PA R I S
1 4 n o v e m b r e 2 0 1 9
HOW ANALYTICS DATA FUELS
EFFICIENT PERSONALISATION
Romain Lhote
Head of data marketing @ L’Equipe
More than 32M French people READ OR WATCH
L'Équipe EVERY MONTH
Since 1998
Since 1946
Since 2000
READER ENGAGEMENT
STRONG RELATION BETWEEN THE BRAND AND OUR READERS
2.8M
users per day
1.5 billion
page views per month
6mins
per visit
80%
from mobile
700
page views per
subscribers per month
Sources: AT Internet, Médiamétrie.
DATA TEAM
ON THE ROAD TO 2020
OUR TEAM HAS GROWN RAPIDLY over THE LAST TWO YEARS
Traffic manager Head of data
Data team in 2015 Data team in 2019
Data Analyst
Data Analyst
Data Analyst Data Analyst
SEO ManagerData Scientist Data Scientist
Dashboard Segmentation PersonalisationScoring
1 2 3 4
What we have done so far
FIRST STEP: DASHBOARD
FROM DASHBOARD TO PERSONALISATION
FROM ‘OLD SCHOOL’ TO AUTOMATED DASHBOARDS
› QLIK
› AT INTERNET
› iOS
› Android
› CRM
› METADATA
› Copy / past from Analyser › Excel via API
2015 2017 2019
SECOND STEP: SEGMENTATION
NO DUDE, DATA IS NOT JUST ABOUT DASHBOARDs
Email
Push
Google Ads
Facebook Ads
Display on the website
FOOTBALL HAS ALWAYS BEEN THE MOST IMPORTANT SPORT
[...] When we have a multi-sport homepage that can be
customised with the sports we want, we will have achieved
our goal.
Disappointing, too many articles about football.
Too much football
Why do you always focus on football?
Football, football, football, football.
Source: Store Android
BUT SOME PEOPLE THINK IT'S TOO MUCH, AND GUESS WHAT… IT’S TRUE
WE UNDERSTAND DATA IS IMPORTANT, BUT WHAT
CAN WE DO?
LET’S USE DATA TO SOLVE THIS PROBLEM OF PERCEPTION…
What people think…
20%
Reality
48%
Source: Internal
PERSONALISATION IS THE
ANSWER!
Skill
improvement
Mastering the
technology and data
WITH A PARTNER DATA TEAM
THIRD STEP: SCORING
WITH A PARTNER OR BY OURSELVES ?
A FEW WEEKS 18 MONTHS
Transparency
Web analytics Data Article metadata
Article
A journalist writes an article
They choose 4 tags
All this data is stored on GCP
Article is live on L’ÉQUIPE
Page view on the app
AT Internet datacentre
Data stored on GCP
Dataflow API call
WE STARTED WITH BOTH WEB ANALYTICS AND METADATA
PERSONALISATION
HERE IS PATRICK. WHAT DOES PATRICK LIKE ON L’ÉQUIPE ?
+
1 2 3
+
Articles he
read last
year
We compare with
what other users
read
We add many
weightings for the
calculation to be
relevant
WE USE THE TF-IDF METHOD
17
Count nbr of tags for all users
DFT = *Document Frequency Tag
Count nbr of tags per user
TF = *Tag Frequency
TF-IDF = (TF x +1)
log (1+n)
(1+DFT)
WELL WELL WELL
86%
Football
ranked #1
We did an
amazing job
But it didn’t
work well
WE APPLIED MANY WEIGHTING PROCEDURES
19
Football
It was not true
Season
News only
Only web analytics
Weighting procedures
Square root of nbr
articles created
1 year
Live, push, home
Questionnaire
WHAT DOES THIS SCORE LOOK LIKE ?
100
65
63
60
14
cyclisme
esport
tennis
football
formule_1
2801
1024
875
205
179
football
cyclisme
tennis
rugby
esport
Nbr of articles I
read
The score
FIRST TEST OFF-SITE
WE HAD GREAT RESULTS THAT HELPED US TO
Opening rate on Pushs vs casual
segmentation
X3,5 X2,5
Opening rate on email vs
casual segmentation
X30 vs general target X27 vs general target
PERSONALISATION OF THE HP
Journalists choiceMy home
100
65
63 60
14
cyclisme
esport
tennis football
formule_1
TO BE CONTINUED...
Thank you!
Head of data marketing – L’Equipe
Romain LHOTE

L'Équipe Customer Success - Using analytics to fuel efficient personalisation

  • 1.
    PA R IS 1 4 n o v e m b r e 2 0 1 9
  • 2.
    HOW ANALYTICS DATAFUELS EFFICIENT PERSONALISATION Romain Lhote Head of data marketing @ L’Equipe
  • 3.
    More than 32MFrench people READ OR WATCH L'Équipe EVERY MONTH Since 1998 Since 1946 Since 2000
  • 4.
    READER ENGAGEMENT STRONG RELATIONBETWEEN THE BRAND AND OUR READERS 2.8M users per day 1.5 billion page views per month 6mins per visit 80% from mobile 700 page views per subscribers per month Sources: AT Internet, Médiamétrie.
  • 5.
  • 6.
    ON THE ROADTO 2020 OUR TEAM HAS GROWN RAPIDLY over THE LAST TWO YEARS Traffic manager Head of data Data team in 2015 Data team in 2019 Data Analyst Data Analyst Data Analyst Data Analyst SEO ManagerData Scientist Data Scientist
  • 7.
    Dashboard Segmentation PersonalisationScoring 12 3 4 What we have done so far FIRST STEP: DASHBOARD FROM DASHBOARD TO PERSONALISATION
  • 8.
    FROM ‘OLD SCHOOL’TO AUTOMATED DASHBOARDS › QLIK › AT INTERNET › iOS › Android › CRM › METADATA › Copy / past from Analyser › Excel via API 2015 2017 2019
  • 9.
    SECOND STEP: SEGMENTATION NODUDE, DATA IS NOT JUST ABOUT DASHBOARDs Email Push Google Ads Facebook Ads Display on the website
  • 10.
    FOOTBALL HAS ALWAYSBEEN THE MOST IMPORTANT SPORT [...] When we have a multi-sport homepage that can be customised with the sports we want, we will have achieved our goal. Disappointing, too many articles about football. Too much football Why do you always focus on football? Football, football, football, football. Source: Store Android BUT SOME PEOPLE THINK IT'S TOO MUCH, AND GUESS WHAT… IT’S TRUE WE UNDERSTAND DATA IS IMPORTANT, BUT WHAT CAN WE DO?
  • 11.
    LET’S USE DATATO SOLVE THIS PROBLEM OF PERCEPTION… What people think… 20% Reality 48% Source: Internal
  • 12.
  • 13.
    Skill improvement Mastering the technology anddata WITH A PARTNER DATA TEAM THIRD STEP: SCORING WITH A PARTNER OR BY OURSELVES ? A FEW WEEKS 18 MONTHS Transparency
  • 14.
    Web analytics DataArticle metadata Article A journalist writes an article They choose 4 tags All this data is stored on GCP Article is live on L’ÉQUIPE Page view on the app AT Internet datacentre Data stored on GCP Dataflow API call WE STARTED WITH BOTH WEB ANALYTICS AND METADATA
  • 15.
  • 16.
    HERE IS PATRICK.WHAT DOES PATRICK LIKE ON L’ÉQUIPE ? + 1 2 3 + Articles he read last year We compare with what other users read We add many weightings for the calculation to be relevant
  • 17.
    WE USE THETF-IDF METHOD 17 Count nbr of tags for all users DFT = *Document Frequency Tag Count nbr of tags per user TF = *Tag Frequency TF-IDF = (TF x +1) log (1+n) (1+DFT)
  • 18.
    WELL WELL WELL 86% Football ranked#1 We did an amazing job But it didn’t work well
  • 19.
    WE APPLIED MANYWEIGHTING PROCEDURES 19 Football It was not true Season News only Only web analytics Weighting procedures Square root of nbr articles created 1 year Live, push, home Questionnaire
  • 20.
    WHAT DOES THISSCORE LOOK LIKE ? 100 65 63 60 14 cyclisme esport tennis football formule_1 2801 1024 875 205 179 football cyclisme tennis rugby esport Nbr of articles I read The score
  • 21.
    FIRST TEST OFF-SITE WEHAD GREAT RESULTS THAT HELPED US TO Opening rate on Pushs vs casual segmentation X3,5 X2,5 Opening rate on email vs casual segmentation X30 vs general target X27 vs general target
  • 22.
    PERSONALISATION OF THEHP Journalists choiceMy home 100 65 63 60 14 cyclisme esport tennis football formule_1
  • 23.
  • 24.
    Thank you! Head ofdata marketing – L’Equipe Romain LHOTE