• Like

Using Advanced Analyics to bring Business Value

  • 161 views
Uploaded on

Two case studies in Digital Advertising

Two case studies in Digital Advertising

More in: Technology , Business
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
No Downloads

Views

Total Views
161
On Slideshare
0
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
10
Comments
0
Likes
1

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1.   Copyright  ©  2013.  Tiger  Analy7cs   Using  Advanced  Analy7cs  to  bring   Business  Value     Two  case  studies  in  Digital  Adver7sing   _________________________   Mahesh  Kumar   CEO,  Tiger  Analy6cs    
  • 2.   Copyright  ©  2013.  Tiger  Analy7cs   Tiger  Analy7cs   •  Bou6que  consul6ng  firm  solving  business  problems  using   advanced  data  analy6cs   •  Focus  areas   –  Digital  adver6sing  and  Social  Media     –  Marke6ng  and  Customer  Analy6cs   –  Retail  and  CPG   –  Transporta6on   •  Offices  in  Bay  area,  North  Carolina,  and  India  
  • 3.   Copyright  ©  2013.  Tiger  Analy7cs   •  Display  Adver6sing  through  Real-­‐6me  bidding  (RTB)   –  Background  on  RTB     –  Business  problem:  improve  CTR   –  Solu6on  Approach  and  Results   •  Credit  Card  Customer  Acquisi6on  via  Facebook  Ads   –  Facebook  ads  plaOorm   –  Business  problem:  op6mal  targe6ng  and  bidding   –  Applica6on  to  credit  card  marke6ng   3 Overview  –  Two  case  studies  
  • 4.   Copyright  ©  2013.  Tiger  Analy7cs   Display  Adver7sing  through  Real-­‐7me   bidding  (RTB)       Case  Study  #  1    
  • 5.   Copyright  ©  2013.  Tiger  Analy7cs     5 Display  Adver7sing  –  Real  Time  Bidding   Milliseconds to bid and load ad … Waiting for ad from ad exchange … Male, 20-30 yrs, NYC Tech user
  • 6.   Copyright  ©  2013.  Tiger  Analy7cs     6 Display  Adver7sing  –  Real  Time  Bidding   Targeted Ads
  • 7.   Copyright  ©  2013.  Tiger  Analy7cs   Real  Time  Bidding  (RTB)  for  Display  Ads   7 Ad-­‐Exchange   Ad-network 2 Publisher (NYT.com) Tech section Ad-network 1 Ad-network 3 Advertiser Advertiser Advertiser Male, 20-30 yrs, New York, Tech user •  The  en6re  process  takes  less  than  500  milliseconds   •  RTB  share  of  online  ads  is  es6mated  to  be  $2B  per  year  
  • 8.   Copyright  ©  2013.  Tiger  Analy7cs   Business  Problem   •  Click-­‐through  rate  (CTR)  predic6on:  Given  a  campaign  line,   what  is  the  predicted  CTR  for  an  impression  based  on   –  User  characteris6cs   –  Webpage  characteris6cs   •  Iden6fy  impressions  with  highest  CTR?   8
  • 9.   Copyright  ©  2013.  Tiger  Analy7cs   Maximizing  the  CTR  is  Cri7cal  For  Cost  Op7miza7on   9 High CTR is good for everyone: users, advertiser, and publisher High   CTR   Relevant   content  for   Users   Revenue   maximiza6on  for   Publisher   Relevant   audience  for   Adver6ser  
  • 10.   Copyright  ©  2013.  Tiger  Analy7cs   Sample  data   10
  • 11.   Copyright  ©  2013.  Tiger  Analy7cs   Data  challenges   •  Challenges   –  More  than  5000  variables   –  Hundreds  of  millions  of  data  points   –  Sparse  and  missing  data   –  Clicks  are  very  rare  (typically  1  click  in  every  3,000  impressions)   11
  • 12.   Copyright  ©  2013.  Tiger  Analy7cs   •  Case  sampling   –  Keep  all  impressions  with  clicks   –  Keep  only  a  random  sample  of  1%  non-­‐clicks   •  This  reduced  the  data  size  by  100-­‐fold,  but  predic6on   accuracy  was  as  good  as  when  using  all  data   12 Reducing  the  number  of  data  sets  
  • 13.   Copyright  ©  2013.  Tiger  Analy7cs   Logis7c  Regression  results   13 - " 50 " 100 " 150 " 200 " 250 " 300 " 350 " 400 " 450 " 1" 1001" 2001" 3001" 4001" 5001" predicted! baseline! K K K K K K 0% 20% 40% 60% 80% All data •  Top  20%  of  data  got  232  out  of  415  (56%)  of  clicks   •  A  liY  of  180%  
  • 14.   Copyright  ©  2013.  Tiger  Analy7cs   14 Insights  –  Final  set  of  variables  
  • 15.   Copyright  ©  2013.  Tiger  Analy7cs   Twier  –  CTR  Predic7on   15 NLP
  • 16.   Copyright  ©  2013.  Tiger  Analy7cs   Credit  Card  Customer  Acquisi7on     Through  Social  Media  Marke7ng       Case  Study  #  2    
  • 17.   Copyright  ©  2013.  Tiger  Analy7cs   Social Media provides rich data to marketers
  • 18.   Copyright  ©  2013.  Tiger  Analy7cs   Ads  on  Facebook   Newsfeed  on  Desktop   Newsfeed  on  Mobile   Right  Hand  Side  on  Desktop   Sponsored  Story   Image source: Facebook
  • 19.   Copyright  ©  2013.  Tiger  Analy7cs   Facebook  Ad  Pla`orm  -­‐-­‐  targe7ng   19
  • 20.   Copyright  ©  2013.  Tiger  Analy7cs   Facebook  Ad  Pla`orm  -­‐-­‐  pricing   20
  • 21.   Copyright  ©  2013.  Tiger  Analy7cs   Case  study:  credit  card  marke7ng   The image cannot be displayed. Your computer may not have enough memory to open the image, or the image may have been corrupted. Restart your computer, and then open the file again. If the red x still appears, you may have to delete the image and then insert it again. The image cannot be displayed. Your computer may not have enough memory to open the image, or the image may have been corrupted. Restart your computer, and then open the file again. If the red x still appears, you may have to delete the image and then insert it again. Cash  Back   The image cannot be displayed. Your computer may not have enough memory to open the image, or the image may have been corrupted. Restart your computer, and then open the file again. If the red x still appears, you may have to delete the image and then insert it again. 1,000,000   Impressions   300   Clicks   3   Applica7ons   1   Approval   Conversions are rare events when compared to clicks. The challenge is to be able to make meaningful inferences based on very little data, especially early on in the campaign. Click-­‐through  rate   0.03%   Conversion  rate   1%   Approval  rate   33%  
  • 22.   Copyright  ©  2013.  Tiger  Analy7cs   Micro  Segments   1  Segment   50  Segments   50  x  2  =     100  Segments   2  Genders   4  Age  Groups   100  x  4  =     400  Segments   25  Interest  Clusters   400  x  25  =     10,000  Segments  
  • 23.   Copyright  ©  2013.  Tiger  Analy7cs   Methodology   •  Iden6fy  high  performance  segments   –  Sta6s6cally  significant  difference  in  ctr,  cpc,  cost  per  conversion,  etc.   –  Use  ctr  as  a  proxy  for  conversion  rate   •  Ac6ons  on  high  performance  segments   –  Allocate  higher  budget   –  Increase  bid  price   23
  • 24.   Copyright  ©  2013.  Tiger  Analy7cs   Data  aggrega7on   Segment  Level  Data   (Sparse  and  Noisy)   Iden7fy  Important  Dimensions     (Using  Sta7s7cal  Models)  
  • 25.   Copyright  ©  2013.  Tiger  Analy7cs   Segment  performance  es7ma7on   Model  Es7mates   Observed  Performance   Prior  Knowledge   Inferred  Performance  
  • 26.   Copyright  ©  2013.  Tiger  Analy7cs   Bidding   Brand  A   Brand  B   Other  Compe77on  for  Ad  Space   Bid:  $1.60   Bids   WIN   Bids  will  differ  by  Ad  and  Micro   segment,  and  will  change  over   7me  
  • 27.   Copyright  ©  2013.  Tiger  Analy7cs   Budget  Alloca7on   •  Increase  budget  for  high   performance  segments  and  reduce   for  low  performance  ones   –  Business  rules  around  minimum   and  maximum  limits  
  • 28.   Copyright  ©  2013.  Tiger  Analy7cs   Methodology   Segment  Level     Observed  Data   Inferred    Performance  Indicators   Based  on  priors,  observed,  model  es6mates   Cost  per   Applica6on   Success   Rate   Dynamic  Budget  Alloca7on   Based  on  inferred  performance  indicators   and  business  constraints   Historical   Campaign  Data   Priors  of   Performance   Indicators   Weighted  Data   Click  vs.  view  through,  card  value,  applica6on   result,  recency,  delay  in  view-­‐through  appls   Cost  per   Acquisi6on   Model  Performance   as  a  func6on  of  targe6ng   dimensions   Model  Es7mates  of   Performance  Indicators   Dynamic  Bid  Alloca7on   Based  on  observed/historical   Bid-­‐Spend  rela6onships     Con7nual  monitoring  and   analysis   Business   Constraints  
  • 29.   Copyright  ©  2013.  Tiger  Analy7cs   Results:  Increased  CTR   29 •  Overall increase in CTR by 50% across more than 100 brands
  • 30.   Copyright  ©  2013.  Tiger  Analy7cs   Results:  Lower  costs   30 •  Overall decrease in CPA of 25% across more than 100 brands
  • 31.   Copyright  ©  2013.  Tiger  Analy7cs   Concluding  remarks   •  Online  and  social  adver6sing  are  fast  growing  areas  with   –  Plenty  of  data   –  A  large  number  of  interes6ng  problems   •  Predic6ve  analy6cs  can  add  a  lot  value  in  this  business   –  Significant  improvement  in  CTR  means  beber  targeted  ads   –  As  much  as  25%  reduc6on  in  cost  of  media   •  Our  solu6ons  are  being  used  by  several  leading  startups  to   serve  billions  of  ads  for  Fortune  500  companies   31
  • 32.   Copyright  ©  2013.  Tiger  Analy7cs   Ques7ons  /  Comments  ?     mahesh@7geranaly7cs.com   www.7geranaly7cs.com   32