16. A data-analysis machine…
• Segmented customer acquisition
• Auction planning
• Personalized auction feeds
• Targeted retention campaigns
What did we build?
17. How much data does Subasta produce?
• 50.000 sessions per day
• 10 pageviews per session
• 50 rows of data per pageview
= 25mio rows/day
= Big data? How many can excel handle?
18. What kind of data?
• Transactional (eg. a sale from own database)
• Behavioral (eg. #sessions from Google Analytics)
• Personal/profile (eg. likes from Facebook)
• Contextual (eg. device or location info or weather conditions
from open data sources)
19. What do we do with the data?
Transform into knowledge…
25. Everybody within reach…
• >90% of our target markets are on FB
• We can make them discover Subasta de Ocio
• But it is not interesting for everyone: you have to be relevant!
Behavioral targeting is key! How?
26. Online and
offline sources
Targeting types
Behaviors
Location
Interests
Demographics
Own
database
Measuremen
t partners
3rd parties
Sample segment
Single
Goes to activities with
friends
Has purchased tickets
before
Interested in
soccer
Living in
Enschede
Example of addressing latent demand
30. Tailor messages to target different micro-segments
4* Hotel in Barcelona starting at 1€
“Visit amazing
monuments &
museums.”
Visit amazing monuments & museums.
Traveller
Oudoors
Culture-lover
31. Tailor messages to target different micro-segments
4* Hotel in Barcelona starting at 1€
“Relive the
romance with
great
restaurants.”
Relive the romance with great restaurants
Traveller
Married
Foodie
32. Tailor messages to target different micro-segments
4* Hotel in Barcelona starting at 1€
“Explore the
buzzing nightlife.”
Explore the buzzing nightlife.
Traveller
18-35
Single
33. Small segments = many many ads…
• 9 segments
x ~8 product categories
x ~20 product types
= 1440 campaigns
• Each with eg 10 messages and 10 images = 144.000 ads…
• …running simultaneously
• …spending money
• …not knowing if it’s worth it
34. • Real-time budget adjustments using Machine Learning algorithms
• Avoid fixed budget:
• Stop marketing to people that don’t respond
• Push campaigns that have a positive ROI
We doubled the results while cutting costs in half: ROI x4 !!
Campaign automation - improving ROI
44. FB campaign spend
Transactional data
FB API
Own dBase
LTV model Operational
dashboard for
campaign
managers
Involvement in
optimizing dash +
owner of results
47. Example of application: Twitter
• AI to test 1000’s of combinations of creatives, images and hashtags
• Push cost effective campaigns, turned off bad ones
• Result: 10.000 new customers at 50% the investment compared to not
applying this algoritm
48. Reasons why companies fail
1. They celebrate vanity metrics
2. They don’t focus on retention first
3. They use the “product is everything” mantra
4. They don’t invest enough in data, analytics & learnings
5. They don’t change and adapt (fast enough)
(adapted from: Brian Balfour/coelevate.com)
49. Consumers drive product
design through data and fast
product-market-fit cycles
“Don’t listen to what consumers
say, listen to what they DO”
50. 5. They don’t change and adapt fast enough
• Market is changing at an accelerated pace…
• …so focus on the things that have maximum impact
51.
52. 1. It will be harder to raise money
2. Valuations will be corrected
Therefore, more important than ever:
Efficiency & control by driving on data
Retention through product-market-fit
53. Key takeaways on how to grow your startup
• Create a learning organization on top of a data foundation
• Focus on retention not product
• Adapt to changes in the market FAST
• Raise money whenever you can
• Resilience and patience…
54. And there is opportunity everywhere…
Trying new things and learning from them
is the driver of growth
56. Possible thesis questions
• Challenge could be a thesis assignment
Others:
• DDDM in startups: making decisions with incomplete info
• Measuring the impact of a points-based loyalty program
• How to stimulate internal knowledge sharing in startups
• Setup the ultimate model to segment product-user combinations
• Auctions vs. posted prices: which context favors one over the other?