Title: Impossible Data Warehouse Situations
1Impossible Data Warehouse Situations
- Portland DAMA
- Sid Adelman
- 818 783 9634
- sidadelman_at_aol.com
2Taken From Impossible Data Warehouse Situations
with Solutions from the Experts
- Sid Adelman
- Joyce Bischoff
- Jill Dyche
- Doug Hackney
- Chuck Kelley
- Sean Ivoghli
- Dave Marco
- Larissa Moss
- Clay Rehm
3Impossible Situations
- Each organization, must wrestle with many of the
very difficult situations that have confounded
other organizations. - The same impossible situations continue to raise
their ugly heads, - You are not alone and your problems are not
unique. - Give hope to the perplexed who see no obvious
solution to their problems. -
4Impossible Situations
- Some of the situations should resonate with those
of you planning to enhance your data warehouse,
to add new data, additional users, and new
applications. - Impossible situation may not yet emerged, but you
definitely see it just around the bend. - You should be able to avoid the situation rather
than having it develop and then needing to fix
it.
5Categories of Impossible Situations
- Management Issues
- Changing requirements/objectives
- Justification/budget
- Organization and Staffing
- User issues
- Team issues
- Project planning/schedules
6Categories of Impossible Situations
- Data warehouse standards
- Tools/vendors
- Security
- Data Quality
- Integration
- Data warehouse architecture
- Performance
7Data Warehouse has a record of failure
- Telling everyone how good its going to be will
be a wasted effort. - The only thing that will convince this
organization is a successful implementation. - The project manager should deliver something of
value and deliver it quickly. - The projects business sponsor should be asked to
tout the success of the project
8Management turnover expected
- Get a backup sponsor
- Solicit a few high ranking sponsors who will be
able to support the project even if the initial
sponsor leaves. - politically powerful
- substantial interest in the success of the
project - are accepting of problems
- sponsors with a long-term and short-term
perspective of what needs to be accomplished.
9User Departments Unwilling to Share Data
- May not be able to get department managers to
share data. - Fear of criticism or that they may be
micromanaged. - Want opportunity to put spin on results
- Few organizations have real sharing.
- The CEO makes clear that everyone is to share
data and is not allowed hold anything back.
10The Operational System is Changing
- Put the development of the data warehouse on hold
until the operational source is reasonably
stable. - Look for commonality between the new and the old
operational systems. - User requirements may be somewhat different.
- The logical modeling will probably require some
changes - Many of the reports and queries will be the same.
- No need to change AA or ETL software.
11DW Objectives Misunderstood
- Develop measures of success, for example
- The data warehouse is useful
- Performance is acceptable
- The data warehouse is cost justified.
- Management is able to get more timely answers to
their questions. - The data is significantly cleaner.
12How to Demonstrate Success
- Start to measure the things that are of interest
to management. - Measure the number of queries/day run by each
analyst. - Measure timeliness (freshness).
- Measure integrity and completeness
- Report each measurement to management monthly
highlighting the differences pre and post data
warehouse implementation.
13User Productivity Justification not Allowed
- Find other areas of benefit like better decisions
for - marketing,
- distribution,
- inventory control,
- quality control,
- supply chain opportunities
- more effective customer control
- Revenue increase
- Cost decrease
14Fair Cost Allocation
- Initial sponsor is entitled to recoup some costs
and allocate part of these costs to the other
divisions. - A fair distribution of the cost could incorporate
costs based on - volume of data,
- number of users
- activity of the users
15Matrix Management?
- Sell the idea of a core team,.
- Problems of the lack of continuity when different
people were assigned - resulting loss of productivity,
- learning curve
- extensions in the schedules
- increased costs.
- Best practices - teams that have direct reports
of DBAs, data administrators, ETL jockeys,
business analysts and query tool administrators.
16Low Level of Readiness
- No technical skills,
- Staff unavailable
- Lack of motivation
- Political infighting,
- Assassins,
- CIO ready to retire and doesnt want to take
risks - The business that neither wants the data
warehouse nor has the money or the inclination to
participate in any data warehouse endeavor.
17Multiple Sponsors Want their Data Warehouse Now
- Cant satisfy all sponsors at once.
- Dont want them to run off and develop their own
data warehouse - Establish a business advisory board/steering
committee. - Representatives from the business, not from IT.
- They would make the high-profile business
decisions - They determine which applications have the most
value to the business and which should be
developed first. - Takes the heat off the data warehouse manager.
18Unrealistic User Expectations
- Set user expectations early and often for
- Schedule,
- Function,
- Performance,
- Availability,
- Data quality,
- Freshness.
19Users Dont Know What They Want
- Develop a proof-of-concept
- Compile stories about what other organizations in
this industry are doing with the data warehouse.
This will often get the users interested and
focused on specific applications and
capabilities.
20No Team Consensus
- Team members who disagree
- Delayed making decisions and moving ahead with
the project. - Total consensus not appropriate
- Someone needs to be given the authority to make
decisions. - Decisions should be made after consultation with
specific team members - Delivering a high quality product on time and
within budget is the goal.
21Consultant Will Solve All Your Problems
- Be skeptical of any such representation
- Ask the consultant how they plan to go about
fixing all your problems - Evaluate their solutions, their capabilities and
commitment to your organization. - Ask the consultant to back up their proposal with
specific and substantial written guarantees in
the event they are unable to solve all your data
warehouse problems.
22Contractors are Gone
- Documentation is poor and out of date. The data
is dirty and there are no controls for data
integrity. The users are unhappy with the
existing data warehouse. - The manager must determine if the existing system
can and should be salvaged.
23Contractors are Gone
- There are a few questions that must be answered
- Are the users using the existing system?
- Are these users dependent on the system?
- How dirty is the data and what are the problems
related to data quality? - Are there any pieces of the system that can be
salvaged? - Where are the problems, what is their impact?
- Are the problems political, perceptual, or
technical?
24Contractors are Gone
- Generate a plan of action and to create a set of
recommendations to management to either fix the
system or to discard it and start over. - If the users are not using the system and are not
dependent on it, throw the system away.
25Knowledge Transfer Not Happening
- Knowledge transfer must be in the contract with
identifiable measurements - Project plan should have included tasks for
knowledge transfer and should have allowed time
in the schedule for knowledge transfer for the
tasks where the consultant had the primary
responsibility and where the client was supposed
to learn how to perform the task.
26How to Best Use Consultants
- Consultants should be brought in to help
- build an architecture,
- establish data warehouse infrastructure
- create standards if none exist,
- recommend a data warehouse methodology,
- develop a project plan,
- develop a project agreement
- select tools.
- Make sure you have people in place to be catchers
for the skills transfer
27Outsource the Data Warehouse?
- In an outsourced situation, contracts are very
clear about what will and what wont be included.
- In DW, users are never able to articulate all
their requirements upfront. Each new request will
require renegotiations and contract changes. - By the time these details are worked out, the
opportunity will be lost. - The data warehouse should not be outsourced!
28Management Doesnt Believe Your Estimates
- Learn from people who have implemented similar
projects - Obtain project plans from others
- Understand the skill levels of those who
performed the tasks. - Consultants can also be provide input but dont
have the same level of credibility.
29Management has Committed to an Unrealistic
Deadline
- Do you want to tell upper management the bad
news now or later? - Bring in selected contractors with specific
skills who can shorten some of the major
activities, - Move some of the major functions or source files
or user groups to subsequent phases and deliver a
subset of the originally agreed upon function.
30The Scope is Expanding
- Do not throw additional people on the project.
- Do not agree to the changes without other
concessions. - Do not say No. IT has a terrible reputation for
being unresponsive to the user community.
31An RFP will Slow us Down
- Keep the RFP as simple and short as possible.
- Only include points you will use for comparing
the vendors. This means only including
mandatory and highly desirable features. - Do not include nice to have and definitely do
not include any blue-sky requirements.
32Only 5 of the intended user set are using the DW
- The training needs to be geared to the users
level of interest and capability. - Give them pre-defined queries and reports and
teach them how to launch those queries and
reports. - Teach them about the data and sell them on their
new-found capabilities. - Evaluate the effectiveness of the training with
course evaluations both immediately after the
class and two-months after the class is
completed.
33The tools have already been chosen
- If those tools wont perform or are otherwise
unacceptable, the selection committee must reject
the decisions. - First determine if the tools will perform.
- See these tools in action at other installations.
- Determine if the tools are running the volume and
complexity of your intended workload - Determine if they are running the data volumes
you anticipate.
34Vendor Goes Over Your Head
- Have a written agreement with the vendor before
the selection process starts. - The vendor needs to know there will be no
reconsideration and if they pursue their course
of action, the vendor will be blackballed from
any further marketing at this company. - Meet with management and explain the evaluation
process, explain how and why the decision was
made.
35How to Eliminate Redundant Data
- Capture the meta data that will help to identify
the redundant data. - Need cost justification for eliminating redundant
data. The company should determine the cost of
creating, storing and maintaining redundant data.
- The cost of eliminating or at least controlling
the redundant data is high. - Use a triage approach to determine which
redundant data to eliminate and which to leave
alone.
36Dirty Data Underestimated
- Get a report card on the data quality and provide
that report card to management - missing values,
- missing values in mandatory fields,
- non-unique values in a field where the values
should be unique, - violation of business rules (e.g., a negative
number of dependents, year of birth greater than
todays date), - invalid data types (e.g., character that should
be packed decimal).
37Want to Avoid Stovepipes
- A stovepipe cannot integrate with any other
system. - No common keys to allow joining of data across
these stovepipes. - Multiple versions of the same data, violating the
idea of a single version of the truth. - Need someone in a position of authority
overseeing all of these data marts to assure the
commonality that will allow joining across the
data marts as well as a unified vision of the
corporate approach to data marts.
38Business Sponsor Wants Real-Time Updates
- Is this an operational system?
- If so, put it on its own track.
- Decision support needs point-in-time.
- Designs for decision support and operational
systems are very different.
39DW Reports Dont Match those in Operational
Systems
- Should they? Whats different?
- Data cleanliness
- Timing
- Absent or added data
- Meaning of the data
- Education of differences
40Inadequate Architecture
- Can the architecture be salvaged?
- Architecture must be able to scale.
- Database size
- Source files
- Number of concurrent users
- Complexity of queries/reports
41Develop a DW Simultaneously with Developing
Source System
- Wait until the operational system is reasonably
stable before working on ETL - Some tasks that can be accomplished in parallel
- Build the infrastructure
- Train and work with the tools
- Build prototypes with sample data
- Develop project plans
- Write project scope agreements
- Create the cost justification
- Analyze risk and plan for risk mitigation
42Data Warehouse is Growing Too Fast
- Implement a cost justification methodology.
Charge the users for the data they request. - Instruct users on the costs and problems of
frivolous requests for data. - With the users, evaluate the necessity of keeping
atomic data as well as summarized data. - Carefully consider the need to replicate data.
- Understand the need for freshness
- Allow sharing of data rather than each user
having their own copy of data warehouse data.
43Summary
- The data warehouse will have impossible
situations - They arent really impossible, just difficult
- Learn from what others have done
- Search out best practices
- Non corrburundum illegitimi