Machine Learning and Predictive Marketing

1

Machine Learning and Predictive

MarketingReality, Pitfalls & Potential

Claudia PerlichChief Scientist

@claudia_perlich

22

Programmatic Advertising

33

Work with Brand

300 Million (US) consumer

100 Billion

bid requests per day

AdExchange

Measurement

Conversion

44

Machine Learning & Predictive Modeling:

Algorithms that Learn from Data

55

Historical Data is the Input for Predictive Modeling

Income Age Buy

123,000 30 Yes

51,100 40 Yes

68,000 55 No

74,000 46 No

23,000 47 Yes

100,000 49 No

66

Machine Learning: Who will buy?

Income

Age

No purchase

Insurance50K

45

Logistic Regression

p(Insurance|37,78000) = 0.53

p(+|x)=

β0 = 3.7

β1 = 0.00013

77

To whom should we show an Mercedes ad?

88

Whatever metric the marketer chooses is what I will

teach the machine to predict …

99

Bid on individuals with high predictions by the modelOnly reach audiences that consistently demonstrate interest in your brand.

SC

OR

ING

CO

NN

EC

T

CAMPAIGN PROGRESS

Conversion

++

+

+

No Longer

in Market

XXAdd to

Prospect Pool

and Serve Ad

+

+

-

-

+

+

1010

Predict who will click on the ad

1111

Predict who will sign up for a test drive

1212

Predict who looks up a dealer location

1313

Data in Programmatic Advertising ?

Income Age Buy

123,000 30 Yes

51,100 40 Yes

68,000 55 No

74,000 46 No

23,000 47 Yes

100,000 49 No

1414

Digital traces of everything: 100 Billion events per day (US)

amazon.com

4/11/16

nytimes.com

2/15/16

50.240.135.41

166. 216.165.92

207.246.152.60

108.49.133.218

38.104.253.134

208.76.113.13

Mlb.com

3/15/16

Etrade.com

4/25/16

mercedes.com

4/10/16

Zip 4

10023-4592

04/15/16

Zip 4

10023-6924

04/20/16

Zip 4

10016-2324

03/10/16

03/14/16

Web App

Phone/Tablet

Desktop

Phone/Tablet

com.mlb.atbat

04/20/16

com.rovio.angry

02/26/16

com.myfitnesspal.android

3/25/16

4/18/16

IDFA

31AB-26FC-

94AE-756B

amazon.com

4/11/16

1515

Predict actions to be taken on site base on all else …

amazon.com

4/11/16

nytimes.com

2/15/16

50.240.135.41

166. 216.165.92

207.246.152.60

108.49.133.218

38.104.253.134

208.76.113.13

Mlb.com

3/15/16

Etrade.com

4/25/16

mercedes.com

4/10/16

Zip 4

10023-4592

04/15/16

Zip 4

10023-6924

04/20/16

Zip 4

10016-2324

03/10/16

03/14/16

Web App

Phone/Tablet

Desktop

Phone/Tablet

com.mlb.atbat

04/20/16

com.rovio.angry

02/26/16

com.myfitnesspal.android

3/25/16

4/18/16

IDFA

31AB-26FC-

94AE-756B

amazon.com

4/11/16

1616

Predict who will use an app

amazon.com

4/11/16

nytimes.com

2/15/16

50.240.135.41

166. 216.165.92

207.246.152.60

108.49.133.218

38.104.253.134

208.76.113.13

Mlb.com

3/15/16

Etrade.com

4/25/16

mercedes.com

4/10/16

Zip 4

10023-4592

04/15/16

Zip 4

10023-6924

04/20/16

Zip 4

10016-2324

03/10/16

03/14/16

Web App

Phone/Tablet

Desktop

Phone/Tablet

com.mlb.atbat

04/20/16

com.rovio.angry

02/26/16

com.myfitnesspal.android

3/25/16

4/18/16

IDFA

31AB-26FC-

94AE-756B

amazon.com

4/11/16

1717

Predict who will go to the dealership

amazon.com

4/11/16

nytimes.com

2/15/16

50.240.135.41

166. 216.165.92

207.246.152.60

108.49.133.218

38.104.253.134

208.76.113.13

Mlb.com

3/15/16

Etrade.com

4/25/16

mercedes.com

4/10/16

Zip 4

10023-4592

04/15/16

Zip 4

10023-6924

04/20/16

Zip 4

10016-2324

03/10/16

03/14/16

Web App

Phone/Tablet

Desktop

Phone/Tablet

com.mlb.atbat

04/20/16

com.rovio.angry

02/26/16

com.myfitnesspal.android

3/25/16

4/18/16

IDFA

31AB-26FC-

94AE-756B

amazon.com

4/11/16

1818

Not interested

Brand action

Logistic Regression

p(sv|x)=

β0 = 0.54

β1 = 0.13

…

β10000000 = -0.82

+ …

Machine learning with ‘Big Data’ …

1919

URL’s indicating a visit

of Mercedes dealership

2020

‘In the market’ signal

2121

Real Estate

2222

High-end fitness ..

2323

TECH-SAVVY FAMILY-ORIENTED

THRIFTY SHOPPERS

• Prepaid Smartphone Shoppers

• Budget Wireless Shoppers

• Holiday Deals Shoppers

• Sweepstakes Enthusiasts

MUSIC FANS

SPORTS FANS

• MLB Fans

• NBA Fans

• Soccer Fans

• NFL Fans

• NCAA Fans

• IT Professionals

• Mac & PC

Shoppers

• Country Music

• Hip Hop

• Streaming Radio

Listeners• Military Families

• Working Parents

• Family Values

Advocates

Insights from models: BMW vs. Audi

CAR & MOTORCYCLE ENTHUSIASTSLUXURY SHOPPER

HEALTH CONSCIOUS

• Luxury Retail Researchers

• High End Camera

• Jewelry & Watch

• Furniture Shoppers

• HDTV Shoppers

• Energy Savers

• Lawn & Garden Enthusiasts

• Nutrition Conscious Eaters

• Heartburn Suffers

• Food Allergies Researchers

HOME DECORATOR

• Motorcycle Enthusiasts

• Aftermarket Accessories

24

Pitfalls:

Things we can predict but should not!

2525

Good prediction can be harmful in an industry

focused on bad metrics

2626

Click Reality

2727

Model learned when people click accidentally:

Fumbling in the dark …

2828

Bots/ ‘Non-Human traffic’

6%

2011

36%

2015

2929

Anything that can be faked will be faked:

bot penetration

3030

Viewabilty?

31

Potential

Programmatic as the new digital focus

group!

3232

amazon.com

4/11/16

nytimes.com

2/15/16

50.240.135.41

166. 216.165.92

207.246.152.60

108.49.133.218 208.76.113.13

Mlb.com

3/15/16

Etrade.com

4/25/16

mercedes.com

4/10/16

Zip 4

10023-4592

04/15/16

Zip 4

10023-6924

04/20/16

Zip 4

10016-2324

03/10/16

03/14/16

Web App

Phone/Tablet

Desktop

Phone/Tablet

com.mlb.atbat

04/20/16

com.rovio.angry

02/26/16

com.myfitnesspal.android

3/25/16

4/18/16

IDFA

31AB-26FC-

94AE-756B

amazon.com

4/11/16

Linking digital and purchase behavior

3333

Measuring incremental impact by creative

3434

Use Predictive modeling to ‘look under the hood’:

People who go to Fast Food restaurants

Delicious:

3535

Use Predictive modeling to ‘look under the hood’:

Gym rats

Self-improvement:

3636

Deep view

into

self

improvement

3737

I can predict which creative is most effective for YOU!

38

@Claudia_Perlich