Data Science-final7

Post on 13-Jan-2017

259 views 0 download

Transcript of Data Science-final7

1 1

The Landscape – Analytics and Data Science

2 2

When I say big data which of these describes what you feel?

3 3

When I say big data which of these describes what you feel?

4 4

When I say big data which of these describes what you feel?

5 5

When I say big data which of these describes what you feel?

6 6

When I say big data which of these describes what you feel?

7 7

When I say big data which of these describes what you feel?

• Well, this talk is NOT about big data, but what it can do for you

• On the way, you might just gain some clarity of terms, and technologies

8 8

Big Data

• The quantity of data allows the five pillars of analytics to become empirical sciences

• If used right, business and medical goals are substantially bettered

• It is not just about knowing more, it is about zeroing in on the truth

• We will talk about they ways people miss the truth, even seeming to use current best practices

9 9

Big Data

10 10

11 11

• I like to start with the business questions, the business and medical practice needs, what leaders of businesses and medicine would most like to do

12 12

The Big Questions

• Who, where, what, how, and how much for each group

13 13

Who, where, what, how, and how much

• Business is about action, doing. Just do it! But what to do, and to whom, and with what, and what channel?

• What are the choices that maximize results and ROI?

14 14

Just do it, to maximize results and ROI

• Global Goal Driven Dynamic Demographics for Proaction Optimization (G2D3PO or just D3PO)

• What Group?• What Actions?• Global Goal: Profit, Customer Satisfaction,

Manufacturing Excellence, Reduced Community Healthcare Costs, etc.

15 15

Global Goal Driven

16 16

Why these arbitrary fixed groupings?

• Groupings, affiliations and interests depend on goals and context

• Arbitrary bounds to ranges are just assumptions that can totally change how statistics come out and our view of the world

• Bad assumptions lead to bad decisions • Data discovered and confirmed foundations

lead to good decisions

17 17

Global Goal Driven

18 18

Case Study – Large Retail Bank – Customer Service Screen

19 19

Goal: Sell Product – Top Pain Found and DefinedOptimal Grouping and Action

20 20

Case Study – Large Retail Bank Targeted Allegiance Maintenance

• Before: 0.07% conversion rate for new product proposals – “shotgun” marketing

• After: 7%– A 100X increase in closure rate– $1B increase in new product the first year – Rising even faster the second year– “Good customer” attrition rate down 6% • $200M annual saving from this churn mitigation

21 21

Case Study – Large Retail Bank

• Almost no lift from “shotgun” marketing approach

22 22

What is new and different? -- Old Profit Lift Curve

• The state of the art predictive approach: Ranking via scores

23 23

What is new and different? – State of the Art Profit Lift Curve

• Global Goal Driven Dynamic Demographics for Proaction Optimization• Spending only as much as needed to acquire

24 24

What is new and different? -- D3PO Lift Curve

• Groupings are generated appropriate for the sale of each product

• The best of multiple offer opportunities is chosen

• The best proposal opportunities first• Product, offering, or discount proposals based

on expected long-term value to company• Cost of acquisition more focused

25 25

Improvements come for three reasons:

26 26

Case Study – Hospital /Managed Care

27 27

Managed Care – Group Based on Goal: Increase population Health

28 28

Hospital /Managed Care – Personalized Recommendations

• Static: Statistical Reports• Interactive Descriptive: BI• Predictive: Learning Algorithms (LA)• Prescriptive: Decision Classes or Optimization• Proactive: Optimizing Groups

29 29

The Progress

• To the business user the technology should just be something that happens in the background

• At the same time, how the recommended decisions are being made should be transparent to the managers

30 30

The Best Thing for Business and Medical Analytics

31 31

This is a talk about:

• How Big Data can enable Data Science to be a true science

• The large opportunities it can offer for generating value for companies and healthcare

• How we only can know, see and forecast because we have assumptions, assumptions that can be wrong

• But what makes the empirical method work is the process of testing and revising assumptions to discover the real world

• What business and healthcare providers want • How BI, and Advanced Analytics depends on assumptions• How easy it is to attribute too much intelligence to artificial

intelligence• True intelligence is bound up with the ability to recognize

and revise assumptions• That methods of grouping are always multiple • How the generation of action classes (or Proaction classes)

to appreciate groups of people (or resources) for a given goal is the method to add great value to business and healthcare

32 32

We will also recognize:

33 33

Revising assumptions changes your world

34 34

Different categorizations for different goals

35 35

Different categorizations for different goals

36 36

Arbitrary Assumptions -- Age vs. Income breaks

37 37

Arbitrary Assumptions -- Age vs. Income breaks

38 38

Arbitrary Assumptions -- Same People Different Understandings

39 39

The interesting cluster cannot even be seen by slice and dice methods

40 40

Both the human eye and LA’s must make assumptions to see at all assumptions that circumstances can reveal

41 41

Both the human eye and learning algorithms can impose readings that make no sense

42 42

Both the human eye and learning algorithms can miss important hidden patterns

43 43

Real Intelligence is knowing how to collect the added data to know what is really going on, like throwing a rock at the lava

• What is Data Science?

44 44

Humans are natural pattern recognizers We project out inner patterns and assumptions on the universe

45 45

So do learning algorithms and many of the modeling methods

46 46

These are random dots

47 47

Our eye just naturally finds meaningless clusters

48 48

Learning algorithms and modeling have built in assumptions too

Training on first half of the flight of a baseball

Attempting to predict 2nd half

49 49

Built in assumptions – Flight of a Baseball – Non-representative data

Error

50 50

Non-representative data

51 51

Non-representative data

52 52

Baseball fit – Error from just assumptions of mathematical form

Error

• Data generated by people, such as in markets or from buying behavior, has far more noise than a physical system such as a baseball’s flight

53 53

54 54

What is the pattern hidden in the noise?

55 55

What is the pattern hidden in the noise?

56 56

What is the pattern hidden in the noise?

57 57

What is the pattern hidden in the noise?

58 58

What is the pattern hidden in the noise?

59 59

What is the pattern hidden in the noise?

60 60

What is the pattern hidden in the noise?

61 61

What is the pattern hidden in the noise?

It is a sine (sign)

62 62

The Landscape – Analytics and Data Science

63 63

Descriptive StatisticsBI, Dashboards, Scorecards, Reporting, Discovery, What-if

64 64

Predictive Analytics – Learning Algorithms – Forecast Techniques

• Taking historical data samples and finding patterns using learning algorithms to project what will happen in the future, or to new individuals to detect opportunities, differences, or abnormalities

65 65

Predictive Analytics

66 66

Network Analytics – Social – Information

• Measuring and creating statistics about the processes that connect individuals or technology in potentially a complex web of interactions

67 67

Network Analysis

68 68

Modeling – Simulation -- Mapping

• Developing a language, potentially mathematical, whose characteristics and relationships have close analogies and structure to something in the real world.

69 69

Modeling

70 70

Optimization

• The process were the best or near best of potentially an infinite number of options are chosen or found relative to a goal

71 71

Optimization

• Network Optimization

• Predictive Modeling

• Predictive Optimization

72 72

Advance Analytics – Any of the above can be combined

73 73

D3PO combines all four advanced analytic methods

74 74

Big data allows us to more carefully follow scientific empirical methods honed over 200 years to find the truth

75 75

More next time about how our LA’s implicit assumptions fail us and how Big Data can help to do it right to get valuable results