View
4
Download
0
Category
Preview:
Citation preview
© 2013 WGSigma Software, Inc. | Confidential
William Guinn
SVP and CTO
San Jose, CA
william.guinn@wgsigmasystems.com
WGSigma Systems
Architecting intelligent real-time systems processing
billions of events a day
1
© 2013 WGSigma Software, Inc. | Confidential
Hundreds or thousands of systems
performing transactions in silos with no
collective intelligence
The Universal Corporate Challenge:
2
BILLING
CORE
PP / RT
TABLES
UPS
Unified
Provisioning
PPOL / RTOL
Price Plan Online
Reference tables
Configuration
CSM Online
Customer Mgmt.User interface
MAF
Mssg Acq
& Format
CH
Customer Hierarchy Mgmt
Discount Engine
BAN Thresholds
& Aggregation
EMS
Error Mgmt System
CDR errors
Corp eCare
Account Management,
Bill View & Payment
LMS
Letter Management
Collection/Notices
ADD
Invoice s
A/R
Accts Receivable
General Ledger
Collections
Risk/Path/Agency
CSM Server
Customer Mgmt Syst.
Primary Business Logic
HP IUM
PPD Middleware
LMPS/NM
Mediation
AG
Activation G/W
WAC
Wireless Activation Ctr
fastACT
Activations Lounge
WOF
Web Order Form
OBOT
One Bill Orders
HUB
Vitria
Conso
One Bill
SAP
Revenue
Reporting
EDW
Business Intelligence
Reports/GARS
WebFocus
BCV
DB Copy
Left Turn Mgr
Order Completion
Cons. Portal Business Portal
Corporate Access
Network
Prepaid (ALU)
CDMA/HSPA
OCS(Openet)
LDAP
LTE
AAA
DEALERS
RETAIL
CSR/ASR WORLD / CFR Viewer
Report Listing
CALL
CENTERS
CSRs
CUSTOMER ACCESS Bills
TrueComp (Callidus)
Commissioning
Scorecard
Dashboard
Extracts
Informatica
Recon Tool
Reconciliation
MAX
Pagers
Single View
CRM
IVR
Call routing
EPC
Enterprise Product Catalog
IDR
Online Bill Manager
CMT
Campaigns
MoneyMap
cVidya RA
SS
Self-Service Back-End
EDIRIS
Payments
Printshop
OneRT
Recom.Offers
ECT
Email Capture
HEAT
Trouble Ticket
BM Portal
Forms
One Source
Content Mngmt
MCO
Mass Changes
IRIS
Equifax
Credit Eval.
IP APPs
PPD PAD
Emergis
CC Payment
RMS
Equipment Mgmt
OEM/ SIMS
Inventory
SAS
BI Tool
API - Link
T
o A
PI
- L
ink
E-Glue
OPS
ARIBA
Empl.Payments
CLEARING
Roaming
MPS
Rating
About 20% of the Customer Systems in a large Telco
Mosaix
Dialer
SMG/VMC
WNP
PLM
Notifications
Corp.sec.
Hot List DB
NM1
Database
DNC
Do not call
FraudX
Fraud
CSM
Interfaces
Billing
Bill Cycles
NM
Number Mgmt
CMS
© 2013 WGSigma Software, Inc. | Confidential 4
Create a collective knowledge of the
operation where processing decisions are
holistically optimized
….Patients
….Customers
…Subscribers
….Account Holders
Real Time Intelligence for Large Scale Environments
The Goal:
03.15.10 © 2013 WGSigma Software, Inc. | Confidential
Closed Loop Event – Condition – Action
Events Decision Engine
Container Container
Actions
Application Server – Elastic Data Grid Based
“Sesame” Performance cache
RDF
Triple Store DB
Event
Ingestion
Scheduled
Events
Semantic
Inference
Engine
and
Business
Rules Bayesian
Belief
Network
Events
Operational Systems Any operational system
Data Sources Any internal or external system
Structured or Unstructured
Event Collector
X Z X
Integration
Framework
Y
Network Business Intelligence
Other
System Other
System Other
Systems
Other
System Other
Systems
5
03.15.10 © 2013 WGSigma Software, Inc. | Confidential
Inferencing
Inference
Engine
and
Business
Rules Bayesian
Belief
Network
Semantic Concept Model
Defines concepts and meaning
Defines relationships between
concepts
Semantic model capabilities
reasoning
The model contains the relationship between
concepts including the dependencies
The model defines the concepts including the high
level business concepts
Semantic Concept Model Machine learning trained models
BBN probabilistic model
Recommendation model
Other Machine learning
algorithms
When a concept is dependent on “machine learned”
information the inference engine manages the invocation and
timing of interfacing
Machine learning
When an event occurs the event handler rule fires for that event
Evaluates the event message
Evaluates the existing ontology
Determines which semantic instances to create or update
When any data changes, the inference engine fires in a “When -
Then” style of computing, updating all “Automatic” concepts.
Custom concept rules are fired if necessary. This creates a chain
of updates
When a “on demand” concept is needed the inference engine
finds and computes all of the dependant concepts
Inference
Event Create semantic concepts from events
Concept Define a custom business
concept A set of custom rules actioning business
policy
Action
Business policy Rules
6
03.15.10 © 2013 WGSigma Software, Inc. | Confidential
Use Case: Customer Interaction Prediction and Best Action
Chronology of events
Interactions
Bills
Orders
Payments
Collections
Charge dispute
Individual
Customer
Pay instructions
Device Activated
Device heartbeat
Subscriptions
Device changes
Transformed into a connected
graph of business concepts
Subjective "good payer"
Patterns "always pays 2 days late"
Trends “improving payer"
Geospatial “within 5 miles of the tower"
Time “within 5 minutes of an outage"
Probability “probably will call about the bill"
Absence of occurrence “missed payment"
Relationship between " friend of a friend"
Events collected in
real time Plan Overage
Bill greater than last month
Prorates
Roaming charges
Third Party Charges
Abnormal fee
Rate increases
Charge dispute
First bill
Past due amount
Pay bill
Customer Cancellation
Reactivation
Device activation
Device education
Device exchange
Device Lost
Device not working
Device resume
Service data not working
Service text not working
Service voice not working
Subscription cancellation
Wrong plan
Predictions & Actions
7
03.15.10 © 2013 WGSigma Software, Inc. | Confidential
• Use in memory elastic data grid
• New hardware nodes can be added and removed dynamically
• Data is replicated to one other node (no single point of failure)
• If a server fails, the backup becomes the primary seamlessly
• Partition grid across objects
Scalability & Resilience
8
• Graphs are partitioned across multiple triple store instances
• A single index instance defines the location of each graph
• Historical data is subdivided into graphs by timeframe (e.g. quarters)
03.15.10 © 2013 WGSigma Software, Inc. | Confidential
Make it Business Friendly
9
Search Map RELATIONSHIP MAP DEPENDENCY MAP TAG PATH Filter Map Library
EVENTS ENTITIES ACTIONS PROCEDURES
Customer
Acquisition Date
Status
Status History
Customer Accounts
Customer Bill
Customer Devices
Customer Action
Entity Name
Composer Demo Application
CONSTANTS +
BUSINESS ENTITIES & CONCEPTS +
Type
Call Frequency
Customer
Customer Accounts
Customer Bill
Customer Problems
Interactions
Subscriptions
Acquisition Date
Status
Status History
Type
Call Frequency
Churn Propensity
Collection Risk
Payment Pattern
+
Customer -
+
- Churn Propensity
Collection Risk
Payment Pattern
Customer Payment Pattern
Customer Payment Pattern
EDIT DEPLOY VIEW
LAST MODIFIED: 02/12/2010 MODIFIED BY: Greg Sloan VERSION: 1.3
DESCRIPTION: Determine the payment pattern for the customer by evaluating payments determining good payer, bad payer, worsening payer or improving payer
2. If all bills have Payment Timeliness
then is good payer Payment Pattern
COMPUTED:Automatically
Find all of the 1. Previous Bills Within 6 months
equal to early
3. If all bills have Payment Timeliness
then is bad payer Payment Pattern
equal to late otherwise
4. If at least 75% Earlier Bills
late
of otherwise are early or on time
and later bills are then is worsening payer Payment Pattern
5. If At least 50% Earlier Bills
On time or early
of Late
and all later bills are then is Improving payer Payment Pattern
are
Customer
Description. The role played by an Individual or Organization in a business relationship with the
service provider in which they intend to buy, buy, or receive products or services from the service p
Customer Problems
03.15.10 © 2013 WGSigma Software, Inc. | Confidential
Technology Stack
Technology Purpose
TopBraid Composer Ontology Modeling
Java Event Pipeline, Decision Engine, Integrations
R Statistical and Predictive Modeling
ETL Event ingestion - Extract, Transform, Mapping Relational to Graph
Red Hat
BPMS & Drools
Forward and backward chaining rules, process definition and
control
Red Hat
HornetQ
JMS Queue persistent multi-threaded input and output (publish-
subscribe)
Gigaspaces Data Grid
Apache UIMA + Solr Natural Language Processing, Semantic Searches, Content
Analytics
Norsys Netica BBN
Franz Allegrograph Triple Store (RDFS)
Apache Cassandra Time series data store 10
Recommended