role of artificial intelligence in business big and ......confidence level usage churn random forest...
TRANSCRIPT
ROLE OF ARTIFICIAL INTELLIGENCE IN BUSINESS
AND TECHNOLOGY. A PROJECTION IN 2030+
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
AI
Big
Data
IIoT
Agenda
o Global Problem
o Define AI
o AI for Business
o AI Framework for Enterprise
o AI Applied in Industry
o Benefits
o Impacts
o Demo
o Q&A
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Objectives
Bringing enterprise view to AI and demonstrate practical business
problem-solving techniques and approach through proven industry
solution.
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Discuss business benefits and impact of AI on people, business,
and technology
Beyond
Theory
We are a ground-breaking cloud based “end-to-end” AI platform
for mainstream businesses to re-merge their businesses to the
digital world. Our AI platform brings digital business live.
Our AI platform has 100’s of industry specific pre-trained deep
learning models to save your time, costs, and risks.
Health, Energy & Consumer
Our AI platform is built on enterprise technologies SAP HANA,
Oracle, Microsoft, HADOOP, Spark, Alluxio and many other open
source technologies.
Enterprise AI is Our Single Focus
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Why Mainstream Business Truly Never
Realizes Measurable Business Value of
Artificial Intelligence (AI)?
Global Problem
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Global Problem (Cont’d.)
The industry for
Artificial
Intelligence,
Machine Learning
or Deep Learning
has been
perceived as an
academic concept
or scientific
research
solutions.
Business truly
never realizes
measurable
business value of
AI.
ProblemThe Real Challenges are:
Lack of skills and
expertise to
translate
complex industry
problems into AI
solutions using
machine learning
and data
capabilities.
No inter-operable
advanced
computing
platform to make
AI, IIoT and big
data
technologies
work together.
Companies think
AI is complex,
time-consuming
and carries high
risk to implement
and adopt.
Lack of
Skills and
Expertise
Lack of
System
Interoperability
No
Business
Alignment
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Define Artificial Intelligence (AI)
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Artificial Intelligence
Problem Solving
Data Analysis & Learning
Decision Making & Perceiving
Act like a human
Human
Intelligence
Define AI
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Take an action and an reaction
Prediction and others
Act Like
a Human
Data Analysis
Decision Making
Take an Action
Prediction
Perceiving
Problem Solving
Others
Narrow
Intelligence
General
Intelligence
Solves many problems
Strong AI
Human-level AI –
Understands situations
and circumstances and
then acts like a human
Define AI (Cont’d.)
Solve a Problem
Focused - Narrowly
Defined Task
Self-driving Cars
Limited Speech
Recognition
Disease
Diagnosis
Super
Intelligence
Constantly learning
and improving itself
could be unstoppable
Current Next Future
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Where is AI headed?
Imitate Human
Level
Act at Human
Level
Beyond Human
Level
Artificial Intelligence
Machine
Learning
Deep
Learning
NLP
(Natural
Language
Processing)
Robotics
Define AI (Cont’d.)
Computer
Vision
AI Enabled Products and Services
…
How to implementing AI?
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Predict patient re-admission rate
Forecast product demand and inventory levels
Artificial Intelligence For Enterprise
+ Revenue - Cost+ Quality
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
AI for Enterprise - Business Needs
P&LChurn
Model
Segmentation
ModelUpsell
Model
Financial
Model
o Predict customer usage (GB) churn, when and why
o Micro segment the churn population
o Predict the best upselling plan for each micro segment
o Predict and forecast financials
o Prevent call drops and increase call quality
Cash
Flow
Integrated AI Models
Business Needs
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Micro
Segmentation Target Marketing
& Upsell
Financial
Forecast
Predict
Churn
Who
K-Means
ClusteringLDA
Logistics
Regression
Why
When
Confidence
Level
Usage
Churn
Random
Forest
Low Data
Usage
Secondary
Connection
High
Customer
Service Call
Confidence
Level
AI for Enterprise-Integrated AI Model for Customer & Finance
Plan A
Plan C
Plan B
Confidence
Level
5% Revenue
Increase
3% Cost Reduction
7% Revenue
Increase
1% Cost Reduction
3% Revenue
Increase
2% Cost Reduction
Confidence
Level
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
How It Works?
John
Current Plan
Plan A
High Data
and Video
Usage
High
Customer
Service Call
Customer Most
Likely Churn in
Next 30 Days
o Plan B
John will get 15% more data
John will pay 5% more than the
current plan
3% Revenue Increase
2% Cost Reduction
AI
Modelo Plan C
John will get 25% more data
John will pay 8% more than the
current plan
5% Revenue Increase
3% Cost Reduction
o Plan C
John will get 35% more data
John will pay 10% more than
the current plan
7% Revenue Increase
5% Cost Reduction
Recommended Plan
Integrated AI Model for Customer and Finance
99% Confidence Level
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Artificial Intelligence Business Value
Artificial IntelligenceStrategic
Vision
+ Revenue
- Cost
Key Capabilities
Our AI Platform Connects Business Goals and
Objectives to Measurable Outcome.o Prevents call drops
o Real-time root cause analysis
on call drops and call quality
o Increases usage - Data, Video,
Voice and Message
o Prevents churn - GB and Plan
o Revenue targeted marketing
and upsell camping
o Optimizes capital spend and
increases operational
efficiency
o 30% Increase Service Quality
o 15% Increase Revenue
o 25% Reduction Cost
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Technical View
Artificial Intelligence For Enterprise
Problem Statement AI Team & Business
Model Selection AI Team
Technology Development AI Team & IT Team
Data Collection AI Team & Data Team
Data Preparation AI Team & Data Team
Data Integration AI Team & Data Team
Data Provisioning AI Team & Data Team
Model Building AI Team
Deeper Business
ML. DL, NLP, Computer
Vision & Statistical
Data, AI, Cloud, Security
and Infrastructure
Data Engineering &
Business Model
Data Engineering &
Business Model
Data Engineering &
Business Model
Data Engineering &
Business Model
AI, Data Modeling &
Business Model
Deeper Business Skills
Deeper Business and Data
Analysis Skills
Data Analysis/Modeling
Data Analysis/Modeling
Data Analysis/Modeling
Data Analysis/Modeling
Data Analysis/Modeling
Model Training AI Team AI, Data Modeling &
Business Model Data Analysis/Modeling
Model Testing AI Team AI, Data Modeling &
Business Model Data Analysis/Modeling
Model Validation, & Acceptance AI Team & Business AI and Deeper Business Data Analysis/Modeling
Model Deployment in Production AI Team & IT Team AI, Data and Technology Data Analysis/Modeling
Model Tuning AI Team AI, Data Modeling &
Business Model Data Analysis/Modeling
Model Re-Training AI Team AI, Data Modeling &
Business Model Data Analysis/Modeling
Model Re-Testing AI Team AI, Data Modeling &
Business Model Data Analysis/Modeling
Model Re-Deployment AI Team AI, Data Modeling &
Business Model Data Analysis/Modeling
…Continue to Monitor
Itera
tive P
rocess
Ownership & Accountability Skill Requirement
- Primary
Skill Requirement -
Secondary
De
ep
er T
ec
hn
olo
gy S
kills
Re
qu
ired
in th
e R
es
pe
ctiv
e F
ield
AI Framework For Enterprise
AI
Platform
Customer GB
Churn
Customer
Segmentation
Call
Drops
------------------
Prevention
Call
Drops
-------------------
Root Case
Analysis
Microservice Architecture for Artificial Intelligence
Get Data
Services
------------------
Data
Collectors
Set Data
Services
------------------
Data
Preparation
AI
Services
------------------
Model
Building
SAP
Oracle |
Microsoft
Device | Sensor
Opensource
Data
AI Libraries
Pre Trained Model
MLComputer
Vision
DL
Data
Data
Data
Mobile
Entry
Cloud
Gateway
Web
Entry
o Loosely coupled services for global telecom industry to quickly deploy Artificial Intelligence at an enterprise scale
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Microservice Architecture for Artificial Intelligence
Data
Collection
Data
Preparation
AI
Model
Building
Get Structured Data()
Get Unstructured Data()
Get Machine Data()
Get Customer Data()
Get Finance Data()
Get Usage Data()
Get Demographic Data()
Get Call Center Data ()
Get Feedback Data ()
Get SAP HANA Data()
Get SAP IQ Data()
Get NAS File Data()
Interoperable Microservices Catalog (Sample)
Set Data Mapping()
Set Data Integration()
Set Data Conversion()
Set Customer Data()
Set Finance Data()
Set Usage Data()
Set Demographic Data()
Set Call Center Data()
Set Feedback Data()
Set SAP HANA Data()
Set SAP IQ Data()
Set NAS File Data()
Get Customer GB Churn()
Get Segmentation()
Get Call Drops()
Get Random Forecast()
Get Decision Tree()
Get K-means Clustering()
Get Train Model()
Get Test Model ()
Get Predict Model ()
Set Train Data()
Set Test Data()
Set Prediction()
Get Services
Set Services
Get & Set Services
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
AI for Enterprise - Advanced Customer Analytics
o Problem Statement o Model Selection
▪ Usage Churn ▪ Gradient Boosting Decision Tree
▪ Random Forecast
▪ Segmentation & Upsell ▪ K-means Clustering
▪ Latent Dirichlet Allocation (LDA)
o Churn - Who, When and Why
o Financial Forecast
o Target Marketing
o Market Basket
▪ Call Drops ▪ Logistics Regression
▪ Boosted Decision Tree
▪ Neural Networko Call Drop Prevention
o Live Data Streaming & Root Case Analysis
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Node 1
Physical Node
Node 2
Physical Node
Node 3
Physical Node
Node 4
Physical Node
Node 5
Physical Node
Node 6
Physical Node
Node 7
Virtual Node
Node 8
Virtual Node
Node 9
Virtual Node
Node 10
Virtual Node
o YARN
o SPARK
o ALLUXIO
Worker/ Executor
Worker/ Executor
Node Manager
ResourceManager
Worker Worker Worker WorkerWorker Worker Worker
Worker/ Executor
Worker/ Executor
Worker/ Executor
Worker/ Executor
Worker/ Executor
ResourceManager
ResourceManager
ResourceManager
Node Manager
Node Manager
Node Manager
Node Manager
Node Manager
Worker/ Executor
Spark Driver
Spark Driver
WorkerAlluxio Master
Alluxio Master
o HDFS Data Data Data DataData Data Data Data Name Name
AI Platform Architectural (Sample)
SAP IQ
SAP HANA
Structure Data Structure Data
Sensor Data Device Data Machine Data
o Zeppelin Rapid Development
o KAFKA Broker Broker Broker BrokerBroker Broker Broker Broker Broker Broker
o AI Model Python, ML Lib
Python, ML Lib
Python, ML Lib
Python, ML Lib
Python, ML Lib
Python, ML Lib
Python, ML Lib
Python, ML Lib
Python, ML Lib
Python, ML Lib
o Tableau Data Visualization Data Analysis Self-servicing
Unstructured Data
AI Platform Architectural Core Components (Sample)
o SAP & HANA o Data Source
o HDFS o Persistent Data Storage
o KAFKA o Live Data Streaming
o ALLUXIO o In-Memory Data Store - Memory Consolidation
o SPARK o In-Memory Data Processing
o Zeppelin o Rapid Design, Development and Deployment
o Python | Scala o Distributed Programming - Microservices
o AL Model o ML & DL Models - Tensorflow, Caffe & Spark ML
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
KAFAKA
Overall AI Solution Flow
SAP HANA & Oracle
HDFS
ALLUXIO
SPARK
ZEPPELIN
TABLEAU
AIC
OM
PU
TIN
G
EN
GIN
E
PY
TH
ON
&
SC
AL
A
MIC
RO
SE
RV
ICE
S
Tensorflow | Caffe | Spark ML
Devices NAS Files
MIC
RO
SE
RV
ICE
S
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Data Integration - Key Components
o DATA SOURCE
o HDFS
o ALLUXIO
o SPARK
o AI MODEL
o TABLEAU Business Analytics
Self Servicing
Data Analysis
KAFKA
SPARK
SPARK
SPARK JDBC | ODBC
SPARK
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Data Volume and Live Data Streaming
# DATA SETS DATA FREQUENCY DAILY DATA VOLUME
1 o Connection o Every 60 Min o 1 GB
2 o Usage - Data o Every 30 Min o 177 GB
3 o Usage - Voice, Video and Messaging o Every 30 Min o 454 GB
4 o DPI (Deep Packet Inspection) o Every 60 Min o 5.5 TB
5 o Financial (Recharge) o Every 30 Min o 190 MB
6 o Type Allocation Code (TAC) o Weekly Once - Monday o 84 MB
7 o Customer Care (Customer Interaction) o Every 60 Min o 200 MB
8 o HSS (SIM Latched Status) o Once in a Day o 25 GB
9 o OMACP (Device Change Status) o Every 60 Min o 250 MB
10 o LSR (CDR) Data o Every 15 Min o 65 TB
150 TBDaily Data Production
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
How It Works? - Step by Step
o Data
Provisioning
o Data
Preparation
o Model Training
o Usage Churn
o Segmentation &
Upsell
o Data
Visualization
1
2
3
4
5
6
o Stream data from data source ( Device,
Sensor) to HDFS using Kafka
o Load data from data source (SAP HANA,
SAP IQ & Files) to HDFS through Spark
o Load data from HDFS to Alluxio using Spark
Ste
p
o SAP HANA, SAP IQ, NAS File and Device
o HDFS
o Spark
o Alluxio
o Kafka
Ste
pS
tep
Ste
pS
tep
Ste
p
High Level Task Implementation Technology
o Load data from Alluxio to Spark runtime for in-
memory data processing
o Real-time data integration, data mapping and
data conversion
o Spark
o Alluxio
o Kafka
o Python | Scala
o Load training data from Alluxio to Spark
o Feature engineering on the training data to
extract features of customer churn and
customer segmentation
o Alluxio
o Spark
o Zeppelin & Python
o TensorFlow & Caffe
o Run predictions on test data using trained
model
o Gradient Boosting Decision Tree
o Random Forecast
o Store model outcome in SAP HANA table
o TensorFlow
o Spark ML
o Caffe
o Python
o Run predictions on test data using trained
model
o K-means Clustering
o Latent Dirichlet Allocation (LDA)
o Store model outcome in SAP HANA table
o TensorFlow
o Spark ML
o Caffe
o Python
o Financial forecast
o Churn reason
o Target marketing
o Zeppelin
o Tableau
o SAP BusinessObjects
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
360 Degree Customer View
o Customer Experience
o Recommendation System
o Facial Recognition
o Human Recognition
o Image Recognition
o Voice Recognition
Incre
ase
Cu
sto
me
r En
ga
ge
me
nt
o Our AI platform also provides pre-trained models and data capabilities to instantly deliver a consistent customer
experience across all industries, including Retail and CPG
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
A PROJECTION IN 2030+
Impact of AI on Business, People & Technology
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
A PROJECTION IN 2030+
Narrow Intelligence
General Intelligence
Super Intelligence
2020 2025 2030
A PROJECTION IN 2030+
Narrow intelligence may not be a main
focus
More focus on general intelligence
Enterprise adoption of AI will significant
increase
Redefined business processes
AI enabled products and services
AI enabled Robot (not general
robot) to optimize and improve
human intelligence
Workforce reduction and new role
creation
Source : Several Research Papers, Market Study and Industry Prediction
Business Adoption
Business Adoption
Business Adoption
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Beyond Theory
Impact of AI on Business, People & Technology
AI
A PROJECTION IN
2030+
Impact of AI on Business, People and Technology Cont'd.
Business People Technology
Changes in
Operating Model
New Business
Opportunities
Reinvented
Business
Processes
Improved
Productivity
Optimized Time and
Cost
Self-driving Issue
Resolution
Interoperability
New Architectural
Framework
Modernized
Infrastructure
Advanced
Computing
Distributed Data
Processing
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Skill Enhancement
Role Change
Manual vs.
Automation -
Reduction
Competition
Social Values
Ex
am
ple
Reduced C
ost
Impro
ved P
roductiv
ity
Incre
ased R
evenue
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Impact of AI on Business, People and Technology Cont'd.
Traditional Approach AI Driven Approach
Software and Application Development
Problem
Statement
Functional & Technical
Specification
Hardcoded Programming
Logic
Test and Deploy
Continue to Monitor
Enhancement
Problem
Statement
Model Selection &
Feature Engineering
Model Building and
Training
Model Test and Deploy
Continue to Monitor
Model Tuning and
Enhancement
Artific
ial
Inte
lligen
ce
Hu
man
Inte
lligen
ce
A P
RO
JE
CT
ION
IN 2
030+
Opportunity | Benefits
Reinvented Business
Improved Productivity
Increased Revenue
Reduced Cost and Time
Optimized Capital Spend … much more
Impact of AI on Business, People and Technology
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Challenges
Do we need a car when a car comes by itself when we
need?
Do we need a driver's licenses when a car drives by itself?
Do we need highway patrol when a car drives by itself?
What would be the role of DMV?
Whom to insure insurance, who owns the liability when a
car drives by itself? … much more
This is just one example
of AI …
Impact of AI on Business, People and Technology Cont'd.
Challenges
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Do we need a car when a car comes by
itself when we need?
Do we need a driver's licenses when a
car drives by itself?
Do we need highway patrol when a car
drives by itself?
What would be the role of DMV?
Whom to insure insurance, who owns the
liability when a car drives by itself?
Impacts
Societal Impact
People Impact
Business Impact
Operational Impact
Financial Impact
Tech
no
log
ica
l
Ad
van
cem
en
t
Let’s Deeply Understand the Future Ahead of us - A PROJECTION IN 2030+
A P
RO
JE
CT
ION
IN 2
030+
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Financial Processes
Impact of AI on Business, People & Technology
A PROJECTION IN 2030+
Example : Financial Close
Group
(US)
Region 2
(LA)Region 1
(AP)
India Singapore Japan
B/SP&L
Financial Disclosure
Current Process
Process occurs every period
Highly time-consuming
System driven and/or manual processes
Five to ten resources spending time
Key Tasks Performed
Eliminate inter company transactions
Consolidate assets and liabilities
based on the ownership
Intercompany transaction
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
AI Impact to Financial Processes
Implementation Cost
Time
Operating Cost
Productivity
~.5 Million
Two Weeks Every Period
100K
Medium
This is Just
for One
Customer
Reusable Not really
Let’s understand the current implementation cost, time, resource & productivity
Now compare this with global market and see the amount of time, cost and
resource spend on this process© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
AI Impact to Financial Processes Cont'd.
AI Model
Identify and eliminate
intercompany
transaction
Book the difference in
an intercompany
transaction to an
elimination account
Mo
del R
ule
s ( D
ata
Patte
rn)
Build a Model ( ex: Random Forest and
Reinforcement learning - Neural Net)
Train a Model
Test and Deploy Model
Just one AI model solves the
entire world problem
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
AI Impact to Financial Processes Cont'd.
Financial Close
A P
RO
JE
CT
ION
IN 2
030+
What patterns (features) we should look at the data
for intercompany elimination process?
What model we need to implement for intercompany
elimination process?
Do we need machine learning or deep learning or
both for implement intercompany elimination?
Let' s do
some
brainstorming
Mo
del B
uild
ing
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
AI Impact to Financial Processes Cont'd.
© Jothi Periasamy. | Confidential & Proprietary | 03/20/2018
Thank You !Jothi Periasamy
Chief Data Scientist
(916)-296-0228