Automated Solutions for Directory Assistance Services

1
Automated Solutions for Directory Assistance
Services

• Fatima Vital
• Director of Marketing
• Philips Speech Processing

2
Content

• Introduction
• Challenges
• Business Cases
• Cost Savings
• New Revenue Stream
• Fielded Installations
• Respons AB/ Telia
• BT
• Telefonica Argentina

3
Directory Assistance Market Segments
4
Directory Assistance Example

• Thank you for calling Directory Assistance.
  Please say the town or city where the person
  lives.

Boston.
Please say the persons last name.
Bryant
Please say the name of the street where the
person lives.
Columbus Avenue
I found the following entry Bryant, living in
Boston, Columbus Avenue, has the phone number 617
26 24 786. Thank you for your call. Good-bye.
5
White Pages

• Standard DA service
• Covers residential and company listings by name
• Company listings account for most of the traffic
  (see left)
• Callers also mostly come from business
  environment
• DA is basically B-2-B!

20
80
Business Listings
Residential Listings
Typical distribution across Telcos
6
Automation of Residential Listings

• Challenging recognition task
• long lists of names with frequently unusual
  phonetics
• But easy task compared with business listings
  because
• dialog is structured
• information items available (name, first name,
  street, )
• Full automation possible
• Ask for as many different items (name, street,
  city, ) as possible
• Use spelling if possible
• Make use of DB
• Dialog success rates
• for automizable calls (caller cooperative, knows
  information, )
• succes rates of 75 - 87.5 have been proven in
  trials with real callers

7
Coverage of U.S. population by last names
8
Automation of Business Listings

• More difficult task compared with residential
  listings
• Different domains require different info items
• small businesses name of owner, street,
• big business branch, part of city
• doctors area of medicine, address
• Dialog can not be strict and uniform as for
  residentials
• Automation is possible by
• Automate only partly top 1000, 5000, 20000,
  listings
• Automate fully by using intelligent search
  techniques (see below)

9
Requirements Towards Speech Recognition Engine

• Very large vocabulary (1 million words in one
  lexicon)
• High accuracy
• Speaker and dialect independent
• Lexicons can be switched on the fly during dialog
• Language can be switched within one application
• Open and closed grammars
• Efficient resource management
• Integration with DA Databases
• Intelligent search algorithms

10
Combination DB use lead to high success rates
Smith Miller Moore Myrick ...
John Jim Juan James Joan

• Various challenging recognition tasks involved
• recognition of large name lists (up to several
  100 K words)
• free spelling
• words with unusual phonetics (foreign names)
• To achieve high dialog success rates
• results of various recognition tasks have to be
  combined intelligently
• DB know-how should be used as much as possible
• Exploitation of previous recognition steps to
  reduce lexicon size of next step

DB
James Moore Joan Moore Joan Myrick ...
11
Optimizing Database Traffic
Database FilteredRecognition - use
intermediate queries to generate reduced
lexicons - e.g. only first names existing
for a given set of last names
Queries
one
multiple
large
medium
Lexicons
12
White Pages Business Listings
Idea Use intellgent search methods from Internet
for doing business white listings
Company DB
Internet
IR Methods
IR Methods
Internet-like Search Engine
Internet Search Engine
ASR
free form input
Closest match
User
Caller
13
Hidden Automation
Upon detection of failure automated system hands
off to (hidden operator) the caller does not
even notice the human intervention. System acts
like Store-And-Forward system. Automation rate
can be increased over time.
Transfer of sound files for failed dialogs over
CTI link (Call whispering)
(Hidden) Operator
Manually transcribed Input
Automated System
Dialog
DB
14
Business Case Cost Savings

• Some Assumptions
• Costs per agent 30,000 / year
• Calls per agent 124,000 / year
• Automation rate 50 / Succes rate 70
• Costs of automated system (over 4 years) 2,000
  / year
• Capacity Utilization rate 80
• 500 channel system
• Lead to results ...
• 0,12 per call
• 96 days to pay-back
• 12 Mio savings / year

which form a pretty convincing business case!
15
Example Portugal Telecom

• Solution
• SpeechPearl and SpeechFinder from Philips
• DB and IVR from PT/ InescTel
• Status
• The RDA service is live since Nov. 2000 the
  other services since Q1 2001

• Customer
• Portugal Telecom
• Description
• White Pages
• Pink Pages
• Reverse DA
• Existing phone number
• Service is used to replace part of the call
  center by an automated system

16
Business Case New Revenue Stream
Traditional Operator Assisted DA
Speech Enabled Auto- mated DA
Internet
CD-ROM

• Low cost, speech enabled automated DA as
  additional revenue stream to address new customer
  segments

17
Is Internet a Threat to Traditional DA Services?

• Most DA operators offer (free) Internet access
• Sharply rising access volumes are reported
• However Calls to (charged) traditional DA
  services are virtually unaffected

Internet is no threat but a complementary market
18
Market Size

• Estimations are 2.5 Billion DA calls / year in
  Europe
• Equals 5 calls per inhabitant
• US comparison 30 / year inhabitant
• Additional (low cost) offerings lead to increased
  demand
• New players
• independent service providers
• mobile operators

There is room for increase of market size
19
Example Respons AB

• New automated white pages service positioned as
  additional service to call center and internet at
  a lower price.
• As many as 45 of low and non-use customers
  preferred and would use the speech enabled
  service
• Additional Revenue

Price
Value added
Basic
A complete Service Portfolio
20
Yellow Pages

• A yellow pages DB is valuable content. Why not
  exploit it over the phone as well?
• Base entry is free (with TTS output)
• Chargeable add-ons
• Output with nice, spoken prompt
• Extended description, including jingle, music
• Can also be created by customer and up-loaded per
  web
• Call forwarding (also possible charge per call)
• Being top of the list in a category/place

21
Global Local Portals
L o c a l I t y I n d e x
V-Home Page
V-Home Page
V-Home Page
V-Home Page
Global Portal
Yellow Pages
Horos- cope
Sport
Stocks
Weather
News
Single No. Access
22
Future steps Location based services

• Location info in DB
• Caller info from mobile network or CLI in fixed
  network
• Route descriptions by voice, fax, sms, possible
• Interesting for car scenario

23
Fielded Installations
Results from field trial done for Stockholm
56 of the customers very satisfied or
satisfied 90 thought the application was easy
to use 62 thought that the search was fast and
efficient 45 of low and non-use customers
preferred and would use speech enabled
service 33 felt uncomfortable talking to a
machine 33 did not like the voice
13
87
Successful
Unsuccessful
24
Fielded Installations

• The Trial
• Three cities
• Two months duration
• 10,000 calls
• 2,700 calls analyzed in detail
• 300 interviews with callers

Other
6
Nothing
Confirmed
22

• Results from Interviews
• 65 customers satisfied with enquiry
• 78 happy to talk to computerized system
• 87 feel automation acceptable in DQ
• 58 willing to pay for service

Successful
72
Results - Valid Calls
25
Fielded Installations

• Customer
• Telefonica Argentina
• Speech-enabled services
• Area code information (launched October 2001)
• New (launched October 2001)
• Extension planned Speech-enabled directory
  assistance portalFully automated white pages
  (residential business listings, governmental
  services)

26
Conclusions

• Very strong business cases for DA automation
• Speech-enabled DA is a challenging application
• Choose the right solution
• Philips is your partner