Call Center Multimodal Voice Search

11  Download (0)

Full text

(1)

Call Center Multimodal Voice Search

Matt Yuschik, Ph.D.

Human Factors Specialist, Human Computer Interactions

Relationship Technology Management

Session A103 – Voice User Interface in Voice Search Voice Search Conference

(2)

© 2008 Convergys Corporation. All rights reserved.

ƒ

75,000 employees

ƒ

85 customer and employee contact, service and data centers

worldwide; focused on optimizing employee and customer

experience

ƒ

Clients in 70+ countries speaking nearly 35 languages

ƒ

$2.8 billion in revenues

ƒ

Listed on NYSE, S&P 500, Fortune 1000

ƒ

A

Fortune

Most Admired Company for eight consecutive years

ƒ

Serve more than half of the Fortune 50 as clients

ƒ

Host more than 1 billion customer interactions annually

ƒ

Support more than 3 million employees and retirees worldwide

ƒ

Billing for 350+ million communications subscribers worldwide

ƒ

Named to the Top 10 for Innovative Use of Technology by

InformationWeek

(2007)

Convergys Corporation

A Global Leader in Relationship Management

Worldwide

Capabilities

A Leading

Public

Company

Key Facts

About

Convergys

(3)

3

Path from Multimodal Agent to End-User Self-service

• Multimodal presentation

• Conversational Interface

• Speaker Verification for security

• Hidden Agent when needed

• Device-Independent • Multichannel experience • Real-Time Decisioning • Customer History • Transactions • Dialogs • Tasks • Subtasks

Point Solutions vs. Solution Path

• 70 k SMEs

• Solutioners

• Observers

• MMUI testers

(4)

© 2008 Convergys Corporation. All rights reserved. 4

Devices and Multimodal Preferences

Input:

What should be spoken?

What should be typed or pressed? Output:

What should be read? What should be heard?

device

landline cellphone PC PDA 3G phone

Input speak x x x x x type x x x x tap x x GPS x Output listen x x x x x listen - TTS x read text x x x x view figures x x x view video x x

What devices and combination of modes are available?

(5)

5

Needs Assessment and Solution- Delivery Service

Navigation, Form Filling, Data Caching, Search

Outside-In Approach = Monitor Agents and Callers

Identify most frequent functions

Monitor Services –> Transaction types / Use Cases

Decompose into Tasks

Sevice Types Other 32% ZIP code 7% Hours and Locations 11% Track 21% Redeliver 4% Hold 3% Rates 5% Address 5% Complaints 12%

Call Distribution

Start with GUI–based Service Handling

Many layers / screens to complete a call

Solution = A Multimodal Agent Tool that includes Voice Search

Caller value-add

Navigation, Data-driven search Better Customer Care Experience

Agent value-add

Improves Agent Productivity and Quality Better Agent Satisfaction, increased retention

(6)

© 2008 Convergys Corporation. All rights reserved. 6

Simulation Results - Longitudinal

E1- H ol d A H T

0

10

20

30

40

50

60

70

1

2

3

4

5

6

7

trial

sec

A A - H ol d A H T 0 10 20 30 40 50 60 70 1 2 3 4 5 6 7 trial sec A2 - Hold AHT 0 10 20 30 40 50 60 70 1 2 3 4 5 6 7 trial sec Longitudinal testing

Gives agents time to learn Identifies time training

Test Conditions –

4 Agents (6 mo -10 yrs experience) 1 UI per day for 3 days

5 Services

7 repetitions / svc – expect leveling No caller involved

UI Types – “wrapper” concept

Existing GUI

Narrow – voice activate existing flow (AA)

(7)

7

Pilot Test – Learning Types

Excellent Learners

“get it” quickly <7 days Level 2 Agents -40 -30 -20 -10 0 10 20 30 40 50 60 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 days se c Level 3 Agents -40 -30 -20 -10 0 10 20 30 40 50 60 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 day se c Level 1 Agents -40 -30 -20 -10 0 10 20 30 40 50 60 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 day se c

Delta AHT for Level 2 Agents (without E1 bias)

-20 -10 0 10 20 30 40 50 60 70 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 day se c Expected Results Normal Learners

accomodate then decrease 10-15 days

Delayed Learners

At risk

(8)

© 2008 Convergys Corporation. All rights reserved. 8

Multimodal AHT- Performance Groups

AVA Group - w/o CTI Agents

-40 -30 -20 -100 10 20 30 40 50 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Days on AVA Ch an g e f ro m B as elin e

Faster(n = 7) Little Change (n = 6) Slower (n = 5)

Mismatched

Agent

Successful

Agent

Comfortable with new technology Shorter tenure

Use MMUI most of the time

Longer tenure

Resistant to technology changes Use of Mute mode

(9)

9

Multimodal UI Design Principles – Voice Search

• Speech and graphics are both active at the same time • Error conditions are handled in both modalities

• Shortcuts for common transactions (mixed initiative dialog)

• Agent and the system are focused on one vocabulary

• Don’t need SLMs or NL for the most part (constrained domain)

(10)

© 2008 Convergys Corporation. All rights reserved. 10

Conclusions and Next Steps

Applications can be built from scratch using VXML or SCXML standards Wrapper approach is valuable for legacy applications

API approach if software has sufficient “hooks” More trials, Component Applications

Voice Pad

Speaker Verification

Evolution to Multi-Modal Self-care devices Application Assessment Process

Determining best places to add another modality (voice) Types of Transactions

Customer service hot-spots Categorizing flows and tasks Training

What should be learned, and when Useful tools and practices

(11)

11

Matt Yuschik, Ph.D.

Human Factors Specialist,

Human Computer Interactions

Multi-Channel Self-Care Solutions

Relationship Technology Management

Convergys Corporation

matt.yuschik@convergys.com

Questions, Comments,

Insights?

Figure

Updating...

References

Related subjects :