3. Agenda
What is behavioural data?
1. What is included? What is missed?
2. Define the granularity
3. What is not Why
4. Causality
5. The lamppost phenomenon
6. Make patterns easier to see
7. Say/Do grids
Q & A
4. What is behavioural data?
Data that describes what people do
In its typical form, data that doesn’t depend on people’s memory
Examples
• Web data (pages viewed, links clicked, search words, ads shown, length of visit etc)
• Digital wake (mobile phone usage, ATMs, payments, WiFi connections, messages,
emails etc)
• Metered data (energy, footfall, traffic flows etc)
• Social Media (Facebook, Instagram, Pinterest etc)
• Biometrics (smart watches, health apps, Strava etc)
• Observations
5. 1
What is
included?
What is
missed?
Are we including unwanted people?
• Do your analytics include people from
other countries, non-customers, bots?
Who is not included in the data?
• PAYG versus contract users
• Cash spenders versus cards/digital
• Non-loyalty card people versus loyalty
card holders
• Non-registered users (e.g. people using
their own eScooters)
6. Where should the armour plating go?
By Martin Grandjean (vector), McGeddon (picture), Cameron Moll
(concept) - Own work, CC BY-SA 4.0,
https://commons.wikimedia.org/w/index.php?curid=102017718
7. 2
Define the
Granularity
Is one data source = one person?
Multiple sources -> one person
• People with two or more phones or two or
more cars
• Multiple browsers
• Multiple profiles
One source -> multiple people
• Household data (energy, water, viewing
etc)
What are the units?
• Tampons – packs or individual items
8. 3
What is not
Why
Not buying/choosing/using something does not
mean it is not wanted
• There could be barriers
Doing something regularly does not mean
people want to do it
• Brands can confuse loyalty with being a
hostage
Member checking
• The gold standard for understanding the
why is through ‘talking’ to people
• But don’t assume they can/will tell you
9. 4
Causality
Causality is often difficult to establish from
behavioural data
ü Weather and switch from bicycles to
other transport
✘ HRT and heart problems
Experiments
• Creating Randomized Control Tests
• Natural Experiments
10. Missing the Real Driver
0
10
20
30
40
50
60
0
10
20
30
40
50
60
70
80
90
Time 1 Time 2 Time 3 Time 4 Time 5 Time 6
SM $ Sales $
Summer
Winter
Winter
Ice Cream Sales
11. Utilising Experiments
Region A
– T1 sales = 100
– T2, TV, sales = 110
– T3, TV & Twitter, sales = 130
Region B
– T1, sales 100
– T2, Twitter, sales = 110
– T3, TV & Twitter, sales = 130
Region C
– T1 sales = 100
– T2, sales = 105
– T3, sales = 110
The counterfactual = some growth would
have happened anyway.
12. 5 The Lamppost Phenomenon
Why are you looking
for the keys under the
lamp, is that where
you lost them?
No, but this is where
the light is!
13. 6
Make
Patterns
Easier to See
Fit trendlines or moving averages
Transform the data
• e.g. per capita, indexing, log scales etc
Check for outliers
Categorize
• e.g. 0 marriages, 1 marriage, 2 or more
18. 7 Say / Do Grids
Do
Say
Check if opiate patients need laxatives
Prescribe laxatives
Buy ethical products
Buy Apple products
19. 7 Tips 1. What is included? What is missed?
2. Define the granularity
3. What is not Why
4. Causality
5. The lamppost phenomenon
6. Make patterns easier to see
7. Say/Do grids