Optimizing IVR/Speech Using Customer Behavior Intelligence Michael Chavez Vice President Client Services ClickFox, Inc.
Jan 09, 2016
Optimizing IVR/Speech Using Customer Behavior Intelligence
Michael Chavez
Vice President Client Services
ClickFox, Inc.
Agenda• Welcome and Introductions
• The Optimization Problem
• Case study #1o Large Fortune 100 Telco carrier Speech/ IVR system
• Case study #2o State Medicare/Medicaid IVR, considering speech
• Questions and Answers
Why Have Analytics?
Customer Service Challenge
Cut Costs,Do More
With Less
Cut Costs,Do More
With Less
Increase Satisfaction,
Deeper Relationships,
IncreasedRevenue
Increase Satisfaction,
Deeper Relationships,
IncreasedRevenue
Creating and managing high-quality self-service channel experiences that meet both goals is difficult
and hard to measure.
EfficiencyEfficiency EffectivenessEffectiveness
Customer Satisfaction by Channel
PhoneCustomers
Face-to-Face
Self-Service
Web
IVR
Fundamental Analytics Problem
• Key metrics
• KBIs
•Drop-offs
•Recognition
•Hang-ups
• Thresholds
• Alerts
WHAT?
Step 1
WHY?
Step 2
What do I do?
Step 3
• Re-scripting
• Tuning
• Menu Restructuring
• Extend automation
• Build new automation
?
Result: Optimization is based upon qualitative assumptions, guesswork and can be extremely costly and time consuming.
“Naming something,” said Alice to the Red Queen, “isn’t the same as
explaining it.”
Lewis Carroll, Alice’s Adventures in Wonderland
Events
Patterns & Trends
Structures e.g.
What happened?
Why did it happen?
What was the cause?
What’s been happening?
Getting to Why
React
Change or improve
Predict
Behaviors
IncentivesSkills
TechnologyMeaning
Culture
ExpectationsScriptsIVR Structure
Experience
The “Black Box”
User Experience
in the IVR
IVR optimization takes place through a cumbersome, qualitative process
Design Documents
Optimization based on qualitative factors and extensive time investment
Call Logs / Reports
Some Assumptions and Guesswork
Extensive analyst hours
CSR Interviews (Qualitative)
Management By “Events”
Metrics Current Period
Last period
Call Volume 450,000 375,000
Overall drop-off to CSR
35% 29%
Incomplete calls/ hang ups
20% 22%
Recognition rates for key modules
89% 95%
Proposition: MBE has limitations because it associates location with causality.
% of executions ending in:
Name
# Executions
Success
Max retry
Max timeout
Hangup
Other
TOTAL 10768 94.6 0.4 0.9 4.0 0.1
HEAR EMERGENCY MESSAGE 1177 (10.9%) 89.5 0.0 6.2 3.9 0.4
MAIN MENU 1156 (10.7%) 95.8 0.3 0.6 3.3 0.0
GET ARRIVAL CITY 722 (6.7%) 97.5 1.1 0.1 1.2 0.0
GET DEPARTURE CITY 721 (6.7%) 95.7 1.4 0.3 2.6 0.0
GET NUM PASSENGERS 676 (6.3%) 98.7 0.4 0.0 0.9 0.0
LIST FARES 551 (5.1%) 83.1 0.2 0.0 16.7 0.0
MBE: “What”, not “why”
Problem: We don’t know why success is measurably lower for one module.
Proposition: Not a “data” problem, but a problem of perspective.
The Need for New Thinking
“The significant problems we face cannot be solved with the same level of thinking we were at when we created them.”
--Albert Einstein
Fundamental Problem of Organic Systems
• Highly complex relationships
• Non-linear
• Cause and effect are distant in space and time
• Leverage is generally not where the problem appears
Getting to “Why”IVR/Speech WEB
Live Agent
How Can WeHelp You?
SS # Account #
Google Yahoo
MSN
Home Page
eNewsletter
Experiences, not events
IVR/Speech WEB
How Can WeHelp You?
SS # Account #
Google Yahoo
MSN
Home Page
eNewsletter
Say “Agent”Say “Agent”
Experiences, not events
IVR/Speech WEB
How Can WeHelp You?
SS # Account #
Google Yahoo
MSN
Home Page
eNewsletter
ABANDON
ABANDON
PRESS “0”
Experiences incorporate “usage memory”IVR/Speech WEB
How Can WeHelp You?
SS # Account #
Google Yahoo
MSN
Home Page
eNewsletter
Press or Say “Zero”
ABANDON
Experiences incorporate “usage memory”IVR/Speech WEB
How Can WeHelp You?
SS # Account #
Google Yahoo
MSN
Home Page
eNewsletter
PRESS “0”
ABANDON
PRESS “0”
What What HappeneHappene
dd& Why?& Why?
PRESS “0”
ABANDON
MBE: “What”, not “Why”
Transfer analysis tells you how people transferred and even where they transferred from.
Did they transfer because of problems at that dialogue or because of an earlier experience?
Reservations – Transfer Analysis
Total Calls Transferred 386
Requested agent 119 (30.8%)
DTMF 0 61 (15.8%)
Said “agent” 58 (15.0%)
Error condition 267 (69.2%)
Technical 95 (24.6%)
Other 87 (22.5%)
Out-of-app request 85 (22.0%)
Total Calls 1277
Total Calls Transferred 386 (30.2%)
Speech Confidence measures
Low-confidence measures direct you to fix recognition or grammar. But what if the problem is related to an overall experience and not this one event?
Main Menu
Recognition Event
Value Raw Text Conf.Reservations Reservations 962schedules schedules 109
0:19.3
0:09.2
Prompt: _ UNKNOWN 0:20.8
22
Often, the cause is the experience, not the dialogue
state.
The Why: much of the drop-off is caused by “error spiraling”.
Offer
Case Study I
Speech Optimization for Fortune 100 Telecom Company
Case Study II
State Medicare/Medicaid Member and Provider Helpline
You need to show connectivity…
To solve the puzzle.
Continuous Optimization
“If something is worth doing, it’s worth doing poorly until you can do it well.”
Robert Fritz
About ClickFox• Founded 2000 in Atlanta• Pioneer in customer behavior intelligence• Continuous optimization services• Top-tier Fortune 500 customers:
Questions & Answers
Michael Chavez – VP Client [email protected]
Mike Kent – Director National [email protected]