1. www.productschool.com
Part-time Product Management, Coding, Data, Digital
Marketing and Blockchain courses in San Francisco, Silicon
Valley, New York, Santa Monica, Los Angeles, Austin, Boston,
Boulder, Chicago, Denver, Orange County, Seattle, Bellevue,
Toronto, London and Online
3. +5000
Alumni Graduated
across 14
Campuses
· San Francisco
· Silicon Valley
· New York
· Los Angeles
· Santa Monica
· Orange County
· Austin
· Boston
· Boulder
· Chicago
· Denver
· Seattle
· Toronto (Canada)
· London (UK)
15. AI product lifecycle
Define
SimplifyDeploy
• User & Business Understanding
• Data & Metric Definition
• Data Preparation
• UX Design, Model Evaluation
• Model Deployment
• Ongoing Measurement &
Learning
13
2
17. Customer Understanding: Jobs Theory
Seeing tasks from a customer vs. product context
1a
Functional
“Help me wake up
with the best coffee
at consistent
quality”
Social
“Give me a place to
connect with my
friends”
Emotional
“Help me treat
myself
at the end of a long
day”
20. CASE STUDY: Travel Chatbots (Mezi, Expedia)
https://www.altexsoft.com/blog/business/chatbots-in-travel-how-to-build-a-bot-that-travelers-will-love/
21. Data Understanding: Preventing Bias1c
https://i.imgflip.com/1w3emg.jpg
• Comprehensive test cases
(represent the real world)
• Data stratification
• Diverse workforce (avoid tech
bro AI)
• Unconscious bias – review
model outputs for correlations
to race and gender
22. Case Study:
What Celeb do you look like?
http://www.playbuzz.com/chloep19/what-female-celebrity-do-you-look-like
• Image understanding
• Face types, Colors, Tones
• Emotion understanding
23. Metric Understanding: Problem & Output
Definition
1d
https://i.imgflip.com/1w3emg.jpg
• Instrumentation
• Data Quality
• Primary vs. secondary goals
• Product vs. Feature
• Standard vs. derived
metrics
24. Task & Domain Selection1e
When to use:
• Diversify dataset
• Generate dataset
• Generate labelled data
• Structure NL responses
25. Outcome: Find your niche
Customer
DataBusiness
NicheArticulate AI value in terms of:
• Agility / Performance / Cost
• Growth drivers
• Brand value / Industry Status
• Risk reduction
• Accessibility
• Customer Delight
• Convenience/usability
28. Data Preparation
4Cs of data quality
• Correct
• Conforms
• Current
• Consistent
• Consolidated
2a
http://4.bp.blogspot.com/
Why is this important:
• Avoid underfitting, remove bias
• Avoid overfitting, reduce noise
29. Designing AI
5 principles of ethical design
• Humans as Heroes
• Honor Diversity
• Balance EQ and IQ
• Know context
• Evolve over time
2b
https://www.microsoft.com/en-gb/ai/our-approach-to-ai
"The AI tools and services we create must assist humanity and augment our
capabilities."
—Harry Shum, Executive Vice President, AI and Research
31. What model will you choose?
• Classifying cheese into Brie, Mexican,
Parmesan, Mozzarella
32. What model will you choose?
• Analyzing weather patterns to uncover
trends
33. What model will you choose?
• Analyzing weather patterns to FORECAST
the next 10 days
34. What model will you choose?
• To find anomalies in your dataset?
• For spam, fraud filtration?
35. Model Evaluation2d
• Iterative process to find the best model for your scenario (your MVP)
• Balance model performance vs. accuracy
Sample feature recovery rate for a matrix completion algorithm
37. Model Deployment
• Scale data collection
• Scale scenario coverage
• Action movie recos for all users vs.
Movie recos for subset of users
• Scale model
• Check outliers and bias
• Visualize outputs
• Model specific
• Optimizations: Bagging, Boosting
3a
https://cloud.google.com/automl/
https://cloud.withgoogle.com/next18/sf/sessions/session/193072
https://www.youtube.com/watch?v=GbLQE2C181U
48. RECAP: CRISP-DM Methodology
Business &
User
Understanding
Data & Metric
Understanding
Data Prep
Model
Development
Model
Evaluation
Deployment
1. Scope your problem -> find your
niche
2. Build the business case for ML / AI
3. Select your ML model
4. Balance model performance and
accuracy
5. Ensure model relevancye to
changing business needs
6. Human powered vs. machine AI