Conventional BI is Broken

 

---

Successful Business Intelligence: Secrets to Making BI a Killer App

(c) 2008 Cindi Howson

Preface

Business intelligence consistently rates at the top of companies' investment priorities. Despite its priority, businesspeople routinely complain about information overload on the one hand and the inability to get relevant data on the other. BI professionals complain about lack of executive support and business users who don't "get" business intelligence. As a technology, BU usage remains modest, with significant untapped potential.

...

My hope for this book, then, is that it is a resource for both business users and the technical experts that implement BI solutions. In order for businesspeople to exploit the value of BI, they must understand its potential. ... When BI is left only for the IT experts to champion, it can provide only limited value. The real success comes when businesspeople take action on the insights BI provides, whether to improve financial performance, provide best-in-class customer service, increase efficiencies, or to make the world a better place.

Recommended Audience

The book is recommended reading for:

• • Businesspeople who feel their company is not making the most optimal decisions or who recognize the data their company has amassed is not being exploited to its potential.
  • Executives who sponsor BI initiatives
  • BI program and project managers
  • Technology experts who are asked to design and implement any aspect of the BI solution
  • Anyone involved with a BI project that is struggling to deliver value

Contents

Chapter 1	Business Intelligence from the Business Side	1
Chapter 2	Techno Babble: Components of a Business Intelligence Architecture	21
Chapter 3	The Business Intelligence Front-End	35
Chapter 4	Measures of Success	53
Chapter 5	The LOFT Effect	71
Chapter 6	Executive Support	89
Chapter 7	D Is for Data	99
Chapter 8	The Business-IT Partnership	115
Chapter 9	Relevance	127
Chapter 10	Agile Development	139
Chapter 11	Organizing for Success	149
Chapter 12	The Right BI Tool for the Right User	165
Chapter 13	Other Secrets to Success	183
Chapter 14	The Future of Business Intelligence	199
	Appendix A: This Successful BI Survey	215
	Appendix B: Recommended Resources	227
	Notes	229
	Index	237

 

 

---

Data warehouses contain data restructured to represent business entities, more closely aligned to business people's cognitive models of those entities. The DWs also provide a secondary store of the data so that the data analysis processing loads do not adversely impact the operational systems.

Analytics platforms have real value, particularly within the traditional software development processes adopted by IT-based analytics approaches.

However, BI became seen as the whole enchilada, the one true way by which any and all of an organization's data is to be used analytically.

from the increasing volumes of undecipherable data the business's systems were generating, has suffered from a legacy of failure that would put to shame virtually every other serious human undertaking.

Surveys regularly indicate that BI programs and projects have more often than not failed to meet, even substantially, their objectives.

Data Warehouses

Data warehouses and marts were introduced to address the difficulties introduced by the various data sources.
Data from different sources can be collected, conformed, and stored
Data structured into forms that are intelligible as business entities of interest; these are both meaningful and somewhat more efficient to query.
Pre-constructed aggregations—OLAP, etc.—constructed to address common reporting needs, eliminating the cost of runtime aggregations at the cost of creating and maintaining these structures.

Data warehouses have real value, making otherwise inaccessible, unintelligible, and unanalyzable data's information content to people who can benefit from it.

Considering its pedigree it's tempting to think that analytics has evolved, matured, and become proficient at delivering on its fundamental obligation: helping people understand the information contained in the data that matters to them. The general position of analytics luminaries and vendors is that this has been and is the case, with some usual caveats about how the availability of new data sources, and analytical techniques and technologies require the adoption of new approaches and tools.

The reality is that far too often analytics has been and continues to be done poorly, especially within enterprise contexts where Business Intelligence has been the dominant, nearly universal paradigm since the 1990s. Conventional BI has consistently failed to deliver success; searching the web for "business intelligence failure" will produce many resources covering BI's problems, often with suggestions for addressing them. Yet, even though BI's shortcomings have been known for quite a long time it has continued to be disappointing.

Are conventional BI's problems so deep that they cannot be overcome? No, they are not; it's possible to conduct analytics in a way that works, and works well. But first, it's helpful to identify the problems with conventional BI that prevent it from succeeding. A review of the "business intelligence failure" resources is informative. Please take your time, we'll wait.

The resources identify numerous problems with BI, with quite a lot of overlap among them. However, and this is the critical point: there is a very strong alignment among these references to a particular paradigm which is accepted as the unquestioned, unexamined, unrecognized framework within which all things exist (that's what paradigms are). The fundamental problem with conventional BI is not that particular elements of the how it's set up and conducted are at fault, but that the paradigm itself is unsuited to the problem domain it's applied to.

In principle BI is capable of encompassing the full spectrum of data-analytical needs, within its realm of operation.
However, as BI became embraced and adopted, particularly within large organizations, it was subjected to conventions that prescribed what it was and how it should be conducted. These conventions are the root causes of much of how BI has strayed from its original promise.

Conventional BI's paradigm includes a constellation of misalignments with analytics' real goals, obligations, opportunities, and constraints. Each misalignment has its own sources including conceptual and operational biases and limitations; in combination they operate in feedback loops, reinforcing one another and resulting in systemic dysfunction.

A brief overview of conventional BI's paradigm.

Scope and scale.

Scope - what data is subject for analysis, and what analyses are appropriate, and from what sources

Scale - how analytics is organized, managed, conducted

get-started cost / cost of entry / buy-in / ante

One of these problems is the conception and practice of data analysis as a technical activity first, and as a human sense-making activity second (or worse). Another is the industrial production model of work focused on building and employing factories to ingest inputs, processes them with completely designed, managed, and operated processes that produce large numbers of similar or identical outputs. Yet another is the hierarchical business organization model that works hand-in-hand with the others to shape the presumptions of who needs to know what and what it takes to inform them.

Scale

Analytics is valuable and worth pursuing at the largest scale: addressing and answering the biggest questions for the most senior people.

There's very much a chicken-and-egg relationship between analytics' perceived purpose and value, and the forces that led to its framing in this manner.

On one hand there's the very real situation that the business systems data has historically been undecipherable to non-technical people, largely due to the combination of relational data modelling and the technically-designed tooling required to access and analyze it, for example the near-absolute dominance of SQL-query based approaches.

This led to the acceptance of the necessity for technical support, along with the expenditure of substantial time, energy, money, and other resources required to obtain information from business data.

Once the need for substantial investments was accepted as a threshold, it became simple to accept that the investments were best justified by providing answers to the biggest questions, those at the top of the executives' strategic concerns.

This fed back into the mix, increasing the demand for more and more data to be included in order to better answer the strategic questions, leading to an escalating spiral of demand for more data along with the ever-larger, more complex and expensive platforms and the staffs required to manage them.

One unfortunate consequence of enterprise analytics' conception and evolution is that the conduct of analytics is too often left up to technocrats, reinforcing the technology-centrism framing, building walls of technology and technocratic bureaucracy that impede rather than facilitate people's access to the information contained in the data that matters to them.

A common manifestation of these flaws is in the process of analytics expressed as some form of "first thing: prepare the data for analysis."

Implicit in this is the idea that analysis is an end-point activity done to make processed data intelligible to relatively passive information consumers.

This is completely the wrong way of thinking. There is no way to do anything meaningful and useful with data unless and until the data is understood, and the only way to understand it is through analyzing it. The follow-on effects of this mischaracterization are structural, pervasive, and once in place extremely difficult, but not impossible, to recover from.

Enterprise analytics' dominant paradigms are briefly described below; they are examined in detail, along with opportunities for remediating their flaws in auxiliary content. For historical reference, see A Brief History of Analytics.

Points:

Relational database design renders data undecipherable for people without the proper combination of technical and domain expertise
Data warehouses are good and valuable for their appropriate purposes, huge problems result when they are seen as the obligatory and only place from which data can be analyzed, a decision often made by technocrats
Data analysis is, or comes to be seen as, the proper domain of technical people. This is often accompanied by the perception that the first step in data analysis is the creation of a query, almost always an SQL query.
The industrial model of work, accompanied by:
BDUF
Too many hands - stovepiped activities, e.g. business analysis - domain expert - data engineer - project manager - developer - tester - user
building report factories
all work data-related gets turned over to IT, often to IT vendors offering lowest-cost workers; this leads to many of the
BI consultancies, including strategic advisory firms, reinforce the paradigm that analytics is a large scale, complex, serious undertaking that can only be approached with a full, robust enterprise-scale, industrial model that involves tremendous planning, management, and technical development efforts in order to achieve meaningful outcomes - this leads to things like project plans with hundreds or thousands of discretely enumerated tasks extending over many weeks or months; it wasn't unusual for lead times of many weeks or months to be accepted as normal before any actual information would be delivered to the business people who needed it.
Aside: IBM's BI Method was centered around a boilerplate process the heart of which was such a project plan, with very limited opportunity to adjust to the specifics of the client's needs and environment.
Predictability: business management craves predictability. Managers need to know how much money, time, energy, personnel, and other resources will be expended when. Consulting managers need to predict their revenue streams - I once worked for a large consulting company where the primary emphasis was the partner's bonus was dependent upon maximizing revenue, achieved via selling the client a large, complex project plan with tasks identified in two hour increments five months out.
Single Version of The Truth — becomes the false ideal, a mirage, the pursuit of which impedes even prohibits the analysis of valuable data unless and until all analyses can be guaranteed to align with one another.
Top-down hierarchical model of enterprise organization, particularly when coupled with the manager/worker separation of work that organizes work into design/monitor/adjust vs execution spheres.
This leads to things like:
data is most profitably used to inform decisions at the top - strategic decisions that steer the organizations
useful information and insights for executives need a comprehensive, consolidated view of the organization and its environment, leading to
the belief that all data should be incorporated into the harvested, consolidated, homogenized, pasteurized, conformed, centralized data warehouse before:
the serious, important decisions can be made
lesser-consequence information delivered to lower-level personnel can piggyback the value-add of the centralized data

Starting with real promise BI became ossified, enamoured with itself and its internal processes. Inflexible in practice, it has become the domain of technocrats who are primarily concerned with building elaborate factories to harvest, devour, store, and, in due time, expose preconstructed artifacts—reports, dashboards, scorecards, etc., that are in turn made available to people for viewing, often with some limited configurability.

 

Business Intelligence has become the vortex where data goes to die.

The failures of conventional Business Intelligence (cBI), the dominant form of analytics since the 1990s, are well documented. These failures are rooted in conventional BI's technological and environmental origins and nurtured by the blind adherence to industrial production operational and management paradigms that are not only irrelevant, but are in many ways antagonistic to effective analytics.

The visible manifestations of conventional BI's large scale industrial paradigm is its strict orientation around the construction of data management and storage platforms fronted by complex report factories where developers wielding arcane technical tools and skills create artifacts, e.g. reports, dashboards, scorecards, etc., for delivery to business stakeholders. Conventional BI treats data as raw material for its enterprise-spanning, industrial strength data harvesting machinery that collects data from source systems, processes it into homogenized forms suitable for answering any and all conceivable questions, and fronts the comprehensive data stores-warehouses, marts, etc., with similarly-scaled analytics platforms that need armies of technical experts to tease information from in the form of pre-constructed, developer-created, reports.

Conventional BI also has a huge blind spot: it fails to recognize that the great majority of people's data-based information needs can be satisfied from locally available data that neither needs nor benefits from being conformed to Enterprise context standards. These people need personal analytics.

If DWBI is good and valuable, how did conventional BI go wrong?

Two main issues are conceptual and operational:

contemplation of the data that's suitable/appropriate for inclusion in the DW
how DBI is conducted, from design to execution
industrial model of work taught to generations of business managers led to the concept that building report factories was the way to do analysis of the data; this requires large scale mobilization of resources, BDUF, and long lead times survey, harvest, model, process, store data to be analyzed, then building reports to provide the analyses identified by business analysts working with business stakeholders

The things that work well with DWBI became to be see as the right, proper, and only

In the world dominated by conventional BI, as is the case with most organizations, even when data is available, personal analytics has suffered from the unavailability of highly effective, moderately priced, human oriented, visual data analysis tools. Even worse, in the most pathological organizations people are actively discouraged, even prohibited, from effectively analyzing the data that matters and is immediately useful to them.

Our ambition is to help correct this situation by providing people with the means to effectively and efficiently distill useful, meaningful, and timely information from their data.

Retire BI, Embrace Analytics

Given BI's current state it's unrealistic to think that it can be rescued from its self-inflicted wounds. Although there's no technical reason preventing BI's rehabilitation, in practice its sheer weight and inertia, and the degree to which its current popular practices have become embedded in their psyches as the One True Way for an entire generation of BI practitioners, BI will continue lumbering forward as long as people adopt it as the default when faced with the need to help people analyze and understand the information that matters to them.

Liberate Information

full advantage of the opportunities presented by effective, efficient analysis of the information

We challenge the commonplace, nearly universal, presumption that information is synonomous with data.

Information the essential element

Facilitate Analysis

Analytics International has its roots in the time when automated business data processing was emerging out of its COBOL report-generating origins. The invention of fourth generation languages enabled nontechnical business people to create their own reports and ushered in a new age, when business users could glean for themselves the information they needed from their data.

The original analytical heuristics are lessons learned helping these people with their data analysis, and their organizations recognize and realize the benefits of offloading analysis from the technical organization.

Analystic: People benefit when they can analyze their own data.