innovation in action - ncr...ncr innovation conference 2017: confidential 29 customer support use...
TRANSCRIPT
INNOVATION CONFERENCE 2017
INNOVATION IN ACTION Suzan Szollar & Marshall Yuan DI Labs
Please use the Innovation Conference Event
App to check-in to this session #S667
NCR Innovation Conference 2017: Confidential
Ram Ridgeway - Director of Retail Support at Wright-Patt Credit
Union
Mariano Maluf – Director of Disruptive Innovation and Platforms,
Financial HW Solutions
Suzan Szollar – DI Labs Product Management
Marshall Yuan – DI Labs Product Management
Speakers
3NCR Innovation Conference 2017: Confidential
Why AI Now? (And… What Is It?) Transforming Physical Space with AI (Smart Everything)
Digitally connected physical space (Mariano)
Vision project – retail example (Shuki)
Transforming Digital Experience with AI Conversational banking (Marshall & Ram)
Building a Foundation Data and Data Science
Visualizing Insights & Taking Action
Agenda
1966 Joseph Weizenbaum created
Eliza, a chatterbot. Meant as a parody
of therapy.
“What I had not realized is that
extremely short exposures to a
relatively simple computer program
could induce powerful delusional
thinking in quite normal people”
Joseph W.
Were We Always Like This?
7
Tech’s Role in Connections and Relationships
Today – tech provides a platform (facebook), or a channel (watch, bots, voice) for communication and information
In an AI ecosystem, tech becomes a responsive, predictive fabric that can enable stronger connections to better outcomes
We used to talk
on the phone Now we
talk to our
phone
AI – What Is It?
General AI
Machine Learning
Classic
Machine
Learning
Neural
Networks
General AI: can do anything equal or better
than a human
Task Based AI: equal or better than a
human at a task
Why Now? • Computing horsepower – GPUs, Quantum
• Big data, real time
• Off the shelf frameworks, new algorithms
Both can uncover unknown
and unexpected
relationships…
Classic Neural• Told “what” to look for
• Needs human intuition
• Doesn’t typically improve
with more data
• Learns by example, shown
“how,” keeps learning
• Shows ingenuity &
invention
Classic vs Neural Network Machine LearningAn example
My daughter wants a small dog.
Can I get her muffin instead?
Classic ML Example – if the “rules” fail, the accuracy of
the model will not improve
Neural Network Example - “Generative Adversarial
Network,” Dueling neural networks:
• One tries to classify dogs
• The other tries to trick the classifier
• Makes the classifier more accurate than a human
The Promise of an AI Ecosystem (an example)Ingenuity, Invention, Inspiration
The Past
Beautiful public spaces
Everything “on the way”
The Present
Transactional spaces
Everything “out of the way”
The Promise of the Future
Experiential “living” spaces
Transportation ingenuity
Interactive signage Manager & Staff AppTeller AutomationTeller Assistant
Easy Authentication
Smart Staff
Mobile Payments
Smart Lockers
Consulting
Online Banking Assisted Service
Teller+Customer Display Active WaitingATM- Interface
BranchMobility
Stage Director
Location based services
AI Smart Edge Ecosystem
Taking Shape - Some Examples
Robo BranchesMulti-Tenant Kiosks
Smart Lockers
To learn more: 3:15PM Wednesday Session – IoT and the Digital Branch Ecosystem
The Promise of an AI Ecosystem
The Past
Banking face-to-face
A human interaction
The Present
Convenience, efficiency
Transactional tech interactions
The Promise of the Future
Connected experiences
Collaborative interactions...
Messaging could become the new mode for business to engage with consumers
Image from TechCrunch “Forget Apps, Now the Bots Take Over”
Fintech Startups are pioneering conversation interfaces to engage users… and FIs
21
2. Standalone consumer fintech
chatbots
1. Fintech exploring chat
interface capability
3. Fintech chatbots moving to
white label for FIs
Abe.aiOlivia Penny
Cleo Finn.ai
Larger financial institutions are following suit with both internally developed solutions…
Bank of America announces
“Erica” to be integrated in mobile
app (not live yet)
Capital One innovation team
goes live with SMS-only
conversation banking
…and with solutions developed by partner capabilities
23
Ally Assist• Personetics
USAA Virtual Assistant• Nuance
• Clinc
Kai• Kasisto
NCR Innovation Conference 2017: Confidential
Will users be willing to talk to a bot? --- Yes
“Call us now”600 clicks/week
FAQ Chatbot500 clicks/first week
WPCU Chatbot Test
NCR Innovation Conference 2017: Confidential 24
Will the bot be able to deflect calls from call center? --- Yes
NCR Innovation Conference 2017: Confidential 25
500 sessions
200 questions
26 calls to call center
Chatbot Call Driver Funnel
87% of those who asked a question don’t call in!
How much work could a bot help automate for service teams?
17%
17%
16%
50%
1. Check Balance CBS for user txninfo
2. Routing#/ Account#
Workflow to set up direct deposit/bill pay
3. Branch hours/ locations
Branchscheduling integration
4. Depositavailability
Core/RDCintegration
5. Reg D limits Core integration
6. Lost Credit Card
Card management
7. Pin Reset CAS?
Initial accuracy for correctly
answered questions
1
NCR Innovation Conference 2017: Confidential
How much more could we improve?
17%
17%
16%
50%
1. Answer incorrectly matched
• Retrain to correct existing
answers
2. Answer not available, but
frequently asked
• Create new answer
Improve accuracy for
incorrectly answered questions2
NCR Innovation Conference 2017: Confidential 27
What do we do with the long tail?
17%
17%
16%
50%
Answer not available, and
infrequently asked
Send to call center or Live Person
Seamless transfer
to live help3
NCR Innovation Conference 2017: Confidential 28
Data we get from conversations:
Familial relationships: “How do I open up a credit card account for my husband. He is not
on my current Wright paTT bank account”
“Can I transfer money from one account to my sons account”
X-sell opportunities: “Can I add a bank account to my profile to pay my loans with?”
“What is my business daily transaction limit?”
“How do I accept popmoney from the app”
Opportunity to detect life stages, demographics, family relationships to provide advice and x-sell at the right moment
NCR Innovation Conference 2017: Confidential 29
Customer Support use case helped answer questions and did
not drive additional calls
Training takes a lot of work
Natural Language Processing needs deeper expertise
Use cases around PFM insights still to be proven
What else did we learn
30NCR Innovation Conference 2017: Confidential
AI - A Many Layered Cake
Data Storage
Computational
Horsepower
Data Science
Platform
Frameworks &
Models
User Experience
“Channel”
User Experience Immersive Gamified Interactive/Chatty Behavioral/Nudgery ….
VR AR Chatbot Voice Mobile …
TensorFlow Caffe PyTorch Keras SciKit Theano….
Jupyter… Zeppelin Python…Java…Scala…R… Apache
Spark…
CPU GPU Quantum…
Permission Privacy Compliance
Feed Insights into Actions via DI or FI Products & Services Logic
Product or Service
Data Sources Internal (Prod Systems, Distributed DBs…) External (Customer, 3rd Party)
Data Lake API Access Distributed DBs…
Data Access
Use Cases
Data Science – Big Picture & Out The Gate
Vision: AI infused products and services that create strong customer
connections and high impact outcomes
Hypothesis: Applying machine learning to customer problems will enable us to anticipate and respond to customer needs better than ever before
Problem: FI-s need to do more with what they have to grow revenue. 60% - 75% of FI income is from interest (credit cards, loans).
Opportunity: FI-s can get new income, differentiate themselves, and grow a bigger credit worthy customer base, by identifying and helping the right customers shift spending behavior to achieve their financial goals.
POC Goals:Understand where machine learning surfaces valuable, actionable insightsUnderstand what data we need to get meaningful resultsShow results - insights in action (via VR/AR, Alexa) to you!
Data Science – First Steps
Questions Customer Problems
Analytical Strategy
Gather &
Clean Data
Pre-processing
Explore & Experiment with Models
Evaluate
Visualize
Focus on Use Cases That:
• Grow interest income and help
customers achieve financial goals
• Detect engagement shifts, early
signals of churn
• Customer lifecycle/lifestyle
opportunities
Actions
New products/services
New Experiences
Prioritization Criteria:
o Business impact
o Mutual value for FI
and customer
o Differentiation
o Actionable
Right strategy?
Good data?
Right data?
Enough data?
What Data: Batch FI file
3 years of debit card and other non credit-card transactions, bill payments on 3rd party CC
No balance, demographics, credit score, credit card transactions, digital banking logging…
Use Case: Credit Card Switching Behavior
What We Did:
Extracted features and labeled data
Clustering on credit card switching behaviors
Predict switching using deep learning model
What We Learned:
Interesting grouping of credit card usage behavior, limited insight into why
86% accuracy on switching prediction
We need better data!
Our 1st Time – 3rd Party Credit Card Switching Behavior
36
3rd Party CC Behavior Groups
StableVisa fansLow: income, spend, loanHigh: Transfers
From shifty to stableDiscover fans with someswitchingLow: loanMedium: income, spend
Big switchersDon't like DiscoverHeavy debit usersHigh: income,expenditure, loans
StableNo favoritesLow: spendMedium: income, loan,savings
Our 2nd Time – Increase Credit Card Usage
37
What Data: Batch FI file 6 months CC, Debit, Credit Card Transactions
Rich demographics
Use Case: Increase FI credit card usage?
Approach:
Understand strongest feature tied to increasing usage
Is it actionable?
Results will be shared at Conference session
What We Did: Identified behavioral groups
Used bag learner approach to classification
Ran attention mechanism to discover top feature that drove behavior
What We Learned: 90% of time is cleaning data
TBD
Insight Product + Experience
Smart Digital Interaction
Human Connection
From Insight to Action
NCR Innovation Conference 2017: Confidential 38
On credit you’ll cross
30% of your limit, it’ll
hurt your credit score.
Purchase on debit?
You have enough and
can still pay your bills
I have upcoming bills,
do I have enough?
CB
CB
Hi Phoebe,
you’re on track
for a car loan,
what car do you
want?
Opportunity:
No Credit
Needs Car
How can I
afford a new
car?
I’ll help you
improve your
credit score with
Credit Buddy
You can earn 10 points toward the
Innovation Game right now!
Please go to “Schedule” in the
mobile app, find this session, and
take a short survey to provide
feedback.
Surveys will remain open for 2
hours.
Check in code is S667
40NCR Innovation Conference 2017: Confidential
Please give us your feedback!
INNOVATION CONFERENCE 2017
THANK YOUDI Labs Innovation In Action
Please use theInnovation ConferenceEvent App to check-in to this session
NCR Innovation Conference 2017: Confidential
AI – What Makes It Special Now?
AlphaGo (Go) vs Deep Blue (Chess)
After just the first two moves of a Chess game, there are 400 possible next moves. In Go, there are ~130,000…
You can’t compute every move
AlphaGo trained on 30 million human moves
Then learned by playing itself
Then it invented new moves
AlphaGo: “The top Go players couldn't get enough of its unique and sometimes transcendent style of play”
Augmented reality in the branchPowered by DI analytics in the cloud
43NCR Confidential - Internal Use Only
https://www.dropbox.com/s/gw9yi20ghk65o
es/DI%20AR%20demo.m4v?dl=0