Leveraging Data Science to power e-
commerce.
2
● About Cars.com & it’s business
Agenda
● Data Science at Cars.com Inc
○ Data Science Challenges
○ Identifying duplicate vehicles
○ Building a core DS team
● Questions
● DS is more than ML, much more!
○ An overall perspective
○ Working example
3
About cars.com
A platform/decision engine that helps you sell/buy a car? NO.
4
About cars.com
We are more than that, we manage the entire lifecycle of your car.
5
About cars.com
How do we make money?
Subscriptions
Model
Advertisements
0 Margins in vehicle sales.
6
About cars.com
Value proposition to users?
A few products.
Place your screenshot here
- Relevant Search Results
7
About cars.com
Value proposition to users?
- Relevant Search Results
- Pricing Tool & Car valuation Place your screenshot here
8
About cars.com
Value proposition to users?
- Relevant Search Results
- Pricing Tool Place your screenshot here
- Car & Dealers’ reviews and
ratings
9
About cars.com
Place your screenshot herePlace your screenshot here
- All of these features are
wrappers over predictive
models; aka data products.
- Lean agile methodology
across the company.
- They have underlying data
pipelines, model building and
AB testing platforms; aka
technical products.
10
Data Science Challenges @Carsdotcom
Start off with a user’s first foot print
Diverse & Relevant Cars
Intuitive & Great Product Exp
Understand user while they
research & browse
Enrich experience with data
products
Help them manage
vehicles they own
Help them sell cars. Enrich
experience with data products
Acquire Users via Marketing
Help them manage
their inventory
Consumer/Buyer
IndividualSeller/Dealer
Facebook bidding
SEO/SEM
Improve SIte Relevance
Relevant Site Ads
No duplicate cars
Data widgets and tools,
Compare your car with
existing cars (Dealers)
Pricing badging,
predicting time to sell,
Similar Cars
Vehicle Valuation, Car
Service Predictions
11
Data Science Challenges
Unstructured data; a lot of it
Decode car context using word embeddings
We learn vectors for each feature or attribute of the
vehicle segments and later aggregate those
vectors to form a vehicle vector
Resulting similar vehicles are agnostic of user (not
dealer) context and behaviour
12
Most people when they talk about Machine Learning at an e-commerce company
Predictive Model
This is how I saw Data Science in 2011
13
In 2013
Predictive ModelData Ingestion
Data Analysis
Modeling Strategy
Data
Transformation
Data Validation
Model
Validation/Evaluation
Models as a service
14
Fetch Data
Write
analysis
scripts
Analyse
Results
Verify
Assumptions
Debug
Codes
Try new
approaches.
Clean it
Verify it
Plot it.
Reflection
OOT Analysis
Effect on KPIs
Local & Global
Sampling
MDI
Modeling
Strategizing
Outliers
Removal
Data Analysis
Acquire Right
Data
Ideation
Finalise Target
Variable
Dashboards
Reports
Insights &
Deep Analytics
A/B Testing
Impact Evaluation
Modeling &
Data Products
Data Science
Process
Business Process
15
The whole picture - DS at Cars.com Inc. 2017
Predictive Model
Integrated Job Management, Monitoring, Data/Model Evaluation/Visualization
Logging & Model Tuner
Data Ingestion
Data Analysis
Modeling Strategy
Data
Transformation
Data Validation
Model
Validation/Evaluation
Models as a
service
Real time Data
Batch Processing
Model Config
Model
Database/Back end
User facing
“
Inspiration
- Proving media is biased towards certain political organizations and their ideologies
16
As a DS, sooner or later you’d want to gather more skills than just ML/AI
Product and Vision
- Empower users with custom data science models to analyze the way media covers current affairs
- Provide insight into crowd-sourced reviews of media organizations & political figures
- Goto tool for data journalists, political strategists and enthusiasts
17
Prototype in about 300 hours
18
Questions?
Reach out to me at apandey@cars.com; @addhyan_pandey;
addhyanpandey@gmail.com
Growing Cars’s Data Science team. Hiring at various levels.
So, if you’re into cars & data science then cars.com is the place for you.

Leveraging Data Science in the Automotive Industry

  • 1.
    Leveraging Data Scienceto power e- commerce.
  • 2.
    2 ● About Cars.com& it’s business Agenda ● Data Science at Cars.com Inc ○ Data Science Challenges ○ Identifying duplicate vehicles ○ Building a core DS team ● Questions ● DS is more than ML, much more! ○ An overall perspective ○ Working example
  • 3.
    3 About cars.com A platform/decisionengine that helps you sell/buy a car? NO.
  • 4.
    4 About cars.com We aremore than that, we manage the entire lifecycle of your car.
  • 5.
    5 About cars.com How dowe make money? Subscriptions Model Advertisements 0 Margins in vehicle sales.
  • 6.
    6 About cars.com Value propositionto users? A few products. Place your screenshot here - Relevant Search Results
  • 7.
    7 About cars.com Value propositionto users? - Relevant Search Results - Pricing Tool & Car valuation Place your screenshot here
  • 8.
    8 About cars.com Value propositionto users? - Relevant Search Results - Pricing Tool Place your screenshot here - Car & Dealers’ reviews and ratings
  • 9.
    9 About cars.com Place yourscreenshot herePlace your screenshot here - All of these features are wrappers over predictive models; aka data products. - Lean agile methodology across the company. - They have underlying data pipelines, model building and AB testing platforms; aka technical products.
  • 10.
    10 Data Science Challenges@Carsdotcom Start off with a user’s first foot print Diverse & Relevant Cars Intuitive & Great Product Exp Understand user while they research & browse Enrich experience with data products Help them manage vehicles they own Help them sell cars. Enrich experience with data products Acquire Users via Marketing Help them manage their inventory Consumer/Buyer IndividualSeller/Dealer Facebook bidding SEO/SEM Improve SIte Relevance Relevant Site Ads No duplicate cars Data widgets and tools, Compare your car with existing cars (Dealers) Pricing badging, predicting time to sell, Similar Cars Vehicle Valuation, Car Service Predictions
  • 11.
    11 Data Science Challenges Unstructureddata; a lot of it Decode car context using word embeddings We learn vectors for each feature or attribute of the vehicle segments and later aggregate those vectors to form a vehicle vector Resulting similar vehicles are agnostic of user (not dealer) context and behaviour
  • 12.
    12 Most people whenthey talk about Machine Learning at an e-commerce company Predictive Model This is how I saw Data Science in 2011
  • 13.
    13 In 2013 Predictive ModelDataIngestion Data Analysis Modeling Strategy Data Transformation Data Validation Model Validation/Evaluation Models as a service
  • 14.
    14 Fetch Data Write analysis scripts Analyse Results Verify Assumptions Debug Codes Try new approaches. Cleanit Verify it Plot it. Reflection OOT Analysis Effect on KPIs Local & Global Sampling MDI Modeling Strategizing Outliers Removal Data Analysis Acquire Right Data Ideation Finalise Target Variable Dashboards Reports Insights & Deep Analytics A/B Testing Impact Evaluation Modeling & Data Products Data Science Process Business Process
  • 15.
    15 The whole picture- DS at Cars.com Inc. 2017 Predictive Model Integrated Job Management, Monitoring, Data/Model Evaluation/Visualization Logging & Model Tuner Data Ingestion Data Analysis Modeling Strategy Data Transformation Data Validation Model Validation/Evaluation Models as a service Real time Data Batch Processing Model Config Model Database/Back end User facing
  • 16.
    “ Inspiration - Proving mediais biased towards certain political organizations and their ideologies 16 As a DS, sooner or later you’d want to gather more skills than just ML/AI Product and Vision - Empower users with custom data science models to analyze the way media covers current affairs - Provide insight into crowd-sourced reviews of media organizations & political figures - Goto tool for data journalists, political strategists and enthusiasts
  • 17.
  • 18.
    18 Questions? Reach out tome at apandey@cars.com; @addhyan_pandey; addhyanpandey@gmail.com Growing Cars’s Data Science team. Hiring at various levels. So, if you’re into cars & data science then cars.com is the place for you.