may 14-15, 2014 | icc excel, london...better rpc and roi: adobe on adobe 22 +66.6% +70.4% 19.5% 5.7%...

© 2014 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential. #AdobeSummit

May 14-15, 2014 | ICC ExCel, London

Learn more at summit.adobe.com/emea #AdobeSummit


Leveraging Analytics Data for Large-scale Ad Optimisation Dr. Abhishek Pani | Sr. Principal Scientist, Adobe

© 2014 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential. #AdobeSummit 3

Data sparsity in

Leverage user behavior data from web analytics tools

big data era

E .g . Ad Optimisation


Abhishek Pani Sr. Principal Scientist, Adobe


Agenda

1 | Adobe Media Optimizer (AMO) and Adobe Analytics Integration

2 | Data in Ad Optimisation

3 | Why and how to leverage Analtyics data in AMO

4 | Examples and results


Agenda






Analytics and Media Optimizer in Adobe Marketing Cloud

7

Media Optimizer Analytics

Tell the story of what’s happening in your business through data using high-performance, real-time analytics across online and off line marketing channels

Best of breed portfolio and rules based ad management with intelligent campaign forecasting and targeted ad delivery for data optimised advertising

Track and understand users’ behavior

Optimise your ads

Site Behavior Data (Site Engagement and Conversion Metrics)

Ads Data (Search Engine Metrics)

Focus on: How Media Optimizer leverages user behavior data from Analytics to enhance

ad optimisation


Agenda






Types of Ads AMO optimises

9

Desktop

Tablet

Mobile


Bid optimisation problem in Search Engine Marketing (SEM)

Bid $4.55 Pay $0.64

Bid $2.89 Pay $0.42

Bid $2.41 Pay $0.36


Ad Optimisation Workflow

11

Clicks

Positions

Impressions

Click Through Rate (CTR)

Cost Per Click (CPC)

Conversion Rate

Revenue Per Click

Optimizer

Publishers (e.g., search engines)

Tracking cookies

3rd party

Advertisers

……

Optimisation Engine Predictive Models Data Collection

Constraints (budget, min bid, max bid,

etc.)

Optimal Decisions

(bids, budgets, ads targeting)


Ad Optimisation Workflow

12

Clicks

Positions

Impressions

Click Through Rate (CTR)

Cost Per Click (CPC)

Conversion Rate

Revenue Per Click

Publishers (e.g., search engines)

Tracking cookies

3rd party

Advertisers

……

Optimisation Engine Predictive Models Data Collection

Optimizer

Constraints (budget, min bid, max bid,

etc.)

Optimal Decisions

(bids, budgets, ads targeting)


Data is…

Ultimate source of signals and premise of everything in ad optimisation

More Data beats Complicated Algorithms

Scalable Algorithms – Low Computational Complexity


Available to Media Optimizer® through integration with Analytics®

Data collection throughout the conversion funnel

14

Keyword searching

Ads impression

Click

Conversion

After-click behaviours Website navigation Time spend Page views Other engagement activities

Feeds from Search Engine

• Tracked by Search Engine Marketing (SEM) vendors

• And/or feeds from the advertisers

Usually not available to SEM vendors


Data availability: Integrated vs. Non-integrated

15

Adobe Others

N/A The Differentiator

Powered by Adobe Analytics®

Site behavior

+ Engagement

Metrics


Agenda






17

Keyword Keyword No.1 Keyword No.2

Clicks 3 3

Site behavior

• Visited 5 pages per click in average

• No bounces • Watched video twice

• Revisited the website later after closing the browser

• ……

• Less than one page view per click

• 2 bounces out of 3 clicks • No video watch

• No revisit • ……

Conversions 0 0

Q: Which “long tail” keyword has a higher revenue per click prediction?

Why site behavior data are critically important Looks more promising

Top funnel

Middle funnel

Bottom funnel

®


Example of engagement metrics available through analytics

18

Revenue Signals

Time spent on the site in the first visit


How to use the data in the predictive models

19

Site behavior and engagement data

from Adobe Analytics

Existing hierarchical revenue prediction

model in Media Optimizer

Revenue prediction model based on

Adobe Analytics data

Ensemble revenue prediction model


Agenda






Better forecasting: Adobe on Adobe (Closer to 100% is better)

21

0%

20%

40%

60%

80%

100%

120%

140%

160%

180%

1 2 3 4 5 6 7 8 9 10 11 12

Acc

urac

y (1

00%

is th

e be

st)

Portfolio

old model

new model

82.26% Accurate

92.03% Accurate


Better RPC and ROI: Adobe on Adobe

22

+66.6%

+70.4%

19.5%

5.7%

RPC ROI Portfolio A Portfolio B

RPC ROI

Pre Post

Pre Post

* * Daily spends are controlled in the analysis. Performance changes may also be partially attributed to other uncontrolled factors, such as seasonality, trending, special event (e.g., promotion, new product announcement), portfolio composition changes, and etc. ** Effects of seasonality and other factors are partially adjusted through performances of similar portfolios in the same periods


0

1

2

3

4

5

6

0

5

10

15

20

25

bid

in $

clic

ks p

er d

ay

clickbidposition7 per. Mov. Avg. (click)

How the new model promote promising tail keywords (“photoshop cost”)

23

0.002

0.02

rpc(revenue per click)_estimation_oldmodelrpc_estimation_newmodelrpc_actual

Switched to new model (May 3rd)

Rev

enue

per

clic

k


Travel site in Asia-Pacific

24

% Change (Post vs Pre) Total Cost -18%

Total Revenue +11% Return On Advertising

Spend (ROAS) +36%

Post (14 days) vs. Pre (14 days) New model switched on Jan 22, 2014


Online retailer in EMEA

25

% Change (Post vs Pre) ROAS +27%

Revenue-Generating Keywords

+13%

Two control portfolios had no performance changes

Post (7 days) vs. Pre (7 days) New model switched on Oct 21, 2013


Performance Analysis – Pre v/s Post

26

65 clients; 380 portfolios Post (14 days) vs. Pre (14 days)

Mean change

Lower bound of 95% CI

Upper bound of 95% CI

ROAS +15.5% +12.4% +18.7% Revenue per click (RPC) +6.5% +3.5% +9.5%

Revenue-generating keywords +5.1% +2.5% +7.6%

* Daily spends and client differences are controlled in the analysis. Performance changes may also be partially attributed to other uncontrolled factors, such as seasonality, trending, special event (e.g., promotion, new product announcement), portfolio composition changes, and etc.


Performance Breakdown by Domain

27

0.00%

5.00%

10.00%

15.00%

20.00%

25.00%

Automotive Ecommerce Education Finance Lifestyle &Entertainment

Retail Technology Travel

Average ROI Lift (%)

Average ROI Lift (%)


Key benefits of the integrated approach

Significant accuracy lift

Guided learning

Discovering “hidden treasures” with limited learning budgets

Improving RPC and ROAS

Managing portfolio risk

28


Agenda






Key Takeaways

Site behavior data is highly valuable for ad

optimisation

1 Adobe leverages

those data through integration and

agorithms

2 1⊕1 >> 2

3


Call to Action

1 | Ensure you have the ability to access site behavior data in your ad optimisation system

2 | Leverage those data through either automated predictive modeling or insights generation


Questions and Answers


Every Session Complete 8 surveys to enter the prize draw

Go HERE, do THIS, and you could WIN!

Download the App

Survey available immediately after

the session

Give us feedback It takes less than a minute

Eight to win

Win a MeCam Record every minute

Win an iPad Air


Abhishek Pani [email protected]


May 14-15, 2014 | ICC ExCel, London

Learn more at summit.adobe.com/emea #AdobeSummit

may 14-15, 2014 | icc excel, london...better rpc and roi: adobe on adobe 22 +66.6% +70.4% 19.5% 5.7%...

Documents