Welcome [tc18.tableau.com]...Driving value at Cars.com: Using spatial data to identify 7MM leads on...

Preview:

Citation preview

Welcome

Driving value at Cars.com: Using spatial data to identify 7MM leads on dealership lots

Jeff Bloomfield

Senior Software Engineer

Cars.com

# T C 1 8

Who Are We?

Cars.com is a leading two-sided digital automotive marketplace that creates meaningful connections between buyers and sellers.

Who Am I?

Jeff BloomfieldSenior Software Engineer

Avid Amateur

Cyclist &

Commuter

Agenda

In this presentation, we’ll focus on:

• Developing processes and partnerships to capture dealer lot polygon data and visitor location data

• Using Python to generate a GEOJSON file

of dealership polygons to join with visitor data

in a data extract

• Building performant geospatial views that provides

dealers detailed information about their visitors

Cars.com revolutionized the attribution game when it released its award-winning Lot Insights dashboard, which shows physical visitors on or near a dealer’s lot.

Overview

Our Dashboard Overview

Basic Stats• 8k Monthly Views & 429

Avg Prints• 5.7 Sec Load Time• 13 Months of Daily Data• 3 Extracts

Business Problem We Needed To Solve

Changing The Value Story

Traditional Leads No Longer Telling The Whole Story We Needed a Way to Quantify Mobile Shoppers on Dealership Lots

Half went on Cars.com from a mobile device for the first time

1 week prior to their dealership visit

1 month prior

3 months prior

>3 months prior

Developing A Proof of Concept

Can We Do It?:

• How to combine data on dual-axis map?

• What data do we use for POC?

Cars-Sponsored Event:

• Mapped Wrigley as polygon

• Employees turned on location services

• Proved out solutions with 100 records

Key Requirements

Marketing:

• How many visitors on/near a lot?

• What other dealers are they visiting?

• How many visitors in a market/DMA?

Technical:

• How to map 30k+ dealer polygons?

• How to capture & process lat/long of 35MM+ monthly searches?

• How to combine 13 months of dealers & visitors into performant dashboard?

Lot Insights

Dashboard

Visitors On Lot

DMA Visitors

Cross Dealer Visitors

Dealer Polygons

Visitor Location

13 Months of Data

Building Our Geospatial Dataset

Defining a Dealership Lot

Defining A Dealership Lot

Determine your polygon criteria

• Cannot cross other dealerships

• Buffer between street & polygon

• Don’t include inaccessible areas

• Centroid determined by largest building on lot

• No jagged lines

Capturing & Maintaining Dealer Polygon Data

Name & Address of

Dealerships to Map

(300-400 monthly)

Identify Trusted

Mapping Partner

Maps Centroid,

Building & Lot

Return Monthly WKT

file to Cars (+ Exceptions)

Load to Dealer

Lot table

Develop a process to capture and update 30k+ dealership lots’ polygon data

* Unique permissions id assigned to each dealer

Capturing & Maintaining Visitor Search Data

1.1MM Daily

Searches on Cars

Apps & Mobile

50% w/Location

Services On

Device lat/long Where

Search Initiated

(Session Only)

Flag 4k Daily

Visitors Within

Dealer Polygon

Load to Lead

Walkin table

Develop a process to capture and update 35MM+ monthly searches

** Heavy lifting done in database for performance

Flag 11k Daily

Visitors .1 Miles

of Centroid

Extracting Dealer Polygons & Lot Visitors

Generating Polygons:

• Tableau geospatial functions not supported in Teradata

• Developed Python script to transform data into geojson file

• Pull unique visitors by dealer and month and roll up total sessions

• All dealer data comingled in single extract

Benefits:

• Eliminates 500k dealer records (avg 15 polygon points)

Dealer_Polygon.py

Designing our Geospatial Dashboard

Create a Dashboard Framework

Use templates and common filters:

• Standardization speeds development & ensures consistency

• Standard parameters allow us to control multiple data extracts & are performant

• Users authenticate to web portal -control access with URL with trusted authentication

User Logs Into

Web Portal

Tableau URL Sets

Parameters

Report Shows

Correct Dealer

CONTROLLING DATA ACCESS

https://...views/LotInsights/15LotInsights?CustomerUUID=82170000779&DMACODE=602

Use Targeted Extracts for Performance

Volume

Spoke – Lot InsightsDealer polygons joined with lot visitors for

last 13 months4.7MM

Spoke – Lot Insights DMA

Percent of mobile searches and percent of

devices with location services on in each

DMA market by month

3.1K

Spoke – Lot Insights Imps

The vehicles that are being viewed through

the app, and the location they are being

viewed from for the last 3 months by dealer

150K

No blending required – all controlled by UUID, Date, and/or DMA parameters

Keeping The Traffic Map Simple

Showing dealers 500K+ monthly visitors on or near their lot:

• Minimize map background – focus on the story

• Created using a dual axis map overlaying visitors on top of the dealer lot polygon

• Defaults to previous month but is users can choose month to view

• Help make it pop by using branded shape file and color palette

• Provide additional context with tooltips

Enhanced Mapping - Other Lots Visited?

Showing dealers what other lots their

visitors are going to find a car:

• Use conditional Sets to identify visitor records

• Use Great Circle Distance formula to determine distance and they had to be ON another dealer’s lot

• Use map radius tool to validate data

• Use Dealer Lat/Longs instead of polygons for

viewability

• Provide additional context with tooltips, and allow users to highlight dealers with actions

Building Geospatial Worksheets

DEMO!!

Future Enhancements?

Lot Insights is constantly evolving:

• Potential Mapbox integration

• Tableau 2018 Point in Polygon functionality

• Showing visitors who also wrote a review on Cars.com

• Integrating visitor data from 3rd party vendors

MAPBOX INTEGRATION

Questions?

Please complete the

session survey from the

Session Details screen

in your TC18 app

Thank you!

#TC18

Developing Our Initial Prototype

Keys To Success:

• Wireframe to glass in two months

• Dedicated team – database architect, dashboard developer, and marketing lead

• Agile approach – released initial dashboard and iterated through changes/enhancements

WIREFRAME PROTOTYPE DASHBOARD

INITIAL RELEASE

Recommended