The Google ITA Software deal

belligerentgooseInternet and Web Development

Jun 26, 2012 (4 years and 11 months ago)

378 views

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


1

The Google


ITA Software deal

How Android voice integration with ITA’s airfare
query structure can drive travel innovation and
create a more effective consumer experience

January 2011

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


2

Introduction

Google’s

$
700

M

acquisition

of

ITA

has

driven

both

good

debate

and

illogical

lobbying

on

risks

to

airfare

search

integrity,

competitor

web

traffic,

advertising

costs,

and

service

to

air

industry

customers
.





T
here

are

positive

opportunities,

of

course,

and

here

we

focus

on

a

specific,

less
-
publicized

one

that

could

have

a

meaningful

impact

in

the

near

future
:

Based on our tests, Google’s Android voice recognition outperforms Bing,
Vlingo

and possibly other
speech solutions in travel
-
focused natural
language
search and is almost ready for mainstream adoption



Integrating with ITA’s airfare query structure (and potentially hotels) is a key opportunity for Google
-
ITA
to drive industry innovation and create a better consumer travel search experience

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


How it could work


the basic process

3

Google
-
ITA could convert spoken natural language terms to voice
objects similar to drop down selections in
booking
engines,
which would then integrate with the fare search query data structure. Hotel search is a potential opportunity as well.

VXML 3.0


Analyzes voice input
using Speech Recognition
Grammar Specifications
(SRGS) to translate to text

SISR


Semantic Interpretation for
Speech Recognition

extracts
key travel terms to convert to
voice
objects

QPX fare search


Voice objects matched to
data structure for fare search
query and result processing

Google

ITA Software


Fly

1
st
-
class

Seattle

to LA

August 25

coming back August 28

on
Virgin America


© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


4


Cutting multiple inefficient and redundant steps out of today’s mobile search and travel intermediary experience




Enabling consumers to bypass OTA and
Metasearch

competitors, who
could not easily replicate
it without
considerable capital
investment




Encouraging
competitors to innovate and find other ways to improve and bring their mobile products to consumers




Prompting Microsoft and Apple to leverage their mobile platforms and alter the intermediary landscape further


Potential high
-
level impact

In

a

sense,

this

is

a

component

of

a

larger

potential

Google

PC

and

mobile

metasearch

play,

but

it

also

impacts

a

key

area

of

consumer

experience

needing

improvement,

and

implications

of

leveraging

the

Android

platform

could

include
:

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Potential
vs

current
mobile search
experience

5

Speak or Type
to search

Compared to today’s mobile search experience, voice search with fare integration would be much faster than inconsistent
text links due to SEM/SEO manipulation, redundant steps,
and
multiple
drop
-
down or typing
fields

Speak or Touch to
select airline

Speak or Touch to
select flight

Airline / OTA
Booking
path

Potential Voice
search to
booking path

Airline / OTA
Booking
path

Speak or type
to search

R
e
-
select


Flight


Inefficient
text links

Re
-
enter

terms

in multiple fields

Flight result
matrix or list

Current mobile
search to OTA
mobile path


Note: Expedia used for demonstration of general steps, not to imply variance in quality relative to other OTA paths


© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Potential
vs

mobile app download experience

6

Open app

Flight result
matrix or list

Flight detail

Choose OTA
or supplier

Comparing to
downloaded mobile apps
, current
metasearch

or OTA apps also require several additional steps, including
multiple drop
-
down or typing fields and additional site selections

M
ultiple text, calendar,
or drop
-
down fields

Note: Kayak used for demonstration of general steps, not to imply variance in quality relative to other
metasearch

apps


Current
metasearch

mobile app path

Potential Voice
search to
booking path

Speak or Type
to search

Speak or Touch to
select airline

Speak or Touch to
select flight

Airline /
OTA

Booking
path

Airline / OTA

Booking
path

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Why timing is right for Google’s opportunity

7

Travel is a strong
candidate for mobile voice
search and multimodal
engagement



Google
-
ITA will have the
tools to execute it

Speech recognition has seemed promising for years,
but has mainly been limited
to electronic voice
menus…so what
technology and consumer forces
are aligning with
Google capabilities to enable this now?

Travel lends well to speech
recognition


demonstrated in
Defense
Dept

tests 1991
-
1995



Smartphone penetration,
processing power, and dual
microphones growing
rapidly



Cloud
enables network
-
based
speech to process larger
vocabularies than embedded apps



Travel planning requires relatively
small grammar
sets

Android currently outperforms
Microsoft and
Vlingo

for natural
language voice search of travel terms



In a short period of time, 25
% of
Android searches are
already Voice



Google
has invested
heavily to train
speech algorithms



Mobile and Tablet search
should be
15% of searches by 2013
-
2014



Consumers rapidly adopting voice in
local search, car navigation, and
platforms like
Microsoft
Kinect

Technology elements

Consumer adoption and
Google advantages

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


8

Competitive implications

Only
Google
-
ITA could achieve proprietary voice and fare technology integration
, but Microsoft and Apple (a threat to enter)
own their mobile and speech platforms and have a key advantage in interfacing voice search directly with consumers.


OTA and
Metasearch

players are at
risk if Google, Microsoft or Apple leverage ownership of mobile
platforms and would also need to
partner with
Vlingo
,
Loquendo
, Nuance or others to embed
speech.

Travelport

ePricing


Amadeus Meta
-
pricer


Sabre


Vayant


Everbread


Expedia BFS

(if productized
)

Fare query platform

Speech platform

Mobile platform

Siri

iPhone

MSFT Speech /
Tellme

Windows Phone 7

Android

Android

ITA Software

Vlingo


Nuance


Loquendo


Promptu


MSFT API


Android SDK

None


Must distribute apps
or gain search traffic
through Android,
WP7 and iPhone
platforms

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Sample testing
shows
Android
advantage
and challenges

9

Sample
-


Fly

1
st
-
class

Seattle

to LA

August 25

coming back August 28

on
Virgin America


To evaluate the readiness of voice search platforms to handle travel
-
specific natural language input strings, we tested 140
searches on Android, Bing (Windows Phone 7), and
Vlingo

speech platforms
(see Appendix for specific
samples).


Our tests
indicate Android is currently more
advanced for travel, while Bing performed fairly well and
Vlingo

was poor.

Platform
Noise level
Key term
success rate
WER
Queries
mean search
time (sec)
mean
words
mean error
words
Android
loud - Starbucks
30%
29%
20
7.4
12.4
3.6
Android
quiet - office
85%
4%
20
3.7
13.2
0.6
Windows Phone 7
quiet - office
70%
7%
20
4.9
13.2
0.9
Vlingo
quiet - office
10%
15%
20
3.5
13.2
2.0
Improved
accuracy and search time is
needed for practical use, especially in noisier environments, which we expect will
occur as
dual microphone devices proliferate and
cloud
technology and speech
training sets across all platforms improve.

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Appendices

10

Appendices

Defense Advanced Research Project (DARPA
) Travel Planning testing



Travel vocabulary size and relative speech complexity indicator



Lenati voice
test
sample data by phone platform



Projected mobile and tablet search query growth 2010
-
2014



Voice technology provider
segments



Voice technology overview






© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Appendix


Travel
planning
ideal
for speech recognition

11

Travel lends
well
to speech
recognition, demonstrated by testing during the Defense Advanced Research Project
(DARPA) from 1991
-
1995

http://www.itl.nist.gov/iad/mig//publications/ASRhistory/index.html

DARPA’s Air
travel
planning test:



U
sed multiple microphones




Word Error Rate (WER) dropped from
20% to less than 3% in 5 years




Measured to be as effective as
human
interpretation

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Appendix


Travel
planning
ideal
for speech recognition

12

Travel

vocabulary

consists

mainly

of

finite

sets

of

origins,

destinations,

dates,

travel

brands,

and

other

key

categories

rather

than

continuous

or

conversational

speech

that

becomes

incredibly

complex

in

vocabulary,

meaning,

nuance,

and

dialect
.



Therefore

travel

likely

fits

in

the

least

complex

speech

recognition

type




And

reduces

the

complexity

of

creating

voice

objects

to

match

fare

search

data

structures

ABI Research

Data set vocabulary size and relative complexity by speech type

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Appendix


Sample
testing results

13

Vlingo
was worse than
expected at this point

Android error
rate
was low, and in most cases, errors did not affect relevant terms needed to create Voice Objects

Spoken search string
Text result
WER
Search Virgin America flights from Seattle to San Francisco August 25 returning August 28
Search Virgin America flight from Seattle to San Francisco August 25 returning August 28
0%
I want to fly to Seattle from Washington DC on October 15, returning October 20
I want to fly to seattle from washington d c on october 15th returning october 20th
0%
I want to fly to Philadelphia from New Orleans on October 15, returning October 21
I want fly to philadelphia from new orleans on october 15th returning october 21st
7%
Fly from Seattle to Anchorage August 25 returning August 28 Alaska Airlines
fly from seattle to anchorage august 25th returning august 28th alaska airlines
0%
Find flight Boston to Ft. Lauderdale August 25 returning August 28 JetBlue
find flights boston to fort lauderdale august 25th returning august 28th jetblue
0%
Search round-trip flight St. Louis to Phoenix January 5 to 9
search round trip flights saint louis to phoenix january fifth 29
17%
Spoken search string
Text result
WER
Search Virgin America flights from Seattle to San Francisco August 25 returning August 28
virgin america flights from Seattle to San Francisco Aug 25th returning August 28th
0%
I want to fly to Seattle from Washington DC on October 15, returning October 20
12 flight to Seattle from Washington DC on October 15th, returning October 20th
27%
I want to fly to Philadelphia from New Orleans on October 15, returning October 21
Iwon to Fly to Philadelphia from New Orleans on October 15th raton October 20th
13%
Fly from Seattle to Anchorage August 25 returning August 28 Alaska Airlines
flights from Seattle to Anchorage Aug 25th returning August 28 Alaska Airlines
0%
Find flight Boston to Ft. Lauderdale August 25 returning August 28 JetBlue
flight Boston to Fort Lauderdale August 25th returning on 20 Eights Jet Blue
17%
Search round-trip flight St. Louis to Phoenix January 5 to 9
round trip flight St. Louis to Phoenix January 5th 29
17%
Spoken search string
Text result
WER
Search Virgin America flights from Seattle to San Francisco August 25 returning August 28
search virgin america flights from seattle to san francisco otis 25th return in august 28
14%
I want to fly to Seattle from Washington DC on October 15, returning October 20
I want to fly to seattle from washington dc on october 15th return in october 20th
7%
I want to fly to Philadelphia from New Orleans on October 15, returning October 21
I want to fly the philadelphia from new orleans on the kerber 15 return in october 21st
20%
Fly from Seattle to Anchorage August 25 returning August 28 Alaska Airlines
fly from seattle to anchorage august 25th attorney in august 28 alaska airlines
8%
Find flight Boston to Ft. Lauderdale August 25 returning August 28 JetBlue
flight boston the fort lauderdale august 25th return in august 28 jet blue
25%
Search round-trip flight St. Louis to Phoenix January 5 to 9
search round trip lake saint louis to phoenix january fifth 29
17%
Windows Phone 7 error
rate
was higher, but in many cases did not affect relevant terms

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Appendix


Voice
will help drive mobile search growth

14


PC search should continue to dominate, but Mobile and
Tablet are also high
-
growth




Google owns estimated 90
% of mobile query share




25% of Android searches are already
Voice


Mobile, Tablet and PC search queries 2010
-

2014

Both Google and Microsoft have invested heavily in
Android and Windows
-
based Voice
recognition in anticipation of Mobile
and Tablet search growth

Sources: Morgan Stanley; Microsoft

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Appendix


Voice technology overview

15

VXML manages dialogue
between
person
and
device, and version 3.0 focuses on
improved media
control
and synchronization
,
identification
and
verification capability, improved extensibility, and
better multimodal input architectures via
use of
XHTML+Voice

VoiceXML

Speech Recognition
Grammar Specifications

Semantic Interpretation
for Speech
Recognition

SRGS identifies the sentence
structure
expected
in the human voice
responses

SISR extracts key information
from
voice inputs to create voice objects for applications, and
examples include
airfsearch

integration with ITA data structures or converting phone
-
based
restaurant delivery order
data into
a format for order
processing
applications

S
peech recognition is based on the Hidden
Markov
Model (HMM), with a
statistical
approach enabling
the
technology to
be “
trained,”
decision
weighting
adjusted
to
improve accuracy, and huge data sets ideally of examples
of all
variations of
a
spoken string.


Since it is unlikely any
two
spoken terms by different people will match,
the
model statistically determines
what was most likely
said,
and the process below takes place to enable spoken terms to convert to use for applications.

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Appendix


Voice technology overview

16

Often
preferred
since response
times
can
be faster than network
-
based
speech
and performance is not
subject to network
connectivity
.


Though phone technology will improve and many applications are suitable for embedded speech, large
vocabularies and spontaneous speech require processing and memory not yet available.

Virtually unlimited
processing
power enables more robust capabilities in handling large vocabulary
requirements
,
complex continuous
speech
processing,
and natural language
inputs.


K
ey challenges revolve
around
latency in
data
transfer
, and
a
quick
and accurate experience
on mobile
devices. However, both should continue to improve consistently as technology improves.

Hybrid systems combine embedded and
network
-
based
processing, offering
an advantage over
a straight
network approach

as the
extraction
performed
on the device
reduces
complexity
that
can increase
errors.

Network
-
based
Speech

Embedded
Speech

Hybrid or
Distributed
Speech

S
peech recognition systems can be network (cloud)
-
based, embedded directly into phone operating systems, or a combination of both
,
each of which carries advantages and disadvantages.

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


Appendix


Voice technology provider segments

17

Microsoft Windows Phone 7


Google
Android


Apple iPhone /
iPad

(
iOS
)

Vlingo



Nuance


Loquendo



Promptu

IBM


AT&T Watson

Mobile platforms with proprietary speech

S
peech providers with mobile apps

Speech technology infrastructure


By owning the phone platforms and default search applications, they
could essentially control the voice search experience and market



But by owning ITA Software, only Google could create seamless voice
search integration with airfare and potentially hotel data structures


Travel
intermediaries or suppliers c
ould partner with them, or they
could develop their own travel
-
specific applications



Must market to travelers to bypass default mobile search applications



Promptu

owns
ProntoTreno
, a voice
-
interactive train services and
booking application in Italy, and Flights2Go, an airline services app


AT&T Watson technology is a key component of
Vlingo



Travel
intermediaries or suppliers c
ould partner with them

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential

2825 Eastlake Ave E #210, Seattle, WA 98102 | (800) 848
-
1449 | www.lenati.com



Point of View


18

About Lenati

We are an award
-
winning boutique consulting firm that designs, implements, and optimizes
marketing and sales solutions for companies that want to build stronger customer connection.


Based in Seattle, Lenati was recently named one of the “Seven Small Gems of the Consulting
Profession” by Consulting Magazine.








About the Author

Jonathan Alford brings over 14 years of experience in consumer travel and lifestyle business,
technology and finance. His specialties include business strategy, online and mobile marketing
strategy, and global finance.


He is a graduate of The Johnson School at Cornell University and the University of Virginia.

About Lenati

© 2011 Lenati LLC All Rights Reserved Proprietary and Confidential