{
 "byKey": {
  "cover": [
   {
    "say": "Welcome back to Product Marketing.",
    "item": 0
   },
   {
    "say": "In this chapter we look at AI marketing and sales automation: what AI can do, how to keep it safe, and how to work out whether it pays off.",
    "item": 0
   },
   {
    "say": "We finish with a worked case, LedgerLeaf, where we do the math together.",
    "item": 0
   }
  ],
  "Principles: The AI Marketing Workflow and Three Lines of Defense#1": [
   {
    "say": "Let's start with a scene. Maya is the marketer at an eight-person company.",
    "item": 0
   },
   {
    "say": "She used to write four ad copies a week, and each one took forty minutes to think up, rewrite, and run past the boss.",
    "item": 1
   },
   {
    "say": "Now she asks AI for twelve first drafts at once, then picks, rewrites, and fact-checks them. That lets her test twelve versions a week.",
    "item": 1
   },
   {
    "say": "But last month it almost went wrong. The AI wrote guaranteed results in three days, for a product with no such effect. Review caught it, luckily.",
    "item": 1
   },
   {
    "say": "That is the theme here. AI makes marketing faster and bigger, but it does not make you less responsible, and mistakes spread faster too.",
    "item": 2
   },
   {
    "say": "One note: every company, person, price, and number in this chapter is a fictional teaching assumption.",
    "item": 3
   }
  ],
  "Principles: The AI Marketing Workflow and Three Lines of Defense#2": [
   {
    "say": "Let's sort what AI actually does in marketing and sales. There are four kinds of work.",
    "item": 0
   },
   {
    "say": "First, generate: first drafts of copy, headline variants, newsletters, social posts, product descriptions, images, and short-video assets.",
    "item": 1
   },
   {
    "say": "Second, analyze and organize. Summarize customer comments, organize interview transcripts, group reviews by theme, and digest competitor information.",
    "item": 2
   },
   {
    "say": "Third, predict and rank, like lead scoring, churn-risk prediction, or recommending the next product that fits.",
    "item": 3
   }
  ],
  "Principles: The AI Marketing Workflow and Three Lines of Defense#3": [
   {
    "say": "The fourth kind is converse and automate.",
    "item": 0
   },
   {
    "say": "Think AI support chat, sales assistants, and emails or messages triggered automatically by customer behavior.",
    "item": 1
   },
   {
    "say": "Each kind has a different risk. With generation, text reads smoothly but may not be true. With prediction, biased data can distort the ranking.",
    "item": 1
   },
   {
    "say": "With conversation, the risk is promising customers things you cannot deliver. So know which kind you are using, and you will know what to guard against.",
    "item": 1
   }
  ],
  "Principles: The AI Marketing Workflow and Three Lines of Defense#4": [
   {
    "say": "Now the standard workflow. It has five steps: input, generate, review, publish, and feedback.",
    "item": 0
   },
   {
    "say": "Input means giving the AI clear material: product facts, price, what you may and may not claim, audience, brand voice, banned words, and examples.",
    "item": 1
   },
   {
    "say": "The more specific your material, the more usable the output.",
    "item": 1
   },
   {
    "say": "Generate means asking for several versions at once, and asking the AI to flag information a human needs to confirm.",
    "item": 2
   },
   {
    "say": "Review is where a person checks facts, promises, legally sensitive wording, and brand voice. The reviewer must have the authority to say no.",
    "item": 3
   }
  ],
  "Principles: The AI Marketing Workflow and Three Lines of Defense#5": [
   {
    "say": "Step three is not optional. AI is so fluent that it lulls people into relaxing, so review with a checklist, not a gut feeling.",
    "item": 0
   },
   {
    "say": "Here are the last two steps.",
    "item": 0
   },
   {
    "say": "Publish means test on a small scale first. Do not go all-in at once.",
    "item": 1
   },
   {
    "say": "Feedback means using data, like click-through rate, conversion rate, unsubscribes, and complaints, to decide which version is better.",
    "item": 2
   },
   {
    "say": "Then write what you learned back into your material library, so the next round is sharper.",
    "item": 2
   }
  ],
  "Principles: The AI Marketing Workflow and Three Lines of Defense#6": [
   {
    "say": "Next, three lines of defense: quality, privacy, and trust.",
    "item": 0
   },
   {
    "say": "Line one is quality. AI can invent numbers, cite sources that do not exist, and get features wrong. So give it only reliable material, ask it to flag uncertain spots, and verify.",
    "item": 1
   },
   {
    "say": "Line two is privacy and compliance. Before pasting a customer list or chat logs into an outside tool, ask whether this use fits what you told customers.",
    "item": 1
   },
   {
    "say": "Can you de-identify the data? How does the vendor store it? Are consent and unsubscribe in place? In the US, think FTC rules, CAN-SPAM, T C P A, and state laws like C C P A.",
    "item": 1
   },
   {
    "say": "If you serve E U customers, think G D P R too. These are general directions to check, not legal advice, so check with counsel.",
    "item": 1
   },
   {
    "say": "Line three is trust. Posing as a real person or misleading with synthetic images costs you more than one sale. Labeling rules depend on platform and regulator, so follow the latest guidance.",
    "item": 1
   }
  ],
  "Principles: The AI Marketing Workflow and Three Lines of Defense#7": [
   {
    "say": "This one deserves its own slide: never use AI to invent reviews or fake testimonials.",
    "item": 0
   },
   {
    "say": "AI can easily write, I've used it for three months and my skin looks better. It sounds real, but no customer said it.",
    "item": 1
   },
   {
    "say": "Reviews, expert endorsements, and before-and-after comparisons must come from real people with real experience, and you need their permission.",
    "item": 1
   },
   {
    "say": "AI can help you find themes in real reviews, polish a real story you have permission to use, or turn thank-you notes into interview questions. It must not make something up from nothing.",
    "item": 1
   },
   {
    "say": "This is not only ethics. In the US, the FTC has rules on endorsements and reviews, many places prohibit misleading testimonials, and platforms will penalize you too.",
    "item": 1
   }
  ],
  "Concepts: Personalization, Segmentation and Lead Scoring#1": [
   {
    "say": "Let's get three terms straight, because they are often mixed up.",
    "item": 0
   },
   {
    "say": "Segmentation splits customers into a few groups, and each group gets a different message. For example, just signed up, active users, and haven't logged in for a while.",
    "item": 1
   },
   {
    "say": "Personalization adjusts content within a group using individual data, like the name, a recently viewed product, or time since the last purchase.",
    "item": 2
   },
   {
    "say": "One-to-one is a design tailored for a single customer, usually reserved for high-value B to B accounts.",
    "item": 3
   },
   {
    "say": "Segment first, personalize second. Good segmentation already delivers most of the effect, so do not aim for a unique version per person at the start.",
    "item": 3
   }
  ],
  "Concepts: Personalization, Segmentation and Lead Scoring#2": [
   {
    "say": "What should you segment on? Traditional segmentation looks at age, gender, and location. In practice, behavior usually predicts the next step better.",
    "item": 0
   },
   {
    "say": "Look at what pages they viewed, like pricing, features, or blog.",
    "item": 1
   },
   {
    "say": "Look at what actions they took: added to cart, started a trial, or invited a teammate.",
    "item": 2
   },
   {
    "say": "Look at how long since they were last here, and how much they spend, meaning order value and purchase frequency.",
    "item": 3
   }
  ],
  "Concepts: Personalization, Segmentation and Lead Scoring#3": [
   {
    "say": "Here are the last two signals.",
    "item": 0
   },
   {
    "say": "How recently they interacted, and how much they spend.",
    "item": 1
   },
   {
    "say": "AI can find groups in large amounts of behavior data and write copy in a different voice for each group. But you must understand the segmenting logic yourself.",
    "item": 1
   },
   {
    "say": "If AI found five groups and nobody can say how they differ, that is a useless segmentation.",
    "item": 1
   }
  ],
  "Concepts: Personalization, Segmentation and Lead Scoring#4": [
   {
    "say": "Personalization has a line between useful and creepy. The goal is to be more relevant, not to make people feel watched.",
    "item": 0
   },
   {
    "say": "The test is simple: will the customer think, that makes sense, because I just did that?",
    "item": 1
   },
   {
    "say": "Reasonable: you looked at this insulated tumbler last week, and there's a sale on now.",
    "item": 1
   },
   {
    "say": "Possibly over the line: using inferences the customer never provided, like guessing a health condition or family change from purchase history, then stating it in an email.",
    "item": 2
   },
   {
    "say": "The more sensitive the data, like health, finances, or minors, the more conservative you should be. Confirm with counsel or another qualified professional.",
    "item": 2
   }
  ],
  "Concepts: Personalization, Segmentation and Lead Scoring#5": [
   {
    "say": "Now lead scoring. Say you get six hundred trial signups a month, and sales can only follow up with one hundred. Who do they call first?",
    "item": 0
   },
   {
    "say": "Scoring turns each lead's attributes and behaviors into a score, and you rank by score. There are two approaches.",
    "item": 1
   },
   {
    "say": "Rule-based scoring means you set the rules yourself. Model-based scoring trains a model on past closed and not-closed deals to predict the chance of closing.",
    "item": 2
   },
   {
    "say": "A model needs enough history. With too little data it is unstable, so beginners should start rule-based, which is transparent and explainable, then compare a model later.",
    "item": 2
   }
  ],
  "Concepts: Personalization, Segmentation and Lead Scoring#6": [
   {
    "say": "Here is an example of rule-based scoring.",
    "item": 0
   },
   {
    "say": "Viewed the pricing page adds twenty points.",
    "item": 1
   },
   {
    "say": "Logged in three days in a row during the trial adds twenty-five.",
    "item": 2
   },
   {
    "say": "Invited a teammate adds fifteen.",
    "item": 3
   },
   {
    "say": "Opened the last two emails adds ten.",
    "item": 4
   },
   {
    "say": "And signing up with a free email address subtracts ten.",
    "item": 5
   }
  ],
  "Concepts: Personalization, Segmentation and Lead Scoring#7": [
   {
    "say": "Let's add it up for one lead.",
    "item": 0
   },
   {
    "say": "This lead viewed pricing, logged in three days in a row, and invited a teammate. That is twenty plus twenty-five plus fifteen, which equals sixty points.",
    "item": 1
   }
  ],
  "Concepts: AI Support, Sales Assistants and Launching an AI Product#1": [
   {
    "say": "Now AI support and sales assistants. What can they do?",
    "item": 0
   },
   {
    "say": "First, answering frequent questions like business hours, shipping costs, return policy, and how a feature works.",
    "item": 1
   },
   {
    "say": "Second, pre-sale advice, like recommending a fitting plan or explaining spec differences.",
    "item": 2
   },
   {
    "say": "Third, order and status lookups, such as checking shipping status once it is connected to your systems.",
    "item": 3
   }
  ],
  "Concepts: AI Support, Sales Assistants and Launching an AI Product#2": [
   {
    "say": "There are two more scenarios.",
    "item": 0
   },
   {
    "say": "Scheduling and intake: booking meetings for sales, and asking about needs up front before handing off to a person.",
    "item": 1
   },
   {
    "say": "And an internal assistant, helping salespeople find product info, past proposals, and answers to common objections during a conversation.",
    "item": 2
   },
   {
    "say": "A good design goal is not to replace people. It is to handle repetitive questions fast, so people can focus on complex and emotional conversations.",
    "item": 2
   }
  ],
  "Concepts: AI Support, Sales Assistants and Launching an AI Product#3": [
   {
    "say": "Design principle one: answer only what it has a basis for. The biggest risk of AI support is answering with total confidence and being wrong.",
    "item": 0
   },
   {
    "say": "Limit answers to the material you provide, like the FAQ, policy documents, and product descriptions. When it finds no basis, it should say, I'm not sure, let me connect you with a team member.",
    "item": 1
   },
   {
    "say": "Forbid promises about price concessions, refunds, delivery dates, and results. Humans decide those.",
    "item": 2
   },
   {
    "say": "And keep the material current. When a policy changes, the documents the AI uses must change too, or it will say outdated things.",
    "item": 3
   }
  ],
  "Concepts: AI Support, Sales Assistants and Launching an AI Product#4": [
   {
    "say": "Design principle two: a clear handoff to a human. Set trigger conditions.",
    "item": 0
   },
   {
    "say": "One, the customer clearly asks for a real person.",
    "item": 1
   },
   {
    "say": "Two, the topic involves refunds, compensation, complaints, legal matters, or safety.",
    "item": 2
   },
   {
    "say": "Three, the customer is clearly upset, or has asked the same question twice or more without it being solved.",
    "item": 3
   },
   {
    "say": "Four, the AI lacks confidence or cannot find a basis. When you hand off, pass along the whole conversation so the customer does not repeat themselves.",
    "item": 4
   }
  ],
  "Concepts: AI Support, Sales Assistants and Launching an AI Product#5": [
   {
    "say": "Design principle three: let customers know who they are talking to.",
    "item": 0
   },
   {
    "say": "Do not disguise AI as a person. Whether you must disclose explicitly depends on region and platform rules, but honest disclosure usually prevents trust problems later.",
    "item": 1
   },
   {
    "say": "Customers should also be able to find a switch to a human easily.",
    "item": 1
   }
  ],
  "Concepts: AI Support, Sales Assistants and Launching an AI Product#6": [
   {
    "say": "Let's work out whether AI support is worth it.",
    "item": 0
   },
   {
    "say": "Suppose an online store gets one thousand inquiries a month, and the AI handles sixty percent on its own. That is six hundred.",
    "item": 1
   },
   {
    "say": "A human takes nine minutes per inquiry, and support staff cost twenty dollars an hour. So cost per inquiry is nine divided by sixty, times twenty, which is three dollars.",
    "item": 1
   },
   {
    "say": "Savings are six hundred times three dollars, or eighteen hundred dollars. The tool costs two hundred a month, so net savings are sixteen hundred dollars.",
    "item": 1
   },
   {
    "say": "But that is the optimistic version. Subtract the time to build the knowledge base, updates, spot-checking labor, and complaints and refunds from AI mistakes. Run a one to two month pilot first.",
    "item": 1
   }
  ],
  "Concepts: AI Support, Sales Assistants and Launching an AI Product#7": [
   {
    "say": "How do you read the metrics?",
    "item": 0
   },
   {
    "say": "Resolution rate is the share the AI handled on its own where the customer did not follow up again.",
    "item": 1
   },
   {
    "say": "Also watch handoff rate, and satisfaction after handoff.",
    "item": 2
   },
   {
    "say": "Watch customer satisfaction, or C SAT.",
    "item": 3
   },
   {
    "say": "And watch the wrong-reply rate, found by sampling conversations.",
    "item": 4
   },
   {
    "say": "Looking only at the share of automated replies misleads. Answering a lot does not mean answering correctly.",
    "item": 4
   }
  ],
  "Case Study: How LedgerLeaf Used AI to Lift Trial-to-Paid Conversion#1": [
   {
    "say": "Now the case study: LedgerLeaf, a fictional company. It is a twelve-person startup with a small cloud bookkeeping tool for freelancers and small studios.",
    "item": 0
   },
   {
    "say": "It costs twenty dollars a month, has about seventy percent gross margin, and loses about five percent of customers each month. Marketing has three people and sales has two.",
    "item": 0
   },
   {
    "say": "They get about six hundred new trial signups a month.",
    "item": 1
   },
   {
    "say": "Sales can seriously follow up with at most one hundred.",
    "item": 2
   },
   {
    "say": "Ad copy can only be tested four at a time per week, and Maya spends about forty minutes on each.",
    "item": 3
   },
   {
    "say": "The boss asks how to raise trial-to-paid conversion without adding headcount. The team tackles two parts: ad copy testing, and lead scoring.",
    "item": 3
   }
  ],
  "Case Study: How LedgerLeaf Used AI to Lift Trial-to-Paid Conversion#2": [
   {
    "say": "Step one: analyze the current situation.",
    "item": 0
   },
   {
    "say": "On copy, four per week means few tests, so learning is slow.",
    "item": 1
   },
   {
    "say": "On sales, the one hundred leads are chosen by signup order, which is effectively random. Followed-up leads convert to paid at about five percent.",
    "item": 1
   },
   {
    "say": "The other five hundred, handled by automated emails, convert at about two percent.",
    "item": 1
   },
   {
    "say": "So current new paying customers per month are one hundred times five percent plus five hundred times two percent. That is five plus ten, which equals fifteen.",
    "item": 1
   }
  ],
  "Case Study: How LedgerLeaf Used AI to Lift Trial-to-Paid Conversion#3": [
   {
    "say": "Step two, decision A: AI-assisted copy testing. Here is how they did it.",
    "item": 0
   },
   {
    "say": "Maya organized the fact material: features, price, and what may and may not be claimed. For example, never write guaranteed tax savings.",
    "item": 1
   },
   {
    "say": "She asked AI for twelve ad headlines and body drafts at once, and asked it to flag items needing human confirmation.",
    "item": 2
   },
   {
    "say": "She checked each with a review checklist at about fifteen minutes. Twelve times fifteen divided by sixty is three hours.",
    "item": 3
   },
   {
    "say": "The original four copies took about two point seven hours. So time is about the same, but tests went from four to twelve.",
    "item": 3
   }
  ],
  "Case Study: How LedgerLeaf Used AI to Lift Trial-to-Paid Conversion#4": [
   {
    "say": "Of the twelve, eight passed review. Four were cut for over-promising or not matching the product's features.",
    "item": 0
   },
   {
    "say": "Each of the eight ran for five thousand impressions. The baseline had a one point two percent click-through rate, which is sixty clicks.",
    "item": 1
   },
   {
    "say": "The best version had one point eight percent, which is ninety clicks. A two-proportion z-test gives z of about two point four seven, so the gap is unlikely to be pure chance.",
    "item": 1
   },
   {
    "say": "But the team stayed careful. Picking the best of eight tends to overestimate true performance, the winner's curse, so they re-tested before raising the budget.",
    "item": 1
   },
   {
    "say": "The lesson: AI's value is not replacing review, but getting more versions to test in the same time. And judge results by data, not gut feeling.",
    "item": 1
   }
  ],
  "Case Study: How LedgerLeaf Used AI to Lift Trial-to-Paid Conversion#5": [
   {
    "say": "Step three, decision B: rule-based lead scoring.",
    "item": 0
   },
   {
    "say": "They avoided a complex model at first, because there were not enough closed deals yet. They validated the rules against the past three months of data.",
    "item": 1
   },
   {
    "say": "The paid rate of the high-score group, fifty or more, was clearly higher than the low-score group, so they went live.",
    "item": 1
   },
   {
    "say": "Fifty or more goes to sales follow-up. Thirty to forty-nine goes to nurture emails. Below thirty gets generic emails only.",
    "item": 2
   },
   {
    "say": "Sales still follows up with one hundred a month, but now it is the highest-scoring one hundred.",
    "item": 2
   },
   {
    "say": "Assume the top hundred convert at fourteen percent, estimated from past data, and the other five hundred stay at two percent.",
    "item": 2
   },
   {
    "say": "New paying customers are one hundred times fourteen percent plus five hundred times two percent: fourteen plus ten, which equals twenty-four. That is nine more than fifteen.",
    "item": 2
   }
  ],
  "Case Study: How LedgerLeaf Used AI to Lift Trial-to-Paid Conversion#6": [
   {
    "say": "Here are the scoring rules the team set.",
    "item": 0
   },
   {
    "say": "Viewed pricing page, plus twenty.",
    "item": 1
   },
   {
    "say": "Logged in three days in a row, plus twenty-five.",
    "item": 2
   },
   {
    "say": "Invited a teammate or assistant, plus fifteen.",
    "item": 3
   },
   {
    "say": "Opened the last two emails, plus ten.",
    "item": 4
   },
   {
    "say": "Signed up with a free email address, minus ten.",
    "item": 5
   }
  ],
  "Case Study: How LedgerLeaf Used AI to Lift Trial-to-Paid Conversion#7": [
   {
    "say": "Step four: is it worth it? Lifetime value, or L T V, is monthly fee times gross margin divided by monthly churn.",
    "item": 0
   },
   {
    "say": "That is twenty times zero point seven, divided by zero point zero five, which equals two hundred eighty dollars.",
    "item": 1
   },
   {
    "say": "Nine more paying customers a month are worth about nine times two hundred eighty, or twenty-five hundred twenty dollars.",
    "item": 1
   },
   {
    "say": "Costs: the tool is one hundred a month, and setup is about twenty hours at twenty-five dollars, five hundred one time. Three months cost eight hundred dollars.",
    "item": 1
   },
   {
    "say": "Over three months that is twenty-seven extra customers, worth about seventy-five hundred sixty dollars. Remember, all of these numbers are fictional assumptions.",
    "item": 1
   }
  ],
  "Key Takeaways: AI Marketing and Sales Automation#1": [
   {
    "say": "Let's do a quick chapter map.",
    "item": 0
   },
   {
    "say": "One, the four kinds of work AI can do, and the standard workflow.",
    "item": 1
   },
   {
    "say": "Two, segmentation, personalization, and lead scoring.",
    "item": 2
   },
   {
    "say": "Three, AI support, sales assistants, and launching an AI product.",
    "item": 3
   },
   {
    "say": "Four, the LedgerLeaf case, with its estimate and honest review.",
    "item": 4
   }
  ],
  "Key Takeaways: AI Marketing and Sales Automation#2": [
   {
    "say": "Key point one: the four kinds of AI work and the workflow.",
    "item": 0
   },
   {
    "say": "The four kinds are generate, analyze and organize, predict and rank, and converse and automate.",
    "item": 1
   },
   {
    "say": "The workflow is input material, generate several versions, human review, small-scale publish, and feedback from data.",
    "item": 2
   },
   {
    "say": "Review with a checklist, and give the reviewer the power to veto.",
    "item": 3
   }
  ],
  "Key Takeaways: AI Marketing and Sales Automation#3": [
   {
    "say": "Key point two: three lines of defense.",
    "item": 0
   },
   {
    "say": "Quality: give only reliable material, flag uncertain spots, and have a person verify.",
    "item": 1
   },
   {
    "say": "Privacy and compliance: scope of purpose, data minimization, vendor terms, consent, and unsubscribe. These are general directions, not legal advice, so consult a professional.",
    "item": 2
   },
   {
    "say": "Trust: do not pose as a person, do not invent reviews or testimonials, and handle AI labeling by platform rules.",
    "item": 3
   }
  ],
  "Key Takeaways: AI Marketing and Sales Automation#4": [
   {
    "say": "Key point three: segmentation, personalization, and scoring.",
    "item": 0
   },
   {
    "say": "Segment first, then personalize, and lean on behavior for segments.",
    "item": 1
   },
   {
    "say": "Rule-based scoring is transparent and useful. Validate it with historical data first.",
    "item": 2
   },
   {
    "say": "For example, twenty plus twenty-five plus fifteen is sixty points. A score is a probability, not a guarantee.",
    "item": 3
   }
  ],
  "Key Takeaways: AI Marketing and Sales Automation#5": [
   {
    "say": "Keep the risks in mind.",
    "item": 0
   },
   {
    "say": "Bias, correlation is not causation, data quality, and privacy.",
    "item": 1
   }
  ],
  "Key Takeaways: AI Marketing and Sales Automation#6": [
   {
    "say": "Key point four: support and sales assistants.",
    "item": 0
   },
   {
    "say": "Answer only what has a basis, and make no promises.",
    "item": 1
   },
   {
    "say": "Define clear handoff triggers, and attach the transcript.",
    "item": 2
   },
   {
    "say": "Watch resolution rate, satisfaction, and error rate, not just the share of automated replies.",
    "item": 3
   },
   {
    "say": "And remember the estimate: six hundred times three dollars is eighteen hundred, minus two hundred, so sixteen hundred dollars.",
    "item": 3
   }
  ],
  "Key Takeaways: AI Marketing and Sales Automation#7": [
   {
    "say": "That sixteen hundred dollars is the optimistic version.",
    "item": 0
   },
   {
    "say": "You still have to subtract the hidden costs, so run a pilot before you scale.",
    "item": 1
   }
  ],
  "closing": [
   {
    "say": "That wraps up this chapter on AI marketing and sales automation.",
    "item": 0
   },
   {
    "say": "Now go do the lab, then take the chapter quiz to check how much you've learned.",
    "item": 1
   },
   {
    "say": "Remember: AI gives you speed, and you keep the responsibility. See you in the next chapter.",
    "item": 1,
    "pause": 1
   }
  ]
 }
}