Download presentation
Presentation is loading. Please wait.
Published by悠跃 平 Modified over 5 years ago
1
The Science of Success: Building Faith in a Data Warehouse
Todd McDermid The Science of Success: Building Faith in a Data Warehouse
2
SQLSaturday Sponsors! Without the generosity of these sponsors, this event would not be possible! Please, stop by the vendor booths and thank them.
3
What is Success? Meeting or Exceeding Expectations
4
Expectations Who determines expectations? How is the DW measured against them? Who decides when the DW has met them? Who can change the expectations… whenever they want? Not You
5
The Good…
6
They Want to Trust You Single Version of the Truth
At best, each stakeholder only trusts their own data You’re here to help other stakeholders trust them too Analytic capability You can give them analysis they need
7
The Bad…
8
You’re an Ignorant Witch
ETL is witchcraft to business users No amount of explanation is going to be remembered Business data is foreign to you You have no idea what “looks right”
9
The Ugly…
10
How can you say your Data Warehouse is right if someone else has to tell you when it’s wrong?
You are fighting with your hands tied behind your back
11
The Common (and Wrong) Solution
12
One-Time Validation Subject-matter experts (SMEs) define success at outset. Once development is complete, SMEs validate. Problems with this: Predefined validation targets get out of date Attempts at flexible validation targets tend to be non-specific Being correct once doesn’t inspire lasting confidence
13
A Real Solution
14
Real Solution Requirements
Has to: Validate frequently Verify specific, unchanging (past) truths Balance to “current” figures Be “accessible” to SMEs
15
Additional Benefits Allows you to tell business users when the DW is wrong Improves communication between you and SMEs about the definition of success Allows for success definition to change – and makes that change measurable
16
Continuous Data Warehouse System Testing
17
Two Testing Styles Static Dynamic Compares DW to unchanging data
Provides guarantees that reporting on older data remain correct Will detect DW changes that cause differences in older data that SMEs don’t often review Will detect unanticipated changes to source systems that affect the past Tests can become outdated Compares DW to another “live” system Provides guarantees that the DW balances to systems of record for current time periods Should detect new scenarios and processes that the business did not communicate to BI team Tests can be extremely difficult or limited depending on “live” system capabilities
18
Static Testing Two parts:
Static data (CSV, …) manually compiled from any source: Not necessarily a direct export! Data Warehouse data dynamically constructed from a query
19
Static Testing Scenarios
During initial development Baseline result sets expected from requirements Over regular use Ensuring quarter-end reporting remains consistent Fix bugs using Test-Driven Development (TDD) Construct a failing test first Keep test forever to prevent regressions
20
Dynamic Testing Two parts:
Live system data dynamically resulting from a query Data Warehouse data dynamically resulting from a query
21
Dynamic Testing Scenarios
Ensure the DW balances with source systems Internal table integrity Many DWs do not define referential integrity Otherwise unenforceable integrity constraints Type 2 dimensions “have no date gaps” and/or “don’t have overlapping effective date ranges”
22
Automate Execution Static Testing Dynamic Testing DW Query
Source Query CSV DW Query Data Warehouse Export Source Data PDF Paper Napkin
23
Examples
24
Two Testing Styles (Review)
Static Dynamic Compares DW to unchanging data Provides guarantees that reporting on older data remain correct Will detect DW changes that cause differences in older data that SMEs don’t often review Will detect unanticipated changes to source systems that affect the past Tests can become outdated Compares DW to another “live” system Provides guarantees that the DW balances to systems of record for current time periods Should detect new scenarios and processes that the business did not communicate to BI team Tests can be extremely difficult or limited depending on “live” system capabilities
25
Third Testing Style: Statistical/Reasonability
What happens if the source system is wrong? Or the business process changes significantly, breaking source system reporting, and DW reporting? Compare DW data with itself: Do metrics reasonably fit within past variances, trends, and correlations?
26
SQLSaturday Sponsors! Without the generosity of these sponsors, this event would not be possible! Please, stop by the vendor booths and thank them.
27
The Science of Success: Building Faith in a Data Warehouse
Thank You! Todd McDermid The Science of Success: Building Faith in a Data Warehouse
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.