#TC18 using self-service analytics zulily to grow 1.5m ... · zulily: a place for fun shopping “...
Transcript of #TC18 using self-service analytics zulily to grow 1.5m ... · zulily: a place for fun shopping “...
using self-service analytics@ zulily to grow 1.5m customers
in one year
#TC18
Sasha BartashnikMarketing Analytics Manager
zulily
Sasha BartashnikManager, Marketing Analytics
@zulily
introduction
How we achieved 1.5m in customer growth
Our approach to marketing optimization
What’s next in marketing analytics
Q&A
agenda
zulily: a place for fun shopping
“
”
The fun of browsing is new ideas on what to buy. I am constantly getting inspired and updating. I
get a real endorphin rush when a new idea gets ignited in me.
routine(regular) shopping
funshopping
we launch
100 new sales every daytypically for 72 hours
1Based on customers that made purchases during 2017 calendar year. 2Based on orders placed from January to December 2017 by customers who previously purchased from zulily.
71%of orders placed on mobile
devices1
zulily today
A typical Costco has 4,500 SKUs;
we launch more than 9,000 product styles daily with minimal inventory
91%of orders placed by repeat
customers2
launch
hundredsof ads
serve
millionsof impressions
convert
tens of thousands
of customers
a day in the life of zulily marketing…
1An active customer is defined as an individual who has purchased at least once in the last twelve months, measured from the last date of a period. Active customers are in millions. All data as of Q2 2018.
6.4M Active Customers1
31% increase in active customers
13% increase in revenue
enabling growth
-
1
2
3
4
5
6
7
Q12014 Q22014 Q32014 Q42014 Q12015 Q22015 Q32015 Q42015 Q12016 Q22016 Q32016 Q42016 Q12017 Q22017 Q32017 Q42017 Q12018 Q22018
Act
ive
Cu
sto
mer
s (m
illio
ns)
Active Customers
self-service analytics
how we achieved growth?
Analytics teamcan move fast
without needingto involve IT
in everyday activities.
Business users get real-time access to key data
without needing to involve analysts
to generate basic insights.
marketing analytics team growth
BQ migration
develop self-service toolset
separate reporting from analytics
Create data science function
building the
Tableau + BigQuerypipeline
data teampushes all data —structured and unstructured,real-time and batch— into BigQuery
marketing analysts & data scientists join multiple data sources using BigQuery’s SQL
analysts develop models on BigQuery data marts using a variety of common data science toolsas well as internalETL platforms
marketers and analysts use Tableau for self-service analyticson data andmodel results stored in BigQuery
zulily Tableau + BigQuery pipeline
key centralized data in Tableau
Last Fiscal Week Efficiency
###
LTVLTV
######
Last Week Stats
Activation RateApp InstallsApp VisitsBounce RateConversion RateEmail ClicksEmail OpensOrders 0dOrders 14dOrder 365dRevenue 0dRevenue 14dRevenue 365dTotal CostVisits
2/5/2018
All data shown is masked to protect proprietary information.
All data shown is masked to protect proprietary information.
intra-day self-service insights
All data shown is masked to protect proprietary information.
holistic view of cross-company metrics
Revenue
Active customers
Brands on site
Order delivery
Cancellations
our approach to marketing optimization
1customer
acquisition
2lifecycle
management
3measuring the
brand
4monitoring
marketing analytics program components
add customer segmentation information
determine visit source
join to multiple traffic sources
collect raw data
foundational data layer with one view1 customer acquisition
use historical dataproduce variables using data
science model predict value days after
acquisition
1 customer acquisition
find existing highvalue customers
Which behaviors distinguish them from others?
Are new customersshowing these behaviors?
To what extent?
understand value of a new customer
What she uses to access zulily
What her first purchase was like
How she engages with us
Where she is located
How she found us
what goes into the zulily predictive model?
thousandsof variables considered
forthe model
hundredsof models tested
dozensof variables
chosen using gradient boosting machine
learning
+85%accuracy in predicting 1
year revenue
1 customer acquisition
$ $$$
1 customer acquisition
optimize new customer acquisition
a/b testing:
creative
timing
landing experience
paid vs unpaid
DR vs brand
shift budget & bids
lift by likelihood to be high value
mo
de
l sco
re
lift (%) in chance of purchase
lean in to these segments for offer
targeting
2 lifecycle management
targeted messaging
Favorite brands
Site and App Engagement
Purchase History
Email Behavior
Love this feature!
Text and Customer Service
using personalization to optimize2 lifecycle management
optimization based on response2 lifecycle management
All data shown is masked to protect proprietary information.
Email Performance Metrics YoY
Sen
d C
ou
nt
Op
en R
ate
Clic
k R
ate
Dem
and
/
Sen
d
Purchase/Churn Propensity Modeling
Attribution Model
Lookalike Modeling
Incrementality
Customer Level Profitability Modeling
Engagement Modeling
building the right models & tools2 lifecycle management
3 brands & partners
how we track the brand
TV response
NPS
customer surveys & VOC
All data shown is masked to protect proprietary information.
4 monitoring
monitoring data science models
Variable X Bin
All data shown is masked to protect proprietary information.
4 monitoring
monitoring inputs to data science models
what’s nextin marketing analytics
the future of zulily… and the industry
Increased integration between marketers
and data scientist– the rise of marketing engineers ex: advanced attribution
Using self-servicephilosophy to bemore efficient,
do “morewith less”
A renewed focus on sustainable growth,
long-term customer relationships
please complete the session survey from the
Session Detailsscreen in your TC18 app
thank you!any questions?