FREM DDM 1
Introductie to Data
Driven Management
Introduction…
- Who am I?
- Who are you all?
- What are we going to cover?
- Overview
- End-product / Brightspace
- Making decisions with data
- Why do we care?
- Approaches
- Data science – terms and terminology
- Practical exercise – survivors on the titanic
- Assignment for next week
Groei van data
Bron: [Link]
Digital Transformation
Goal:
• From ad-hoc and labour-intensive
processes → smart automated efficient
processes → cheaper, more flexible, ils
faster and data-driven et a
d
• Compliance, governance and security ore
• Sometimes we also care about a –m
better work environment & social ality
r qu
values
ighe
– h
Data science: data
• The smart part ore
M
• Automating decisions
Making Data-Driven Decisions – Data-Driven Management
Decision
Governance Data-driven
needed Decision
decision
(Problem)
Data Insights How do we get useful
Teams Data Analysis (rapport, insights out of data?
dashboards)
Data
Data-driven management - maturity
Characteristics: Data &
Level Culture & Decision-Making Example
Technology
Data scattered across silos, Decisions mainly based on Local business that
1. Ad Hoc / Starting often in Excel; low quality or intuition; data is used occasionally reviews
lack of standardization occasionally customer feedback
First tools appear (e.g.,
Managers see the value of Municipality that analyzes
2. Data-Informal / simple dashboards); some
data, but usage is satisfaction surveys but does
Experimental KPIs defined, but not widely
inconsistent not use them structurally
adopted
Reliable datasets, Data is standard in meetings
Marriott (data-driven pricing
dashboards widely and planning; decisions are
3. Data-Driven / Integrated & occupancy), [Link],
accessible, experiments evidence-based; usage still
evidence-based medicine
(A/B testing), clear KPIs somewhat siloed
Real-time integrated data, Data = DNA: every Amazon (algorithmic
predictive models widely employee uses the same recommendations),
4. Data-Transformed /
applied, self-service BI, metrics; culture of Facebook/Meta (data central
Predictive
processes for acting on experimentation and in product development),
insights learning; data lakes TPS
Good cases….
Data-Driven Project Description Strength Outcome
Netflix – Personalizes viewing Smart use of behavioral Higher customer
recommendation experience based on data and machine satisfaction and longer
algorithm watch history. learning. viewing time.
Saved millions of liters
UPS – route Optimizes delivery Efficient use of real-
of fuel and reduced
optimization (ORION) routes using big data. time and historical data.
CO₂ emissions.
Personalized playlists Combines collaborative Increased user
Spotify – Discover
based on listening filtering with content engagement and
Weekly
behavior. analysis. platform growth.
Data-driven pricing Better occupancy rates
Airbnb – dynamic Adjusts prices based
models with for hosts and higher
pricing on supply and demand.
transparency. revenue.
Bad Cases…
Data-Driven Project Description Problem Outcome
Predicted flu outbreaks Overfitting and Strongly overestimated
Google Flu Trends based on search influenced by media flu cases → misleading
behavior. hype. information.
Analyzed purchasing Unintentional disclosure
Target – pregnancy Lack of context and low
behavior to predict of private information →
prediction privacy awareness.
pregnancy. reputational damage.
Unsafe
Trained on synthetic
IBM Watson for AI system that advised recommendations →
data instead of real
Oncology cancer treatments. waste, potentially
patient data.
dangerous outcomes.
Algorithms that predict
Based on historically Increased discrimination
Predictive Policing crime and guide police
biased data. and over-policing.
surveillance.
Making Data-Driven Decisions – Data-Driven Management
Decision
Governance Data-driven
needed Decision
decision
(Problem)
Data Insights How do we get useful
Teams Data Analysis (rapport, insights out of data?
dashboards)
Data
Geschiedenis en domeinen van data gedreven beslissingen
Time / Domain Paradigm Type of Data Goal Validation / Evidence
17th–20th century – Data Analysis Explain and
Controlled, clean, Statistical significance,
Natural Sciences, (Age of demonstrate causal
experimental data confidence intervals
Medicine Science) relationships
Operational
Mid-20th century – Structured, Optimize systems and
Research Mathematically proven
Industry, Logistics, numerical, fully allocate resources
(Lean / TPS) optimum, sensitivity analysis
Finance defined data efficiently
(Industrial Age)
Discover patterns,
21st century – Data Science Large, messy, often Generalizable usefulness:
predict, and make
Business, Government, (Information secondary and does it work on new / unknown
decisions under
Digital World Age) uncertain data data?
uncertainty
Data Science
“Data science comprises the principles, processes, and techniques for
understanding phenomena via the (automated) analysis of data.”
— National Institute of Standards and Technology (NIST, Big Data Interoperability
Framework, Vol. 1, 2015)
“Data science is the study of data and how to extract meaning and insight
from it.”
— Association for Computing Machinery (ACM, Data Science Task Force Report,
2021)
“Data science is the ability to take data—to understand it, process it, extract
value from it, visualize it, and communicate it.”
— Patil, D. J., & Hammerbacher, J. (2012). Data Scientist: The Sexiest Job of the
21st Century. Harvard Business Review.
Terminologie
Making Data-Driven Decisions – Data-Driven Management
Decision
Governance Data-driven
needed Decision
decision
(Problem)
Data science gets useful
Data Insights
insights out of data…
Teams Data Science (rapport,
dashboards)
If you do it right ;)
Data
Making Data-Driven Decisions – Data-Driven Management
Decision
Governance Data-driven
needed Decision
decision
(Problem)
Data Insights
Teams Data Science (rapport,
dashboards)
Data
Data Science – How do we do it?
Split – training & test
Build the model
Training
Training / Model
Data
building
Big Data:
1k-100k+
rows
Compare and choose the How good is the
best model with new
Test Data data (vs random)?
Why?
Insights
(rapport,
Data Science - Focus on results – not method dashboards)
Making Data-Driven Decisions – Data-Driven Management
Decision
Governance Data-driven
needed Decision
decision
(Problem)
Data Insights
Teams Data Science (rapport,
dashboards)
Data
End-product
What are we going to do?
1. Pick a business problem, FREM related, that data can help with
2. Find some relevant data
3. Find patterns in the data to get insights
4. Prepare a report that communicates this back to the people with the problem to
inform their thinking
Triangular outputs:
• Story document – the report – Business problem to insights for the decision
stakeholders – inform their thinking….
• Design document – how you did it, for the nerds
• The video – walk us through it and show you did it / understand
Titanic Data – Bored of talking, lets do some – literally in the deep
end
• Built: 1909–1912 in Belfast by Harland & Wolff
• Maiden voyage: 10 April 1912, from
Southampton to New York
• Collision with iceberg: 14 April 1912 at 23:40
• Sank: 15 April 1912 around 02:20
• Passengers and crew: about 2,224 people on
board
• Survivors: approximately 710
• Ship length: 269 meters, at the time the
largest in the world
Can we find patterns that predict the survivors – is there evidence for a story in the data?
The data – the Titanic Passenger list
[Link]
Variable Meaning / Description Type of Data
Survived Survival status: 0 = no, 1 = yes. Binary / categorical
Pclass Passenger class: 1 = first class, 2 = second class, 3 = third class. Ordinal (1, 2, 3)
Age Age in years (with decimals for children, e.g., 0.83). Numeric (float)
SibSp Number of siblings and/or spouses aboard. Integer
Parch Number of parents and/or children aboard. Integer
Ticket Ticket number or code. Text
Fare Ticket price (fare paid). Numeric (float)
Cabin Cabin number(s). Text
Embarked Port of embarkation: C = Cherbourg, Q = Queenstown, S = Southampton. Categorical
Open it and look at it, what do we have?
Google Colab
Super-simple environment for doing data-science via python
Why do we use python?
1. It turns words into actions and automates the work
2. It gives us access to lots of libraries of machine learning techniques
Go to: [Link]
Sign in or create an account
Once signed in – new notbook…
End-product
What are we going to do?
1. Pick a business problem, FREM related, that data can help with
2. Find some relevant data
3. Find patterns in the data to get insights
4. Prepare a report that communicates this back to the people with the problem to
inform their thinking
Triangular outputs:
• Story document – the report – Business problem to insights for the decision
stakeholders – inform their thinking….
• Design document – how you did it, for the nerds
• The video – walk us through it and show you did it / understand
Data Science proces
CRISP-DM - CRoss-Industry Standard Process for Data Mining
- Created: end of the 90s by DaimlerChrysler,
SPSS, NCR en OHRA
- Adoptie:
[Link]
Data Gedreven Beslissingen