enterprise-data-everywhere

- Bill Peer, Kiran Kumar Kaipa, Shyamala Sadananda, and Swaminathan Natarajan
Abstract
The proliferation of new analytic processing capabilities, the deafening marketing
hype of Big Data, and the radical dropping of processing power barriers have
brought focus on a problem that has long existed in the area of information
harvesting: data resides in lots of places and in lots of forms. There is no singular
solution or approach available today that allows all the information latent in an
enterprise to appear and be exploitable by all systems in the enterprise. That is,
there is no ideal way to handle the enterprise knowledge value-chain. This paper
provides an articulation of a decision framework for identifying the“best fitting
approach”for an enterprise’s“data everywhere”challenge, exploring the common
models of data foraging, data virtualization, data consolidation, and information
fabrics. The viewpoint of this paper is based on a common usage of enterprise-
wide data: Business Intelligence (BI). Within the realm of BI, this paper further
refines specific usage scenarios that many of our forward looking clients expect:
advanced analytics and self-service BI.
Knowledge value chain approaches: A decision framework
WHITE PAPER

Introduction
The proliferation of new analytic
processing capabilities, the deafening
marketing hype of Big Data, and the radical
dropping of processing power barriers
have brought focus on a problem that has
long existed in information harvesting:
data resides in lots of places and in lots
of forms. Companies that address the
problem are creating a new generation
of business applications that support
customer service, risk management,
multichannel integration, loyalty
management, regulatory compliance,
marketing, business performance
management, and other critical functions
[3, 2]. These applications require cross-
functional data, with varying semantic
meaning, to be blended in near real-time.
This has created an increased demand for
data access and integration solutions [3].
Increasing volume of data, faster data
consumption and decision-making, and
dealing with integration cost and timelines
makes data integration a complex
challenge. This challenge acquires further
complexity due to diversity of data sources
External Document © 2015 Infosys Limited
and data freshness requirements [3, 2].
Compounding things even further is the
growing diversity of the user population of
the synthesized, blended data knowledge
products. With so many dimensions to
the problem, companies are seeking
more flexible ways of making integrated
data available to constituencies, people,
applications, and systems – in ways that
control costs and complexity by keeping
customized application code to a minimum
[3, 2].
As of this writing, there is no singular
solution or approach available that
allows all the information latent in an
enterprise to appear and be exploitable
by all systems in the enterprise. There
are, however, numerous approaches to
address various aspects of the knowledge
value chain problem and they go by the
names of data lakes, information fabric [1],
data federation, data virtualization, data
warehouse augmentation, and more.
Vendors, research organizations and
the like offer varying approaches and
perspectives on“the best approach”for
blending enterprise data. Our experience
at Infosys in implementing these solutions
in a variety of contexts highlight that
while there is no panacea available, there
is a decision-making criteria that can be
followed to help an organization pick
the“best fitting”answer for a particular
scenario and need. By“best fitting”we
mean a solution that is more effective at
realizing the desired outcome under one
set of conditions than another. This paper
explores these conditions to help our
clients make the best choice.
This paper is organized as follows. First,
a baseline of terms and concepts is
established such that the reader and
authors can carry forward the dialog.
Second, the“data everywhere”problem
is briefly discussed, noting the different
critical dimensions that must be
considered. This is followed by a discussion
on two common scenarios, advanced
analytics and self-service business
intelligence (BI). The paper finally presents
a decision-making framework that will help
enterprises identify which approach best
fits their situation.

Terms and Concepts
The terms and concepts covering the
domain of enterprise information
exploitation are constantly evolving. There
is a multitude of definitions from academia,
consultants, vendors, and historical
precedents. This paper defines a set of
working terms, definitions, and concepts
that contain and restrain the knowledge
being presented. It is not the intent of this
paper to usurp working definitions that
may be in use by any one group, but it is
to make the idea conveyance possible. If
there are differences in the definitions with
the reader’s knowledge base, the authors
ask the reader to grant an exception for the
duration of this paper.
Basic Terms
• Data is a set of values. The numbers 5,
15, 32, 38, and 41 are data.
• Knowledge is the awareness of patterns
of information. If the information
“Lottery Numbers 5, 15, 32, 38, and 41”
is known to appear every three weeks,
then there is knowledge.
• Business intelligence (BI) is the
awareness of business operational
information.
• Analytics is the art and science
of discerning patterns in data and
information.
• Knowledge value chain (KVC) is a
set of processes or activities by which
a company refines data such that it
becomes knowledge. Each step in the
process is known as a Knowledge Value
Link.
Basic Concept
An enterprise is laden with data and
information. By applying analytics one
can create knowledge from these assets
by finding patterns of interest. This
knowledge can be showcased through
reports, dashboards, or as inter-system
actions such as ordering more stock in a
particular geographic location. Knowledge
creation and knowledge sharing requires
people, processes, and technology. The
entirety of this knowledge creation and
sharing is collectively known as Knowledge
Value Chain (KVC). Different approaches
to realizing different parts of the KVC give
rise to different implementations seen in
the field.

The Problem
In today’s fast moving world, business
decision-makers, customers, and trading
partners need to act on rapidly changing,
difficult to harmonize information in
near real-time to generate value as
part of the KVC [3, 2]. The need for near
real-time response, combined with the
sheer number and complexity of the data
sources, creates a new set of data access
and integration challenges across the
enterprise [3, 2]. Enterprise data assets
are typified by different technologies in
underlying data sources and ownership
issues with lines of business data [3, 2].
Information in an enterprise can be
scattered across systems, technologies,
and geographies in various formats. There
are numerous reasons for such distributed
and heterogeneous information, such
as varying project funding models,
organizational structure constraints,
mergers and acquisitions, partner
integration requirements, regulatory
constraints, legal restrictions, and more
[3]. Regardless of the reason, blending
such widely distributed data is difficult to
accomplish. Even if one overcomes this
challenge, who is to say that the labeled
data being blended has the same format?
Typically, the information required for
blending is present in heterogeneous
formats thereby making the consolidated
access of this information difficult [3]. For
example, many companies today have
a variety of data marts and warehouses
for data consolidation activities as well as
operational data stores for transactional
purposes [3]. While it may be of value to
blend the information in these separate
systems for some analytical purpose, the
information contained within may be of
different formats, each optimized for its
particular need (e.g. business reporting
on one hand and shipment tracking on
another). Even if one overcomes this
challenge, who is to say that the unified
formatted information has the same
semantic meaning?
It is common for organizations to have
multiple stores with labeled data names
that are identical but whose semantic
meaning is not identical. For example, an
accounting department may have a stored
piece of information called“Sales”and
the sales department may have a stored
piece of information called“Sales.”. In the
accounting world, this is specifically net
sales and it refers to a company’s revenue
earned from the sale of a product or
service while in the sales world it refers
to a specific transaction where money is
exchanged for ownership of a good or
service. If these subtle differences are not
recognized, blending these commonly
labeled informational elements will result
in polluted synthetic information. Even if
one overcomes this challenge, who is to
say that the harmonized information is
on compatible technology platforms that
allow interchange?
Any company that has had digital systems
for more than a year is guaranteed to
have different platforms of information
technology underpinning its landscape.
These varied platforms have varied
protocols, varied requirements on
information interchange, and varied
performance capabilities. These variances
introduce a swath of engineering
challenges.
To cull a unified blend of information
based on the aforementioned challenges
introduces significant latency, as not only
do the technology capability variances
in underlying technologies have to be
accommodated, but information must
be harmonized, varying formats need to
be resolved, and physical spread must
be traversed. Even if one overcomes
these collective challenges, a slew of
new challenges and requirements come
forth from the new synthesized (blended)
information.
Dealing with synthesized information
in the enterprise requires a revisiting of
policies, access rights, data stewardship,
and provisioning. Individual elements
of information in isolation may be fine
to operate on, analyze, and report, such
as a social security number, but the
moment it is blended with first and last
name, the synthesized element is subject
to personally identifiable information
constraints. Such cases are particularly
challenging to identify when dealing with
technical data that may be subject to
export control laws.
All the aforementioned challenges beget
four broad categories of operational
challenges to be addressed:
1. Information service level agreements:
The right information needs to be
delivered to the right system in a timely
and consistent manner, irrespective of its
source. Many applications require real-
time data retrieval abilities [3, 2].
2. Data integration: Data which can
belong to different heterogeneous types
needs to be transformed, integrated
and aggregated to create intelligent
information for applications, systems,
and platforms [3, 2]. Sometimes this
requires creating integrated views of
data from different data sources [3]. The
variance in the business semantics from
one business unit to another creates an
additional challenge for data integration
and interleaving.
3. Data stewardship and provisioning:
Data architects and administrators must
protect the integrity and security of data
sources while making them available
for consumption by applications across
the enterprise [3, 2]. Added to these
concerns is the growing list of regulatory
compliance requirements that require
more careful auditing of enterprise
information usage [3, 2].

4. Information access rights: Information
security administration and
administrators must ensure that the
right people and right systems only
have access to the right information,
including dynamically created synthetic
information. If user access to a particular
atomic component of some synthesized
data is revoked, this must cascade to all
other systems that leverage the blended
data.

Possible Solutions
The fundamental problem faced by
large organizations is that distributed
data in variant forms is difficult for an
enterprise to exploit. This macro problem
is often referred to as the enterprise data
integration challenge. The fundamental
challenge stems from two basic issues:
1. Knowing what data is where (finding);
2. Knowing what information can be
derived from the data (blending). As with
most things, these issues can be solved
with people, processes, technology, or
(more likely) some combination. There are
a multitude of approaches within each of
these three dimensions.
The reader is cautioned that the usage of
the simple terms finding and blending
is only meant to make the concept more
consumable. Neither of these notions are
trivial when it comes to implementation;
each requires critical tradeoff
considerations along the triad of people,
processes, and technology. For example,
when blending there are many approaches
that can be adopted to help data make
its life cycle progression into actionable
insights as part of the KVC. While in some
cases, people serve as primary blender
(such as with statistical analysis), in other
cases machine acts as blender (such as in
machine learning).
It may not be surprising to learn that the
solution choice is often driven by the
constituency driving the need. Some
companies first face this problem with BI
(often when doing enterprise reporting)
and solve the problem with one method,
while some address this problem first with
advanced analytics (finding patterns that
cross organizational boundaries), resulting
in a different choice. Other first drivers
include transactional personalization
(targeting offers based on demand
moments), omni-channel (the unified
singular transaction experience), and so on.
The bottom line is that there is no single
solution as of this writing that fits every
situation perfectly.
To solve this fundamental challenge of
enterprise data integration, a number of
approaches have emerged over the past
few years, each addressing the problem
based on a particular need (often driven
by the project’s funding source). Below is
a brief summary of each method along
with key points associated with the
approach to help in decision-making. The
reader is reminded that the vocabulary
used in defining each basic solution type
isn’t presented as the definitive name for
an approach, but is rather used to serve
a common pattern to help convey the
decision-making criteria that this paper is
highlighting.
Data Foraging
This approach to enterprise data
integration relies on each consumer of data
to source and pull data on his/her own.
Following this approach places all finding
and blending burdens on the consumer
(be it a digital system or a human being)
along with all data movement and
orchestration activities. It is not uncommon
for this approach to be aided with some
form of online accessible data dictionary
or even a data encyclopedia to aid the
data hunters in their quest to find data
that meets a given need. Many exploratory
efforts begin in this model but eventually
morph into, or become inclusive of, one
of the other solution patterns defined. In
this approach the entirety of the KVC rests
with the person (or system) doing the
data foraging. As the variety of discrete
informational entities under study grows,
this solution becomes untenable. The
wider the range of different elements that
are to be considered, the more difficult this
solution becomes.
Data Consolidation
integration creates consolidated pools
of data, separate from the originating
data stores. These amalgams of data
create a pool of data teaming with new
blend possibilities. These consolidated
repositories of data are characterized by
data movements including synchronization
actions and duplicated data (e.g. data
exists in the source data store and in the
consolidated store). There is a possibility
of inconsistent data with this approach
as the source system must first change
and then this change cascades to the
consolidated pool with the time interval
being an important decision driver and
engineering concern. By placing all the
data together, the finding problem is
made easier and the processing paradigm
of moving algorithms to the data can
be exploited; a great value add with the
explosion of Big Data paradigm processing
technologies. Normally this approach is
implemented as a physical store, but it
could be implemented purely in memory
Information
Fabric
Data
Virtualization
Data Consolidation
Data Foraging1
3
2
4
Figure 1 Four key knowledge value chain approaches

as well. Depending upon which capabilities
are expected from the data consolidation
approach, the label applied to it varies
from augmented data warehouse to data
lakes, data pools, or data marts. In this
approach, the wider the range of source
systems to be considered, the more
difficult this solution becomes as large
troves of data are constantly shuffling
around and maintaining which systems are
to be pulled when (e.g. orchestration and
synchronization) becomes a nightmare to
manage.
Data Virtualization
integration leaves all data in their source
data stores, and it simply provides a proxy
that takes in-bound requests for data and
routes them on behalf of the caller to the
source system. This creates a rich pool
without moving data until it is actually
requested, creating consistency guarantees
(see exception that follows). This approach
creates some data retrieval latency due to
the extra routing but is often ameliorated
to some degree with some form of memory
based caching which brings forth a set of
tradeoffs. In the cache implementation
case, the possibility of inconsistent data
(e.g. different values in the source system
and virtualized data) arises but leads to
creation of a localized cache model with
the net effect of faster overall data retrieval.
For example, if a source system is located
half-way around the world but a virtualized
data store is local and implemented with
a cache, it will be quicker to get the data.
Depending upon what is expected of this
approach (transactional abilities, canonical
forms of data, read and/or write, etc.), it
can be labeled as data virtualization or
data federation. However, the more the
consumer base grows and the needs vary,
the more difficult it is to meet all needs
consistently under varying loads.
Information Layer
integration not only unifies data stores, but
also provides the capability to integrate
operational systems. Many operational
systems in the company have transient
data that is of value to the enterprise, but
the data is ephemeral (e.g. an intermediate
processing step). The information fabric
approach adds the capability to refine data
such that it is available to the enterprise
and its systems as information. It is a
software solution that enables applications
to access both raw and integrated data
from multiple, heterogeneous, and
distributed data sources and systems while
hiding the complexity of the disparate
data sources [3]. Instead of moving data
or creating new stores of integrated
data, an information layer creates a loose
federation of multiple existing data sources
and provides a single, virtual data source
through which people, applications, and
systems can access data [3]. In other words,
an information layer creates a“data service”
or“data veneer”that allows applications
and end users to treat a broad variety of
multiple data sources as if they were one
large single source of information [3]. As
such all capabilities of access rights, lineage
tracking, provisioning, etc. are included.
While the tactical implementation
approach to information layer can vary, it
is most common for the implementation
to be based on some form of data
virtualization. According to the technology
research company Forrester, besides other
aspects information layer also comprises
the ability to do transactional processing
and conduct roll-backs on information
changes. [1] However, as the service
level expectations and variations grow,
this solution becomes complex as varied
implementation approaches are required.

Knowledge Value Chain
Decision Making Framework
The four basic solution models for KVC
implementation discussed in this paper
provide a different set of trade-offs that
must be considered when picking an
approach. In practice, we find that most
enterprises have two (or more) of the four
solutions as driven by different business
and operational requirements, resulting
in a hybrid solution to the KVC challenge.
However, the question that still remains
unanswered is which particular solution
is the“best solution”for a particular“data
everywhere”challenge?
The KVC approach decision-making
framework this paper proposes identifies
three dimensions that must be considered:
Constituency, containment, and
composition.
The primary informants of“best fit”are
driven by the study of expectations of the
KVC as represented by the following three Cs:
Constituency: The user population of the
KVC must be defined. This could be people,
applications, and/or other systems. Who is
the constituency of the value chain?
Containment: The universe of data or
information that is considered to be a
part of the KVC must be defined. What is
contained by, or within, the value chain?
Composition: The approach (people,
processes, and/or technology) for each
knowledge value link progression must be
discerned. How is each link composed?
The KVC decision framework takes each
of these dimensions and explores them at
three different levels. First, at a macro level,
we study the variety of expectations of the
KVC itself in a given enterprise. Second, we
begin decomposing these expectations of
the KVC and its individual links. Third, we
ask a collection of specific questions that
will inform our decision-making flowchart.
While not required, a“current situation”
analysis can play a part in selecting the
future strategy based on existing (and
already paid for) capabilities. To this end,
a KVC subsystems capabilities can be
discerned from the Figure 3 graphic where
each individual capability required for KVC
is articulated.
Knowledge Value Chain: Variety
At the highest level, the variety of sources
and usages that the KVC must service
drives the“best approach”decision-making
framework. The variety of information (i.e.
the scope) to be blended is compared with
the variety of usage (i.e. the constituency
of users, be it people or systems).
Figure 2 shows the outcomes we’ve found
at the macro level.
Knowledge Value Chain: Expectations
The constituency, containment, and
composition dimensions must be explored
for their individual expectations. To this
end, the following questions must be
answered:
• How durable do we want the results of
the KVC to be? Some progressions of
the KVC are ephemeral while others are
intended to provide long-term results
that inform the enterprise. Discerning
typical usage is important and a
reflection of the constituency.
• How self-contained and self-consistent
are the data (or informational) elements
that will be included in the KVC? This
drives data integration, data cohesion,
and data harmonization complexity
concerns.
• How much shall technology address the
KVC? This helps identify options that can
be used in the composition of the value
chain.
Knowledge Value Chain: Details and
Specifics
The third level of analysis along the three
dimensions manifests the following
specific considerations that an enterprise
would have to look at.
Cost: The cost incurred for building and
maintaining the solution. This would
also include cost incurred for procuring
infrastructure, continued licenses and
maintenance contracts. Based on the
budgets available for resources (skills,
infrastructure and continued support) the
solution needs to be identified.
Timeliness: Time taken for consolidation,
cleansing, processing, augmenting,
persisting and accessing data for business
operations is one of the most critical
parameters or dimensions for the decision
matrix framework. Most solutions options
depend on the timeliness of data for
business decisions. Ease, frequency of data
access, and quick response to business
scenarios are critical for success of most
business operations. Timeliness and
frequency also depend on how often the
client system pulls data. For BI systems the
data would be accessed very frequently
Figure 2 Knowledge value chain variety.
Data
Consolidation
Information
Fabric
Data
Foraging
Data
Virtualization
Data
Consolidation
Data
Virtualization
Data Foraging
Variety of Consumer
InformationScope
Low High
High

Some other areas to be considered in
this level of depth include configuration,
technology standards in the enterprise,
technology adaptability, resource
constraints (including knowledge worker
pool), and data availability.
Knowledge Value Chain: Subsystem
Capability Requirements
It is instructive and informative to view the
requisite capabilities and potential tie-ins
between the solution options to help form
the final choice. This enables two types of
analyses. First, if an inventory of existing
capabilities is being done, the decision
framework can be used to identify what is
possible within a given scenario. Second, if
a particular solution model is picked then
the sourcing and implementation patterns
can ensure the required capabilities
are present. The functions also help us
identify and evaluate the right tools and
technology vendors for the enterprise.
Figure 3 Knowledge value chain decision framework
(data size would not be big), whereas for
analytics data access is not frequent but
the data size could be big.
Ownership: Not all data sets are available
for persistent storage. Some of the data
sources owners limit the duplication of
data in other storage systems due to
regulatory/ legal or contractual reasons.
The ownership of the data is solely limited
to the original stewards and owners of the
data.
Impact: Impact on the underlying database
and to other business systems using the
database could cause slow down. For
example, a frequent access of the order
book could slow down the transaction
processing system.
Throughput rates: Throughput rates
required by the workload need to be
considered carefully.
Stability: Stability of data requirements
from the consumer workloads needs to
be checked. For example, in case of BI
reports the nature and composition of the
report can be modeled easily, but for some
analytics workload the requirements are
not defined and are more exploratory in
nature.
Quality: Quality of data is another critical
decision parameter for the framework. For
example, self-service BI will need good
quality data and thorough cleansing and
augmentation before business operations
can explore, analyze and visualize the data.
Some of the analytics model, however,
can be built on noisy data to identify the
patterns of inconsistency.
Volume: Volume of data sources used,
information processed, and the need
for storing the processed information is
yet another concern. If the time taken to
process the information is more, we might
want to avoid data processing repetition
even if it was easy.

Legend:
Do we have
resources (people,
process, and technology)
allocated to this
program?
Do we have
clarity in the
business and operational
requirements for BI/
analytics?
Do we have
any regulatory/legal
or contractual reasons
to not persist
the data?
Do we need the BI/
analytics responses on
demand?
Is performance a
very critical factor for
the BI and analytical
applications?
Does the data
need aggregation,
augmentation, cleansing,
and/or other data
quality activities?
Do downstream
applications need data/
information access via
services?
Do we need
transaction write back?
Is the data
semi-structured
or unstructured?
Do we need to augment
data from numerous
social and web
portals?
Are users
asking for data in
an ad-hoc manner? Is
the data requirement
more exploratory
in nature?
Is the data manually
manageable?
Is the data available
in real time?
Can the data be
persisted in a separate
storage?
Data Foraging
Constituency Containment Composition
Data Consolidation Data Virtualization Information Fabric Hybrid Solution
Are we
constrained by time
for coming up with
BI/analytical
decisions?
Decision Flow
No
Yes
Yes
Yes
No
No
No
No
No
Yes
Yes
Yes
Yes
No
Yes
No
No
Yes
Yes
No
No
No
Yes
Yes
No
Yes
No
Yes
Yes
No
No
Yes
Figure 4 Flowchart to identify“best suited”solution approach in knowledge value chain
The functions can be distinctly categorized
into the following five groups based on the
implemented approach:
Data Sources: This function includes
all data sets that are required by the
enterprise to create information or
meaningful decisions. This could include
relational databases, third party market
analyst data or data augmented from
external sources, data on the cloud, manual
data or excel sheets created by business
users, data streamed via other internal/
external sources, or data from social media
or web portals.
Data Consolidation: Data can be
consolidated with the help of tools and
technologies. The functions enabled during
consolidation lead to the creation of data
lakes, data marts, and/or data warehouse.
Data Virtualization: Data virtualization
provides an abstraction layer for the source
data which can be directly accessed by the
BI and analytical applications.
Information Fabric: The information
layer or information fabric converts
data into information by leveraging
the functionalities of data quality
along with transactional management
capabilities of writing back data into the
data consolidation layer. Also integral
to this layer is the capability to provide
the information to the downstream
applications via data or information
services.
BI Applications and Analytics: BI
applications and analytics provide the
final insight to the enterprise data. Data
can be foraged from data sources by
manually massaging, integrating or
consolidating data for BI applications
and analytical capabilities. Advance
analytical and BI capabilities can be best
derived by leveraging data consolidation,
virtualization and information fabric
approaches.
Knowledge Value Chain Best Fit
Decision Flow
The enterprise can further envisage the
“best”solution with the help of a decision-
making flowchart by answering some basic
questions across the various informants
(three Cs) and dimensions as mentioned
in the earlier sections. The reader needs to
note that Figure 4 is a pragmatic flowchart
to identify the solution approach“best
suited”but nevertheless needs thorough
evaluation for your particular enterprise
environment and constraints. Also to be
considered are the tools and technologies
(that are ever evolving) available for this
solution. In cases where a straightforward
single approach is not possible, we might
need a hybrid solution.
Do we have
restrictions from
source owners to
run queries and
processes on their
infrastructure?
Do I need the data
for advance analytical
models, simulations
and processing?

Usage Scenarios
This section illustrates two possible usage
scenarios that leverage the advanced
analytics and self-service BI approaches.
Advanced Analytics
In this scenario, we cover data needs for
data science teams that are engaged
in development of advanced analytics
models. Data science team members are
qualified power users who are frequently
engaged in all aspects of the data life cycle,
from defining the source for data to data
formats and data quality aspects.
Figure 5 Knowledge value chain with advanced analytics human links
Possible solutions for enterprise data (data
foraging, data consolidation, data virtualization,
information layer/fabric)
Modeling (advanced analytics algorithms and models, service
API to integrate model to business applications)
Insight delivery (advanced visualizations, interactive analysis)
Integrate
and Manage
Operational
Data
Model
Development
Validation and
Deployment Insight Delivery
Users/Applications
Knowledge Supply
Chain
Other Data
Sources
Analytics Data
Access,
Transform,
Explore
Data Scientist Data Scientist Data Scientist Business Analyst
Data Engineer Data Engineer Data Engineer Executives
Data Analyst Data Analyst Business
Applications
Data Scientist
Data Engineer
EDW/Marts

The data intensive activities that a typical
data science team goes through while
building and deploying advanced analytics
models are listed below:
Exploratory Data Analytics is the most
data intensive phase that is iterative and
runs through the entire data science life
cycle. It starts at the finding phase where
data scientists work with analysts and
data engineers to explore all relevant
data available (to address the business
problem), identify patterns, formulate
hypothesis, and strategize the analytical
models that would be needed to deliver
the solution. The team access the data in
real-time to create an interactive solution
for accessing large data sets.
Data Cleansing and Transformation is
closely knit with the exploratory analysis
phase where analysts and data engineers
blend the data to derive information,
formulate hypothesis, and develop
analytical models. As the requirement is
not real-time, this activity is carried out as
a batch process. However, the solution has
to provision clean and transformed data for
real-time (interactive) access.
Analytical Model Building is experimental
and iterative. In this phase analysts work
with data scientists to test out their initial
hypothesis. They start by working on a
representative data sample (and later
on the entire data cube created for the
analytical model) where the team develops
and evaluates the model, analyzes the
performance, fine-tunes the models and
finally selects the model(s) that should be
deployed in production.
Analytical Model Deployment is done
when the data scientists are satisfied with
the models developed in the experiments
that were carried out in the model
development phase. In this phase data
scientists work with data engineers to
embed and run the model on the entire
data set to carry out the analysis and
deliver results based on the type of the
business problem. Depending on the use
case, the model may need to be deployed
as a service and interact with applications
in real-time to enable data-driven
decisions. The“next best offer”is a good
example of this. Alternatively, the model
may be deployed as a batch program to
analyze data, such as sales forecast, in
offline mode.
Analytical Model Maintenance is the
process of maintaining and managing the
model life cycle including reconstructing
sets of ethereal data. This aids regulatory,
governance, legal, and contractual
requirements.
Insight Delivery - Visualization,
Dashboard and Reporting is the most
crucial part of the entire data science
life cycle. Articulating the findings and
insights that the analytical models have
unearthed in an easy to understand format
is necessary for delivering the business
value to human beings as well as to digital
systems. It is therefore important for the
solution to deliver insights to business and
business systems in a form where they can
access, interact and analyze the findings
and take data-driven business decisions.
While all data science teams would like to
have an infinitely scalable, comprehensive
data platform with data from all sources in
the enterprise at their disposal, the reality
is dictated by different factors such as
cost, time, effort, compliance issues, etc. In
reality, a combination of the methods listed
in this document are deployed by IT teams
to meet the needs of advanced analytics.
To decide what all methods are required,
following are the constraints and factors
that need to be considered:
• The amount of variance between
informational elements drives the
program scope. The scope of the
analytics program (all business
functions, limited business functions,
and/or business informational domain
elements) is a driving factor in
determining the methods that need
to be adopted for addressing the“data
everywhere” problem.
• Potential benefits of specific advanced
analytics models determine the overall
spend that can be made in data, without
impacting the return on investments.
The more impactful models should have
correspondingly higher investments in data.
• Longevity of the advanced analytics
models determine the effort that needs
to be spent to ensure availability of
up-to-date and valid data across the
life cycle. For a long running model it
is important to have all relevant data
aggregated in one place, with data
quality and data lineage capabilities built
in. This would help with validations on
the efficacy of the model across different
time periods and also with tracing back
important decisions that were taken
during the life of the model.
• No matter how formal and rigorous the
planning process is, it is difficult to avoid
ad hoc requests for advanced analytics
model. This brings in an aspect of agility
that needs to be provided in sourcing
data with reasonable data quality.
• In some cases there is a short time to
market for the analytics model which
makes it even more important to adopt
agility and enable the data scientists
to source and provision data for the
analytics model.
Self-service BI
In this usage scenario, we discuss one
of the frequent scenarios for business
operations, managers and business users.
Business users prefer to explore and access
data themselves for BI and reporting.
They require functional capabilities that
could help them build reports, scorecards,
dashboards, and exploratory visualizations
using self-service tools and wizards.

Self-service BI, as the name suggests,
is used to drag/drop or create quick
visualizations, reports, and dashboards in
limited period of time by using wizards
so that the users can generate BI reports
themselves. Users use the underlining
integrated data from heterogeneous
sources brought together using our four
solutions approaches (data foraging, data
consolidation, data virtualization, and
information fabric).
Business users can leverage self-service BI
environments for their daily BI activities
along with business decisions and
operations activities. They could:
• Run a semantic (using natural language)
and quantitative search to discover data
available in the enterprise, and/or blend
enterprise data with data available in
the public domain, to make meaningful
insights. This discovery and exploration
could further lead to standard
operational BI reports that can be
presented to executives and managers
at set regular frequencies using job
schedulers.
• Request for purchase or loading of
data not found in the integrated data
environment. Data extracts and sets can
be requested and rendered with minimal
IT involvement. For data sets that are not
available to the enterprise, self-service BI
helps as an investigative and exploratory
mode of analysis to further enhance and
blend data.
• Run predefined and ad hoc queries on
the data sets. Build quick and custom
visualizations and analysis to resolve
immediate business problems and needs.
• Collaborate with other users to create
data mash-ups across multiple data
sources.
• Enable social BI features such as
community rating, collaborative
metadata enrichment, and more.
Besides functions such as collaboration,
integration, data mash-ups, data
visualizations along with data lineage,
metadata search, self-service BI also
provides an easy, simple and intuitive user
interface (UI) that can be made available
on a desktop as well as on smartphones,
tablets and other mobile devices. This
makes information available to business
users even when they are on-the-go,
thus making it an integrated BI platform
Integrate and
Manage
EDW/Marts
Access
Possible solutions for enterprise data (data foraging, data
consolidation, data virtualization, information layer/fabric)
Easy UI (office support, advanced visualization, portal, search,
analytical model)
Collaborative UI (collaboration, usage tracking)
Mobile UI (access on mobile devices)
Executives
Analytics Builder
Analytics Builder
Analysts/ModelersInformation
Producer
Information
Producer
Information
Consumer
Information
Collaborator
Decide
Information Supply Chain
Users
Analyze and
Publish
Workgroup
Data
Discover and
Enhance
Operational Data
Other Data
Sources
Figure 6 Knowledge value chain with human BI links
with ease of authoring, modeling and
publishing to the end user without IT
support.
Typical user groups that leverage the
information supply chain would include
data engineers (who are aware of the data
sources that supply the required data), BI
authors (who are capable of leveraging
the self-service intuitive UI and wizards to
publish reports, visualizations, dashboards
and scorecards for their managers) and
business executives (who leverage the
outputs for decision-making).

Conclusion
Data integration can be a significant effort
whether you are engaged in building new
data-intensive applications, adapting a
packaged application to a new context,
or are trying to create a single point of
access for your enterprise data [3]. More
companies are turning to a multitude
of solutions which can shorten data
integration projects and lower data
management and maintenance costs over
time [3]. The information fabric approach
can simplify data provisioning, access
and integration, thus shortening the data
integration time frame and enhancing
productivity by enabling developers
to spend their time concentrating on
developing actual application logic [3].
However, it requires a tremendously
skilled systems integration capability,
in-depth data theory, and more. On the
other extreme, those who need integrated
data can be left to their own devices. In
the modern landscape, this is simply not
a long term option. Therefore, without
conducting in-depth situation analysis,
the data consolidation solution is the most
pragmatic. Many of the efforts required
in data consolidation activity can roll into
either data virtualization or the information
layer.
Citations
1. August 8, 2013. Noel Yuhanna and Mike
Gilpin.“Information Fabric 3.0”. Forrester
Research
2. 2011.“Infosys Gradient: An EII Solution”.
Infosys
3. April 2005.“Infosys Gradient: Enabling
Enterprise Data Virtualization”. Infosys

About the Authors
Bill Peer is Principal Technology Architect
at Infosys Labs. He has over 20 years of IT
experience. His focus area is IT strategy
for business competitive advantage, with
hyper specializations in innovation, large
scale global enterprise architecture, and
emergent technology exploitation (such as
‘Big Data’systems).
Kiran Kumar Kaipa is Senior Consultant
at Infosys Labs. He has over nine years
of experience in the IT industry where
he has worked in both consulting and
technical roles. His current focus area is
Big Data analytics with specializations
in data munging, data analysis and data
visualization.
Shyamala Sadananda is Senior Architect
at Infosys Labs. She has over 15 years of
professional experience in systematic
innovation, architecture, consulting, and
solution implementation. She primarily
anchors engagements with focus on
emerging technologies in the areas of data
virtualization, analytics, data visualizations
and business intelligence.
Swaminathan Natarajan is Principal
Product Architect at Infosys Labs. He has
over 17 years of experience in the software
industry and has worked in various
functions such as product engineering,
R&D and technology consulting. His
current focus area is Big Data analytics
with specific interest in data munging and
unstructured data analytics.

© 2015 Infosys Limited, Bangalore, India. All Rights Reserved. Infosys believes the information in this document is accurate as of its publication date; such information is subject to change without notice.
Infosys acknowledges the proprietary rights of other companies to the trademarks, product names and such other intellectual property rights mentioned in this document. Except as expressly permitted,
neither this documentation nor any part of it may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, printing, photocopying, recording or
otherwise, without the prior permission of Infosys Limited and/ or any named intellectual property rights holders under this document.
About Infosys
Infosys is a global leader in consulting, technology, outsourcing and next-generation services. We enable clients, in more than 50 countries, to stay a
step ahead of emerging business trends and outperform the competition. We help them transform and thrive in a changing world by co-creating
breakthrough solutions that combine strategic insights and execution excellence.
Visit www.infosys.com to see how Infosys (NYSE: INFY), with US$8.25 B in annual revenues and 165,000+ employees, is helping enterprises renew
themselves while also creating new avenues to generate value.
For more information, contact askus@infosys.com www.infosys.com

enterprise-data-everywhere

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to enterprise-data-everywhere (20)

enterprise-data-everywhere