Upload
others
View
21
Download
0
Embed Size (px)
Citation preview
Agile Analytics
January 2020
PwC
1. What is Agile Analytics 03
2. Model Development Lifecycle 05
3. Planning Analytics Projects 08
4. Organising for Agility 10
5. Summary 12
Agenda
2
January 2020Agile Analytics
PwC
Agile
adjective
1. able to move quickly and easily.
2. relating to or denoting a method of project management, used especially for software development,
that is characterized by the division of tasks into short phases of work and frequent reassessment and
adaptation of plans.
We are going to focus on Definition 1
Source Oxford English Dictionary
Agile is many things to many people
3
January 2020Agile Analytics
PwC
So how do we make Analytics quick and easy
Quick
Traditional analytics has depended on a standard model development lifecycle. This is underpinned by:
• Data scientists developing models
• Focus on developing the “Best” model (highest goodness of fit)
• Each model is designed and built to highest standard, deployed and then retrained (usually after 12 –
18 months)
Easy
Usually modelling is a specialist role. Requiring many different complex skills
• Data & feature engineering
• Extensive coding experience
• Vast knowledge of different modelling techniques
Agile Analytics January 2020
4
What if all this is wrong?
PwC
Model Development LifecycleExample – Crisp DM
Agile Analytics January 2020
5
Source DataScienceCentral.com
Traditional Approach
Typically MDL takes at least 9-12 weeks
Most of the time is spent
• Data Engineering
• Feature Engineering
• Model coding
• Packing data prep and model code
• Testing code package
Build once – Deploy once – retrain in a year
PwC
Automated Modelling tools – Rapid Modelling
Automated Modelling Tools
Automated tools are not new – many have been around nearly a decade
They build models and engineer features with little/no user intervention
As computing power becomes cheaper their abilities have increased
Models are NOT as good as a trained data scientist
But they are quick and their output code is bug free (less test time)
Time to Decision is now more important than accuracy for many, but
not all applications
Used correctly Automated tools are:
• Faster
• Cheaper
• More accurate
• Support the Citizen Data Scientist
Agile Analytics January 2020
6
Source: kdnuggets.com
PwC
Agile Analytics January 2020
7
Traditional Approach
Accuracy Effort
Jan 0.00% 20
Feb 100.00% 0
Mar 90.00% 0
Apr 81.00% 0
May 72.90% 0
Jun 65.61% 0
Jul 59.05% 0
Aug 53.14% 0
Sep 47.83% 0
Oct 43.05% 0
Nov 38.74% 0
Dec 34.87% 0
Rapid Approach
Accuracy Effort
80% 5
72% 0
65% 0
80% 5
72% 0
65% 0
80% 5
72% 0
65% 0
80% 5
72% 0
65% 0
Automated Modelling Tools Vs Traditional Modelling
Average 57.18% 72.27%
PwC
January 2020
8
How do we USE
Predictive AnalyticsDATA
ANALYSE
RESULTS DECISION
ACTION
EVALUATE
OUTCOMES
A more agile approach to modelling
PwC
January 2020
9
How do we PLAN
the use of Predictive
Analytics
DATA
ANALYSE
RESULTS DECISION
ACTION
EVALUATE
OUTCOMES
A more agile approach to modelling
PwC
September 2018
10
4 main Analytical Organisational Models
1. Fractional• In this model no central analytics function exists, each function
performs its own analysis
• Information is difficult to reconcile and often varies hugely
2. Centralised• One central unit responsible for all Analytics, requests are submitted from the
business and prioritised accordingly
3. Hybrid• A more mature version of the centralised model,
• Business users are brought in as part of the project team
• Projects are selected based on business case
4. Community• Most advanced of the models
• CoE supports departmental or SME led Analytics functions
• CoE acts as an internal consulting function, sets standards, manages vendors, sets strategy
• Ensures high standards are used throughout the organisation
If we are working differently do we need to organise differently?
PwC
September 2018
11
Fractional Centralised Hybrid Community
Structure: Fractured, each business unit has its own analytical capability, no co-ordination or standards.
Culture: Information is power, not shared, no collaboration, focus on me.
Standards: Generally only adherence to standards is due to regulatory requirements, and then only the most important ones
Information Processes: Not defined, no centralised control or repositories
People: Analytics is not a career, rather something that a few people do as part of other responsibilities
Infrastructure: Mainly PC based, high reliance on desktop tools, particularly excel. Any sharing of data or information is via email
Structure: One central function for analytics. Requests are submitted and results sent back.
Culture: Even with central function there is little collaboration, each department focuses on its area. CoE is a service centre
Standards: Adhere to all required regulations, but do not view standards as a positive
Information Processes: Sharepointpredominant, each function has its own site, no inventory is maintained
People: Analytics is a career, but only in the CoE. Specialist technical skills are hired for this specialist role, little requirement for business acumen
Infrastructure: Server based, but data is still relatively ad-hoc, little metadata or semantics
Structure: Central function, but with strong business relationships. Business SMEs join teams for projects and rotate back out
Culture: Collaboration is at the core of everything. CoE works closely with the business and understands business processes and constraints
Standards: Both regulatory and industry best practice standards used.
Information Processes: Information portals are common, some self-service reporting for simple analysis
People: Some lines of business consider data and analytical skills as part of the recruitment process, but it is still a “nice to have” skill
Infrastructure: Server and cloud based, usually a data warehouse in place, strong metadata management
Structure: CoE enables citizen data scientists in business units. CoE works as an internal consulting group
Culture: Innovation is everywhere, “fail fast” approach is typical. Information is an asset and is valued
Standards: Standards are regarded as good practice. CoE sets standards and best practice for the analytics in the organisation
Information Processes: Subject centric, searchable, commentary enabled information portals abound
People: Analytics is a career, with defined structures both in the CoE and the business. Managers are expected to have a good understanding of analytics.
Infrastructure: Mobile devices are typical. Data lake with strong, searchable semantics and lineage
Maturity LevelLow High
Analytical Maturity
PwC
Agile Analytics January 2020
12
Summary
To achieve Agile Analytics requires more than just a project delivery methodology
1. Right tools – automated tools to reduce Time to decision.
2. Rapid development approach – less initial accuracy, better results by retaining more often
3. Plan the project with a focus on business actions
4. Organise to support Fast decisions
pwc.com
Thank [email protected]
© 2020 PwC. All rights reserved. Not for further distribution without the permission of PwC. “PwC” refers to the network of member firms of PricewaterhouseCoopers
International Limited (PwCIL), or, as the context requires, individual member firms of the PwC network. Each member firm is a separate legal entity and does not act as
agent of PwCIL or any other member firm. PwCIL does not provide any services to clients. PwCIL is not responsible or liable for the acts or omissions of any of its
member firms nor can it control the exercise of their professional judgment or bind them in any way. No member firm is responsible or liable for the acts or omissions of
any other member firm nor can it control the exercise of another member firm’s professional judgment or bind another member f irm or PwCIL in any way. Please see
www.pwc.com/structure for further details.