BUILDING AI SOLUTIONS @ SCALE
Dr. Arati DeoManaging DirectorAI Practice Lead, India
Copyright © 2018 Accenture All rights reserved. 2
WE ARE IN AN UNPRECEDENTED PERIOD OF TECHNOLOGY INNOVATION
Mainframe Client-Server and PCs
Web 1.0 eCommerce
Web 2.0, Cloud, Mobile
Big Data, Analytics, Visualization
IoT and Smart Machines
Artificial Intelligence
Quantum Computing
Quantum Computing
1950 Turing Test
Artificial Intelligence
Mainframe
Client Server and PCs
Web 2.0, Cloud, Mobile
IoT and Smart Machines
Big Data, Analytics, Visualization
Web 1.0 Ecommerce
Copyright © 2018 Accenture All rights reserved. 3
WHAT IS ARTIFICIAL INTELLIGENCE (AI)?More natural interactions between people and machines
Multiple Technologies that enable computers to
• Sense (e.g., computer vision, audio processing or sensor processing)
• Comprehend (e.g., natural language processingor knowledge representation)
• Act (e.g., inference engines, predictions or expert systems)
• Learn and self-tune (e.g., machine learning, deep learning)
Expert Systems
Computer Vision
Inference Engines
Machine Learning
Robotic Process
Automation
Deep Learning
Sensor Processing
Knowledge Repre-
sentationMini Bots
EmotionRecognition
Gesture Recognition
Ontologies
Neural Networks
BiometricsNatural
Language Processing
Video Analytics
Copyright © 2018 Accenture All rights reserved. 4
WHAT’S DIFFERENT ABOUT AI SOLUTIONS
AI Development
Lifecyle
Business Requirement to AI
Problem Defn.
Model Development
Production Data Pipeline
Ongoing monitoring & evaluation
Data Collection
Model Deployment
Heavy Data Dependence
Probabilistic Solutions
More complex to debug and
support
New, less-understood technology
Choice of solution is non-unique
5
AI@SCALE SOLUTION – KEY CHARACTERISTICS
Proven ROIVolume
Reliable
Repeatable Evolutionary
Operating on volumesbeyond pilot/prototype
ROI proven on production volumes
Engineered for reliableoperation and quality
Repeatable methodology for other
similar solutions
Enables evolution to increasing automation
Copyright © 2018 Accenture All rights reserved. 6
BUILDING AI@SCALE – 1/5 – BUSINESS-VALUE FOCUSED APPROACH
#1
BusinessValue-focused
• Strong Industry and Technology partnership to experiment and evolve the solution– Conversion of business problem to the right AI problem
– Right use of the AI outputs in downstream processes
– Ongoing ROI measurement
• Maintain a problem-first approach, find the simplest solution– Leverage human-computer interactions to simplify
Copyright © 2018 Accenture All rights reserved. 7
PROCESS AUTOMATION - QUOTE AMENDMENT (HIGH LEVEL FLOW)
Systems
Sales Rep
Advisor
Advisor asks if the seller wants to create an alternative quotes with increased Service Levele.g. - “Would you like to create alternative quotes with increased service level?”
Create Alternative quotes
Navigates to ‘Deal Info’ tab on Opportunity with Quote
AI Engine
SFDC
SAP
For making changes to an individual quote advisor redirects seller to [SSQ Link]
No
Advisor creates alternate quotes for all the quotes under consideration
Create alternate quotes for all these
and change
SL?
Advisor identifies the opportunity related to the account and also recognizes the quotes which have identical SLs and may need same SL upgrade
Advisor recommends the next (logical) increased level of service based on current SL
Advisor adds the new Service Level to all the quotes and confirms the creation of alternate quotes
Advisor moves to the next change,
if applicable
Advisor recommends the Service Level to be added or lists all the Service Levels to choose from
Seller feeds in the Service Level
Quotes are on SFDC opportunity, viewable to the seller through the UI. Changes are made in the Service Level field.
Yes
Seller confirms
Copyright © 2018 Accenture All rights reserved. 8
INDUSTRY AND TECHNOLOGY PARTNERSHIP THROUGHOUT DEVELOPMENT LIFECYCLE
AI Development
Lifecyle
Business Requirement to AI
Problem Defn.
Model Development
Production Data Pipeline
Ongoing monitoring & evaluation
Data Collection
Model Deployment
Copyright © 2018 Accenture All rights reserved. 10
BUILDING AI@SCALE – 2/5 – DIVERSE ECOSYSTEM FOR AI DEVELOPMENT
#2
Diverse Ecosystem
• Diverse skills employed for development
• Partnerships with academia to deliver thought leadership and innovative solutions– Community approach to team development
• Relationships with key technology partners and start-ups
Copyright © 2018 Accenture All rights reserved. 11
DIVERSE & EVOLVING ROLES FOR AI DEVELOPMENT
SHAREDDEDICATED
DEDICATED & SHARED
Designs and implements the new user interfaces and visualization mediums
needed for AI systems
Visualization Designer/Engineer
Sustain AI assets and platforms; recruit AI talent; create training
programs for upskilling for AI skills
AI DeliveryLeads
Data Engineer
Designs and builds the data pipelines for data processing and
ingestion to the AI system
Business Analyst
Contributes in converting business problem to relevant AI
problem; user acceptance testing
Builds overall framework for AI deployment, designs the architecture
for AI production setup
Ai ArchitectAI/ML Scientist
Develops model design; conducts various experiments to determine best
model: creates final best model
AI/ML Engineer
Implements code and services to process data and generate desired outputs from final model in
production system
Copyright © 2018 Accenture All rights reserved. 12
AI Ecosystem and partnerships
AI technology radars Global alliances
Accenture Investment
$600 million - including 185+ Related Patents and Applications
Total # of practitioners
2200+
University / Research partnerships
• Stanford• MIT
• DFKI • Turing
institute
Global Labs
• UK Liquid Studio
• Dublin AI COE• Tech Labs Palo
Alto
• Tech Labs Sophia
• IDC Innovation Centre
Copyright © 2018 Accenture All rights reserved. 13
BUILDING AI@SCALE – 3/5 – USE TECH-AGNOSTIC FRAMEWORKS
• Develop integrated solutions that leverage best-of-breed products
• Create processes to continuously evaluate new technology and software solutions
#3
Technology-agnostic
Copyright © 2018 Accenture All rights reserved. 14
Technology-agnostic Architecture Framework for AI
Custom Services
Sources
Doc Types
Channels
Te
ch
nic
al S
erv
ice
s
AI T
ra
inin
g
System
Enterprise
Systems
Business Services
Data Services
AI Interface Services
AI Services
AI Platforms/Libraries
ACIP
Fraud Detection
Service
Feedback Service
Answer Service
Orchestration
Services
Natural Language
Processing
Speech
Recognition
Image
Processing
Data
Insights……
AI Platforms (Azure, IBM Watson, …) AI Libraries (Python, R …)
Data
Transformation
Data
MaskingOCR
Persistence
ServicesData Lake
Social Media
Enterprise
Systems
Enterprise
Free Text
XML
XML
</>
Email/Fax/Voice Mail
Health Advisor
Re-admission Service
Cross Industry
Services
Smart Metering &
Outage Resolution
Predictive Network
Operations Service
Product Sale
Prediction Service
Visualization Services
Copyright © 2018 Accenture All rights reserved. 15
Solution
Business Context
CASE STUDY – COMBINATION OF MULTIPLE TECHNOLOGIES
• The Client (an APAC Broadcasting Network) has requirement to identify new telecom poles as part of an acquisition process• The project team has 18 resources are working to identify the poles manually. This incurs heavy cost and is also time consuming.• Each pole is located in Google Maps manually and the type of pole (Utility Pole or Telecom Pole) is visually identified• Sample Images
Benefits
• Automatic identification of different types of poles (despite differences in backgrounds, angles, resolutions, distances)
• Over 35% automation of image classification problem for pole types (and significantly lower manual intervention)
• The images are downloaded using Google Street API• Automated image identification using Pattern Matching with
Google’s Tensor Flow Machine Learning Framework
• The Inception Model of Tensor Flow is based on Convolutional Neural Networks
• Final layers of inception model are re-trained with pole images to classify poles based on patterns
Telecom Pole Utility Pole
Copyright © 2018 Accenture All rights reserved. 16
BUILDING AI@SCALE – 4/5 – THINK INDUSTRIAL SOLUTION (NOT PROTOTYPE!)
• Data and technology platforms built to scale– Flexible and Robust Data pipeline is crucial
• Industrialized services and cloud capabilities optimized for AI delivery
• Repeatable and sustainable development processes
#4
IndustrializedDevelopment
REPEATABLE DEVELOPMENT METHODOLOGY
Program Management
Project Management
Change Enablement
Service Introduction
De
plo
y R
ele
as
e
Se
rv
ice
De
liv
ery
AI Value
Targeting
Initiation
AI
Pil
ot
Pil
ot
De
plo
ym
en
t
Pil
ot
Op
era
tio
n
Solutioning
Co
nfi
rm
Co
nfi
rm
Project Execution
Sprint
Management
Release 1..n
Copyright © 2018 Accenture All rights reserved. 18
Fraud Detection Solution at Scale
Courtesy: FICO Fraud Manager
Copyright © 2018 Accenture All rights reserved. 19
BUILDING AI@SCALE – 5/5 – STRATEGIC CHANGE MANAGEMENT INCLUDED
• Evolve the level of automation through incremental steps
• Design approach that puts humans at the center of the solution– Incorporate change management plans in the development plan
• Involve the end users in design and measurement
#5
Strategic Change Management
Copyright © 2018 Accenture All rights reserved. 20
AUTOMATION JOURNEY
Give Me the Facts
Give Me a Suggestion
Do This Task for Me
Take Responsibility Until
I Tell You Otherwise
Take Responsibility andDon't Let Me or Anyone Else Mess It Up
General Information
Specific Information
Advisory Guidance
Opt-In Automation
Overrideable Automation
Nonoptional Automation
Help Me as I Go
Courtesy: Gartner
Copyright © 2018 Accenture All rights reserved. 21
AUTOMATION THRESHOLD TO OPTIMIZE BUSINESS VALUE
Source: Internet
Automation Threshold
Copyright © 2018 Accenture All rights reserved. 22
BUILDING AI@SCALE – RECAP
#1 #2 #3 #4 #5
BusinessValue-focused
Diverse Ecosystem
Technology-agnostic
IndustrializedDevelopment
Strategic Change Management
Volume Proven ROI Reliable Repeat-
ableEvolution
-ary AI @ SCALE
Thank You