Scan to ask a question
press any key to close
Moving through the Talk
Turn your phone sideways for a better view
How many of you chose Applied Statistics because you genuinely love statistics?
How many of you chose it because someone told you it has good career opportunities?
How many of you know exactly what job you want when you graduate?
|
What is likely to happen?
What could go wrong?
If it goes wrong, how much will it hurt?
What should we do?
The world needs someone who knows how to make better decisions when there is no perfect answer.
More now than ever.
Chapter 2
Driver
Vehicle
OBD device
Driving data
Insurance policy
Age, gender, address, occupation, prior claims. Then the car — its brand, its model, its age.
Her answers ran through actuarial rate tables. A pricing model did the deciding.
She was sorted into a risk class — grouped with every driver who shares her demographic data.
Alice’s risk class
One price — the group’s average expected cost
But they don’t drive alike
The careful subsidize the risky
Alice pays ฿18,000 — but her real risk is ฿12,000. She covers ฿6,000 for the class.
Illustrative figures.
The OBD-II device
The device logs every trip Alice makes — where, when, and how far.
Alice
MX 5662Trip
Home → Office
Distance
12.4 km
Time
09:14–09:50
Duration
36 min
The same trip — now with every hard brake, sharp turn, and speeding moment.
Alice
MX 5662Trip
Home → Office
Distance
12.4 km
Time
09:14–09:50
Duration
36 min
Driving events
The same class — Alice and five others
Now each driver gets her own price
The price matches how she actually drives
Everyone pays their own way — no subsidy
Alice pays ฿12,000 — her own risk, and no one else’s.
Illustrative figures.
Reducing risk, or just offloading it?
Where does the shed risk go?
Whose behaviour — or whose context?
Can you ever outdrive your past?
P1 asks a different question using different math.
Where are you heading — for you?
Baseline — your own average over your first three months
Current — your recent average, smoothed
A/E Ratio — current ÷ baseline
Rating Factor — clamped between 0.70 and 1.00.
Next premium — your current premium × the rating factor
Three months later
Three months later
Three months later — a full year in
What can a negative-only sensor actually reward?
What's the worst the clamp allows?
What turns a demerit tally into encouragement?
The credential I lacked did not limit the thinking I could do. But the job is far from completed.
Gaming the onboarding window
A driver could deliberately drive badly for the first three months, then improve against a bar they set low on purpose.
Consistently great drivers earn nothing
A driver who's always been excellent has no room to improve — and gets no reward for staying excellent.
Whose baseline is it?
Three people share one car. The sensor doesn't know which one was driving.
What happens on lapse and rejoin?
Does a fresh start mean a fresh, gameable baseline?
Privacy is a separate cost
Continuous monitoring changes how people drive — and how it feels to be watched — in ways a pricing formula doesn't capture.
The ratchet's real cost
Every discount earned is permanent. Someone has to actuarially price the cost of a bet that only ever pays out.
Life changes, not just driving changes
A new job, a new baby, a new car. The model can't tell worse driving from a worse situation.
Sensor error isn't driver error
GPS drift and phone-based tracking have their own noise. The model treats every reading as ground truth.
Regulatory filing risk
A rate that only ever falls still has to be filed and proven adequate to a regulator who's never approved a model like this.
Chapter 3
Seven areas of work. Thirty-five skills. No single person covers all of it — no single person should.
Business
Product
Programming
Math and Stats
Data Collection and Modeling
Data Analysis and Prediction
Data Visualization and Communication
Without a job title, who are you?
Average is crowded. Good is crowded. Where isn't?
What did the other nine do that you didn't?
What are you the only one of?
Recidivism tools still score reoffense risk against demographic and history base rates, not the person in the room. A fairer model tracks behavior change since intake — not where you started.
Life and health insurers still price you against a population mortality table — same age, same price, regardless of your habits. A fairer model would watch your own trend instead of the group average.
Standard credit scores need years of history to build. Gig workers, migrants, and the young don't have that history — so they're invisible, not just risky. A fairer model reads cash flow instead of credit years.
A percentile rank only shows where you sit today, not the direction you're actually heading. Two students, same rank, very different trajectories. A fairer model grades long-term momentum, not one snapshot.
Your version of P1 is somewhere in front of you.
Thank you
