditized. What needs special training today may be a standard feature tomorrow, so knowing how to use one model is hard to turn into a lasting advantage.",
    "item": 2
   },
   {
    "say": "Quality has a ceiling. On some tasks A I gets you eighty to ninety percent of the way, and the last part is what customers care about most.",
    "item": 3
   },
   {
    "say": "And in areas like personal data, health, finance, or law, an error costs far more than the savings. Rules differ by state and industry, so check with counsel. This is not legal advice.",
    "item": 3
   }
  ],
  "Concepts: AI-Native, AI-Enhanced, and the Three Levers#1": [
   {
    "say": "Let's tell A I-enhanced apart from A I-native. The first way to classify an A I business model is by the role A I plays in the product.",
    "item": 0
   },
   {
    "say": "A I-enhanced means the product already exists and has customers, and A I is a feature added to it. Think of Lineleaf Books, a fictional bookstore chain adding A I reading recommendations to its app.",
    "item": 1
   },
   {
    "say": "Stores, memberships, and shipping don't change. Take away the A I and the product is still there, just a bit worse.",
    "item": 1
   },
   {
    "say": "A I-native means the core value depends on A I. Instant Meet, also fictional, turns a meeting into a transcript in twelve languages and organizes action items in real time.",
    "item": 1
   },
   {
    "say": "Doing that by hand would cost far too much to offer at this price. Without A I, the product could not exist.",
    "item": 1
   }
  ],
  "Concepts: AI-Native, AI-Enhanced, and the Three Levers#2": [
   {
    "say": "These two types mean different things for the business, so let's compare them across five dimensions.",
    "item": 0
   },
   {
    "say": "A I-enhanced starts from existing customers and revenue. A I-native starts from a new customer job or a new price tier.",
    "item": 1
   },
   {
    "say": "The main benefit of A I-enhanced is lower cost, better experience, and higher retention. A I-native can create a new category and new use cases.",
    "item": 2
   },
   {
    "say": "The main risk for A I-enhanced is limited payoff, because it can be seen as a free extra. For A I-native, model cost and vendor dependence can decide survival.",
    "item": 3
   },
   {
    "say": "Pricing and moat differ too. Enhanced features are often bundled into existing plans, while native products can charge by usage or outcome. Neither type is better than the other.",
    "item": 4
   },
   {
    "say": "Enhanced has customers and cash flow, so experiments are cheap. Native may open a new market, but risk is more concentrated. A common path is enhanced first, native later.",
    "item": 4
   }
  ],
  "Concepts: AI-Native, AI-Enhanced, and the Three Levers#3": [
   {
    "say": "Let's walk through that table row by row.",
    "item": 0
   },
   {
    "say": "Row one, the dimension headers: A I-enhanced on one side and A I-native on the other.",
    "item": 1
   },
   {
    "say": "Starting point: existing customers and revenue, versus a new customer job or a new price tier.",
    "item": 2
   },
   {
    "say": "Main benefit: lower cost, better experience, and higher retention, versus creating a new category and new use cases.",
    "item": 3
   }
  ],
  "Concepts: AI-Native, AI-Enhanced, and the Three Levers#4": [
   {
    "say": "Next, the two rows about risk and pricing.",
    "item": 0
   },
   {
    "say": "Main risk: for enhanced, limited payoff and being seen as a free extra. For native, model cost and vendor dependence decide survival.",
    "item": 1
   },
   {
    "say": "Pricing: enhanced is often bundled into existing plans, so it's hard to charge extra. Native can charge directly by usage or by outcome.",
    "item": 2
   }
  ],
  "Concepts: AI-Native, AI-Enhanced, and the Three Levers#5": [
   {
    "say": "The last row is the moat.",
    "item": 0
   },
   {
    "say": "For A I-enhanced, the moat is existing customer relationships, data, and channels. For A I-native, it's workflow, accumulated data, brand, and trust.",
    "item": 1
   }
  ],
  "Concepts: AI-Native, AI-Enhanced, and the Three Levers#6": [
   {
    "say": "A I-enhanced has some common traps. Here are three to avoid.",
    "item": 0
   },
   {
    "say": "One: the feature looks cool, but customers won't pay more for it, so it only adds cost.",
    "item": 1
   },
   {
    "say": "Two: A I is added but the process doesn't change. The manual steps stay, so efficiency doesn't really improve.",
    "item": 2
   },
   {
    "say": "Three: A I is put in the step customers care about most, with no review mechanism. That's a risky place to skip a safety net.",
    "item": 3
   }
  ],
  "Concepts: AI-Native, AI-Enhanced, and the Three Levers#7": [
   {
    "say": "Now the three levers. The value A I brings to a business model can be grouped into automation, personalization, and scaling.",
    "item": 0
   },
   {
    "say": "Automation hands repetitive knowledge work from people to machines. Track labor time per unit of work and unit cost. Think auto-sorting support tickets or organizing meeting notes.",
    "item": 1
   },
   {
    "say": "Personalization shows different content to each person. Track conversion rate, retention, and average order value. It used to take lots of labor, and A I makes it affordable.",
    "item": 2
   },
   {
    "say": "Scaling lets the same team serve more customers or markets. Track customers served per employee and the number of languages or regions covered.",
    "item": 3
   },
   {
    "say": "The levers can stack, but be clear which one you mainly make money from, because each is measured differently and has different risks.",
    "item": 3
   }
  ],
  "Concepts: The AI Value Chain, When Not to Use AI, and Feasibility Checks#1": [
   {
    "say": "To see how A I makes money, start with the value chain. In simplified form, it has three layers.",
    "item": 0
   },
   {
    "say": "The foundation model layer trains and provides general-purpose large models. It's capital-intensive, and earns from A P I usage fees, subscriptions, and licensing.",
    "item": 1
   },
   {
    "say": "The tools and platform layer sits between models and applications, with cloud computing, retrieval, evaluation tools, and workflow frameworks. It earns through usage billing and subscriptions.",
    "item": 2
   },
   {
    "say": "The application layer faces end customers and solves a specific problem, like support, legal, marketing, or education. Revenue can come from subscriptions, per-seat pricing, usage, or outcomes.",
    "item": 3
   },
   {
    "say": "These layers don't earn separately. The application layer pays part of its revenue upstream, so upstream prices and policies directly affect your gross margin.",
    "item": 3
   }
  ],
  "Concepts: The AI Value Chain, When Not to Use AI, and Feasibility Checks#2": [
   {
    "say": "Each layer has its own strengths and risks.",
    "item": 0
   },
   {
    "say": "Foundation models have pricing power while they're technically ahead, but the investment is enormous and capability is easily caught up and commoditized.",
    "item": 1
   },
   {
    "say": "Tools and platforms are sticky because developers rely on them while building, but the layers above and below may build the same feature.",
    "item": 2
   },
   {
    "say": "Applications are closest to customers and context, and can accumulate data. But their costs are tied to upstream, and they're easy to copy.",
    "item": 3
   },
   {
    "say": "If you build at the application layer, you're not competing on the strongest model. You compete on who understands customers' work better. Chapter 9 returns to this.",
    "item": 3
   }
  ],
  "Concepts: The AI Value Chain, When Not to Use AI, and Feasibility Checks#3": [
   {
    "say": "Here's the same table, one row at a time.",
    "item": 0
   },
   {
    "say": "The columns are layer, strength, and risk.",
    "item": 1
   },
   {
    "say": "Foundation models: pricing power while technically ahead, against enormous investment and easy commoditization.",
    "item": 2
   },
   {
    "say": "Tools and platform: developers rely on it, so it's sticky, but layers above and below may build the same feature.",
    "item": 3
   }
  ],
  "Concepts: The AI Value Chain, When Not to Use AI, and Feasibility Checks#4": [
   {
    "say": "And the last row.",
    "item": 0
   },
   {
    "say": "Application: closest to customers and context, and it can accumulate data. The risks are costs tied to upstream, and being easy to copy.",
    "item": 1
   }
  ],
  "Concepts: The AI Value Chain, When Not to Use AI, and Feasibility Checks#5": [
   {
    "say": "Now an important question: when should you not use A I? Sometimes the best answer is not yet, or only as an assistant.",
    "item": 0
   },
   {
    "say": "First, when errors are very costly and hard to detect, like medical diagnosis, legal opinions, or safety-critical systems. A I can assist, but a qualified professional must take final responsibility.",
    "item": 1
   },
   {
    "say": "Second, when the task has clear rules and fixed inputs. Ordinary code or a spreadsheet can be completely correct, so you don't need a probabilistic A I.",
    "item": 2
   },
   {
    "say": "Third, when there's too little data or the data is poor. Without enough real cases, you can neither test nor improve.",
    "item": 3
   }
  ],
  "Concepts: The AI Value Chain, When Not to Use AI, and Feasibility Checks#6": [
   {
    "say": "Three more situations where you should think twice.",
    "item": 0
   },
   {
    "say": "Fourth, when you need full explainability and auditability. A I output is sometimes hard to justify, so be especially careful in tightly regulated settings.",
    "item": 1
   },
   {
    "say": "Fifth, when the customer is buying the person, like counseling companionship or high-end consulting. Customers value a real person's attention and accountability.",
    "item": 2
   },
   {
    "say": "Sixth, when volume is small and not repetitive. Build and upkeep would cost more than the labor you'd save.",
    "item": 3
   }
  ],
  "Concepts: The AI Value Chain, When Not to Use AI, and Feasibility Checks#7": [
   {
    "say": "The seventh case is about data.",
    "item": 0
   },
   {
    "say": "If you cannot let data leave, and you have no compliant data-handling arrangement, don't put confidential or personal information into outside services.",
    "item": 1
   },
   {
    "say": "Saying let's not use A I here isn't being timid. It's a business judgment, and putting limited resources where payoff is largest and risk is manageable is real competitiveness.",
    "item": 1
   }
  ],
  "Case Study: Cloudline Support Adds AI, Full Numbers#1": [
   {
    "say": "Now let's work through the case study with full numbers. Cloudline Support is a fictional company.",
    "item": 0
   },
   {
    "say": "It provides outsourced support for thirty small e-commerce stores, handling about twenty thousand tickets a month with ten agents.",
    "item": 1
   },
   {
    "say": "Each agent costs about two thousand dollars a month, fully loaded, and Cloudline charges clients one dollar eighty per ticket. Every number here is a design value, not data from a real company.",
    "item": 1
   },
   {
    "say": "The owner wants to adopt A I, and the first question is, which tickets can actually go to A I?",
    "item": 2
   }
  ],
  "Case Study: Cloudline Support Adds AI, Full Numbers#2": [
   {
    "say": "First, let's see the before-A I numbers.",
    "item": 0
   },
   {
    "say": "Labor cost is ten agents times two thousand dollars, so twenty thousand dollars a month.",
    "item": 1
   },
   {
    "say": "Divide that by twenty thousand tickets and the cost per ticket is one dollar.",
    "item": 2
   },
   {
    "say": "So gross profit per ticket is one dollar eighty minus one dollar, or eighty cents. That's a gross margin of about forty-four point four percent.",
    "item": 3
   }
  ],
  "Case Study: Cloudline Support Adds AI, Full Numbers#3": [
   {
    "say": "And the monthly total.",
    "item": 0
   },
   {
    "say": "Twenty thousand tickets times eighty cents gives sixteen thousand dollars of gross profit per month. That's the baseline we want to beat.",
    "item": 1
   }
  ],
  "Case Study: Cloudline Support Adds AI, Full Numbers#4": [
   {
    "say": "Step one is to sort the tickets into tiers. The team samples past tickets and sorts them into three types.",
    "item": 0
   },
   {
    "say": "Type A is lookups, about forty percent: order status, shipping progress, store hours. Clear rules and a fixed data source.",
    "item": 1
   },
   {
    "say": "Type B is processing, about forty-five percent: returns, address changes, re-shipments. These need policy judgment, but there are templates to follow.",
    "item": 2
   },
   {
    "say": "Type C is complex, about fifteen percent: complaints, disputes, and anything needing a calmed customer or cross-department coordination.",
    "item": 3
   },
   {
    "say": "The point is not to treat customer support as one task. Split it by risk, and decide for each type whether to use A I.",
    "item": 3
   }
  ],
  "Case Study: Cloudline Support Adds AI, Full Numbers#5": [
   {
    "say": "Step two is the feasibility check, using four gates.",
    "item": 0
   },
   {
    "say": "Gate one, data. There are three thousand closed tickets with correct answers, and personal information has been masked. That passes.",
    "item": 1
   },
   {
    "say": "Gate two, quality threshold. Type A needs at least ninety-five percent correct. Type B needs at least seventy percent of drafts usable as-is or with minor edits. Type C goes to people.",
    "item": 2
   },
   {
    "say": "In testing, Type A got one hundred ninety-four of two hundred correct, or ninety-seven percent. Type B had one hundred fifty of two hundred usable, or seventy-five percent.",
    "item": 2
   },
   {
    "say": "Gate three, human review. Type A gets a five percent spot check, about five cents per ticket. Type B is reviewed by an agent before sending, about thirty cents. Type C stays human at one dollar.",
    "item": 3
   },
   {
    "say": "Gate four is economics, which is the next step. The decision: Type A fully automated with spot checks, Type B A I draft plus human review, Type C stays human.",
    "item": 4
   }
  ],
  "Case Study: Cloudline Support Adds AI, Full Numbers#6": [
   {
    "say": "Step three is to compute the unit cost. Remember, all of these are design values, and real costs vary by vendor, version, and usage.",
    "item": 0
   },
   {
    "say": "The assumptions: inference costs five cents per ticket, monthly fixed cost is three thousand dollars, and Type A errors happen three percent of the time at ten dollars each. Build cost is twelve thousand dollars.",
    "item": 1
   },
   {
    "say": "Variable labor is fifteen percent times one dollar, plus forty-five percent times thirty cents, plus forty percent times five cents. That adds up to thirty and a half cents.",
    "item": 2
   },
   {
    "say": "Add inference and the variable subtotal is thirty-five and a half cents. Add twelve cents of error cost and fifteen cents of fixed cost. Total unit cost is sixty-two and a half cents.",
    "item": 2
   },
   {
    "say": "Monthly cost falls from twenty thousand to twelve thousand five hundred dollars, saving seven thousand five hundred. Twelve thousand divided by seven thousand five hundred is one point six months to pay back.",
    "item": 3
   },
   {
    "say": "But that holds only if the price stays the same, quality meets the threshold, and customers don't leave.",
    "item": 3
   }
  ],
  "Case Study: Cloudline Support Adds AI, Full Numbers#7": [
   {
    "say": "Let's recap those assumptions and results as formulas.",
    "item": 0
   },
   {
    "say": "Inference cost averages five cents per ticket, because every ticket goes through A I sorting and drafting.",
    "item": 1
   },
   {
    "say": "Monthly fixed cost is a thousand-dollar platform fee plus two thousand dollars of labor to maintain the knowledge base and prompts, so three thousand dollars.",
    "item": 2
   },
   {
    "say": "Type A auto-replies have a three percent error rate, and each error costs about ten dollars to fix.",
    "item": 3
   },
   {
    "say": "Put it together and gross profit per ticket is one dollar eighty minus sixty-two and a half cents, or one dollar seventeen and a half cents. That's a margin of about sixty-five point three percent.",
    "item": 3
   }
  ],
  "Key Takeaways: How AI Changes Business Models#1": [
   {
    "say": "Let's recap the chapter. Here's the map of what we covered.",
    "item": 0
   },
   {
    "say": "First, which boxes of the business model A I touches, mainly cost structure and value proposition.",
    "item": 1
   },
   {
    "say": "Second, A I-enhanced versus A I-native, which have different positioning and different risks.",
    "item": 2
   },
   {
    "say": "Third, the three levers: automation, personalization, and scaling.",
    "item": 3
   }
  ],
  "Key Takeaways: How AI Changes Business Models#2": [
   {
    "say": "The rest of the map.",
    "item": 0
   },
   {
    "say": "Fourth, the A I value chain: foundation models, tools and platforms, and applications.",
    "item": 1
   },
   {
    "say": "Fifth, when not to use A I, and the feasibility check covering data, quality threshold, human review, and economics.",
    "item": 2
   },
   {
    "say": "Sixth, the Cloudline case: unit cost, sensitivity, and the pricing decision.",
    "item": 3
   }
  ],
  "Key Takeaways: How AI Changes Business Models#3": [
   {
    "say": "Key point one is cost structure.",
    "item": 0
   },
   {
    "say": "A I swaps part of the payroll for inference cost, review labor, and upkeep cost.",
    "item": 1
   },
   {
    "say": "Marginal cost is not zero, since more usage means more inference cost. Prices and features vary by vendor and version, so check the official documentation.",
    "item": 2
   },
   {
    "say": "Total unit cost is variable labor plus inference plus error cost plus a share of fixed cost.",
    "item": 3
   }
  ],
  "Key Takeaways: How AI Changes Business Models#4": [
   {
    "say": "Key point two is value proposition and price.",
    "item": 0
   },
   {
    "say": "A good A I value proposition states the result, the quality boundary, and how errors are handled, not just the technology.",
    "item": 1
   },
   {
    "say": "An efficiency gain is not profit. Prices can be pushed down, so look at price, cost, and volume together.",
    "item": 2
   },
   {
    "say": "Model capability gets commoditized. The moat comes from data, workflow, integrations, brand, and trust.",
    "item": 3
   }
  ],
  "Key Takeaways: How AI Changes Business Models#5": [
   {
    "say": "Key point three covers enhanced, native, and the three levers.",
    "item": 0
   },
   {
    "say": "Enhanced adds A I to an existing product. Native doesn't exist without A I. A common path is enhanced first, native later.",
    "item": 1
   },
   {
    "say": "Automation saves labor time, personalization fits the individual better, and scaling lets the same resources serve more.",
    "item": 2
   },
   {
    "say": "Every lever has limits: exceptions, data and consent, and errors being copied.",
    "item": 3
   }
  ],
  "Key Takeaways: How AI Changes Business Models#6": [
   {
    "say": "Key point four is the value chain.",
    "item": 0
   },
   {
    "say": "The foundation model layer is capital-intensive, and earns from A P I fees, subscriptions, and licensing.",
    "item": 1
   },
   {
    "say": "The tools and platform layer offers development, retrieval, evaluation, and governance tools, billed by usage or subscription.",
    "item": 2
   },
   {
    "say": "The application layer is closest to the customer, but its costs are tied to upstream and it's easy to copy.",
    "item": 3
   }
  ],
  "Key Takeaways: How AI Changes Business Models#7": [
   {
    "say": "Key point five is when not to use A I, and the feasibility check.",
    "item": 0
   },
   {
    "say": "Don't use it, or use it only to assist, when errors are costly, rules are clear, data is thin, you need full explainability, customers want a person, volume is small, or data can't leave.",
    "item": 1
   },
   {
    "say": "The four gates are data, quality threshold tested on real samples, human review, and economics.",
    "item": 2
   },
   {
    "say": "And design error handling first: who finds an error, how it is fixed, and how it is prevented.",
    "item": 3
   }
  ],
  "closing": [
   {
    "say": "That's the end of this chapter. You now have a clear picture of how A I reshuffles cost, value, and risk in a business model.",
    "item": 0
   },
   {
    "say": "Before you move on, try the lab and run the four gates on a process from your own work.",
    "item": 0
   },
   {
    "say": "Then take the chapter quiz to check how much you've learned. I'll see you in the next chapter.",
    "item": 1
   }
  ]
 }
}