U.S. patent application number 11/401953 was filed with the patent office on 2007-10-18 for system and method for applying predictive metric analysis for a business monitoring subsystem.
Invention is credited to Pawan Raghunath Chowdhary, Shubir Kapoor.
Application Number | 20070244738 11/401953 |
Document ID | / |
Family ID | 38605944 |
Filed Date | 2007-10-18 |
United States Patent
Application |
20070244738 |
Kind Code |
A1 |
Chowdhary; Pawan Raghunath ;
et al. |
October 18, 2007 |
System and method for applying predictive metric analysis for a
business monitoring subsystem
Abstract
Predictive metric analysis for business management is divided
into build time, corresponding to the business owner view of the
enterprise, and run time, corresponding to the information
technology view of the enterprise. The build time consists of a
predictive model and a monitoring model. These models go through
transformation processes to the components of the run time. The run
time components are a Metric Value Prediction Service (MVPS), which
receives as input predictive model transformation and outputs
predicted metric values, and a monitoring engine, which receives as
input monitoring model transformation, the predicted metric values
and business events from the business process. Various analytical
engines can be plugged in to provide the predictive capabilities.
Input is provided to a framework from various business systems
which results in predicting the value of the metrics across the
future time horizons.
Inventors: |
Chowdhary; Pawan Raghunath;
(Montrose, NY) ; Kapoor; Shubir; (Shrub Oak,
NY) |
Correspondence
Address: |
Whitham, Curtis, & Christofferson, P.C.
Suite 340
11491 Sunset Hills Road
Reston
VA
20190
US
|
Family ID: |
38605944 |
Appl. No.: |
11/401953 |
Filed: |
April 12, 2006 |
Current U.S.
Class: |
705/7.31 ;
705/7.37 |
Current CPC
Class: |
G06Q 10/0637 20130101;
G06Q 10/063 20130101; G06Q 10/06375 20130101; G06Q 30/0201
20130101; G06F 2216/03 20130101; G06Q 30/02 20130101; G06Q 30/0202
20130101 |
Class at
Publication: |
705/010 |
International
Class: |
G06F 17/30 20060101
G06F017/30 |
Claims
1. A method for providing predictive modeling capabilities to allow
intelligent business performance management comprising the steps
of: using a computer to generate a meta model consisting of
business metrics organized as a hierarchy with each metric having
the ability to be associated with time as a look-ahead dimension;
using a computer to transform the meta model to service interface
definition and related artifacts; using a computer to transform the
meta model into performance warehouse meta-data schemas to allow
persistence of historical data; and using a computer to receive
requests for metric predictions and to use analytical techniques to
service the requests in real time.
2. The method of claim 1, wherein said meta model also comprises
business metrics categorized within a predictive metric context to
be used as input for predictive analysis.
3. The method of claim 2, wherein said meta model also comprises
trigger conditions describing how and when said predictive analysis
will be triggered.
4. A system for providing predictive modeling capabilities to allow
intelligent business performance management comprising: a computer
generating a meta model consisting of business metrics organized as
a hierarchy with each metric having the ability to be associated
with time as a look-ahead dimension; a computer transforming the
meta model to service interface definition and related artifacts; a
computer transforming the meta model into performance warehouse
meta-data schemas to allow persistence of historical data; and a
computer receiving requests for metric predictions and using
analytical techniques to service the requests in real time.
5. The system of claim 4, wherein said meta model also comprises
business metrics categorized within a predictive metric context to
be used as input for predictive analysis.
6. The method of claim 5, wherein said meta model also comprises
trigger conditions describing how and when said predictive analysis
will be triggered.
7. A computer-readable medium for providing predictive modeling
capabilities to allow intelligent business performance management,
on which is provided: instructions for using a computer to generate
a meta model consisting of business metrics organized as a
hierarchy with each metric having the ability to be associated with
time as a look-ahead dimension; instructions for using a computer
to transform the meta model to service interface definition and
related artifacts; instructions for using a computer to transform
the meta model into performance warehouse meta-data schemas to
allow persistence of historical data; and instructions for using a
computer to receive requests for metric predictions and to use
analytical techniques to service the requests in real time.
8. The computer-readable medium of claim 7, wherein said meta model
also comprises business metrics categorized within a predictive
metric context to be used as input for predictive analysis.
9. The computer-readable medium of claim 8, wherein said meta model
also comprises trigger conditions describing how and when said
predictive analysis will be triggered.
Description
BACKGROUND OF THE INVENTION
[0001] 1. Field of the Invention
[0002] The present application generally relates to business
monitoring systems and, more particularly, to a model driven
approach to enhance existing business performance models with
predictive modeling capabilities.
[0003] 2. Background Description
[0004] For an enterprise to be competitive, the ability to perform
predictive analysis on large amount of data is very important to
analyze a trend, discover the paint points, and/or discover new
opportunities. Most companies today implement various Business
Performance Management solutions including Business Intelligence
techniques that help determine the current state of the business.
This is achieved by defining metrics or key performance indicators
organized in a hierarchy through the various vertical and
horizontal silos of the organization. Data and events received in
real time are persisted in a data mart and are used to provide
historical analysis summarizing what has happened in the past. In
other words, historical analysis can reveal who the best customers
were last month and who they were this month. This kind of
traditional analysis cannot reveal what will happen in the future.
Predictive analysis discovers meaningful patterns and relationships
in data separating the signals from the noise thereby helping in
improved decision making. Business process monitoring models
currently lack the ability to incorporate the meta-data for such
predictive models, which restricts the models' ability to capture
such metrics. There are currently limited modeling capabilities and
supporting tools to capture the metric definition, relationships,
dimensions, semantics and their management. The available tools
today limit metrics modeled as hierarchical structure with value
dependency. This implies that the existing models are also not
sophisticated enough to capture the relationship of time as a
dimension to allow for look-ahead prediction of the metrics such as
dynamic systems models, time-series based models, forecasting,
propensity and scoring models. Combining predictive analysis with
organization business process and performance management provides
insight into critical business issues and enables proactive
decision making and risk management amongst other benefits.
[0005] There are many subsystems available that provide prediction
capabilities. Among these are general purpose systems such as data
mining tools and system dynamics. Data mining tools provide scoring
models and predictions based on historical data; however, data
mining tools do not provide metric values but can determine
qualitative relationships. It is objective in nature. System
dynamics use continuous modeling to predict values of metrics based
on the specified time dimension; however, system dynamics is
subjective since it is based on the user's perception of the metric
network and relationships.
[0006] The integration of these predictive systems with standard
business process monitoring and management systems has always been
a challenge. Business monitoring systems are built using metrics
catering to the current and in some cases historical aspects of the
business whereas predictive models look ahead in time. The current
invention features a novel mix of both these capabilities in order
to provide a system and method for predictive metric analysis to a
business monitoring subsystem
SUMMARY OF THE INVENTION
[0007] According to the present invention, there is provided a
business performance metamodel comprising business metrics
organized as a hierarchy across the operational, tactical and
strategic levels of an organization. A metamodel is a model that
describes a language in which models can be expressed. A metamodel
spanning the operational level of an organization is comprised of
transaction metrics organized in a hierarchy. For instance, in the
transportation industry, OnTimeDelivery constitutes one of the most
important transactional metrics. It is derived from two lower level
transactional metrics: Shipment Dispatched and Shipment Arrived.
The difference between the shipment arrival time and the shipment
dispatch time generates the value for the OnTimeDelivery metric. A
metamodel spanning the tactical level of an organization is
comprised of tactical metrics which are aggregate by nature and
have an inherent relationship between themselves and their
corresponding lower level operational metrics. For instance,
Efficiency is tactical metric which is usually generated across a
department or a hub and aggregated over the OnTimeDelivery metric
of every shipment for that department or hub. A metamodel spanning
the strategic level of an organization is comprised of C-level
strategic metrics which have an inherent relationship amongst
themselves and their corresponding lower level tactical metrics.
For instance, customer satisfaction is a strategic metric that the
C-level executive measures themselves against and is derived from
the efficiency of the department or hub amongst other tactical
metrics. Most of the existing business performance meta models do
not consider time as a first class element and therefore allow for
modeling the current state of the business but not future
predictions of any of the metrics. The current invention considers
time as a first class element associated with each predictive
metric which defines the horizons for prediction. A predictive
metric is derived from a regular metric with the additional
association of time for look ahead predictive capabilities. The
invention also provides the modeling of relationships between
metrics that are needed for the predictive analysis, thereby
describing the expected changes in behavior of these metrics over
time. These metrics are organized in a predictive metric context.
The invention further provides the modeling of the trigger
conditions that describe when and how the predictive analysis
should be performed. Users create business performance models
enhanced with predictive capabilities based on the specified meta
model. The business performance model is then transformed into a
platform independent IT model. A platform independent IT model is a
description of the solution, independent of the platform on which
it executes. There are three core models which serve as the
platform independent IT model--predictive model interfaces,
warehouse model and solution composition model. The predictive
model interfaces are used as inputs to the MVPS (metric value
prediction service). This service serves to interface data and
metrics with various analytical solutions to perform the predictive
capabilities. Metrics modeled within the predictive metric context
are passed in from the business monitoring subsystem based on the
trigger conditions also described in the model. The solution
composition model orchestrates the business performance solution
consisting of the business events and metric hierarchy and
interfaces with the MVPS. The warehouse model serves to persist the
current and predicted state of the business that is exploited by
the business monitoring subsystem as well as the MVPS.
[0008] There are various analytical techniques that can provide the
predictive capabilities within the context of an end to end
solution. Some examples are data and text mining, system modeling
and dynamics, time series analysis and forecasting etc. Each
technique is chosen based on the analysis of the data along with
the business process currently being monitored coupled together
with the overall organization strategy.
[0009] For example, an insurance company looking to improve
healthcare and reduce costs would implement a data mining based
predictive model which examines claims data and predicts individual
usage of healthcare services over the next insured period. The
metrics involved in the analysis and historical data provided
usually don't have any fixed patterns and therefore an important
component of the prediction would be to devise a pattern from the
historical claims data. The predictive metrics in this scenario
would be claim type (i.e. dental, medical, pharmaceutical,
disability etc) and costs for every member. The trigger would be
tied to every claim being submitted by the particular member which
implies that the predictive model will be executed every time a
member files a claim with the insurance company. The MVPS would be
responsible for interfacing with the data mining based predictive
model to generate the prediction of the claim types and costs the
results of which would be sent back to the monitoring engine and
warehouse for further analysis.
[0010] Another example where predictive analysis would be useful is
in the electronics industry where the predictions of customer
ordering behavior can significantly allow responding in ways to
avoid deleterious outcomes, i.e., migrating supply to upward
trending demand before pending orders emerge. For this purpose, a
time series based forecasting system would be applicable that would
factor in seasonality, order skews, product life-cycles and
repetitive order trends inorder to make its prediction. The
predictive metrics in this scenario would be future weekly order
loads. The trigger would be tied to the daily demand-supply
decision cycles implying that the predictive capability would be
executed daily, possibly tied to a scheduler. The MVPS would be
responsible for interfacing with the forecasting system to generate
the daily trends the result of which would be sent back to the
monitoring engine and warehouse for further analysis.
[0011] The present invention thus provides a computer-implemented
method, a machine-readable medium instructing a computer, and a
system to provide predictive modeling capabilities to allow
intelligent business performance management comprising: generating
a meta model consisting of business metrics organized as a
hierarchy with each metric having the ability to be associated with
time as a look-ahead dimension; a computer transforming the meta
model to service interface definition and related artifacts; a
computer transforming the meta model into performance warehouse
meta-data schemas to allow persistence of historical data; and a
computer receiving requests for metric predictions and using
analytical techniques to service the requests in real time. The
meta model may also comprise business metrics categorized within a
predictive metric context to be used as input for predictive
analysis and may further comprise trigger conditions describing how
and when said predictive analysis will be triggered.
BRIEF DESCRIPTION OF THE DRAWINGS
[0012] The foregoing and other objects, aspects and advantages will
be better understood from the following detailed description of a
preferred embodiment of the invention with reference to the
drawings, in which:
[0013] FIG. 1 is a description of the various levels and their
relationships in a regular business performance model. Also
depicted are predictive extensions at all levels in the model.
[0014] FIG. 2 shows an example of the metrics at the various level
of the models.
[0015] FIG. 3 is a data flow diagram illustrating the predictive
metric management implemented by the invention;
[0016] FIG. 4 is data diagram illustrating in more detail the
predictive model; and
[0017] FIG. 5 is a block diagram showing the solution runtime
process of the invention.
DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT OF THE INVENTION
[0018] Referring now to the drawings, and more particularly to FIG.
1, Business Performance Model and Predictive Extensions, there is
shown the predictive metric management implemented by the
invention. At the top of this diagram are the components of the
build time, corresponding to the business owner view of the
enterprise. This consists of a predictive model 105 and a
monitoring model 101. The predictive model is shown to derive the
information on top of the monitoring models. Typically, a metric of
interest for the predictive analysis can be chosen along with
context metric for predictive analytics. The System then assists
user to add additional information as shown in FIG. 4.
[0019] The monitoring model generally contains metrics that needs
to be monitored and corresponding context metrics. These metrics
are typically grouped into three levels such as Operational 104,
Tactical 103 and Strategic 102. The grouping of metric helps in
determining the level of users and the latency of the metric
calculation.
[0020] For example FIG. 2, An Example of the Levels in a Typical
Business Performance Model, illustrates a hierarchy of metrics that
spans across the three levels as mentioned above. Typically the
leaf node metrics always belongs to the Operational level. Shipment
Dispatch 201 and Shipment Arrive 202 are metrics that contains the
atomic data and hence shown in the figure as part of Operational
level. The On Time Delivery 203 metric derives its value from 201
and 202 metric. This metric again represents data for the user that
typically belongs to this level, such as Operations manager, hence
this metric also belongs to the Operational level. Similarly based
on the user profile and importance the rest of the metric
Efficiency 204 belongs to Tactical level and Customer Satisfaction
205 belongs to Strategic level.
[0021] FIG. 3, Predictive Metric Management, illustrates the high
level view of the predictive management System. Build time 301
illustrates the build time activity when first monitoring model 302
gets designed. Once monitoring model 302 is available, one selects
the metrics of interest for the predictive model 303. These models
go through transformation process 304 to the components of the run
time 305, corresponding to the information technology (IT) view.
These components are a Metric Value Prediction Service (MVPS) 306,
which receives as input the predictive model transformations 304
and outputs predicted metric values 310, and a monitoring engine
307, which receives as input monitoring model transformation, the
predicted metric values and business events 308 from the business
process 309 (i.e., transactions). How to extend a monitored metric
into predictive model and over all view of the run time process is
shown later.
[0022] FIG. 4, Predictive Model, illustrates the process of
extending a metric in the monitoring model and its corresponding
context into predictive model. As shown in FIG. 4, the predictive
model is described as follows. The predictive annotations, PA, is a
4-tuple, <PM, PMC, TC, T>, where PM is a set of predictive
metrics 401 {pm.sub.1, pm.sub.2, . . . , pm.sub.n} whose values
will be predicted based on changes in the business environment, PMC
is the PredictiveMetricContext 402 consisting of a group of metrics
{m.sub.1, m.sub.2, . . . , m.sub.m} which constitutes the context
around which the prediction will be done, TC are the trigger
conditions 403 {tc.sub.1, tc.sub.2, . . . tc.sub.m} that cause the
monitoring engine to trigger a prediction request to the prediction
service, and T is the time bounds 404 for the model consisting of
{inittime, finaltime, timestep, timeunits}. "inittime" is the time
the prediction starts. It defaults to 0 unless specified otherwise.
"finaltime" is the time the prediction ends. "timestep" defaults to
1 unless specified otherwise. "timeunit" is a predetermined unit of
time; e.g., year, qtr, month, week, day hour, minute second. Time
Buckets, TB {tb.sub.1, tb.sub.2, . . . , tb.sub.y}, are generated
by the system where y=(finaltime-inittime)/timestep and every
tb.sub.1 is a timebucket belonging to the timeunit. For every
pm.sub.i, the Prediction Engine will generate {pmtb.sub.1,
pmtb.sub.2, . . . , pmtb.sub.y} where y is the number of time
buckets generated by the system.
[0023] FIG. 5, Building Solutions with MVPS, illustrates the
processing steps in order to build a solution for Metric Value
Prediction Service (MVPS). Typically one starts at step 501 with
annotating the metrics of interest in the monitoring model with the
predictive model annotations as explained earlier. The second step
at 502 is to transform the monitoring model consisting of monitored
metrics and its corresponding contextual metric elements into the
representation acceptable by the monitoring engine. The third step
at 503 is to transform the predictive annotations into appropriate
service interface definitions defined by MVPS. The fourth step at
504 is to transform the monitoring model and predictive annotations
into a warehouse model to persist historical data for analysis
purposes and also to assist various analytical engines with the
appropriate data. The fifth step at 505 is to define the analytical
capability to predict metric relationships and attribute values
(e.g., data mining, systems dynamics, etc). The monitoring model
and predictive annotations are used as input. The sixth step at 506
is to build a service implementation definition for the analytical
engine and based on the interface definitions. The seventh step at
507 is performed at runtime, as a result of which the monitoring
engine sends a request to the prediction engine when metrics
associated to the predicted metrics are changed. At the eighth step
at 508, the prediction engine predicts the future values of the
metrics and sends it back to the monitoring engine. Finally, both
predictive metric values and monitored values are persisted in the
data warehouse 509 that was generated earlier.
[0024] While the invention has been described in terms of a single
preferred embodiment, those skilled in the art will recognize that
the invention can be practiced with modification within the spirit
and scope of the appended claims.
* * * * *