Another installment and iteration of my talk on predictive applications, automated decision making and why cognitive biases prevent us from making the best decisions at scale
4. — Holger Kisker, Forrester Research
“Even after more than 20 years of
using BI, they still base nearly 45%
of business decisions on
qualitative decision factors
instead of quantitative, fact-based
evidence.“
20. • Drill-down analysis … misunderstood or
distorted
• Metrics dashboards … contradictory and
confusing
• Monthly reports … ignored after two
iterations
• In-house analyst teams … overworked
and powerless
How Data-Driven Decisions
REALLY work
CO M M U N I C AT I O N
B R E A K D O W N
32. • Business rules are like programs – written by
non-programmers
• Business rules can be contradictory,
incomplete, and complex beyond
comprehension
• Business rules have no built-in feedback
mechanism:“It is the rule, because it is the rule”
Business rules are Programs,
just not very good ones.
33. — Mark Twain
“It ain’t what we don’t know
that causes trouble, it’s what we
know for sure that just ain’t so”
46. — Daniel Kahneman
“All of us would be better
investors if we just made fewer
decisions.”
47.
48. How we are making decisions
(Like the big apes we are)
Anchoring effect
IKEA effect
Confirmation bias
Bandwagon effect
Substitution
Availability heuristic
Texas Sharpshooter Fallacy
Rhyme as reason effect
Over-justification effect
Zero-risk bias
Framing effect
Illusory correlation
Sunk cost fallacy
Overconfidence
Outcome bias
Inattentional Blindness
Benjamin Franklin effect
Hindsight bias
Gambler’s fallacy
Anecdotal evidence
Negativity bias
Loss aversion
Backfire effect
49.
50. • Abraham Lincoln and John F. Kennedy were both
presidents of the United States, elected 100 years
apart.
• Both were shot and killed by assassins who were
known by three names with 15 letters, John Wilkes
Booth and Lee Harvey Oswald, and neither killer
would make it to trial.
• Lincoln had a secretary named Kennedy, and
Kennedy had a secretary named Lincoln.
• They were both killed on a Friday while sitting
next to their wives, Lincoln in the Ford Theater,
Kennedy in a Lincoln made by Ford.
51. K-Means Clustering
Naive Bayes
Support Vector Machines
Affinity Propagation
Least Angle Regression
Nearest Neighbors
Decision Trees
Markov Chain Monte Carlo
Spectral clustering
Restricted Bolzmann Machines
Logistic Regression
Computers making decisions
(cold, fast, cheap, rational)
52. • A machine learning algorithm is a system that
derives a set of rules based on a set of data
• It is based on systematic observation, double-
checking and cross-validation
• There is no magic, just data – and without data
there is no magic either
Machine Learning means
Programs that write Programs
56. — Randall Munroe
“Correlation doesn’t imply
causation, but it does waggle its
eyebrows suggestively and
gesture furtively while
mouthing‘look over there’”
57. — Warren Buffett
“I checked the actuarial tables,
and the lowest death rate is
among six-year-olds, so I
decided to eat like a six-year-
old.”
58. More than half of the apps on
a typical iPhone home screen
are predictive applications.
60. 1
Categorizing Analytics
Descriptive
• Focused on gathering and
collecting data
• Key challenges: data volume
and data variety
• Key outcome: hindsight
• Examples: reports, dashboards
• Answers “What happened?”
Predictive
• Focused on understanding
and explaining data
• Key challenges: data velocity
and complexity
• Key outcome: insight
• Examples: prediction models
• Answers: “Why did it happen
and what will happen next?”
Prescriptive
• Focused on anticipating and
recommending action
• Key challenges: execution
• Key outcome: foresight
• Examples: decision support,
predictive apps
• Answers: “What should we do?”
2 3
61. A
Categorizing Analytics
Explicit
• Analytics are a key visible
feature of the program
• Programs are used by trained
analysts and data scientists
• Regular interaction during
business hours
Integrated
• Analytics are included in
another program
• Analytics are consumed in-
context by business users
• Frequent, but irregular
consumption during business
hours
Automated
• Analytics are invisibly part of a
complex process
• Decisions are made and
executed in the process
• Constant and ongoing
optimization 24/7
B C
62. Analytic Application Matrix
2
3
B
C
+
+
=
=
Predictive Integrated
AutomatedPrescriptive
Decision Support
systems for infrequent
strategic decision-
making
Predictive Applications
for massive, automated
decision-making in
operational processes
82. If you ordered 8,5 cases, you
would waste a lot of meat,
the ideal order amount is 4
cases.
83. Predictive Apps in a Nutshell
Batch and streaming data ingestion, batch
and streaming delivery (with real-time option)
Reduce risk and cost » increase revenue and profit
Trend Estimation Classification Event Prediction
Optimize Returns
Collect Data Predict Results Drive Decisions
84. — John Maynard Keynes
“When my information
changes, I alter my conclusions.
What do you do, sir?”
85. One Common Platform for
Predictive Applications
Your own and third-
party data, easily
integrated via API
Link
Build Machine
Learning and
application code
Build
Automatically run
and scale ML models
and applications
Run
Monitor and inspect
resource usage and
model quality
View
Your data stored in
high-performance
database as a service
Store
86. — Kevin Kelly
“The business plans of the next
10,000 startups are easy to
forecast: Take X and add AI”