data warehousing & business intelligence at bmw financial services where we are & how we got...
TRANSCRIPT
Data Warehousing & Business Intelligence at BMW Financial Services
Where We Are & How We Got Here
October 22, 2008 pg 2
Presentation Goals Share the what’s, how’s, and why’s of DW
and BI at BMWFS Give others an opportunity to learn from our
experience Hear your ideas and opinions Share, share, share
October 22, 2008 pg 3
Topics Company Overview Business Drivers Delivery Process Architecture, Infrastructure Data Modeling Challenges Tools Data Governance & Data Quality Organization Strategy
October 22, 2008 pg 4
- BMW FS Overview -
October 22, 2008 pg 5
Company Overview About BMW Group Financial Services
Established in the U.S. in ‘93 to support sales and marketing efforts of BMW of North America.
Offer wide range of leasing, retail and commercial financing and banking products tailored to meet the needs of BMW customers
Offer financing to BMW dealers to expand dealership capabilities and enhance operations.
Expanded into others markets and countries. Continue to evolve beyond a captive finance unit. For example we offer finance products for non-BMW customers and P2P.
3 locations (OH, NJ, UT) 1,000 headcount (associates + contract/temp) 700,000 active accounts. Halloween is a very big deal
DW/BI program started in 2003
October 22, 2008 pg 6
Our BusinessProvide attractive financing products to dealers and
customers: Help the Sales company move cars off the lots Generate profit and revenue for Financial Services
Key Measures Customer sat Residuals Bookings Profitability Delinquency Penetration
October 22, 2008 pg 7
What We Have in Common Many disparate data sources Rapidly changing business needs Impact from current economic conditions IT isn’t nimble enough (business perception) Some shadow IT in the business
October 22, 2008 pg 8
What Makes BMW FS Unique Deeply entrenched static reporting paradigm Business on it’s own when it came to reporting Data wasn’t being leveraged to its fullest, but business results
were still healthy and strong Strict technology blueprint (Microsoft, Intel) Tactical funding model. We are not a “build it and they will
come” organization. Rigorous release process Majority of IT resources are contract / consultant. Transitioning from a nimble medium-sized company to a less
flexible, large corporation.
October 22, 2008 pg 9
- Business Drivers -
October 22, 2008 pg 10
Problems We Set Out To Solve (’03) Give our business access to the data it needs for analysis &
non-operational reporting. Data must be reliable, integrated, historical, 1-day latent. Deliver & support appropriate BI tools. Collect, maintain, and deliver critical business metadata. Approval to help the business when they have questions. Demonstrate how the right data, delivered at the right time, to
the right people, having the right data analysis skills, can significantly move the needle on business results.
Business is used to approaching IT with a solution in mind. Create a culture where the “what’s” come before the “how’s”.
Deliver value every step of the way.
October 22, 2008 pg 11
What We’ve Achieved (’08) Business Processes We Support
Financial Reporting Account Profitability
Operations Collections Lease End
Sales & Marketing Vehicle Logistics Dealer Bonus Program Front Office (used by Sales force
and Dealer channel) Customer Retention / Loyalty
Risk Credit Risk Residual Risk
200+ total users
Services We Provide Answer questions and resolve
issues with supported query, analysis, & reporting tools
help the business find data in EDW and Bengal (operational reporting database)
Recommend best tool(s) for analysis & reporting needs
Optimization & tuning Share tool and data analysis
tips, techniques, best practices
October 22, 2008 pg 12
- Delivery -
October 22, 2008 pg 13
Delivery ProcessProjects
Initial request w/high level
scope
Initial IT estimate
Business go / no-go decision
Final IT estimate
Business go / no-go decision
Maintenance
IT reserves capacity, project scheduled for
future release
Ticket w/ high level
scope
IT Estimate Business owners prioritize tickets for
next release
IT reserves capacity, tickets scheduled for a
future release
• Heavy emphasis on delivering projects on time and on schedule• IT capacity and budgets very transparent• Lot of process around managing capacity (supply) and business requests (demand)• If demand > supply, business may need to prioritize• IT capacity takes into consideration production support + admin / overhead• DW/BI team delivers ~30 projects and ~100 maintenance requests / year.
October 22, 2008 pg 14
SDLC Have one, but it only addressed
transaction processing solutions.
DW/BI team defined process & collateral for data solutions. Matured over time. Integrated with core SDLC in ’07. Analysis & Reporting
Requirements ETL, BI, Data Model Design Estimating Model Business Rule Validation, Source
Data Quality Verification
October 22, 2008 pg 15
Releases• All platforms follow a common
release schedule. Very efficient.
• DW/BI platform tied to these dates for the simple reason that we have to react to changes in our sources.
• However it’s not easy to knit iterative BI development into this schedule.Key users may not be available at the right times
Harder to deploy changes off cycle
Harder to manage DW/BI capacity & budget
October 22, 2008 pg 16
Testing1. Unit
Responsibility of each developer Verify individual code components using low volume, sample data
2A. System Full volume production data Test cases with expected results. Verified by Build Team and BA’s. Also used for performance testing
2B. Regression Full volume production data loaded separately through old and new ETL process SQL used to concatenate columns & mechanically compare row images Focus on a few high risk DW fact tables each release High effort to build regression scripts but very re-usable & efficience
3. UAT Full volume production data Test cases with expected results. Verified by the business. Goal is 0 defects, we often come close
DW/BI platform goes beyond the IS standard for testing. Higher up-front effort but worth the effort.
October 22, 2008 pg 17
Business Requirements Super critical...but it’s hard to find best
practices. Feels like an area of opportunity for the data management profession.
Detailed requirements don’t guarantee a successful project. But we can’t be successful without it.
Review of our approach:
BI Requirements Template
October 22, 2008 pg 18
Lessons Learned Chunk and iterate projects whenever possible. Note:
We’re still trying to figure out the best way to marry iterative development to a fixed release schedule.
Good requirements have value. They can (and should) evolve during the delivery process, but a baseline is important.
Start with “what’s” before getting into the “how’s”. In other words define the business questions & problems before defining the solution.
Quality is key to keeping the end user’s confidence.
October 22, 2008 pg 19
- Architecture, Infrastructure -
October 22, 2008 pg 20
Where We Started (2003) Organization was luke-warm to a grand DW/BI
implementation. “Why can’t you put all the data in one big table?” Big bang approach did not fit tactical funding model
or culture. Majority of business was getting data from an un-
architected near real-time reporting database. DW/BI Team made a conscious decision:
Start small, deliver business value quickly and frequently Grow organically, but make sure every step is on the path
to an enterprise solution.
October 22, 2008 pg 21
Evolution2003: Daily reporting database to support Dealer Bonus Program.2004: Separate monthly analysis & reporting database for Risk.2005: Databases consolidated into “EDW” with first architected relational
marts for Front Office Reporting.2006: Higher value solutions ex. Customer Retention & Loyalty.
Additional relational marts deployed.2007: First enterprise launch of true BI for Front Office Analytics.
Additional relational marts, first cubes and semantic layers deployed.2008: More demand for BI. Also significant focus on hardware upgrade.2009: Commitment for mission critical BI initiatives (ex. Collections /
Delinquency Analysis, Pricing Analytics).
Incr
easi
ng b
usin
ess
valu
e
October 22, 2008 pg 22
Design Basics Data Warehouse
Not star schema, closer to 3NF Snapshot history Updated nightly
Data Marts Relational marts are star schema, deviating if/when it
makes sense Semantic layers & cubes are also marts
Nightly batch window a major challenge No architected ODS…yet.
October 22, 2008 pg 23
CAN New Biz
IBVLE
ad hoc querying, analysis, standard reportingcontrolled access
for Infobahn, DRM, Ghostfill
CAM
D DDD DM
Extract/Transform/Load
Staging
EDB(leases, loans)
Canada(vehicle sales)
VCS(extended service)
AuctionNet(luxury
auction data)
Bank(CC apps & contracts)
NA (vehicle sales)
APPRO (credit apps)
Safeguard (gap
coverage)
Excel(custom rollups & groupings)
FIN
D
Enterprise Data Warehouse
Siebel (DRM data)
VPSO
M
Credit Risk
M
Extract/Transform/Load
FOA
D
FOA cube
LEARiskFCST
M D
Davox (Dialer data)
HSOB
HIST cube
D
So
urc
es
Wa
reh
ou
se L
aye
rM
art
La
yer
Acc
ess
La
yer
SL SLSL
Current DW Architecture
October 22, 2008 pg 24
IB
ad hoc querying, analysis, standard reporting
controlled access for Infobahn,
DRM, Ghostfill
D
Extract/Transform/Load
Staging
EDB(leases, loans)
Canada(vehicle sales)
VCS(extended service)
AuctionNet(luxury
auction data)
Bank(CC apps & contracts)
NA (vehicle sales)
APPRO (credit apps)
Safeguard (gap
coverage)
Excel(custom rollups & groupings)
Enterprise Data Warehouse
Siebel (DRM data)
Extract/Transform/Load
CSOBFCST
M D
Davox (Dialer data)
So
urc
es
Wa
reh
ou
se L
aye
rM
art
La
yer
Acc
ess
La
yer
cubes
SL’s
D
HSOB
cubes
SL’s
Target DW Architecture
Consolidate, consolidate!• Easier to answer more complex,
higher value business questions• Less dependency on IT• Less data redundancy, more
efficient• Low cost to get here
October 22, 2008 pg 25
Lessons Learned Not enough emphasis on mart usability.
Inconsistent design approaches (normalized, denormalized, star schemas, etc.)
Some structures hard to query Having a good foundation (EDW) makes it
easy to evolve & adapt the marts.
October 22, 2008 pg 26
- Data & Data Modeling Challenges -
October 22, 2008 pg 27
“The Dead Zone” Had our fair share of unpleasant data surprises. Original requirement from the business:
Need “Total FS Accounts as of PM, MTD, YTD” Need “Total NA Sales as of PM, MTD, YTD”
During testing we discovered: FS and NA have different fiscal calendars for internal
reporting & tracking. Uncovered another 5 distinct fiscal calendars Start/end dates for some fiscal periods change over time. Some measures combine metrics associated with different
fiscal periods.
October 22, 2008 pg 28
Headache Time Had to:
Figure out which measures are associated with each fiscal calendar
Design a process that tracked start/end dates for each distinct fiscal period
Allowed updates to the calendar Some measures combine metrics associated with different
fiscal periods. Formula: FS Penetration = FS Contracts / NA Sales FS Contracts are measures through the last calendar day of the month NA Sales are measured into the first week of the following month So what is FS Penetration on November 2nd?
October 22, 2008 pg 29
1st Half of Solution – Calendar Table Simple table to track
cutoff dates for the various fiscal periods.
October 22, 2008 pg 30
2nd Half of Solution – Elegant ETL Logic to stop accumulating measures used in multiple
fiscal periods, until the end of the last period.
Calendar
date
FS Fiscal period
FS contracts
October
NA Fiscal Period
NA Sales
October
Prior Month
Penetration10/31/08 October 10 October 100 [Sept’s number]
11/1/08 Nov 10 October 110 [Sept’s number]
11/2/08 Nov 10 October 120 [Sept’s number]
11/3/08 Nov 10 November 120 10/120 = 8%
FS month end is 10/31NA month end is 11/3
frozen thru 11/3
October 22, 2008 pg 31
Lessons Learned Not easy to find all the landmines via a
typical source data assessment. More detailed requirements and business rule
modeling may have caught it. Business SME was already part of the team!
At the end of the day we’re dependent on analysts asking the right questions of the business, and the business offering the right information at the right time.
October 22, 2008 pg 32
- Tools -
October 22, 2008 pg 33
BI Toolset At one time Crystal Reports was the only supported tool, hence it became
entrenched. It was the solution to every problem. When it was time to add BI capabilities, we evaluated several
products/platforms. Didn’t make sense to spend months & months on a “super” evaluation.
Vendors & technology changing too rapidly. Goal was to make an informed selection and get started, not find the
“perfect” BI platform. B.O. was a logical choice
Synergy with our Crystal platform Strengths aligned with current & near future needs Web Intelligence and Voyager have been deployed Dashboard pilot to “get smart” Universes still can’t span databases….ugh.
October 22, 2008 pg 34
OperationalReporting
DataMining
PredictiveAnalytics
StrategicReporting
Forecasting
Lev
el o
f C
om
ple
xity
Functional Needs
Com
ple
xA
vera
ge
Sim
ple
Scorecards &Dashboards
Ad HocQuerying
DimensionalAnalysis
Anticipating Demand
Where we started(pre-
DW/BI)
added viacurrentDW/BI program
Expected in the next 6-18
monthsneed to “get smart here”
Silo solutions in the biz
(unsupported)
October 22, 2008 pg 35
Other Tools / Technology Metadata
Metacenter by Data Advantage Group Currently not integrated with ETL or BI toolsets. Using it to deliver highest value business metadata.
ETL Informatica. Upgraded from v7 to v8 in Q2.
Database SQL Server 2005
Hardware Quad-core servers running Windows 2003 EE
October 22, 2008 pg 36
October 22, 2008 pg 37
Lessons Learned Generally happy with our technology choices but there is constant
pressure to drive down cost. For products licensed by CPU, there may be different interpretations of
how terms translate them to multi-core processors. Make it clear during your negotiations.
Bundled products (ex. SSIS) and open source offerings are maturing. For cost reasons we’re keeping an eye on them.
A “light” metadata implementation can be a good place to start. We have a large, unmet need for access to operational data for near real-
time analysis. Business doesn’t see value in an ODS. Looking at logical data integration tools in the Sypherlink / Altosoft category.
Don’t try to shut down silo solutions or rogue tools i.e. no empire building. They exist because they meet a need. Focus on delivering value, marketing accomplishments, and being a trusted partner. As others see value there will be less resistance.
October 22, 2008 pg 38
- Data Governance & Quality -
October 22, 2008 pg 39
Governance Not mature in this area. No formal process, but still
effective. We know who the subject matter experts are Rely on this group to define the “single version of the
truth” (business rules, definitions, etc.) No challenge we haven’t been able to resolve, easily
Gap: coordinating OLTP changes before they impact downstream systems, including EDW. No automated way to do this today. Relies on people & process. Misses occur.
October 22, 2008 pg 40
Data Quality Also not mature in this area but, again, still
effective. Most systems are new and internally developed.
Data quality is generally good. EDW has some rudimentary data quality checks.
We know more is needed. Philosophy is not to cleanse data on the way into
the EDW. If data is bad, fix the source.
October 22, 2008 pg 41
Lessons Learned Governance is important, but we see it being
critical when more areas of the enterprise are sharing data
October 22, 2008 pg 42
- Organization -
October 22, 2008 pg 43
DW/BI Team Staffing Team Structure
Team Lead (1) Architect (1) ETL Developers (4) BI Developers (1) Business Analysts (2)
Database Architect (1) Database Developer (1) Infrastructure (< 0.5) End User Access
Services (0.5)
Technical skills are important, but the key to a successful team is finding people that know when and how to collaborate.
October 22, 2008 pg 44
A Day In The Life Building, testing, implementing new data and functionality
(project & tickets) Defining, designing, estimating new requests Production support / break fix End User services and support Strategic activities (reference architecture, technology
evaluations, etc.). Not enough time for this! Project / maintenance split is about 50 / 50. Typical
distribution of maintenance activities:
Admin/Other14%
Enhancements28%
Prod Support27%
End User Services25%
Release8%
October 22, 2008 pg 45
- Strategy -
October 22, 2008 pg 46
Where To? Confident we have a solid DW/BI foundation. Gap: some parts of the business aren’t
leveraging it, or don’t see the benefit of going beyond basic reporting. Opportunity here.
It is time to help the organization mature into a data driven enterprise. Mostly organizational and political, not much technical.
October 22, 2008 pg 47
Changing Hearts & MindsLevel 1
Analytically Impaired
Organization has some interest in analytics
Top Mngmt Support
Level 2Localized Analytics
No
Yes
Functional areas tacklelocal needs
Top Mngmt Support
No
Yes
Terminal stage. Analytics not part of culture
Level 3Analytical Aspirations
Executives committed. Resources & timetable for building broad capabilities.
Level 4Analytical Company
Enterprise-wide analytic capability in development. Corporate priority for execs.
Level 5Analytical Competitor
Organization routinely reaps benefits of enterprise-wide
capability. Ongoing support & renewal.
CurrentState
To-BeState
from “Competing on Analytics: The New Science of Winning” by Thomas Davenport
October 22, 2008 pg 48
So
lutio
n
De
live
ry
(SD
LC
)
Te
ch.
Bu
sin
ess
G
oa
ls
& D
rive
rs
ML 2 ML 3 ML4
Org
. &
C
ultu
reS
erv
ice
s &
S
up
po
rt
ML 1 ML5
SDLC Exists but Does not take BI Into Account
• Proactive, Repeatable, & Reusable Process for End User Services and Training
• Shared Accountability (via SLA’s) with Vendors
• Enterprise Level Strategy (IS & Business) for Backup/Disaster Plans Implemented
• Communication extends to External FS Divisions/Partners
• Optimized Organization, Processes and Roles Measured and Refined
• Effective Metrics/Levers & Consistent Process Turn Value into Project Opportunities
• Method to Incorporate Emerging Technology
• Ongoing Measurement of SDLC Effectiveness
• SDLC Process, Tools and Methodologies Refined
Reference Architecture Does Not Exist
No Data Consistency Across Stores
Understand Major Levers & KPI’s at a Departmental Level
Defined Consistent Measures
No Dedicated Change Management Team
Resources Lack Analytical Skills
Partial Sponsorship
Limited Set of End User Services
Inconsistent Backup and Disaster Recovery Plans
Inconsistent Use of SLA’s
• SDLC Updated with BI Processes, Tools and Templates
• Best Practice Driven Model Defined
• Partial Implementation of SDLC in Projects
• Reference Architecture Established
• Best Practices to Model Data Defined
• Tools Selected and Reused
• Executive Support and Understanding of Levers & KPI’s
• Enterprise-Wide Sponsorship
• New Roles, Positions, Development Plans & Compensation Models Defined
• Communication Plan Developed
• IIG Factors Used to Forecast Projections and Ensure Data Quality
• Full Commitment to Service Offering and Training
• Consistent Use of Backup/Disaster Plans & SLA’s
• Full Implementation of SDLC in Projects
• Incorporate Appropriate Tools and Organization Support
• Reference Architecture Practiced and Enforced
• Data Consistent Across Stores
• Metrics/Levers in Place to Drive Consistent Enterprise Processes
• Change Management Team Staffed• Resources Deep in BI/Analytical Skills• Communication Plan Implemented
• Funding Allocated Based on Projections
• Strategy Developed for End User Services and Training
• Enterprise Level Strategy (IS Only) for Backup/Disaster Plans Implemented
Current State To-Be StateCapability By Maturity Level
October 22, 2008 pg 49
Other Unmet Challenges Master Data Management
Dealer Number Customer Number
ODS or other near real-time solution
October 22, 2008 pg 50
Wrap Up Questions? Thank you for listening.
A person who never made a mistake never tried anything new. - Albert Einstein
I'm sorry this presentation is so long, but I did not have time to make it shorter.- Mark Twain
October 22, 2008 pg 51
About Eric Juttner Information Delivery Team Lead (DW/BI platform) in the IS Department at
BMW FS.
DW/BI consultant and project manager at IBM Global Services / Business Consulting Services from 1996 – 2003. Also a member of the team that defined the IGS worldwide DW/BI methodology and SDLC.
Started I/T career and was introduced to data management and data warehousing at Aetna Life in Casualty in the mid-1980's.
Experiences include leading many large data warehouse / data integration / business intelligence projects in Banking and Finance, K-12 Education, Retail, State & Local Government.
Originally from Connecticut, Eric moved to Ohio in 1996 and lives in Lewis Center with his wife and 3 children.