The Value of Data
• Goal • Create awareness of the value of data • Create awareness of how to value data
• Audience • IT Managers • CFO • CEO, CxO • Consultants
Presenta=e van Henk Scholten, hscholten@bi-‐team.nl
Agenda
• Recognizing the Value of Data • Infonomics
• Informa=on & Value • What is Informa=on
• Measuring the Value of Informa=on • How to Measure Anything • Gues=ma=on • Behavorial Economics & Cogni=ve Illusions
• Recognizing the Value of Data
Question / Observation
• Is the Data present on the Balance Sheet in your organisa=on?
• Are IT Costs found on the Profit & Loss sheet?
• How does management take decisions regarding
• Safety • Opera=onal Costs • Strategic Investments
Remarkable
While collec=ng, storing, transforming and understanding data clearly costs money, the Data is not to be found as an Asset on the Balance Sheet of Companies. The change in value is not on the P&L
Infonomics
Informa<on Economics
is key in the new Business Reality
Evergreen Example: Wallmart
Walmart shared the real-‐=me POS data with suppliers to create partnerships that allowed Walmart to exert significant pressure on manufacturers to improve their produc=vity and become ever more efficient.
As Walmart’s influence grew, so did its power to nearly dictate the price, volume, delivery, packaging, and quality of many of it’s supplier’s products.
Example: Google Ads Google conquered the adver<sing world with nothing more than applied mathema<cs
It didn’t pretend to know anything about the culture and conven<ons of adver<sing – it just assumed that beOer data, with beOer analy<cal tools, would win the day
Google applied analy=cs to massive, detailed data sources to iden=fy what works without having to worry about why it worked…
• Google's popular online e-‐mail service … may not charge for its Gmail accounts. But the company is s=ll collec<ng payment in the form of massive amounts of personal informa<on about the people who use it.
• Google reported $16.86 billion in revenues for the last quarter of 2013 alone. One way it makes money from Gmail is by automa<cally scanning and indexing messages and using the data it mines to show relevant ads to its users.
• "The basic premise of Gmail is, we'll give you a robust e-‐mail service and in exchange we want to display ads alongside our e-‐mail and we're scanning your e-‐mail to decide what ads are most relevant,"
h`p://edi=on.cnn.com/2014/03/31/tech/web/gmail-‐privacy-‐problems/index.html?hpt=hp_c2
• When people send and receive messages using a free e-‐mail service, they are sharing details about their interests, who their connec<ons are and what their finances look like.
• What many consumers don't consider is that companies such as Google can create a comprehensive profile of each user based on informa<on from different products such as search, maps, e-‐mail and Google+, its social network.
• All the major e-‐mail providers, including Microsog Outlook and Yahoo, benefit one way or another from offering a free service.
• "Nothing in life is free, and as a result it is important for people to understand what value they bring to a free service of any kind," said Behnam Dayanim, a partner at the law firm Paul Has=ngs LLP in Washington.
h`p://edi=on.cnn.com/2014/03/31/tech/web/gmail-‐privacy-‐problems/index.html?hpt=hp_c2
The Value of Facebook
• The value of Facebook is related to the amount of data unpaid volunteers have put into their database
• Market capital 2013-‐12-‐01: $ 115,401,470,391 • Ac=ve users: 1,110,000,000 • Value per user: $ 103,96 • Wallstreet pays $ 104,-‐ for the informa=on you have entered about your private life
How do you decide about security measures and their costs without knowing the value of the data?
Result
If we can put a value on providing informa<on • We can set a realis<c budget • We have a norm to evaluate the value when the informa<on is provided
Evidence Based Management
“You can't op<mize it un<l you can measure it”
Value of Informa=on Three reasons why informa=on can have value
1. Reduce uncertainty about decisions 2. Affect behaviour of others
– Change behaviour of others – Reduc=on in uncertainty of behaviour
3. Market value – Selling informa=on to others / Bartering / Sharing – The reduc=on in uncertainty the data offers to others
• Add to this: Discover new aspects of reality – Discover new business models
How to Measure Anything, Douglas W. Hubbard, 2nd edi=on, p. 99
Creating Value from Data
Stages of Information Provision
First • Business Data, 1980’s • Point of Sales scanner, Call Detail and Credit Card • Insight in customers changed the balance of power between producers and retailers of consumer packaged goods (CPG)
• Wallmart and Tesco became more important than Procter & Gamble and Unilever
Second • Personal Business Data, late 1990’s • Web click data made webshops win from brick & mortar • The retailer could now manipulate the consumer on the personal level
Big Data, Understanding How Data Powers Business, Bill Schmarzo, Wiley
Stages of Information Provision
• Third: NOW • Linked Data • Social Media, Mobile Apps, Machine and Sensor-‐based Data
• Customer interests, passions, affilia=ons and associa=ons
• Op=mize customer engagement • Shape the products and services • Rewire the value crea=on processes
• More to come: Internet of Things, DNA, wearable compu=ng, facial & expression recogni=on, behavior in virtual reality, etc.
Big Data, Understanding How Data Powers Business, Bill Schmarzo, Wiley
Arbitrary
• The value of data may seem arbitrary
• But the value of everything is arbitrary
• What is the value of Buildings?
• What is the value of Machinery?
• What is the value of Brands?
• Yet this is all to be found on the balance sheet
• Risk, Deciding and Informa=on
Value of Informa=on
Three reasons why informa=on can have value
1. Market value 2. Affect behaviour of others 3. Reduce uncertainty about decisions
Source: How to Measure Anything, Douglas W. Hubbard
Value of Informa=on in the context of decision support
The existence of risk and the desire to reduce it define a possible value of informa=on
Source: How to Measure Anything, Douglas W. Hubbard
Risk
• Risk is the probability (uncertainty) of loss • Deciding is about reducing uncertainty and risk • Informa<on is about reducing uncertainty
• Measuring/Infoma<on leads to beOer decisions that lead to risk reduc<on
• The value of informa<on is the value of the risk reduc<on
Source: How to Measure Anything, Douglas W. Hubbard
Daniel Bernoulli Groningen, 8 februari [O.S. 29 januari] 1700 – Bazel, 17 maart 1782
• Daniel Bernoulli was born in Groningen, in the Netherlands, into a family of dis<nguished mathema<cians. The Bernoulli family came originally from Antwerp, at that <me in the Spanish Netherlands, but emigrated to escape the Spanish persecu<on of the Huguenots. Aber a brief period in Frankfurt the family moved to Basel, in Switzerland.
• Daniel Bernoulli published in 1738 of Specimen theoriae novae de mensura sor<s (Exposi<on of a New Theory on the Measurement of Risk), in which the St. Petersburg
paradox was the base of the economic theory of risk aversion, risk premium and u<lity.
• One of the earliest aOempts to analyze a sta<s<cal problem involving censored data was Bernoulli's 1766 analysis of smallpox morbidity and mortality data to demonstrate the efficacy of vaccina<on
Wikipedia
Valua=on of Risk
h`p://www.ted.com/talks/dan_gilbert_researches_happiness.html
Uncertainty and Core U=lity Theory
h`p://www.ted.com/talks/dan_gilbert_researches_happiness.html
Uncertainty
Possibili<es with each a Probability
Informa<on
Win Loose Win Loose
The Probabili<es of Possibili<es Value of Risk Reduc<on => Value of Informa<on
Informa<on
Dis<nc<on that Reduces Uncertainty
Zero Informa<on
Prior Knowledge
Informa<on consists of Prior Knowledge of the Possibili<es and their Probabili<es. Informa<on influences the expecta<on of the outcome
Consider two possibili<es, 1 bit, e.g. outcome of a match: -‐ Leb wins
-‐ probability without prior knowledge: 50% -‐ probability with prior knowledge: 99,999%
-‐ Right wins -‐ probability without prior knowledge: 50% -‐ probability with prior knowledge: 0,001%
• If the possibility that neither will win is also allowed in the game, then we need two bits to code the outcome.
Win Loose
Probability is expecta<on founded on par<al knowledge George Boole
Classical randomness is superficial
• You need randomness, some uncertainty that something will happen, to let you describe what you want to describe. Once you have a probability that something might happen, then you can define informa<on. And it's the same informa=on in physics, in thermodynamics, in economics.
www.theguardian.com/science/2010/mar/07/vlatko-‐vedral-‐interview-‐aleks-‐krotoski www.theguardian.com/science/video/2010/mar/05/bright-‐idea-‐vlatko-‐vederal
Decoding Reality • Vedral examines informa=on theory and proposes informa<on as the most fundamental building block of reality. He argues what a useful framework this is for viewing all natural and physical phenomena.
• In building out this framework the books touches upon the origin of informa=on, the idea of entropy, the roots of this thinking in thermodynamics, the replica=on of DNA, development of social networks, quantum behaviour at the micro and macro level, and the very role of indeterminism in the universe.
• The book finishes by considering the answer to the ul=mate ques=on: where did all of the informa<on in the Universe come from? The ideas address concepts related to the nature of par<cles, <me, determinism, and of reality itself.
Decoding Reality -‐ the universe as quantum informa<on, Vlatko Vedral, Oxford Univ. Press, 2010
Shannon’s Measure of Informa=on
• The sum of the probabili=es of all possibili=es
• Logarithmic nature – the effect of rela=ve influence
Think of Bernouilli’s Theory on the Measurement of Risk
Shannon’s Measure of Informa=on Shannon’s Measure of Informa=on (SMI) is a complex looking straighxorward calcula=on that looks a lot like Bernouilli’s calcula=on of the value of risk
Shannon
Bernouilli
Only using 0’s and 1’s
Dis<nguishability
Dis=nc=ons drive Decisons
• We care to decide, it ma`ers what we decide, because we make a dis=nc=on in value of situa=ons
• Dis5nc5on in Value: – Difference in Apprecia5on of Value – Axiology
– The Essence of Informa5on
Birth of Informa=on
1 -‐ Randomness
2 -‐ Value
3 -‐ Dis<nc<on
4 -‐ Informa<on
Conclusion on Informa=on
• Informa=on is rooted in probability (randomness) and perceived dis=nc=on in value
• The informa=on content of a message (data) is equal to the reduc=on in uncertainty (Shannon)
• The state of the receiver defines whether data is informa=on
What is the Value of the Informa<on provided by Traffic Lights? Just by suppor<ng the right choice, based on the expecta<on of outcome that ar<fically is produced
Informa=on on Informa=on
• The Informa<on, James Gleick
• Decoding Reality, Vlatko Vedral
How to Measure the Value of Informa=on
Value of Informa=on
Reduce uncertainty about decisions
• Deciding “wrong” means that you would have decided different, given some informa=on you did not have
• The cost of being wrong is the difference between the choice taken and the op=mal choice given the extra informa=on
Source: How to Measure Anything, Douglas W. Hubbard
Perfect Informa=on
• No source of informa<on can be worth more than the value of perfect informa<on
• The value of perfect informa=on is rela=vely easy to calculate, so start by calcula=ng this
• Next calculate the value of the best alterna=ve without perfect informa=on
• The difference is the highest possible value of the relevant informa=on
Visualisa=on of Value of Perfect Informa=on
Decision Tree Sogware
How To Measure Anything
2007
2010 Third Edi=on 2014
Probably the Most Important Knowledge for Business Intelligence Prac<<oners Specific aOen<on for the Value of Informa<on
Clarifica=on Chain
1) If we care about something we can observe it. We care only about (direct or indirectly) detectable things Ask: What are the (indirectly) specific detectables of the object we care about.
2) If something is detectable, it can be detected as an amount, or range of possible amounts. It can be expressed as more or less of something.
3) If the quality of the object can be detected as a range of possible amounts, it can be measured.
If you don’t know what to measure, measure anyway. You will learn what to measure (David Moore, 1998 president American Sta=s=cal Associa=on)
How to Measure Anything, Douglas W. Hubbard
Anything can be measured
• If a thing can be observed in any way at all, it lends itself to some type of measurement method
• No maOer how “fuzzy” the measurement is, it’s s<ll a measurement if it tells you more than you knew before.
• Those very things most likely to be seen as immeasurable are, virtually always, solved by rela<vely simple measurement methods
How to Measure Anything, Douglas W. Hubbard
Universal Approach to Measurement
• Define a decision problem and the relevant uncertain<es
• Determine what you know now – Describe the current uncertainty
• Compute the value of addi<onal informa<on – What is the value of reducing risk in the decision
• Apply the relevant measurement instrument(s) to high-‐value measurements
• Make a decision and act on it
Source: How to Measure Anything, Douglas W. Hubbard
Measuring and Shannon
Measurement is a type of informa<on
Shannon`s law is applicable: Informa<on is reduc<on of uncertainty (entropy)
The numeric result of a measurement is an expression of the uncertainty and its reduc<on. The measured quality itself does not have to be expressed as a number
Decisions depend on reduc<on of uncertainty
How to Measure Anything, Douglas W. Hubbard
Applied Informa=on Economics
The AIE approach addresses four things: • 1. How to model a current state of uncertainty
• 2. How to compute what else should be measured
• 3. How to measure those things in a way that is economically jus=fied
• 4. How to make a decision
How to Measure Anything, Douglas W. Hubbard
How to Measure Anything – First Edi=on, Douglas W. Hubbard
How to Measure Anything – Second Edi=on, Douglas W. Hubbard
Value of Real Informa=on • EOL -‐ Expected Opportunity Loss : Chance of being wrong 5mes Cost of being wrong • EVPI -‐ Expected Value of Perfect Informa<on is the EOL before measurement of the choosen alterna=ve
• EVI -‐ Expected Value of Informa<on is EVPI =mes the expected uncertainty reduc=on
Introductory Example: -‐ Decide on an ad campaign that will cost € 5 and can bring € 40 (add zeros to taste)
-‐ Calibrated experts put a 40% chance of failure on the campaign
-‐ EOL when approved = cost x chance = 40% x € 5 = € 2
-‐ EOL when rejected = cost x chance = 60% x € 40= € 24
-‐ The default (without measurement) decision is to approve, so the value of perfect informa=on is € 2
How to Measure Anything, Douglas W. Hubbard
Epiphany Equa=on
Realis<c; not a choice between total success or total failure • Decide about the best and worst bounds of the 90% confidence interval
– Best Bound (BB) is best possible outcome (high for income, low for costs) – Worst Bound (WB) is worst possible result (low for income, high for costs)
• Get (calculate) the Treshold, the break even outcome, or neutral result • Calculate: Rela<ve Treshold = (Treshold – WB) / (BB – WB) • Use the nomogram the find the Expected Opportunity Loss Factor (EOLF) • Compute the Expected Value of Perfect Informa<on (EVPI=maximal value)
EVPI = EOLF / 1000 * OL per unit x (BB – WB)
Worst Bound Best Bound
Treshold
B A = Conf. Interval 90%
Rela<ve Treshold = B / A
How to Measure Anything, Douglas W. Hubbard
Expected Opportunity Loss Factor Chart
How to Measure Anything, Douglas W. Hubbard
Par=al Uncertainty Reduc=on
• In reality we will not be able, or it will not be feasable, to totally eliminate uncertainty • For real applica=ons the concept of Expected Cost of Informa<on (ECI) is added • This is harder to calculate, the chart show some simple rules of thumb
How to Measure Anything, Douglas W. Hubbard
Measuring Business Values
• Most measured and reported variables in business have zero value for taking decisions
• Only a few things maOer, but maOer a lot
• A 100% CI (confidence interval) is oben – not needed for business and personal decisions – too wide as input for business decisions – too expensive to narrow down to it
• Note the Context of Observa<ons – Timestamp is mandatory – Locale – Geo loca<on
Source: How to Measure Anything, Douglas W. Hubbard
Measurement Inversion
The economic value of measuring a variable is usually inversely propor<onal to how much measurement aOen<on it usually gets Oben things that get measured don`t maOer as much as what is ignored • People measure what they know how to measure and what they believe is easy
to measure • Managers like to measure things that are more likely to produce good news • When organisa<ons are used to surveys, they may not think about other ways of
measuring. The same is true for data-‐mining, etc.
Source: How to Measure Anything, Douglas W. Hubbard
• Gues=ma=on
Gues=ma=on • Gues5ma5on; an es=mate made without using
adequate or complete informa=on
• First step in a measuring process – Impression of What to measure – Impression of How to measure – Idea of the Economic Impact & Viability to measure
• Result may be good enough for a given “treshold to decision”
• Evaluate numbers that are presented by comparing to an es<ma<on
Source: How to Measure Anything, Douglas W. Hubbard
Uncertainty
“All exact science is based on the idea of approxima<on. If a man tells you he knows a thing exactly, you know you are speaking to an inexact man” Bertrand Russell
“Measurement: A quan<ta<vely expressed reduc<on of uncertainty based on one or more observa<ons”, Douglas W. Hubbard
Source: How to Measure Anything, Douglas W. Hubbard
A`en=on
Es<ma<ons are NOT Assump<ons • An Assump<on is a statement we treat as true regardless of whether it is true
• Assump=ons are used in determinis<c accoun=ng, planning and forecas=ng
• Modelling with ranges and probabili<es does not build on statements that are `taken for true`
• Probabilis<c projec<ons lead to beOer results than determinis<c methods
• Example: Bayesian probabilis<c popula<on projec<ons for all countries, Proceedings of the Na=onal Academy of Sciences of the USA, Adrian Ragery et. al., july 5, 2012
Confidence Interval
Confidence Interval (CI) (Bayesian/Subjec=vists seman=cs is in use here) • A range that has a par<cular chance of containing the correct answer • A Confidence Interval quan<fies uncertainty, it is a measure of uncertainty • A confidence interval of 90 % is usable input for most business decisions • Overconfidence is sta=ng the interval too narrow • Forecasts that provide a single number and not a confidence interval, are
near useless. All numbers that are used as evidence should state a CI.
100% 90%
Es<ma<ng & Measuring = Defining & Narrowing the Confidence Interval
Source: How to Measure Anything, Douglas W. Hubbard
Reduce an Interval to a Number
Source: Gues=ma=on 2.0, Lawrence Weinstein
•
Techniques
• Fermi magnitude of order es=ma=on – Break down in known and/or guessable numbers
• Delphi technique – Wishdom of crowds
• Sampling rule of five – 93,75% change that the median will be between the max and min of
only 5 truely random and representa=ve observa=ons • Bayesian Sta=s=cs • Monte Carlo simula=on
– Run scenario’s based on confidence interval input – Calculate the probability of all possibili<es
Es=ma=ng
• Decompose to a calcula=on of beOer known numbers (Ask Fermi Ques<ons:) – Popula=ons, percentages, frequencies, probabili=es
• Reverse and Avoid the Anchor effect – Start with an “absurd” wide range, than eliminate values – Regard both bounds as separate values
• Get to the point that you are 95% confident of the two bounds • Iden<fy 2 Pros and Cons for the validity of the es=mate • If you seem to have no idea, widen the range <ll it touches an idea
• It is not likely we will care about a subject that has infinity as upper and lower bounds
Source: How to Measure Anything, Douglas W. Hubbard
Fermi Es=ma=on Technique
• Order of magnitude es=ma=on • Produce a quan=fied answer within iden=fied limits. • Zeroing in on the answer by iden=fying the upper and lower bounds of the probable answer
• Over-‐ and under es=ma=ons cancel out • The average of two guesses is more accurate than either guess alone. The average of more guesses is more accurate.
• The Fermi technique is taught with examples. • Most people understand the explained examples, but s=ll are not applying the technique on new ques=ons
hOp://www.na-‐businesspress.com/JABE/Jabe105/AndersonWeb.pdf
Fermi Es=ma=on Technique • Essen<al is the construc<on of an Es<ma<on Formula by determining its Factors
• Iden<fying the factors is the crux of the technique 1. Start out with an es<ma<on formula that consists of at
least two factors. 2. For some factors we can produce a numerical value within
an order of magnitude by es<ma<ng it, looking it up or because we know it. Es<mate the upper and lower bounds and reduce the interval to a number as described before.
3. Factors for which we can not produce a number have to be broken up in other factors
4. Break up factors un<l we have a formula with only factors that we can produce values for
5. Try to simplify the formula by elimina<ng factors that cancel out 6. Calculate the answer
hOp://www.na-‐businesspress.com/JABE/Jabe105/AndersonWeb.pdf
Delphi Technique • In 1907 Francis Galton published about his surprise that the crowd at a
county fair accurately guessed the weight of an ox when the mean of their individual guesses was calculated.
• In 1948 the Delphi technique was developed at the Rand Corpora<on by Olaf Helmer, Norman Dalkey, and Nicholas Rescher.
• In 1970 Barry Boehm and John Farquhar expanded it into the Wideband Delphi technique that involves greater interac<on and more communica<on between those par=cipa=ng
• The Delphi technique is a proven and reliable way to obtain an es<mate
• Experts answer ques<onnaires in two or more rounds. • Ager each round, a facilitator provides an anonymous summary of the
experts’ forecasts as well as the reasons they provided for their judgments. • Experts are encouraged to revise their earlier answers in light of the
replies of other members of their panel. • The process is stopped ager a pre-‐defined stop criterion (e.g. number of
rounds, achievement of consensus, stability of results) • The mean or median scores of the final rounds determine the results
Sampling: Rule of Five
Rule of Five: There is 93,75% change that the median will be between the max and min of only 5 random observa<ons The samples must be truely random and representa<ve
The rule of five is good for a first approxima<on
With specific methods the uncertainty can be reduced further
Source: How to Measure Anything, Douglas W. Hubbard
Monte Carlo Simula=on
• Expressing informa=on in ranges allows for real risk analyses • Adding or mul=plying different distribu=ons usually is
unsolvable, that is why a “brute force” approach is needed • Solu=on:
– Randomly generate a great amount (thousands) of probable scenario’s
– Input is confidence interval and the shape of the distribu<on
• Most used are 90% confidence interval with normal distribu=on
– Compute the outcome of each scenario – Es=mate the distribu=on of the outcome
• This type of analyses was named “Monte Carlo Simula<on” by Stanislaw Ulam. This is how Enrico Fermi and other scien=sts worked out important problems in nuclear physics.
• About a third of surveyed Monte Carlo simula<ons is run on es<mated data, almost all models use some es<mated data
Unexpected Possible Measurements
• Measuring with very small random samples can help to diminish a current great uncertainty
• Measuring the popula<on of things that you will never see all of, or es<mate undetected events
• Measure when many other, even unknown, variables are involved
• Measure the risk of rare events • Measure subjec=ve preferences and values • Measure the risk aversion, preferences and a{tudes of decision makers
Source: How to Measure Anything, Douglas W. Hubbard
Use the right tool for the job
Es=ma=ng
• Guess<ma<on, Lawrence Weinstein
• Guess<ma<on 2.0, Lawrence Weinstein • How Many Licks?, Aaron Santos
• Street Figh<ng Mathema<cs, Sanjoy Mahajan – Free download: h`p://mitpress.mit.edu/books/full_pdfs/Street-‐Figh=ng_Mathema=cs.pdf
• How to solve it, G. Poyla • Back-‐of-‐the-‐envelope physics, Clifford Schwarz
The World in Numbers
• Turning Numbers into Knowledge: Mastering the Art of Problem Solving, Jonathan G. Koomey
• What the Numbers Say: A Field Guide to Mastering Our Numerical World, Derrick Niederman and David Boyum
• Super Crunchers: How Anything Can Be Predicted, Ian Ayres
• The major stumbling block
Behavioral Economics Cogni<ve Illusions
2002 Nobel Memorial Prize in Economic Sciences
Prospect Theory, Kahneman & Tversky
h`p://mee=ngchange.wordpress.com/2013/04/03/guide-‐to-‐behavioural-‐economics/
WYSIATI -‐ “What You See Is All There Is”
Compare Cogni=ve Illusions to Visual Illusions
1. Pre-‐AOenta<ve Thinking
2. Efforted Thinking We only do it when it seems necessary to spend the energy
Visual Framing
Too much data can create un-‐necessary paOerns
Comparing Sizes
Beyond Will
Müller-‐Lyer Illusion
Knowing the visual illusion does not change the pre-‐a`enta=ve experience
Only a reference frame counters the Posi<ve and Nega<ve Bias
Calibrated Es=mates
Calibra<on is a comparison between measurements by an known reference measuring device and a device under test
Calibrated Es<mates • Few people are naturally good at es<ma<ng • 80% of people overes<mate their capacity to es<mate • Calibra<on of confidence by self-‐assesment
– Trivia tests: give 90% Confidence Interval (CI) range and give Confidence Interval for yes/no answer
– Equivalent Bet: Price on correct answer or gamble with chance to win = CI
• Learn techniques to compensate for specific es<ma<ng biases; Cogni<ve Illusions
• Training has a significant effect on ability to es<mate – 75% of people can be nearly perfectly calibrated in a half-‐day of training – Intelligent Intui<on; Repe<<on and Feedback
Source: How to Measure Anything, Douglas W. Hubbard
Calibrated Es=mates
• Learn to Handle BIAS • Learn through Repe<<on and Feedback • Place Equivalent Bets • Consider 2 reasons to be confident and 2 reasons you could be wrong
• Avoid Anchoring, think of it as a range ques<on
• Reverse anchoring by star<ng with extreme wide bounds
Source: How to Measure Anything, Douglas W. Hubbard
Bi-‐Direc=onal Processing
Mind to Eye: Propose, Direct and Reinforce
Eye to Mind: Recognize and Build PaOerns
Object Features Shape PaOerns
Pet Furry Friendly
?!
Source: Informa<on Visualisa<on, Percep<on for Design, 2nd Ed, 2004, Colin Ware
Tacit & Dileberate Systems
Educa<ng Intui<on, Robin M. Hogarth, University of Chicago Press, p 196
PCS Preconscious
Screen
S<mulus Object or Thought
“ACT”
Long Term Memory
Working Memory
Ac<on Output
Feedback: Kind or Wicked
Intelligent Intui=on and the Brain
Evidence Based Approach
VR
Intelligent Intui=on
Behaviour
Sensory Input
Error Processing
Actual Situa<on
Teaching Signals
Emo<on &
Thought Integra<on
Response Inhibi<on & Rapid Associa<ve Encoding
Conscious Thought
Expected Situa<on
Emo<on
External Events
Feedback System
Intui<on
VR The Brain is Ac<ng on a Model of the World
H. Scholten, 2013
Train your Brain
Training our Expecta<ons; our Emo<on and Intui<on
Ac=onable
Taking the right decisions (for instance on the value of data) is primarily dependent on the ability to handle cogni=ve illusions. You can learn this by training.
Some Popular Books on Cogni=on
• You are not so smart, David McRaney -‐ h`p://youarenotsosmart.com • The Invisible Gorilla: And Other Ways Our Intui<on Deceives Us, Christopher Chabris, Daniel
Simons -‐ h`p://invisiblegorilla.com • Predictably Irra<onal, Revised Intl: The Hidden Forces That Shape Our Decisions, Dan Ariely
• The Upside of Irra<onality: The Unexpected Benefits of Defying Logic at Work and at Home, Dan Ariely
• How We Decide, Jonah Lehrer (is descredited)
More References
• Educa<ng Intui<on, Robin M. Hogarth
• Rewire your Brain, John B. Arden
• Mean Genes, Tery Burnham & Jay Phelan • Cogni<ve Dissonance, fiby years of a classical theory, Joel Cooper
• On Being Certain, Robert A. Burton
• Concluding
Concluding
• The Value of Data can be Measured to Usable Precision
• Valueing Data can be Learned • Cogni=ve Illusions are the major stumbling block (Behavorial Economics)
Ac=onable
1. Recognize that the process of valueing data is important
2. Put a value on data 3. Start the conversa=on with
management and accountants
Basic References
• How to Measure Anything 2`nd ed., Douglas W. Hubbard
• Thinking, Fast and Slow, Daniel Kahneman
Think about the Value of Your Data