Overview
Presentation metadata and guiding principles
Title
How Math Powers AI in Everyday Finance
Version
v3.0 — Final talk plan
Audience
High school students (ages 16–18), math focus. No prior AI or programming knowledge assumed.
Duration
45 minutes (42 min content + 3 min Q&A buffer)
Key Themes
- Patterns
- Predictions
- Data to Decisions
Speaker
Prof. Dr. Jörg Osterrieder
Timing Philosophy
Every section has a primary time allocation and built-in flex. The talk is designed so that cutting any single section loses breadth but not coherence. Interactive moments are the pressure valve: expand them when ahead of schedule, compress them when behind. The 3-minute Q&A buffer at the end absorbs any remaining overrun.
Storytelling Arc: Five Acts
The emotional journey from surprise to inspiration
1
Hook & Reveal
0:00 – 7:00“You already use AI finance”
Surprise, curiosity
2
The Pattern Engine
7:00 – 19:00“Math finds what humans miss”
“Whoa, that’s clever”
3
Predictions & Decisions
19:00 – 32:00“From data to action”
Empowerment
4
The Bigger Picture
32:00 – 40:00“Power, fairness, and your future”
Reflection
5
Close & Call to Action
40:00 – 42:00“Math is your superpower”
Inspiration
Buffer
42:00 – 45:00 — Q&A and overflow. If no questions arise, revisit the strongest audience moment from the session or share one more career path.
Running Gag — BankBot
Our AI mascot’s journey from overconfident to wise
BankBot is a cartoon AI character that appears throughout the talk as a comic-relief thread. It starts overconfident (“I analyzed your breakfast. You are 94% human.”), gets humbled by edge cases and false positives, and ends the talk wearing a graduation cap, saying “I am 73% confident. But I defer to the human.” The arc mirrors the audience’s own learning journey: from awe to understanding to critical thinking. Each BankBot moment also serves as a micro-break between heavier math content.
| Stage | Section | Personality | Quote / Action |
|---|---|---|---|
| 1 | Sec 1 | Overconfident | “I analyzed your breakfast. You are 94% human.” |
| 2 | Sec 4 warmup | Cocky | “Too easy. Next.” |
| 3 | Sec 3 | Stressed | (sweating on the decision boundary) |
| 4 | Sec 3.5 | Growing | “FRAUD!” → “Most things are fine” → “Let me calculate…” |
| 5 | Sec 4 | Humbled | “FRAUD DETECTED!” / “…suspicious flowers.” |
| 6 | Sec 7 | Observant | “Based on my analysis, you need 47 houseplants.” |
| 7 | Sec 8 | Critical | (shaking head at social media data) |
| 8 | Sec 10 | Wise | “I am 73% confident. But I defer to the human.” |
Math Concepts Index
Every mathematical idea in the talk, where it appears, and whether the formula is shown
| Concept | Where | How Presented | Formula Shown |
|---|---|---|---|
| Probability / Conditional Probability | Fraud detection (Sec 3) | Updating beliefs when new evidence arrives | $$P(\text{Fraud} \mid \text{Data}) = \frac{P(\text{Data} \mid \text{Fraud}) \cdot P(\text{Fraud})}{P(\text{Data})}$$ |
| Statistics (mean, variance, outliers) | Spending patterns (Sec 3) | Bell curve describes “normal”; outliers trigger alerts | $$f(x) = \frac{1}{\sigma\sqrt{2\pi}} e^{-\frac{(x-\mu)^2}{2\sigma^2}}$$ |
| Classification / Decision Boundaries | Fraud vs. legit (Sec 3) | The line where the AI switches from “OK” to “suspicious” | Conceptual only (threshold on sigmoid output) |
| Sigmoid / Logistic Function | Under the Hood (Sec 3) | Squishes any number into a probability between 0 and 1 | $$\sigma(x) = \frac{1}{1 + e^{-x}}$$ |
| Linear Relationships & Regression | Credit scoring (Sec 6) | The simplest prediction: draw a straight line through data | $$y = mx + b$$ |
| Weighted Averages / Scoring | Credit scores (Sec 6) | Different factors carry different weight — just like school grades | $$\text{Score} = w_1 x_1 + w_2 x_2 + \cdots + w_n x_n$$ |
| Cosine Similarity / Distance | Recommendation engines (Sec 8) | Measure the angle between two people’s preference vectors | $$\cos(\theta) = \frac{\mathbf{A} \cdot \mathbf{B}}{|\mathbf{A}| \cdot |\mathbf{B}|}$$ |
| Feedback Loops & Iteration | How AI learns (Sec 4) | Guess → check → adjust → repeat | Conceptual (gradient descent named, not derived) |
| Dot Product | Recommendations (Sec 8) | Multiply matching preferences, add them up to measure similarity | $$\mathbf{A} \cdot \mathbf{B} = a_1 b_1 + a_2 b_2 + \cdots + a_n b_n$$ |
History Vignettes
Short mathematician stories woven into the talk for human connection
| Mathematician | Dates | Section | Duration | Connection |
|---|---|---|---|---|
| Thomas Bayes | 1701–1761 | Sec 3 | 45 s | His theorem powers modern fraud detection. Published posthumously. |
| Carl Friedrich Gauss | 1777–1855 | Sec 3 | 30 s | Discovered the normal distribution (bell curve). Child prodigy who summed 1–100 in seconds. |
| Florence Nightingale | 1820–1910 | Sec 2 | 40 s | Data visualization pioneer. Her “coxcomb diagrams” changed hospital policy. |
| Ada Lovelace | 1815–1852 | Sec 3.5 | 30 s | First programmer. Saw that math and computation were one, 180 years before ChatGPT. |
| Frank Rosenblatt | 1928–1971 | Sec 3 | 40 s | Built the Perceptron (1958) — the first artificial neuron. Ancestor of every neural network today. |
| Andrey Markov | 1856–1922 | Sec 5 | 35 s | Analyzed Pushkin’s poetry for sequential patterns. His Markov chains now power credit scoring and autocomplete. |
| Abraham Wald | 1902–1950 | Sec 4 | 45 s | WWII survivorship bias: armor the spots with NO bullet holes, because those planes never returned. |
| Lovelace + Nightingale callback | — | Sec 9 | 15 s | Math has always needed diverse thinkers. Different perspectives ask different questions. |
Cartoon & Humor Index
Every laugh line and visual gag, mapped to the timeline
| # | Section | Type | Description |
|---|---|---|---|
| 1 | Sec 1 | BankBot intro | “I analyzed your breakfast. You are 94% human.” |
| 2 | Sec 2 | Analogy | Cafeteria “mystery meat” probability — what are the chances it’s edible? |
| 3 | Sec 3 | Visual gag | Gauss’s hair matches the bell curve shape |
| 4 | Sec 3 | Cartoon | AI Nightmare — the model sees fraud everywhere |
| 5 | Sec 3 | BankBot | Sweating on the decision boundary |
| 6 | Sec 4 | BankBot panels | Four-panel sequence: confident → wrong → confused → improving |
| 7 | Sec 5 | BankBot false positive | “FRAUD DETECTED!” … “suspicious flowers.” |
| 8 | Sec 6 | Cartoon | Fortune teller with a crystal ball (credit scoring analogy) |
| 9 | Sec 6 | One-liner | Weighted average horror: “Homework is only 10%?!” |
| 10 | Sec 8 | BankBot | “Based on my analysis, you need 47 houseplants.” |
| 11 | Sec 10 | BankBot finale | Graduation cap — “I am 73% confident. But I defer to the human.” |
Detailed Sections Breakdown
All 10 sections with key messages, content, math, history, humor, and engagement techniques
1
The Opening Hook
0:00 – 3:00 (3 min)AI in finance made decisions about you today — probably before you ate breakfast.
- Content
- “50–200 AI decisions before breakfast” factoid; reveal that banking, insurance, payments all use AI; introduce BankBot
- Math Concepts
- None yet — pure engagement
- History Vignette
- —
- Humor / Cartoon
- BankBot introduction: “I analyzed your breakfast. You are 94% human.”
- Engagement
- Hand raise — “Raise your hand if you KNEW that AI made 50–200 decisions about you before breakfast.”
2
“You Already Think Like an AI”
3:00 – 7:00 (4 min)Pattern recognition is not alien — you do it every day.
- Content
- Bridge from everyday pattern recognition to how AI does it; cafeteria analogy; introduce probability informally
- Math Concepts
- Probability concepts introduced informally
- History Vignette
- Florence Nightingale (1820–1910) — data visualization pioneer, 40s
- Humor / Cartoon
- Cafeteria “mystery meat” probability analogy
- Engagement
- Cafeteria analogy — “How do you decide if the lunch line is worth it?”
3
Fraud Detection — “Catching the Weird Stuff”
7:00 – 15:30 (8.5 min)Fraud detection is pattern recognition plus probability.
- Content
- Step-by-step Bayesian reasoning walkthrough; normal distribution for “normal” spending; sigmoid function to squish scores; decision boundaries
- Math Concepts
- Bayes’ theorem, normal distribution, sigmoid function, decision boundaries
- History Vignettes
- Gauss (bell curve, 30s), Bayes (updating beliefs, 45s), Rosenblatt (first neuron 1958, 40s)
- Humor / Cartoon
- Gauss hair matches bell curve; AI Nightmare cartoon; BankBot sweating on the decision boundary
- Engagement
- Step-by-step Bayesian reasoning walkthrough with audience
4
“How Does the AI Learn?”
15:30 – 17:00 (1.5 min)AI learns through guess → check → adjust → repeat.
- Content
- Feedback loops; gradient descent (named only, not derived); connect to video-game learning analogy
- Math Concepts
- Feedback loops, gradient descent (named only)
- History Vignette
- Ada Lovelace (1815–1852) — first programmer, 30s
- Humor / Cartoon
- BankBot learning panels: confident → wrong → confused → improving
- Engagement
- “Think of it like learning a video game — you die, adjust, try again”
5
Interactive — “Spot the Fraud”
17:00 – 21:00 (4 min)The AI’s job is harder than it looks.
- Content
- Warmup example + 3 voting scenarios: Alex buys a guitar in another city; Tomoko buys 50 gift cards at 3 AM; Karla has small charges in 4 countries
- Math Concepts
- Applied Bayes, false positives, decision under uncertainty
- History Vignette
- Abraham Wald (1902–1950) — survivorship bias (WWII bombers), 45s
- Humor / Cartoon
- BankBot false positive: “FRAUD DETECTED!” … “suspicious flowers.”
- Engagement
- Audience vote — Fraud or Legit? Hands up for each scenario
6
Credit Scoring — “Your Financial Report Card”
21:00 – 28:00 (7 min)Credit scores are weighted averages — same math as your school grades.
- Content
- Weighted average formula; linear regression; connect school grades to credit factors; simplified “calculate your own score” exercise
- Math Concepts
- Weighted average, linear regression
- History Vignette
- Andrey Markov (1856–1922) — sequential patterns / Markov chains, 35s
- Humor / Cartoon
- Fortune teller cartoon; weighted average horror (“Homework is only 10%?!”)
- Engagement
- “Calculate your own credit score” exercise with simplified weights
7
Interactive — Live Demo
28:00 – 32:00 (4 min)AI is learnable, buildable, and accessible.
- Content
- 3-tier demo approach: (1) pre-recorded screencast (safest), (2) live app demo (if tech permits), (3) static slide walkthrough (fallback). Callbacks to all prior formulas.
- Math Concepts
- Callback to Bayes, sigmoid, weighted average, linear regression
- History Vignette
- —
- Humor / Cartoon
- BankBot observes: “Based on my analysis, you need 47 houseplants.”
- Engagement
- Thumbs up/down — predict the AI’s output before it appears
8
Recommendation Engines
32:00 – 34:00 (2 min)Same math as TikTok/Spotify powers financial recommendations.
- Content
- Dot product for similarity; cosine similarity for normalizing; connect to how music/video recommendations work, then pivot to financial products
- Math Concepts
- Dot product, cosine similarity
- History Vignette
- —
- Humor / Cartoon
- BankBot recommends 47 houseplants
- Engagement
- “You and your friend both like savings accounts and crypto — how similar are you?”
9
Interactive — “Design Your Own AI”
34:00 – 37:00 (3 min)Understanding AI means understanding the human choices behind it.
- Content
- USE IT / SKIP IT rapid-fire exercise — audience votes on which data an AI should use for credit decisions (income? social media? zip code? GPA?)
- Math Concepts
- Callback to weighted averages — what weights would YOU choose?
- History Vignette
- Lovelace + Nightingale callback (15s) — diverse thinkers
- Humor / Cartoon
- BankBot shaking head at social media data
- Engagement
- Shout response — “USE IT” or “SKIP IT” for each data type
10
The Bigger Picture + Closing
37:00 – 42:00 (5 min)Math is a superpower. AI needs ethical thinkers.
- Content
- Career paths (data scientist, quant analyst, AI ethics officer, fintech founder, risk analyst); ethics reflection; “What will YOU build?” closing question
- Math Concepts
- Recap of all formulas; emphasis on math as the common thread
- History Vignette
- Lovelace + Nightingale callback — math has always needed diverse thinkers
- Humor / Cartoon
- BankBot Final Form: graduation cap, “I am 73% confident. But I defer to the human.”
- Engagement
- “What will YOU build?” — closing question left with the audience
Full Timing Table
Complete minute-by-minute breakdown
| Section | Time | Duration | Content |
|---|---|---|---|
| 1. Opening Hook | 0:00 – 3:00 | 3 min | 50–200 AI decisions factoid, BankBot intro, hand raise |
| 2. Think Like an AI | 3:00 – 7:00 | 4 min | Pattern recognition, cafeteria analogy, Nightingale vignette |
| 3. Fraud Detection | 7:00 – 15:30 | 8.5 min | Bayes, bell curve, sigmoid, decision boundaries; Gauss / Bayes / Rosenblatt vignettes |
| 4. How AI Learns | 15:30 – 17:00 | 1.5 min | Feedback loops, gradient descent (named), Lovelace vignette |
| 5. Spot the Fraud | 17:00 – 21:00 | 4 min | Interactive voting (3 scenarios), Wald vignette, BankBot false positive |
| 6. Credit Scoring | 21:00 – 28:00 | 7 min | Weighted average, linear regression, Markov vignette, calculate-your-score exercise |
| 7. Live Demo | 28:00 – 32:00 | 4 min | 3-tier demo, formula callbacks, thumbs up/down predictions |
| 8. Recommendations | 32:00 – 34:00 | 2 min | Dot product, cosine similarity, 47 houseplants |
| 9. Design Your AI | 34:00 – 37:00 | 3 min | USE IT / SKIP IT exercise, Lovelace+Nightingale callback |
| 10. Bigger Picture + Closing | 37:00 – 42:00 | 5 min | Ethics, careers, BankBot finale, “What will YOU build?” |
| Buffer / Q&A | 42:00 – 45:00 | 3 min | Questions, overflow, or revisit strongest audience moment |
Backup & Flexibility Notes
What to cut, what to expand, and how to handle the unexpected
Running Short?
- Expand the Live Demo (Sec 7) with a second example
- Add a deeper Wald anecdote in Sec 5
- Extend USE IT / SKIP IT with more data types
- Let Spot the Fraud debate run longer
Running Long?
- Cut the Markov poetry detail in Sec 6 to a single sentence
- Shorten Sec 8 (Recommendations) to one formula instead of two
- Skip the sigmoid derivative (bonus formula) entirely
- Compress the Wald anecdote to one sentence
Tech Failure During Demo?
- Fall back to pre-recorded screencast (Tier 1)
- Then to static slide walkthrough (Tier 3)
- Never waste audience time troubleshooting
Audience Very Engaged?
- Let the Spot the Fraud voting (Sec 5) run longer — debate is valuable
- Add a 4th voting scenario if time permits
- Expand the “Design Your AI” exercise with additional data types
Audience Quiet?
- Lean harder on BankBot humor
- The false-positive flowers joke is a reliable laugh line
- The 47 houseplants recommendation always gets a reaction
- Switch from hand raises to thumbs up/down (lower social barrier)
Slide Count Estimate
Approximately 32 slides total, averaging ~1.3 min/slide. Heavier sections (Fraud Detection, Credit Scoring) use 5–7 slides each; lighter sections (Opening, Closing) use 2–3.