Building a data-driven future ThoughtWorks Live 2014
Jonas Jaanimagi (REA Group) Jennifer Smith (ThoughtWorks)
Introduction to REA Group
Introduction to realestate.com.au
* Nielsen Online Ratings, October 2012 ** Nielsen Consumer & Media View, Survey 9, 2012
realestate.com.au is one of Australia’s most popular websites
Who are our users?
Who will you find at realestate.com.au?
A diverse mix of ages and families
58% 42%
Gender Age 78 % main grocery buyer
17% singles living alone or with others
28% Couples with no children
42% Families with children
16%
35% 33%
15%
14-24 25-39 40-54 55+
A Month of Property Seeking with REA
* Omniture Site Catalyst, October 2012 ** Nielsen Online Ratings, October 2012 *** Internal listings data
13,566 EMAILS SENT TO AGENTS
POOL IS THE MOST POPULAR KEYWORD SEARCHED
1,565,978 UNIQUE BROWSERS USE A MOBILE
617,794,790 PHOTOS OF PROPERTIES
ARE VIEWED
830,700 NEW VISITORS
65,651 INSPECTION TIMES
SAVED
51MINUTES IS THE AVERAGE TIME SPENT ON OUR SITE
92,436,903 PAGE VIEWS WITH A
TABLET 878,531
PROPERTY DETAILS
PRINTED
3,195,000 UNIQUE AUDIENCE
97,903 NEW LISTINGS IN
BUY
459,187 PROPERTIES SENT TO
FRIENDS
How do users access the site?
8%
9%
10%
11%
12%
13%
14%
15%
16%
17%
monday tuesday wednesday thursday friday saturday sunday
Desktop Mobile Phone Tablet
How do audiences engage with realestate.com.au?
Adobe Site Catalyst, Device Type Report, March 4th to 31st March 2013
Visits
Device Usage by Day of Week
Empty Nesters • Baby Boomers /
Silent Generation • No kids at home • High level (70%
+) home ownership
• Downgrading to smaller property / lifestyle change
What property cycle are people in?
* Residential Consumer Segmentation May 2012 * Residential Consumer Housing Affordability & Sentiment Index Study June 2012 * Consumer Purchase Intention Study BUY April 2012 * Consumer Retire Insights Nielsen CMV Survey 4 2012
Share
Rent
Buy
Sell
Invest
Lifestyle
Retire
Buyers • Mid 30’s • Married, no kids yet • Moderate to high
household income ($70k+ pa)
• Intend to buy house within 5 years
• Just over 50% own property already
Sellers • Baby Boomers • Married with a
couple of kids • Live in the suburbs • Currently paying off
debt (credit cards, home loan)
• Moderate to high household income ($70k+ pa)
Renters • Singles & Couples • Mid twenties • Low to moderate
household income (<$70k pa)
• Live in suburbs close to the city
• 82% don’t own property
Sharers • Single • Early twenties • Looking to live
in the metro area, close to the city
• Sharing a 2 bedroom place
Investors • Aged 35 years
and older • High household
income (>$100k) • Looking for
properties priced <$500k
Retirees • 2.3m Aussies
already retired • Over 50%
planning renovations
• 1 in 3 retirees planning travel domestically & internationally
That ‘D’ word…
Small data can drive big outcomes
We must combine insights and data
That ‘D’ word…
That ‘D’ word…
Web Analytics: A trace of consumer activity
2013-10-23 09:00:22 | Searched for 1 bedroom units in North Fitzroy
2013-10-23 09:01:11 | Viewed property 1
2013-10-23 09:01:24 | Viewed image carousel
2013-10-23 09:02:50 | Clicked mail agent button
2013-10-23 09:03:36 | Viewed property 2
What activities would identify first home buyers?
Searching for low prices? 1 or 2 bedroom properties? “Cheap” suburbs? First home buyer developments?
Applications of machine learning
Handwriting/speech recognition
Stock market analysis
Medical diagnosis Bioinformatics
Fraud detection
Search engines
http://en.wikipedia.org/wiki/Machine_learning#Applications
… and first home buyer prediction?
How do we train our algorithm to detect first home buyers?
Take a survey
Not first home buyer First home buyer
Take a survey
Not first home buyer First home buyer
Machine learning in action: predicting first home buyers
Survey Responses
Web Analytics Data
What does our model think makes a first home buyer?
Searching with a low price band Sharing on social media Looking at property inspection times NOT searching for 4 car spaces NOT searching with a high price band
Predicting first home buyers
Anonymous Consumer
Web Analytics Data
Predicting first home buyers at scale
Predicting first home buyers at scale
Do first home buyers click more?
Ad targeting experiment: Who clicks more?
Continuing the cycle
Tweak model & Adjust experiment
Analyze effect Inspect methodology What do we change?
Just one small piece of the puzzle!
• Better, stronger models! • Diversify segments: general movers, investors • Find further uses beyond ad targeting • Unsupervised learning: what patterns exist
purely in the data?
Taking things further
• Start with an informed idea of your consumers • Get data scientists, developers, ad folks working together closely • Start small, learn from failure and stay skeptical • Creating value as early as possible
If you try this…
Thanks... any questions?