The document proposes a model for using data and analytics to generate insights that drive business processes and decisions. It outlines moving from a past model where data was a byproduct to a new model where data drives decisions. However, simply hoping data will provide answers is unrealistic given limited data scientists. The document advocates combining business knowledge with data to validate, optimize and predict outcomes, focusing first on "solid ground" opportunities before riskier exploratory efforts. It presents a structured approach involving missions, recipes and runbooks to formalize problem-solving and make the process clear and repeatable. The goal is a feedback loop where insights inform new questions and improvements.
Boost Fertility New Invention Ups Success Rates.pdf
Lightning talk on the future of analytics - CloudCamp London, 2016
1. Our big data lake strategy
will deliver digital insights
using cloud technology
Today, lots of companies are releasing press releases saying things like this.
2. Dear data
Tell me what to do
However, they might be secretly hoping for something more like this.
“Don’t give me facts to think through. Don’t give me options. Just tell me my next best
action.”
3. ? ? ? ? ?
? ? ? ? ?
? ? ? ? ?
? ? ? ? ?
? ? ? ? ?
Tell me what to dos Data Scientists
Eek!
These expectations are likely to be dashed, because of the imbalance between the
number of questions we hope will be automagically answered, and the number of
data scientists we have, who can combine business knowledge with an understanding
of what our data can and can’t do.
4. Market & operational knowledge
drove
business process
which created
data
Old model : data as by-product
As we’ve moved from this model ...
5. Data
drives
market and operational decisions
which creates
business processes
New model : data as oracle
… to this model (aka being ‘data driven’) ...
6. Market and operational
knowledge
Data (internal and
external)
Business
process
Insight
Business Case
Business Case
… we’ve struggled to find a middle ground that’s combines knowledge and data to
improve or create business processes - and the results are measured to support
doing more (or less) of something.
8. Where insight is operationalised
To the
business
Known
Human
capital
Human
capital +
Analytic
models
Unknown
Analytic
models
Unknown Known
To the data
Most of the excitement we hear in the data analytics world is about the opportunities
in the bottom right square in this quadrant.
9. Insight opportunities
To the
business
Known
Knowledge,
ideas,
dogma, etc
Validation
Optimisation
Prediction
Unknown
Ideas and
data not yet
created /
realised
Answers
waiting to
be found
Unknown Known
To the data
The idea of the next big discovery is why lots of firms fund exploratory data science
safaris in the hope of uncovering hidden value.
10. Validation Do or do not
Optimisation Make ‘doing’ better
Prediction What may happen
Insight opportunities
But there are plenty of opportunities on more solid ground, which we shouldn’t ignore.
11. Opportunity cost
Advantage from insight
Costofinsight
No
thanks
Yes
please
Not only are we likely to find that they’re cheaper to obtain, and more valuable in the
long run.
12. Opportunity cost
Advantage from insight
Costofinsight Unavoidably
inevitable
But also, they’re less prone to involve high costs over long periods of time, which is
the equivalent of betting everything on 21, spinning the roulette wheel and only
getting lucky occasionally.
13. Optimisation example
To the
business
Known
Combine
this ...
… with
this ...
Unknown
… to explore
this.
Unknown Known
To the data
Maybe, (just maybe) there’s a path to extract ‘value waiting to happen’ from data in a
safer way than ‘big bang’ gambles.
14. Optimisation example
To the
business
Known
We have X
people ...
… achieving
this.
Unknown
How could
we add
more value
in less time?
Unknown Known
To the data
Which can also give us a chance at unlocking the value that ‘big data analytics’
marketing is promising.
15. A structure for insight
Question ?
Missions
Result
Action Action
Situation Situation Situation
Recipes
--- --- --- --- --- --- --- --- ---
--- --- --- --- --- ---
--- --- ---
Run books
Here’s what that model may look like.
We start with a high value question.
Then we’ll structure a set of missions that think through business knowledge and data
components of a question. For example, if we’re looking for ‘prediction’ … what’s our
situation; what actions could we take to change it; then what result(s) would that lead
to?
By formalising our approach to solving this problem, we set up a relationship between
a mission and the relevant data, models and techniques used to complete it.
By formalising these ‘recipes’ and their ingredients we commoditise the interrogation
of data that’s selected for a mission. And to stretch the cooking analogy, we also avoid
relying on expert chefs (data scientists) for everything; instead we write a cookbook.
The end result is a ‘run book’ for business process, which anyone can turn to, to
understand: what problems we’re solving; how; and with what result?
17. Scope Where did we look?
Was that the right place?
Info What did we get?
Is it the right thing?
Coverage How complete is it?
Is that enough to decide?
Accuracy and Precision
We can now make our decisioning clear to all sides about how a request to ‘tell me
what to do’ has been thought through.
And we can invite ideas from the people involved in asking and answering the
question, to make sure the value of the answer (aka: ‘Here’s what you should do’) is
as clear as possible.
18. All models are wrong
Some are useful
As George Box once said, all models are wrong, some are useful.
If you’ve any ideas on how I could make this model less wrong, please let me know!