teaching computers to think like decision makers: the next revolution in the data sciences

25
Teaching Computers to Think Like Decision Makers Mark Zangari CEO, Quantellia LLC San Francisco University May 23, 2014 [email protected] 303 717 4221 Copyright © 2014 Quantellia LLC. All Rights Reserved.

Upload: quantellia

Post on 11-Aug-2014

244 views

Category:

Data & Analytics


0 download

DESCRIPTION

“Big Data” and analytics have revolutionized "micro-decisions", those myriads of tiny decisions that follow a similar pattern, are made frequently, but each of which has relatively low risk and low value (e.g. the cross-sell to “things we might also like” that almost every e-commerce checkout page displays using our purchase history and possibly other data). By contrast, "macro-decisions" are less frequent, but higher-stakes. They are more complex and also need to take risk into account. Software support for macro decisions today is usually provided as “Business Intelligence” or “Dashboards”, both of which typically derive aggregate statistics from existing data, and present these in ways that are “meaningful” and “insightful” to humans. However, once the data has been presented, the synthesis and evaluation tasks at the core of the decision-making process are left to the human decision-maker. This is despite a large and well-accepted body of research (most notably by Kahneman and Tversky) clearly demonstrating that humans systematically lack the ability to perform such tasks accurately. A significant and as-yet untapped opportunity therefore exists for augmenting the existing BI paradigm with new data science techniques developed to assist decision makers. This presentation introduces the “Decision Intelligence” approach which transfers the decision-related inference tasks from human intelligence to machine intelligence. The approach includes a structured framework for decomposing decisions so they can be represented as computable models. Using simulation and optimization techniques, these models generate data sets to which existing BI tools can be applied, giving decision makers the ability to generate data from “possible futures” and to evaluate decision and their outcomes in familiar, existing environments. Mark is a leader in innovative research, software development and services delivery, and business development in the academic and commercial sectors for over two decades. He is co-founder and CEO of Quantellia, a leading Data Science innovator and developer of the award-winning World Modeler software. From 2000-2010, he held the position of CTO at Spatial info (now Synchronoss) where he co-founded the company’s US operations, and led technical operations. Prior to this, he was the architect of StatPlay, software developed jointly at La Trobe University and the University of Melbourne that explored how computer visualizations affect people’s innate abilities to perform statistical reasoning. Mark has also worked as a systems engineer for EDS (now HP) and Anderson Consulting (now Accenture). In 1994-5, he held a British Council Post Graduate Bursary at the University of Cambridge in the UK and from 1996-2000 was an Honorary Fellow at the University of Melbourne. He is the author of numerous publications and has frequently made speaking appearances.

TRANSCRIPT

Page 1: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Teaching Computers to Think Like Decision Makers

Mark ZangariCEO, Quantellia LLC

San Francisco UniversityMay 23, 2014

[email protected] 717 4221

Copyright © 2014 Quantellia LLC. All Rights Reserved.

Page 2: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Robert McNamara• Secretary of Defense (1961-68)• Ford Motor Co. (1946-61)• USAF “Statistical Control” (1943-46)

Page 3: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Data System Analysis

Decision

Page 4: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

http://sunsite.berkeley.edu/FindingAids/dynaweb/calher/jvac/figures/j12EB-644A.jpg

http://www.whatswrongwiththeworld.net/office-interior-1940s.jpg

http://www.biega.com/bcbphotos/biega-engineer.jpghttp://lcweb2.loc.gov/service/pnp/hec/28300/28336r.jpg

Data Acquisition…

Data Storage…

Data Mining…

Analytics…

Data

Page 5: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

http://sunsite.berkeley.edu/FindingAids/dynaweb/calher/jvac/figures/j12EB-644A.jpg

http://www.whatswrongwiththeworld.net/office-interior-1940s.jpg

http://www.biega.com/bcbphotos/biega-engineer.jpghttp://lcweb2.loc.gov/service/pnp/hec/28300/28336r.jpg

Data Acquisition… Data Mining…

Analytics…

Data

Data Storage…

Page 6: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

DataInstrumented

Code / Sensors

DataManagement

Analytics

Presentation

System Analysis

Decision

DataInstrumented

Code / Sensors

DataManagement

Analytics

Presentation

Big Data

Business

Intelligence

Demarcation between automated (computer-centric) and manual (human-centric) information processing.

Gap between computer and humanbridged by Data Visualization.

Page 7: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Units Cost Per Unit1-100 $12.00

101-500 $10.00

501-1000 $9.00

1001-10000 $7.50

10001+ $6.00

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Sale

s Vol

ume

/ M

arke

t Siz

e

Retail Price

Base Demand

0.00%

1.00%

2.00%

3.00%

4.00%

5.00%

6.00%

7.00%

8.00%

9.00%

$0$5

00$1

,000

$1,5

00$2

,000

$2,5

00$3

,000

$3,5

00$4

,000

$4,5

00$5

,000

$5,5

00$6

,000

$6,5

00$7

,000

$7,5

00$8

,000

$8,5

00$9

,000

$9,5

00$1

0,00

0$1

0,50

0$1

1,00

0$1

1,50

0$1

1,75

0

Pct.

Incr

ease

in D

eman

dMarketing Spend

Marketing Driven Demand Uplift

Manufacturing Unit Cost by Volume

The Product Manager’s Decision:

To maximize profit…a) How many units do I order from the

manufacturer?b) What retail price do I charge?c) How much of my profit do I re-invest

in marketing?

(Mkt Size = 50,000)

Page 8: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Even with all the data you need, and clear visualizations, making good

decisions is still very hard to do.Why?

Data System Analysis Decisionü

Page 9: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Because:

a) Humans are not good at runningSystems in their heads.

b) Unlike Data, there is little mainstream computerized support for modeling and analyzing Systems.

(But let’s see if we can change that…)

Page 10: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Build a Computable Systems Model Visually

• Attributes• Dependencies

The Product Manager’s Model and quickly

Page 11: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Identify Model Elements:

• Outcomes / Goals“What are we trying to achieve?”

• Levers“What can we control?”

• Externals“What affects our outcomesthat we can’t control?”

Build a Computable Systems Model Visually

Page 12: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Identify DependenciesDependencies“How are A, B and C related to X, Y and Z?”IntermediatesWhen outcomes are not directly related to levers or externals.

Build a Computable Systems Model Visually

Page 13: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Quantify DependenciesDependencies“How are A, B and C related to X, Y and Z?”

Build a Computable Systems Model Visually

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Sale

s Vol

ume

/ Mar

ket S

ize

Retail Price

Base Demand

Expressions

External Data Sources / AnalyticsSketch Graphs

Page 14: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Quantify Dependencies

Dependencies“How are A, B and C related to X, Y and Z?”

Build a Computable Systems Model Visually

Models also provide a systematic way to assess the impact of uncertainty, sensitivity, precision and risk on the decisions they support.

Page 15: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

While humans are not good at processing systems models, we are much better at analyzing and designing them. This leads to a natural human-computer partnership.

Build a Computable Systems Model Visually

Page 16: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

The Product Manager’s Decision:

a) How many units do I order from the manufacturer?

b) What retail price to I charge?c) How much of my profit do I

re-invest in marketing?

… to maximize profit?

But wait, there’s more.

38,000$15

7%

Page 17: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

The Product Manager’s Decision:

Most decisions are made not justto optimize outcomes, but to managerisk.

A bi-product of the optimization search is data that can be used to:• Assess sensitivity of the desired

outcome to particular levers and externals.

• Assess downside risk associated with each positive outcome.

Opportunity envelopeRisk envelope

Gradient shows sensitivity

Page 18: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Some Interesting Structural Characteristics of Models…Build a Computable Systems Model Visually

Feedback Loop

Page 19: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

… Lead to Important Behaviors.Equilibrium and Transient States

• Real-life systems, even if they are stable, are not static, but in a steady state or equilibrium.

• When such systems are perturbed, they oscillate, or experience a transient.

• Effective decision makers need to be able to understand the effects their decisions will have both on the transient phase and on the new equilibrium.

Build a Computable Systems Model Visually

Equilibrium with price at $12

Price raisedto $15

New equilibrium with price at $15

Transientphase

Page 20: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Data System Analysis

Decision

Big Data / Business Intelligence:

Page 21: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Data

System Analysis

Decision

Decision Intelligence

Analyze system

Build model

Integrate Data to specify dependencies

Search the space of decision leversand externals to determine

optimal outcomes and risk profiles

Gap between computer and humanbridged by Data Visualization of Decision Variables, not the Input Variables as before.

Page 22: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Decision Intelligence:

• Gives decision makers what they need most, and they cannot get from Business Intelligence: help answering the question “If I make this decision, then what will be the likely results, and what risks am I exposed to?”

• Provides a framework for the most effective use of existing data and analytics tools in a given problem.

• Provides visual and other artifacts that assure team alignment and act as a form of “institutional memory”

Page 23: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

New Kinds of Visualizations

• Familiar data visualizations still have their place in Decision Intelligence, but note that the “axes” are now more meaningful to decision makers as each represents an “actionable” quantity.

• In addition, there is a powerful role for new dynamic System Visualizations.

Page 24: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Call to Action:

Now that the “Big Data” problem is mostly solved, we need invest our talents to return to the “Big

Picture”.

We must develop software tools and methodologies that integrate data and systems to

produce the kinds of insights real users really need.

Page 25: Teaching Computers to Think Like Decision Makers: the next revolution in the data sciences

Download a free trial of World Modeler from www.quantellia.com

Mark ZangariCEO, Quantellia LLC

San Francisco UniversityMay 23, 2014

[email protected] 717 4221

Copyright © 2014 Quantellia LLC. All Rights Reserved.

Thank You.