22
Product Launch Analysis

ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

Product  Launch  Analysis  

Page 2: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

2

Introduc)ons  

– Semphonic  provides  full-­‐service  web  analy<cs  consul<ng  and  advanced  online  measurement  to  leading  Enterprises  in  the  United  States.  Semphonic’s  prac<ce  is  heavily  focused  on  advanced  digital  customer  analy<cs  in  the  warehouse  and  with  Web  Analy<cs  solu<ons.  Clients  include  American  Express,  Charles  Schwab,  Genentech,  JP  Morgan  Chase,  Kohler,  MarrioJ,  MicrosoK,  Nokia,  Samsung,  Turner  and  Walmart.  Gary  blogs  at  hJp://semphonic.blogs.com/semangel  

Gary  Angel  Co-­‐Founder  and  President  of  Semphonic,  the  leading  independent  web  analy<cs  consultancy  in  the  United  States.    

Page 3: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

The  Problem  

Page 4: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

4

Product  Launch  

What  we  were  dealing  with:  

A  new  sub-­‐category  of  electronics  

Significant  Mass  Media  and  Social  Media  Efforts  

A  soK-­‐launch  trial  with  a  small  set  of  “fans”  

 

Our  goal:  

To  quickly  (with  two  weeks)  measure  the  top  factors  driving  true  consumer  aZtudes  about  the  product  

 

Page 5: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

The  Upfront  

Page 6: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

6

Our  Pre-­‐Launch  Prepara)on  

Because  of  the  short  turn-­‐around  on  the  project  we  tried  to  do  significant  pre-­‐prepara<on  for  the  launch  analysis.  This  included:  

Descrip<ves  on  exis<ng  social  media  chaJer  in  the  larger  category  

Analysis  of  the  soK  launch  conversa<ons  

Crea<on  of  an  ini<al  classifica<on  scheme  for  the  product  

 

Page 7: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

7

Basic  Descrip)ves  

Pre-­‐launch  culling  of  feature  sets  gave  us  a  sense  of  the  broader  market  and  helped  us  tune  our  classifica<on  system.  

0%  

10%  

20%  

30%  

40%  

50%  

60%  

70%  

80%  

%  of  T

opic  Con

versa)

ons  

Conversa)on  Topic  by  Manufacturer  

Top  Laptop  Feature  Men)ons  

Any  Top  Feature  

BaJery  

Dimensions  

Hard  drive  

Processor  

Screen  quality  

Screen  size  

Touch  pad  

Weight  

Wi-­‐fi  

Page 8: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

8

SoG  Launch  Analysis  

The  short  soK-­‐launch  was  for  enthusiasts  only  –  so  it  wasn’t  a  useful  sample  –  but  it  did  help  us  see  how  conversa<ons  might  trend.  

Page 9: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

9

Ini)al  Classifica)on  

We  started  with  Clarabridge’s  pre-­‐built  Compu<ng  classificaitons:  

Where  appropriate,  pre-­‐built  classifica<ons  exist,  they  greatly  simplify  the  ini<al  startup.  

Page 10: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

The  Process  

Page 11: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

11

Profile  Crea)on  

NMIncite Buzzmetrics •  We used previously created and tuned profiles to create the initial data pulls from the API. •  Why this step is critical to the process:

•  If you miss relevant posts it will likely be systematic – ruining subsequent analysis

•  We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims

•  For one-time analysis, heavy use of exclusions in profile rules is acceptable. For ongoing analysis this can create maintenance problems.

Page 12: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

12

External  Cleaning  

Data Cleansing •  We scrubbed product sellers and “owned” and campaign media •  Eliminated extraneous text (“Source:” labels for example) •  Eliminated URLs •  De-Duplicated Full Text Matches after these 3 steps

•  Clarabridge will de-duplicate but you get charged for loaded verbatims

Page 13: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

13

Classifica)on  Tuning  

Defini)ons  

•   Each  area  is  defined  by  a  set  of  keywords  and  then  further  classified.  

•   Here,  for  example,  Form  Factor  is  divided  into  Style  and  Quality  Words.    

•   Each  sub-­‐division  then  has  a  Brand  classifica<on  so  that  we  can  compare  how  oKen  (and  how  favorably)  our  key  Brands  were  men<oned  within  the  group.  

•   This  is  important  because  posi<ve  sen<ment  might  be  aJached  to  a  comparison  (i.e.  X  has  great  style.  Y  is  rather  boring).    

Page 14: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

The  Pres)ge  

Page 15: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

15

Establishing  Baselines  

At  the  top-­‐level,  it  is  important  to  know  how  oKen  different  features  are  men<oned.  Especially  for  a  new  “category”  or  “sub-­‐category”  there  can  be  real  learnings  even  at  this  basic  level.  

Reliability  Gameplay  

Product  -­‐  Boo<ng  Applica<on  Availability  

Netbook  Men<on  Speed  

Tablet  Men<on  Cloud  

Pricing/Value  Laptop  Men<on  Connec<vity1  

Product  -­‐  Hardware  Use  and  Care  

Product  -­‐  Form  Factor  and  Style  Product  -­‐  Opera<ng  System  

Brand  

0   500   1,000   1,500   2,000   2,500   3,000   3,500  

Category  Men)on  Counts  

Page 16: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

16

Segmenta)on  of  the  Audiences    

Wordcloud for Laptop Subset Wordcloud for Tablet Subset

Academic Focus Battery Focus

Page 17: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

17

Presen)ng  the  Whole  Picture  

GeZng  a  reasonable  sen<ment  analysis  took  LOTS  of  cleaning:  

Page 18: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

18

Trending  

One  significant  advantage  to  this  approach  is  that  it  allowed  us  to  trend  the  Product  Launch  over  subsequent  weeks.  

Page 19: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

The  Post  Mortem  

Page 20: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

20

Goods  and  Bads  

The  good:    Clarabridge’s  sample  classifica<ons  made  the  ini<al  setup  much  easier  

  The  itera<ve  classifica<on  process  made  it  fairly  easy  to  tune  the  analysis  

  The  pre-­‐analysis  helped  A  LOT  to  get  the  final  analysis  done  on  <me    The  analysis  was  impacqul  in  tuning  the  post-­‐launch  social  campaigns  and  cost-­‐effec<ve  enough  to  do  repeatedly.  

The  Bad:    Extrac<on  of  the  data  was  frustra<ng  –  the  API  is  throJled  so  we  had  to  make  repeated  calls  

  GeZng  the  sen<ment  classifica<ons  clean  took  A  LOT  of  work  –  much  more  itera<on  than  we’d  hoped  

  Tuning  sen<ment  words  was  manual,  arduous  and  subjec<ve  

   

Page 21: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

Discussion  

Page 22: ProductLaunch(Analysis( - Text Analytics Worldanalysis • We use Clarabridge on a per-verbatim basis so there is a real cost to loading useless verbatims • For one-time analysis,

22

Thank  you!    Gary  Angel  President  Semphonic  415-­‐884-­‐2511  [email protected]