5. Enterprise & External data
External data is usually wide and shallow. Enterprise data is narrow and deep.
data-driven home search
transportation
socio economics -unemployment-
24. Price-plan
2. We are social animals
1. Black or white
and
Gb
£
data plan (Gb)
# contacts
%customers
25. Understanding reasons to leave and motivations to stay
customer
experience
OOB spend
spend on
roaming
4G at home
26. Homophily and influence effects can be measured and predicted!
Don’t forget the social factor!
Jun Ding et al. Alone in the Game: Dynamic Spread of Churn Behavior in a
Large Social Network a Longitudinal Study in MMORPG
27. Data is not numbers, it’s people.
What you can’t predict, you must at least see!
This customer feedback can go unnoticed if only structured feedback
and scores get looked at.
29. Within the data science community
No organisation is perfect,
but some good practices help [1/2]
30. Invest in standardisation
Standardisation enables sharing and collaboration: lowering barriers, increasing
expectations)
data format
platform
tools
frameworkmethodology
33. Solid software development practices: know your code!
http://thecuriouscan.com/learn-from-the-costliest-mistakes-in-history/
June 4, 1996 Ariane 5 rocket launched by the European Space Agency exploded
just 37 seconds after its lift-off
7 billion dollars development of the rocket
The cost
500 million dollars estimated value of the
destroyed rocket and its cargo
34. Solid software development practices: know your code!
http://thecuriouscan.com/learn-from-the-costliest-mistakes-in-history/
A software programming error!
A 64 bit floating point number relating to the horizontal velocity of the rocket with respect
to the platform was converted to a 16 bit signed integer.
The number was larger than 32.767, the largest integer storable in a 16 bit signed integer,
and thus the conversion failed.
The reason for the blast?
35. Model accuracy is important, but it’s not the only thing
Prefer interpretability at the beginning, then upgrade models.
36. Start simple, then iterate
• think it (reduces the risk)
• build it (as fast as possible)
• ship it (gradually roll out to all users)
• tweak it (continuously improve)
Prefer high precision of one product instead than many sophisticated products.
The real risk is building solutions that no one needs:
D.J. Patil, Data Jujitsu: The Art of Turning Data into Product
37. No organisation is perfect,
but some good practices help [2/2]
With the rest of the organisation
38. • Engage with stakeholders from day 1
It is a bidirectional direction thing:
- a data science team must know the business priorities
- the whole organisation needs to understand and engage with data driven
results, stories and their value
39. • Engage with stakeholders from day 1
It is a bidirectional direction thing:
- a data science team must know the business priorities
- the whole organisation needs to understand and engage with data driven
results, stories and their value
• Remove barriers & enable data connectivity
40. • Engage with stakeholders from day 1
It is a bidirectional direction thing:
- a data science team must know the business priorities
- the whole organisation needs to understand and engage with data driven
results, stories and their value
• Remove barriers & enable data connectivity
• Agree on implementation and performance metrics