Predictive Analytics disrupting
Product Development
Rudradeb Mitra | https://www.linkedin.com/in/mitrar/
My life journey
• Born in a lower middle class family in India.
• Published first research paper on AI in 2002.
• Went on to work in Germany, Belgium and Scotland as AI
researcher. Published 10 papers on AI.
• Masters from Univ. of Cambridge.
• Lived in 9 countries. Built 4 startups - 2 in US, one in UK and
one in Belgium
Since September 2016
• Mentor - FI, Google Launchpad.
• Write (Startups, Sales, AI,…)
• Speak (In 2017 spoke at 13 events in 5 countries)
• Consult various companies around the world on technology and
sales
What I do now?
Why here?
• Love meeting people from all over the world.
• Share what I know.
• Learn from you all.
Today’s talk
• What is Predictive Analytics?
• Case studies on using Predictive Analytics for Product
development
• What’s store in future?
Goal: Show something that you can start working on from tomorrow
Predictive Analytics
What is missing?
How does it fit into product development?
Imagine..
• If you can predict user behavior and future
needs of users
• If you can predict future bugs
• If you can predict value of your product
Two Case studies
Case study I
Predicting user behaviour and adapting the product
Analyzing 3M Instacart Orders
• Prior: ~3.2m orders
Instacart Data
product_id
user_i
d
eval_set order_number order_dow order_hour_of_day days_since_prior_order
2539329 1 prior 1 2 8
2398795 1 prior 2 5 7 15
473747 1 prior 3 7 12 20
22544786 1 prior 4 1 7 21
4215438 1 prior 5 3 15 28
2295261 1 prior 6 2 7 19
2295261 1 prior 7 6 20 20
2550362 1 prior 8 5 14 14
1187899 1 prior 9 2 16 0
2168274 1 prior 10 2 8 30
1501582 1 train 11 1 11 10
Predict what customer will order
One approach to Prediction -
Supervised Neural Networks
Training
wij
wjk
An example
Hidden layer Output layerInput layer
0.5
0.5
0.5
0.5
0.5
0.5
Training
Input: 0 1 0
Output: 0.75
An example
Hidden layer Output layerInput layer
0
1
0
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.25 + 0.25 0.5
Desired output
0.75
Error: 0.25
Training
Input: 0 1 0
Output: 0.75
An example
Hidden layer Output layerInput layer
0
1
0
0.5
0.5
0.5
0.5
0.5
1.0
0.5 + 0.25 0.75
Back propagation
0.5
0.5
Training
Input: 0 1 0
Output: 0.75
One approach to Prediction - Neural Networks
One approach to Prediction - Neural Networks
word2vec- Predicting next word
word2vec
Hey,
you
are
What
doing
Input words
Predicted next word
Current approach to prediction for instacart
2539329
2398785
473747
2550368
1187899
word2vec
(Product ID)
(Predicted product ID)
What is missing?
Instacart Data
product_id
user_i
d
eval_set order_number order_dow order_hour_of_day days_since_prior_order
2539329 1 prior 1 2 8
2398795 1 prior 2 5 7 15
473747 1 prior 3 7 12 20
22544786 1 prior 4 1 7 21
4215438 1 prior 5 3 15 28
2295261 1 prior 6 2 7 19
2295261 1 prior 7 6 20 20
2550362 1 prior 8 5 14 14
1187899 1 prior 9 2 16 0
2168274 1 prior 10 2 8 30
1501582 1 train 11 1 11 10
Time is important!
Recurrent Neural Networks
Long Short Term Memory
Long Short Term Memory
A new approach to Prediction
Day 1 Product x Product y Product z
Day 2 Product a Product b Product c
Day 3 Product e Product x Product z
Day 4 ? ? ?
Day 5 ? ? ?
Input Prediction
New Product
Day 4
Day 5
Day 6
Future Products
Case study II
Predicting future needs and adding features
ROAD VIKINGS
Driver’s app
Record a trip Trip feedback
Using friends network
Unsupervised NN - Clustering
Hidden layer
Output layer
Acceleration score
Speed score
Mobile usage score
Friend’s median
score
Bad
driver
Good
driver
Unsupervised learning Clustering (SOM)
New Product
Bad drivers Good drivers
• High Incentive + penalty
• Insurance
• Gamification to improve
behaviour
• Gamification for fun
and engagement
ROAD VIKINGS
Driver’s app
Record a trip Trip feedback Goals & challenges Rewards
Results
What is in future?
Bug fixing - SOM
• Like crime, bugs are
also around similar
regions of code.
• Dependencies of code.
• New code likelihood for
bugs.
What is in for you all?
• Can you incorporate some of the aspects in your product?
• Build a product to do predictive analytics - Billion dollar market.
Go and build something
awesome!
Rudradeb Mitra | https://www.linkedin.com/in/mitrar/

Predictive Analytics disrupting Product development