1. 11
Know Your Customer:
Using Machine Learning to Improve Sales
Conversions and Marketing Campaigns
Rajat Arya – Director, Sales
rajat@dato.com
@rajatarya
2. 22
Hello, my name is…
Rajat Arya
Director, Sales (also Dato employee #1)
(software engineer, distributed systems, NBA and movie nerd, learning
data science)
3. 33
Intelligent applications create tremendous value
…but are slow to build & require large specialized teams
Recommenders
Lead Scoring
Churn Prediction
Multi-channel Targeting
Auto-Summarization
Fraud detection
Intrusion Detection
Demand Forecasting
Data Matching
Failure Prediction
4. Core blockers to innovators
• Mapping business task to ML problem requires experts
- For example certain recommender systems require matrix factorization…
• Painful to evaluate, improve & combine ML models
- Enormous amount of time on low-value integration, feature engineering & validation
• Multiple systems to deploy & manage ML in production
- Custom build everything: deployment, monitoring, online experimentation,….
6. 6
Dato’s Machine Learning Core Tenets
• Maps business tasks to machine learning routines
• Eliminates bottlenecks to production
• Simplifies iteration & understanding
Create Value Fast
• Easily combine any variety of features & ML tasks with any data
• Platform components are open, reusable, & sharable
• Easily extend & integrate with other frameworks
Flexibility to Innovate
• Make ML safe & consumable for the enterprise
• Easily deploy, manage, and improve ML as intelligent micro-services
• Adapt to a changing world that drifts from your historical data
Intelligence in Production
8. import graphlab as gl
data = gl.SFrame.read_csv('my_data.csv')
model = gl.recommender.create(
data,
user_id='user',
item_id='movie’,
target='rating')
recommendations = model.recommend(k=5)
cluster = gl.deploy.load(‘s3://path’)
cluster.add(‘servicename’, model)
Agile ML Example: create a live machine learning service
Create a Recommender
5 lines of code
Toolkit w/auto selection
Deploy in minutes
10. Poll: Getting to know you
1. What do you do?
2. Are you using Lead Scoring today?
10
11. 1111
Intelligent applications create tremendous value
Recommenders
Lead Scoring
Churn Prediction
Multi-channel Targeting
Auto-Summarization
Fraud detection
Intrusion Detection
Demand Forecasting
Data Matching
Failure Prediction
12. Lead Scoring : Use what you know about
your customers to maximize your sales &
marketing efforts.
13. Teams that implement Lead Scoring see a 77%
lift in ROI.
Lead Scoring : Motivation
http://sherpablog.marketingsherpa.com/b2b-marketing/lead-gen/lead-scoring-tips/
14. Teams that get Lead Scoring right have a 192%
higher average qualification rate.
Lead Scoring : Motivation
Aberdeen Group
15. Lead Scoring : Practical Definition
Inefficient customer acquisition is costing
your business money.
Your teams have limited resources
(money, people, & time)
Lead Scoring enables sales & marketing teams to prioritize
incoming leads to maximize their efficiency in gaining new
customers.
16. Lead Scoring : Practical Results
Once your teams are scoring leads, you can expect:
1. Higher conversion rates
2. Shorter conversion cycles
3. Increased revenue
Metric Before After
’Qualified’ Leads 1,000 600
Opportunity win rate 25% 40%
Average Revenue per sale $50,000 $62,500
Total Revenue $25MM $32MM
17. Lead Scoring : Without Machine Learning
Belief & Intuition about customers:
We are hot with the youth segment, we should target them.
Or your customers are price-sensitive which overlaps with youth.
We should be reaching out to people within an hour of signing up.
Being timely in 1st contact is critical.
Does data back this up? Maybe 4th day is equally effective.
18. Lead Scoring : With Machine Learning
Benefits of Machine Learning for Lead Scoring:
• Leverage historical data about customers
• Learn patterns of behavior and customer profile that indicate
propensity to convert (quickly)
• Understand what attributes of a user indicate their likelihood
to become a customer
• Predict probability of conversion of new lead, prioritize
accordingly
19. Lead Scoring : Machine Learning Process
Supervised Machine Learning workflow:
Historical
Data
• Split train/test
datasets
• Customers &
non-
customers
Train ML
Model
• Use the
attributes of
customers
• Use
behaviors of
Deploy
• Predict
likelihood to
convert on
new leads
20. Lead Scoring : Machine Learning (Advanced)
• Incorporate Time as a feature (ex. when did a customer take
an action, how much time elapsed between actions, how
many total actions, how many actions per week)
• Transform customer attributes to more meaningful data (ex.
age age range, zip code state, time of day
morning/evening)
• Predict when a customer will convert (ex. Bob will convert in
next 7 days with 80% probability)
21. Lead Scoring & Customer Segmentation
Customer Segmentation is learning the
common attributes of your customers
and splitting them accordingly.
Better target each segment.
Predict which segment a new lead
belongs to utilize that for prioritization or
conversion strategy.
22. Poll: Data Science at your workplace
1. Does your team have data scientists or
developers?
2. Are you using Machine Learning in
production today?
22
24. Thank you!
Want to find out how to incorporate lead
scoring into your organization? Ping me
Coursera ML Specialization
http://coursera.org/specializations/machine-learning
twitter: @rajatarya, email: rajat@dato.com
Notes de l'éditeur
Notes:
Didn’t reiterate intelligent applications
Didn’t go into the building blocks of intelligent apps
Didn’t talk about why this is painful
Didn’t hit the plethora of applications possible
Didn’t bring it back to what they care about most “did I answer your questions, do you see how we would fit/be used by your company”
Didn’t talk about the future w/many models in prod
Microservices seemed to low level
Less filler talk – ANSWER THE QUESTION!
Applications with data, it’s not professional, looking for better ways
Didn’t land how these applications cut across groups
How do you compare? Where is SAS?
Do you handle compliance – explaining predictions (important for compliance?)
Bring collateral & handouts
Intelligent microservices wraps models/analyses in a consumable service accessible & consumable by anyone across the enterprise
Empower businesses not about create, stay competitive, destroy,
Move this up
Poll:
What do you do?
Sales
Marketing
Product Development
Information Technology
Human Resources
Operations
Other
Are you using Lead Scoring Today?
No, here to learn more.
Yes, with Marketo
Yes, with Salesforce
Yes, with Tableau
Not sure, I think so.
Not sure, I don’t think so.
Poll:
Does your team have data scientists or developers?
Yes, a full team of data scientists.
Yes, a full team of developers.
Yes, a mixed team of both.
No, but my engineering team does.
No, but my R+D team does.
I don’t have a team.
Are you using ML in production today?
Yes, for real-time predictions.
Yes, for batch predictions.
No, but on the roadmap.
No, not sure.