Upload
emerce
View
1.627
Download
6
Embed Size (px)
Citation preview
page : &num
1
KLM - Emerce PBACross channel attribution past the last click
How to redistribute performance “weight” over media?
Frank Pon
December 2nd
page : &num
2
KLM Promotions Team Goals
page : &num
3
Steer customers via most (cost) effective e-acquisition channel
Social mediaMeta Search E-mail BannersAffiliates Search engines
page : &num
4
… but more importantly, approach/philosophy …
• Goal: pure conversions, sales, revenue
• Primarily tickets
• Central approach, (mostly) central execution and budget
• Close cooperation with local establishments and budgets
• Central expertise centres for advice and execution
• Optimisation across channels & countries
• High impact: central Affiliation activities in limited countries, E-mail 70 countries, etc
page : &num
5
KLM Promotions Team Organization
page : &num
6
KLM Promotions Team working together
K L M P ro m o tio n s T e am"E -com m e rce u m fe ld "
E m ily H a rd yM a na g e r S e arch E n g in e M a rke ting
A rn o N ijla ndM a n a ge r D isp la y M arke ting
E m ilie H a rd ickM a n a ge r E -m a il M arke ting
A rja n G roo tve ldM a n ag e r M e ta S e arch M a rke ting
T e re sa B arte ldsM a na g e r A ff ilia t ion
O rch id e a K ad a ruskanC a m pa ig n M a na g er
N ie ls Jo n kerJu n io r C am p a ig n M a na g er
F ra nk P onS e n io r P ro m otio n s M a na g er
B o b L an ge rakW e b A n a lyst
F ra n s P o ld e rva a rtW e b A n a lyst
T ja llin g S m itD ire c to r E -a cq u is it ion
Jo o s t W ilb rinkE -d eve lop m en t m a n ag e r E -acq u is it ion
S a n dra M a asE -d eve lop m en t m a n ag e r E -acq u is it ion
R o b Z w e rinkD ire c to r E -de ve lop m e nt
C h ris S pra n g e rsS e n io r E -sa le s M an a g er
?D ire c to r E -sa les
G u id o va n T ilE -co nve rge n ce
M a rtijn van d er Z eeV ice P re s ide n t E -co m m e rce
A ir F ra nce K LM
page : &num
7
KLM Promotions Team Results
page : &num
8
Acquisition results: shareFor first 6 months 2010 on average 66% of all KLM bookings is attributable
to one of the acquisition channels
Source: DART, KLM only, apr – aug 10, fully de-duplicated on last click basis Top 13NOTE: Actual acquisition share is expected to be considerably higher: 1. Only structural acquisition activities are included: bookings resulting from campaigns (e.g. world Deals) are not included2. Tracking problems with email since EBT7 (May 09) -> share likely higher.
Channel share
Natural bookings (34.0%) 35,580
Affiliation (3.8%) 4,000
Metasearch (4.8%) 5,012
E- mail (11.7%) 12,296
Display (19.6%) 20,538
Search (26.0%) 27,271
page : &num
9
E-acquisition share growth
*Top 13 results
E-acquisitie share
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
Apr/13
May/13
Jun/13
Jul/13
Aug/13
Sep/13
Oct/13
Nov/13
Dec/13
Jan/14
Feb/14
Mar/14
Apr/14
May/14
Jun/14
Jul/14
Aug/14
maand
sh
are Totaal
Lineair (Totaal)
page : &num
10
But how to get from
?
page : &num
11
The «last touch» model, default attribution rules
The last exposure before conversion gets the full credit, with clicks taking priority over impressions
In other words, no matter how many different ads a consumer has previously been exposed to, the last click before purchase is attributed 100% of the revenue
page : &num
12
page : &num
13
Analysis of pre-booking behavior (30 days)
Technique DART Floodlight Tag# Reported bookings 1 (last click counted)Payment Last click (where applicable* )Future budget allocation
Budget according to weight in decision making process -> attribution model
*Display view (14 days) unless click
page : &num
14
Dig Deeper
page : &num
15
Considering touchpoints prior to last touch?
% of touchpoints in the last 14 days prior to conversion /visitignored in the “last touch” approach
1st elements in 2009 showed a potential double digit gain in ROI through reallocation of budgets across channels and countries
KL
M-A
rte
mis
20
10
, F
eb
. 0
1 –
Ap
r.1
5
page : &num
16
Towards a KLM proprietary attribution model
Overall objectives
Defining methodological guidelines, implementation rules
• to attribute more “correctly” conversion credit to all channels analyzing cross-channel tracking data,
• eventually, to create the framework for an optimization of holistic online strategy including investment allocation across channels
page : &num
17
Adding up an attribution model to include all ROI drivers
page : &num
18
Modeling test period, input variables selection
KLM data base2010 Feb 01-Apr 15th 10+ millions cookies
Sampling 176,461 cookieswith a tagged activity on KLM web site @ each step of the EBT
Clustering the digital “journeys”Operational segments according to the outcomeAnalysis to determine key criteria to measure “assisting power” for each channel
Statistical modeling of individual paths and overall –aggregated time series to quantify contributions to response from each online channel, their interactions
page : &num
19
We need more “mad” scientist!
To break the rules & learn…
page : &num
20
Each cookie described on a 6-dimension scale
Channel (s)consumer was exposed to
/ interacted with
Depth of on-site interaction measured as EBT max level reached
Frequencyor number of contacts of each typeprior to on-site activity
Recencyor number of contacts per period of time prior to on-site activity
Interaction typee.g. click vs. exposure for display ad, keyword
entered for Search…
on-site visits number of visits, max EBT
level reach at each of them
Engagementwith KLM
Contacts, interactions with online ad vehicles
page : &num
21
Key factors in the “online consumer path”Multi-channel interactions at individual level
In the 9 markets, 39% of purchases with Search as last touchpoint had an exposure to a Display ad in the 14 days prior to purchasing
KL
M-A
rte
mis
20
10
, F
eb
. 0
1 –
Ap
r.1
5
page : &num
22
Key factors in the “online consumer path”Multi-channel interactions with a recency effect
Purchase with Search as last touchpoint with previous display exposure
KL
M-A
rte
mis
20
10
, F
eb
. 0
1 –
Ap
r.1
5
page : &num
23
Key factors in the “online consumer path”Multi-channel interactions with a frequency effect
Purchase Search as last touchpoint with previous display exposure Distribution- nb of Display contacts prior to purchase
KL
M-A
rte
mis
20
10
, F
eb
. 0
1 –
Ap
r.1
5
page : &num
24
What the heck does this tell us?
What ingredients are we missing?
page : &num
25
Key factors in the “online consumer path”Different interaction types
In the 9 markets, 77% of purchases with Search as last touchpoint are done through “navigational” keywords – Ad Group KLM
KL
M-A
rte
mis
20
10
, F
eb
. 0
1 –
Ap
r.1
5
page : &num
26
Key performance measurement factorsNumber of visits on the web site
Cross-analyzing number of days with an on-site visit and eventual purchasing – Example of UK data
KL
M-A
rte
mis
20
10
, F
eb
. 0
1 –
Ap
r.1
5
Visit without purchase(95,6%)
Eventual purchase(4,4%)
page : &num
27
EBT7- Pre-purchase steps
1
23
4
56
7
8
Interactio
n
dep
th 1
Interactio
n d
epth
2
page : &num
28
Re-allocation algorithm: inputs & KPIs
Navigational Search
Searchother
keywords
Affiliation
Meta-search
Last touchpoint
Other interaction in the last 7 days
Assisting drivers
Display in the last 14 days
Display in the last 7 days
Display in the last
2 days
DisplayClick
DisplayExposure
KLM onsite performance measurement •Number of visits (in # of days)
•“Depth” of each
•Eventual purchase
Exposure to KLM display ads# of contacts (grouped)
Interaction with KLM
e-mailing
page : &num
29
What did we learn … earn!
page : &num
30
Re-assigning credit to online tactics
Based on online path patterns: weighted according to contribution to customers response (visits, multiple visits, purchases) – not to be generalized as fitted to BY0910-Q4
page : &num
31
Potential gain in incremental bookings via re-allocation across channels and countries +5 to +8%
Based upon 2010, Feb. 01-Apr. 15 only
Potential gain at same “offer attractiveness”
page : &num
32
Key take-outs!
1.) Although online marketing is perceived as sexy, innovative and creative, some of the real value is in good old boring analysis
2.) Don’t limit yourself to existing industry standards and models, dig deeper!
3.) Don’t stop being sexy and explore “single channel” options like BT to the fullest in parallel
..... Model = “ a simplified version of the truth”
.... so keep on testing & experimenting
page : &num
33