2. Modeling of Brain Functions
unit and connection
in the interpretive network
unit and connection
in the gating network
unit and connection
in the top-down bias network
Layer I + 1
Layer I
Layer I +1
3. The Power of UsingArtificial Intelligence
• Main Types of Machine Learning
• Unsupervised learning: discovering hidden
properties of data
• Supervised learning: classifying new data from
known properties
• Reinforcement learning: making the best decisions
now to maximize long-term reward
4. Why useArtificial Intelligence
• Optimize decision making process
• Increase reliability of predicting future outcomes
• Generate actionable information (as opposed to
more diagnostic data)
• Automate analysis
• Improve business performance
5. Well known applications
Unsupervised learning: Google search, Amazon
”if you like X…”, Netflix “Top picks for you”
Supervised learning: BoA automatic fraud
detection, Apple Siri, Microsoft WhisperID
Reinforcement learning: Deep Blue, IBM
Watson, Geico instant online quotes
2017 PACE CONVENTION & EXPO | APRIL 2-5, 2017 | TAMPA, FL
6. A.I. needed in Call Centers
Tsunami of data in Call Centers
Call Volume – billions of calls made per day
Metadata – terabytes of information generated
per day
Business Outcomes – defined and discovered
daily
How does it all Inter-relate?
How can humans analyze all of this data fast and
efficiently?
7. Artificial Intelligence in Call Centers
Visibility and Insight sorely needed in customer/agent
conversation
Big data <==> huge number of dimensions to base
decisions
Advanced Digital Signal Processing can create
features directly from unstructured customer/agent
conversation
End result: a decision-making policy to take in new
data (e.g. calls) and associate them with anticipated
business outcome
9. Other Analytics
Script
Adherence
Business
Performance
Speech analytics works well for script adherence &
compliance
… but struggles to capture meaning or
feelings behind the words
Script adherence addresses compliance issues
… but doesn’t solve for business performance
Speech to text – Apply NLP – Present to analysts - Decipher results – Measure success – Repeat trial
10. A.I. Applications in Call Centers
Artificial Intelligent
Predictive Models
0 0.2 0.4 0.6
Emotion
strength
Neutral
Boredom
Joy
Sadness
Emotions found to have the single greatest
impact on customer decisions & customer
experience
12. Data Fusion
Applying structure to unstructured data
Link data and identify relationships across
multiple sources
Identify descriptive features
13. Goal-based Predictive Analytics
Phone Call
is Placed
Features
Extracted383
Behavioral
Predictions
Made
• Customers retained
• Agents improved
• Business grows
14. Predictive Model building process
• Understand the operation and business objective
• Clean and import data
• Perform exploratory data analysis
• Extract relevant features
• Choose experiment parameters
• Train and validate classifier
• Interpret results
• Refine and select optimal model
• Automate application
19. Managing Quality
”Quality” is notoriously
difficult to define &
measure
Applying Big Data
techniques, able to
model based on
company specific culture
20. Agent Performance Models
Help identify who is in the most urgent need of
coaching and training
Create an objective measurement of call quality across
agents
Allow us to tie agent incentives to an objective call
quality metric
Assist with boot camp retraining and remediation
21. Agent Retention
Intervene with
productive agents at
high risk of leaving
Combining risk
assessments with
business performance
allows the user to
decide what best suits
the company’s needs
22. New hire Models+
Identify emotional and behavioral attributes of successful
and unsuccessful hires
Form effectiveness assessments of each applicant by
augmenting existing test data with recorded audio
Recorded audio is a mix of structured interview questions
and scenario based role play that can be compared with
scoring criteria
Provide risk factors to assist in decision making process
24. Leverage Existing Data & Emotions
At last it is now possible to capture the feelings and
emotions of prospects, customers and agents using
Artificial Intelligence
No longer need to rely solely on human
observation, subjective opinions and random
sampling
Artificial Intelligence can tell you how customers
and agents are feeling before they act so you can
intervene with intention