data (science) . · pdf filedata science is outgrowing its adolescence roi customer...

14
DATA (SCIENCE) GOVERNANCE. Bart Hamers Dexia Co-founder BDSC

Upload: phamnhi

Post on 06-Mar-2018

215 views

Category:

Documents


3 download

TRANSCRIPT

Page 1: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

DATA (SCIENCE) GOVERNANCE.

Bart Hamers Dexia Co-founder BDSC

Page 2: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

DATA SCIENCE IS OMNIPRESENT

Marketing • Customer

segmentation • LTV • Cross & upselling • Churn

Risk Management • Credit Risk • Market Risk • Operational Risk

Markets • Pricing • Trading • High Frequency

Trading

Security & Fraud • Intrusion detection • Anti Money

Laundering • Rogue Trading

HR analytic • Employee churn • Burn-out prevention • 360° evaluations • Career planning

Page 3: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

DATA SCIENCE IS OUTGROWING ITS ADOLESCENCE

ROI

Customer satisfaction

Efficiency gains

Better insights

Regulation • BCBS 239, Solvency 2, Basel

3, MQR, …

Privacy

Security

Risks

Data Vision & Strategy

Knowledge

Wh

y w

e a

re d

oin

g it

? W

ha

t is ho

ldin

g u

s ba

ck?

The regulatory text influence all aspects of data science modeling.

Page 4: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

HOW SHOULD WE DEAL WITH THIS?

•  The results of all data science initiatives produce new information and data.

•  All ‘traditional’ principles of data quality management and data governance remain applicable.

DEFINE YOUR DATA SCIENCE GOVERNANCE

Page 5: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

1.  Data (science) should focus on the end-user’s needs.

2.  Data (science) should be well managed, it should be transparent who has the authority to create, modify, delete, use and control the data science initiatives.

3.  The (data) science results should be trustworthy.

4.  All data (science) should be easily available for the end-users

5.  Data (science) should be fit-for-purpose.

6.  Data (science) initiatives should be globally managed in order to be lean, agile and forward looking.

MY 6 PRINCIPLES OF DATA (SCIENCE) GOVERNANCE

Page 6: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

MY 6 PRINCIPLES OF DATA (SCIENCE) GOVERNANCE

1. Data science initiatives should focus on the end-user’s needs.

•  What is the business problem we are trying to solve? •  Will the data science solution provide a measurable

improvement and how will this be evaluated?

Page 7: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

MY 6 PRINCIPLES OF DATA (SCIENCE) GOVERNANCE

2. Data science should be well managed, it should be transparent who has the authority to create, modify, delete, use and control the data science initiatives.

•  Apply data governance principles to data science in order to create policies and install trust. •  Ownership, stewardship, end-users,… •  Ownership is at business side!

•  Write guidelines about who and how the data science results can be used without constraining the usage.

Page 8: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

MY 6 PRINCIPLES OF DATA (SCIENCE) GOVERNANCE

3. The the results of data science should be trustworthy.

•  Guarantee the data quality used by the models. •  More (big) data is not a solution for bad quality data.

•  Test and backtest the result of your model frequently. •  Test your results on accuracy but also precision and

stability. •  Test both quantitatively and qualitatively. •  Take into account the time dimension and expiration date

of the results. •  Use the know data quality metrics

Page 9: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

PRINCIPLES OF DATA (SCIENCE) QUALITY?

Recency

Volatility Timeliness

Inter-relational

Time

Intra-relational

Co

nsis

ten

cy

q  Time: the time dimension of the data science q  Volatility: characterizes the frequency with which

data vary in time and models need to be refreshed.

q  Timeliness: expresses how current the models are for the task at hand

q  Recency: how promptly are DS results updated. (outdated information)

q  Accuracy: the closeness between real-life phenomena and its representation

q  Validity : the semantic meaning of the data science results. Are the results following the business logic

q  Comprehensiveness: ability of the user to interpret correctly the data science results

q  Metadata: Is there formal description of the data science wrt technical, operational and business information.

q  Can the data science results easy be understood by non-technical users.

q  Consistency: Captures inconsistencies between similar data attributes in data

q  Inter-relational: captures of the violation or conflicting opinions of the data science results on the same data

q  Intra-relational: captures of the risk of conflicting results of different models on different data.

q  Completeness: degree to which concepts are not missing

q  Risk related to the violation of the model perimeter.

q  Operational Risk : Is the data secured in terms of human and IT errors?

q  Human aspects: ad hoc human manipulation, unfollowed regulations and hierarchical access levels

q  IT aspects: unrealistic implementation

Page 10: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

MY 6 PRINCIPLES OF DATA (SCIENCE) GOVERNANCE

4. All data science results should be easily available for the end-users

•  Data science you not be something magical for the happy few.

•  A data driven company is only created by sharing the data results at all levels of the company. •  Marketing predictions •  Sales predictions •  Risk and finance forecasting •  Business process optimization.

Page 11: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

MY 6 PRINCIPLES OF DATA (SCIENCE) GOVERNANCE

5. Data science should fit-for-purpose.

•  Never forget Occam’s razor!

•  Be aware of the risk of over-fitting!

•  Start simple, gradually improve complexity.

Page 12: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

MY 6 PRINCIPLES OF DATA (SCIENCE) GOVERNANCE

6. All data science initiatives should be globally managed in order to be lean, agile and forward looking.

•  Do not create data science silos. •  Share your experience, systems, methodologies and

data. •  Create data sandboxes. •  Define a forward looking data strategy linked to your

business plan. (data is not collected overnight.)

Page 13: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

THE 7TH PRINCIPLE…

7. Data science is an experimental science.

•  Follow the rules •  Formulate a clear hypothesis •  Setup an experiment with a control group. •  Test based on data. •  Evaluate and if necessary refine your hypothesis.

•  It is a company culture. •  Introduce controlled and planned experiments

throughout all aspects of your business not (only data science)

•  Accept failure, but learn from it.

Page 14: DATA (SCIENCE)  . · PDF fileDATA SCIENCE IS OUTGROWING ITS ADOLESCENCE ROI Customer satisfaction Efficiency gains Better insights Regulation • BCBS 239, Solvency 2, Basel

DATA (SCIENCE) GOVERNANCE

Bart Hamers,

Dexia

Co-founder Brussels Data Science Community

LinkedIn: be.linkedin.com/in/hamersbart

Twitter: @BartHamers

Disclaimer: The views and opinions expressed in this article are those of the author and do not necessarily reflect the official policy or position of Dexia.