Upload
kalea
View
49
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Understanding Analytics. Keeping up with the Quants & Lifting the mist. Dr Andrew McCarren. What we start with?. Getting a clear picture. Lifting the Mist. What is the question? No exact answers? Assumptions? Variation (the same inputs don’t always give us the same answers) - PowerPoint PPT Presentation
Citation preview
Understanding Analytics
Keeping up with the Quants & Lifting the mist.
Dr Andrew McCarren
What we start with?
Getting a clear picture
What is the question? No exact answers? Assumptions? Variation (the same inputs don’t always give
us the same answers) Vast amounts data. Is it clean? How do we present our inferences?
Lifting the Mist
Leads the data analysis/ Data capture Interprets the needs of the organisation Understands the data and the analysis Can speak a common language
What is an analyst?
40% of decisions are made on gut instinct. Statistical predictions consistently out
perform gut Extensive evidence that having experts is
good but experts using analysis is much better
Expert intuition is better only when there is no data and little time to get the data.
Analytics VS Gut
+ Cigna health insurance◦ Using phone calls to reduce the amount of time in
hospital of its clients◦ Used analytics to determine which illness had
reduced time in hospital through phone call intervention
◦ Saved money by focusing staff on the right strategy with regard to phone calls
Problem solving with Analytics
- AIG ◦ Didn’t listen to the quants with regard to the risks
the company were taking with over leveraged CDS
◦ Cost AIG billions and effectively put the planet into a tail spin.
Problem solving with Analytics
Analytics – ‘always’ been around (since 5000BC) - tablets found recording the amount of beer workers were consuming.
WW2 – Focus on supply chain and target optimisation. Advent of Operations Research
UPS created a ‘statistical analysis group’ in 1954 70’s: Intel employ statisticians to develop line
optimisation Howard Dresner at Gartner defines “business
intelligence” 2010: Analytics begins to blend with decision
management
History of Analytics
Faster computers ◦ Processing power
Ability to store vast amounts of data.◦ Cloud, hadoop
Better visual analytics◦ Dashboards◦ Graphics◦ More user friendly solutions (Excel, SAS, Cognos
etc)
Improvements?
Academic Vs Real World◦ The interpretation is not always easy to understand
or communicate The world requires data faster and wants real
time solutions, Mathematical Modelling is not intellectually
easy. There is so much data
◦ Which data do we use?◦ Structured vs non-structured data.
Are our assumptions right?
Problems
People not Knowing what they want Quants not been given a clear mandate by
the organisation Rapid change in operational and delivery
technologies Lack of standards.
Culture
Data◦ ‘Quality’ , clean data
Enterprise◦ Management approach/systems/software
Leadership◦ Passion and commitment
Targets◦ Get the right Key Performance Indicators/metrics
Remember, what gets measured gets managed Communication
◦ Training/visuals
What’s needed?
Training Professionalism Define metrics/KPI Ask the right question Pick the right projects Engage management and get their
commitment Show the benefits Make the results clear
Leadership
What are other industries doing today that we could do tomorrow◦ Pharma randomised tests◦ Retail/online price optimisation◦ Manufacturing real time yield reporting
Systems◦ What do we have and can we get data from it?◦ Is our data on different platforms ?◦ Can we merge our data?◦ Can we interrogate our data in an intelligent and efficient
manner?
Looking Outside the box
Stage 1◦ 1. Problem recognition◦ 2. Review of previous findings
Stage 2◦ 3. Modelling◦ 4. Data Collection◦ 5. Data Analysis
Stage 3◦ 6. Results presentation
Quantitative Analysis 3 stages-6 steps: T. Davenport
1. Problem Recognition – Usually starts with broad hypothesis – “We are spending to much money on market research”
2. Review previous findings – Research the area. What are others doing?
Frame the Problem
3. Modelling/ Variable selection 4. Data Collection.
◦ Precision/ measurement capability◦ Qualitative/ Quantitative◦ Structured/unstructured
5. Data analysis◦ Types of stories-descriptive vs Inferential analysis
Solve the problem
6. Results ◦ Presentation and Action◦ Academic not equal to ‘Normal’ Interpretation◦ A Picture Tells a thousand Words
Results
1 2 3 4 5 6 7 8 9 11(blank)05
1015202530354045
Total
Total
Results presentation and action◦ Not normally focused on by academics. But
beginning to change. Need to tell the story with narrative and pictures.
Communicating and Acting on Results
Engineer wants to change printers on board manufacturing because boards are being sent wrong way on the line. ◦ Stopped them spending a fortune on replacing printers world
wide.
Line installation stopped from going wrong.◦ Line approval was stopped until machine gave stable results.
Pharmaceutical industry clinical trial on cancer patients and their reaction/adverse events to a drug.◦ Obsession with significance testing
Examples of Success & failure
CSI Solve a problem Solve a long term problem with analytics MAD Scientist – conducting experiments Survey the situation Prediction – use past results to tell the
future What happened –Straight forward reporting,
descriptive statistics (accounts, CSO)
Types of analytical stories
Choice of measurement device critical◦ Weigh up the ROI of the options and the results
that can be got from it.◦ 27k simple single measurement device versus
350k for XRAY machine for measuring fat on Pigs.◦ What are using the data for?
Stability/Accuracy/Consistency and interpretation of Measurement is critical.◦ Wrong measurement gives wrong conclusions◦ How does one translate language into numbers?
Measurement Problems
Learn the business process and problem Communicate results in business terms Seek the truth with no predefined agenda. Help frame and communicate the problem,
not just solve it Don’t wait to be asked
What non-Quants (Deciders) should expect of Quants
Form a relationship with your quant (Don’t lock them in a room)
Give access to the business process and problem
Focus primarily on framing the problem not solving it
Ask lots of questions, especially on assumptions.
Ask for help with the whole process
What Quants should expect of Non-Quants (Deciders)
Machine Learning Voice, Video, text Personalised Analysis
◦ i.e. what is *this particular* consumer likely to buy at this point in time when presented with these particular choices
Automotive Modelling◦ The models adapt themselves to update analysis
The future?
Building the capability takes a huge amount of time and resources◦ Barclays 5 year plan on ”Information – based
customer management” The big companies believe in it. Communication & Culture is key to success. Every organisation has vast amounts of
data they are not using.
It takes time
Assumptions about the data?
Failures to adapt models◦ Proctor and Gamble run 5000 models a day
Wrong interpretation of the models
Mistakes
Follow the 6 steps Always question the data
◦ Where did they come from◦ How were they measured?◦ Are the data stable?◦ Examine outliers/unusual events
Understanding the problem always takes away the mist.
Communication is key to success. Organisation needs a Culture/ Leadership to
succeed in analytics.
Conclusion
Thank You