Answer lab best practices in research and design for voice user interfaces

November 2017
© AnswerLab 2017. All Rights Reserved
Great Voice Experiences
Start with Listening:
Best Practices in Research and
Design for Voice User Interfaces

2
Table of Contents
INTRO
01 02 03 04Pages 3-4 Pages 5-13 Page 14-26 Page 27-29 Page 30-35
• Riseofthesmartspeakers
• Keyfindings
• Stateofsmartspeaker
devices
• Currentstateofsmart
speakerapplications
• Whatdoesthismeanfor
brands?
• Gettingfound
1. Failgracefully
2. Doyourhomework
3. Beagoodhost(i.e.,
settingexpectations)
4. Designforcontextand
continuity
5. Usenaturalcommands
6. Maketheuserfeelheard
(i.e.,confirmations)
7. Havesomepersonality
• What’snextforVUIs&
smartspeakers
• Abouttheresearch
• Smartspeakerresearch
team
The Current
Landscape
Best practices for
designing voice
interactions
What’s next for
VUIs & smart
speakers?
Appendix

3
Whether you call them “smart
speakers,” “voice-enabled speakers,”
or “voice assistants,” there’s no
denying we’re witnessing the birth of
a major new technology category.
Although voice technology has been
around since the 1950s, Amazon
introduced the first voice home
assistant, Echo, to the US in 2015
with Google following suit in late
2016. While these devices currently
fall short of our dreams of our own
JARVIS (personal assistant to
Marvel’s Ironman), it’s estimated that
over 35M Americans already use a
smart speaker at least 1x/month—a
129% increase over 2016*. The
global market is expected to grow to
$2B by 2020.** In just the past few
months, Amazon has introduced a
slew of exciting new Alexa-powered
technology. Google introduced
Google Home Mini and Google
Home Max. Apple announced their
first smart speaker, HomePod. Sonos
and Harman Kardon have been
introducing speakers powered by
Alexa, Google Assistant, and
Microsoft’s Cortana. The blogs swirl
with rumors of numerous other
companies preparing to enter the
fray, including Samsung, and
Facebook.
The cause of this flurry of activity?
Voice-enabled technology opens up
new possibilities for users. Finally we
will be able to interact with
technology using the most natural
interface of all: speech. No longer
tethering us to screens, voice
interfaces allow technology to
become both more present and less
intrusive. And yet, internet jokes
abound about smart speaker failures.
If you’ve spent time living with a voice
assistant, you know first-hand they
have their limitations. We may
daydream sci-fi visions of
conversational interfaces, but we live
in a world where voice technology is
still struggling to deliver satisfying
command-and-control transactions.
So what’s going on with smart
speakers, what’s working well, what
needs improvement, what does the
future hold, and what should brands
consider when surveying this new
landscape? At AnswerLab, questions
like this are our reason for being. To
answer them for ourselves and our
clients, we conducted multi-method
research with smart speaker owners
to understand not just what they
want, but how speaker owners
currently behave and why.
Rise of the
smart speakers
“Human-Computer
Interaction takes big leaps
every once in a while. The
next generation of speech
interfaces is one of
those leaps.”
–Tim O’Reilly
*Alexa, Say What?! Voice-Enabled Speaker Usage to Grow Nearly 130% This Year." EMarketer. May 08, 2017. Accessed October 02, 2017.
https://www.emarketer.com/Article/Alexa-Say-What-Voice-Enabled-Speaker-Usage-Grow-Nearly-130-This-Year/1015812
**Gartner Says Worldwide Spending on VPA-Enabled Wireless Speakers Will Top $2 Billion by 2020." Gartner. Accessed October 02, 2017.
http://www.gartner.com/newsroom/id/3464317

4
KEYFINDINGS
We surveyed 1000 smart speaker owners and spent time in the homes of 10 owners to explore smart speaker use cases,
opportunities for improvements, and design best practices.
As one participant succinctly put it: “If it doesn’t make it easier, we’re not going to talk to this thing just because we can. If it’s more
complicated than a website or app, then forget it.”
• In general, most participants were very satisfied
with their devices and use them frequently.
• Participants have fairly low expectations of the
current technology: participants were happy with
their smart speakers, but:
• Over 70% have experienced problems or
frustrations.
• 25% do not think the designers of these
experience thought about the needs of people
like them.
• Third-party voice applications are confusing
users:
• Many participants were unclear on the distinction
between voice applications and native
functionality.
• 20% aren’t using voice applications.
• Participants don’t have one place they go for
information on voice applications.
• Brands need to step up their game. As with all
early-stage technologies, users will only blame the
technology or themselves before they start to blame
the app developers for a limited time.
• Security is a concern; privacy, less so. Most smart
speaker owners in our study did not have strong
privacy concerns about the 'always on' nature of the
devices, but many did have security concerns
regarding unauthorized access to information they
wanted to keep secure.
• Design best practices do exist. Best practices in
designing for voice interfaces will continue to evolve,
but clear lessons are emerging and following them will
help brands ensure they get the most value from their
investments in voice interactions.
• The most important tool available to digital leaders and
practitioners--both to avoid costly mistakes and to
create compelling experiences--is user research.
Great voice experiences start with listening.

The Current
Landscape
State of Smart Speaker
Devices
State of Smart Speaker
Applications
What does this mean
for brands?

6
State of Smart
Speaker Devices
Smart speaker ownership is growing exponentially.
We discovered that over one-
third of those surveyed
received their smart speaker
as a gift. Why is this
important? Since this group
is less invested in their smart
speaker (literally), they are
also less invested in learning
the technology. They scored
lower in measures of use,
knowledge, and satisfaction.
This makes it even more
important that those
designing for voice do so
with less technically
sophisticated/more
passive users in mind.
63% plan to purchase
another smart speaker
77% have suggested
buying a smart speaker
to people they know
50% say they use it
more now than they did
during their first month of
ownership
Almost 50% of those
surveyed began using a
smart speaker within the
past 3 months
69% use their smart
speaker at least once
per day

7
State of Smart
Speaker Devices
82% agree their smart
speaker is easy to use
80% agree that their
smart speaker has the
features they need
86% are satisfied or
very satisfied with their
smart speaker
26% have technical
issues
25% don’t think it was
created by someone who
thought about the needs
of people like them
Smart speaker owners love their devices.
Usability scores are high, but 1 in 4 people felt left out or
frustrated to some degree.

8
State of Smart
Speaker Devices
These high satisfaction scores coupled with evidence of frustration or technical challenges support what we heard in interviews:
current expectations of the technology are fairly low. Participants we interviewed often shrugged off problems as the sort of
inconveniences one puts up with in new technology. In fact, 28% of the smart speaker owners we interviewed who purchased their
device themselves did so to experiment with new technology. This was the second most popular reason for getting a smart speaker,
after entertainment purposes (30%). Expectations may be low, but we don’t expect them to remain that way for long.
Much has been made of privacy and security concerns with smart speakers. Indeed, this may well keep some from purchasing these
devices in the first place. But among the smart speaker owners we studied, most did not have strong privacy concerns about the
'always on' nature of the devices. In interviews, some described it as “weird” or “a little creepy,” but they followed this assessment by
saying that the convenience afforded by the technology outweighed any concerns they might have. But while most did not have
privacy concerns, many did have security concerns. When surveyed about problems or concerns, only 13% included concerns about
privacy, but 34% did not agree with the statement “I feel my smart speaker interactions are secure.” In interviews, participants often
expressed reluctance to link their smart speaker to financial accounts or other personal data points because they were worried about
unauthorized access to information they wanted to keep secure.

9
Once the concept of third-party
integrations was explained, over 80%
of those surveyed said they did in
fact use voice applications. The
remaining 20% use only the out-of-
the-box functionality that is a part of
each device’s operating system. To
better understand the current state of
voice applications, we looked at how
users discovered them, how they
learned about them, and how they
used them.
State of Smart
Speaker
Applications
While Amazon calls their third-party
integrations “skills” and Google calls
them “actions,” we’re going to refer to
this category as “applications.”
Roughly 20% of those surveyed had
never heard of skills or actions and
another 30% had heard of them but
were unclear on what they were.
Most smart speaker users don’t think
in terms of skills, actions, or
applications—they think in terms of
what they want to do.
72% said they were
likely to add a new
application in the next
3 months
42% of application users
said they seek out new
applications more now
compared to during their
first month of ownership,
indicating a desire to
expand their voice
assistant’s functionality

10
Current State of
Smart Speaker
Applications
One statistic that really grabbed our attention is 63% of those who added voice
applications said they had encountered some form of problem or frustration with the
applications they use. The most common frustrations with applications:
24%
17%
33%
38%
31% 33%“It’s the same as picking up
my laptop and going to their
website. If the website’s
crummy, I’m going to blame
the company, I’m not going
to blame my laptop.”
–David, participant
We’ll get into the specific challenges users face in ‘Best Practices for Designing
Voice User Interfaces,’ but for now, suffice it to say that brands who fail to deliver a
satisfying experience at each touchpoint will suffer damage to their reputation across
all touchpoints.
Again, smart speaker owners are happy with the experience, but encounter
challenges frequently. Currently, the top two reasons people assume they
experience problems with the technology are their own mistakes (37%) and
limitations of the current technology (27%). But this won’t last. As technologies
mature and design conventions become established, users are less likely to blame
the technology or themselves and more likely to blame the app developers.
19% said the skill
often does not
understand them
19% said the initial
setup of skills are difficult
22% said it’s difficult
remembering the exact
wording for commands

11
• Voice is great for simplifying
complicated requests. Tasks that
require taking out one’s phone,
opening an app, and going through
multiple steps of simple input in a
graphic interface can sometimes be
completed with a single sentence in
voice. Consider, for example, setting a
timer or playing a specific album.
• Voice is great for hands free contexts.
Voice opens up new opportunities for
interacting with users where they are—
cooking, driving, doing home
maintenance, caring for a baby,
gardening, etc.
• Voice is great for bringing people
together. We heard numerous stories
of families gathering around their smart
speaker to play games, hear jokes,
and play music. While many tech
devices seem to isolate us further from
one another, smart speaker
applications have the ability to engage
with a group.
What does this
mean for brands?
• Voice is not good for tasks requiring
complex outputs. Consider what it
would be like to use voice for
comparison shopping on an
ecommerce website, scanning
industry news to determine what of it
is relevant to you, or reviewing charts
and graphs intended to support
decision-making.
• Voice is not good for tasks requiring
complex inputs. Long forms and/or
inputs with multiple variables can
quickly become overwhelming.
Imagine completing your tax forms
using voice alone. Not fun.
• Voice is not good for situations
where auditory privacy is necessary.
Participants in our study said they
were not comfortable discussing most
financial matters, health concerns, or
personal details in shared spaces.
And many contexts (e.g., the office)
are not places where we can
speak freely.
How brands should
approach developing
voice applications
“If it doesn’t make it easier,
we’re not going to talk to this
thing just because we can. If
it’s more complicated than a
website or app, then forget
it.”
-Jenna, participant
Our research makes clear that customers and prospects expect big brands to have a
presence in smart speakers. Customers want to interact with brands in ways that are
relevant to them and that leverage the unique opportunities of the platform.
However, brands should first ask themselves whether voice is the right interaction for
the task.
New communication technology rarely replaces previous technologies. At least not
initially. Rather, it adds a layer on top of existing technologies, filling in the gaps and
extending their reach. Voice interfaces can’t do some things as well as graphic
interfaces and things they do better.
Where voice fails:Where voice succeeds:

12
What does this
mean for brands?
How brands should
approach developing
voice applications
• Voice applications are another touchpoint—ensure your
brand’s experience is consistent across devices and
platforms.
• Do not simply “port” your web or app experience to smart
speakers. There is no need to try and provide all the same
functionality in a voice application that you do through other
interfaces.
• Consider your web, app, or product’s functionality through the
lens of what voice does well and what it does not do well.
• Conduct research to learn what users want to do with voice
and build for that.
• Remember: if the smart speaker experience doesn’t reduce
friction as compared to existing methods, users are unlikely to
use it.
AnswerLab recommends:

13
Getting found
How do application
users learn about
those applications?
Getting your voice application found is currently a challenge. Many of our survey
respondents were unclear on the distinction between native applications and third-
party integrations. In our in-home interviews, even those who did understand the
distinction sometimes struggled to successfully launch the third-party voice
application. Most, when trying to order through the Domino’s Pizza voice app, were
instead given a list of nearby restaurants.
Survey respondents who had enabled third-party applications found them through a
variety of means [chart below]. Participants found no single source for learning about
and finding applications. Further, the traditional model of an “app store” doesn’t work
well over voice. As mentioned previously, it is not easy for users to scan large or
complex sets of information in voice. And as we discovered in our research, many
users don’t frequent the smart speakers’ mobile apps.
In short, you can’t rely on device manufacturers to get your voice application noticed.
If you’ve committed the resources to developing an application for this platform,
make sure you’ve committed resources to marketing it effectively.
0
20
40
60
80
100
A company's
website (i e ,
Dominos com,
Spotify com)
38% 33% 33% 31% 24% 17% 8%
A company's
mobile app
(i e , Dominos
mobile app,
Spotify mobile
app)
Amazon com
or Madeby
google
com/home
Alexa app
or Google
Now
Emails
from
Amazon or
Google
I asked the
smart
speaker
Other
(please
specify)
Question: Where did you go to determine if a skill/action existed for your smart
speaker? (select all that apply)
Total sample; Unweighted; base n = 813

Best practices for
designing voice
interactions
1. Fail gracefully
2. Do your homework
3. Be a good host
(i.e., setting
expectations)
4. Design for context
and continuity
5. Use natural
commands
6. Make the user feel
heard (i.e.,
confirmations)
7. Have some
personality

15
Best practices for designing
voice interactions
In addition to our research on users’ wants, needs, and behaviors around voice technology in general, we also surveyed them about
specific challenges and conducted in-home usability tests. In this section, we provide an overview of the things every brand needs to
consider when approaching a VUI/smart speaker project.*
*For a detailed guide, we recommend Cathy Pearl’s Designing Voice User Interfaces (O’Reilly Publishing, 2016).

16
1. Fail gracefully
Branded Interaction on Device: Would you like to track your order or place a
new order?
Alysha: Yes.
Device: I’m sorry, can you please repeat what you said?
Alysha: Yes.
Device: I’m sorry, I didn’t get that. Can you please repeat what you said?
Alysha: [shrugs] No?
Alysha: [quietly to interviewer] I’m irritated now. I don’t even remember
the question.
[loudly to device] Yes.
Alysha: OK, Google, stop.
“To err is human…” and we can expect no more from technology designed and built
by us. The key is to fail gracefully. Users will forgive many technical limitations and
errors if the system responds in a way that helps them to understand what happened
and what to do next. We start with failing gracefully for two reasons: First, the way
any digital technology handles fail states is critical to users’ perception of the
experience—and this is especially true with the intimacy of voice. Second, virtually
all the remaining guidance is this section is built around avoiding these very errors.
In the example above, Alysha didn’t realize she was being asked whether she wanted to track an order or place a new order, she
thought she was being asked if she was interested in either. This specific error could have been addressed with how the query was
written (e.g., “Which would you like to do, track an order or place a new order?”). Alysha’s frustration could have been ameliorated
by repeating the question after the second error instead of expecting her to remember a query she was clearly struggling to
answer. Further, the experience would have been less frustrating, and more human, by varying the error response (e.g., “Forgive
me, I don’t understand,” or “Can you say that again?”). Ultimately however, no matter how carefully scripts are drafted and tested,
things can—and will—go wrong.

17
When errors occur:
1. Fail gracefully
Never suggest the user is
at fault. Apologize and
try to help.
Finally, if there is an error loop occurring (as with Alysha’s interaction above), don’t
keep repeating it. After several failed attempts, offer a friendly apology and release
the user from a frustrating back and forth. For example, "I'm sorry for my limited
intelligence. Smart people at [brand] are working to make me better every day
though! In the meantime, you may want to go to their website or mobile app."
Look for opportunities to not speak.
A soft chime and/or change of indicator light can tell users they weren’t
understood without the system telling them so. This is gentler, takes less time,
and avoids too many “I’m sorry”s.
Be helpful.
• Be specific about what went wrong and what the user can do to
resolve it.
• Ask clarifying questions. “I think you said ‘X,’ is that right?,” “Did you
mean…?”
• Offer suggestions. “I can’t find X, would you like information on Y?” “I’m not
able to do that, but there may be an application that can. Would you like to
search for voice applications that can help with that?”
Be humble. Never suggest the user is at fault. Apologize and try to help.
Be human. Consider using light-hearted messages where appropriate (as
noted in best practice #7, humor can lighten the mood when errors occur, but
proceed with caution. Humor should only be used when you know the user is
likely to be in a low stress situation and engaged in a low risk task.)

18
2. Do your
homework
“This doesn’t make any
sense. They should’ve talked
to me first. I could’ve told
them this wasn’t going to
work.”
–Amanda, participant
The most effective approach to avoiding errors, both in execution and strategy, is to
do your homework. One-quarter of our survey respondents didn’t agree that smart
speaker interactions were created by someone who thought about their needs. This
was even more pronounced among respondents who were older, less tech-savvy, or
women. And the best way to know about someone’s needs is to observe and listen.
Far too often, user research is confused with usability testing and only comes at the
end of the development cycle. But how do you know you’re solving the right problem
in the first place? How do you know that your idea for addressing that problem is a
good one, that you’ve anticipated all the use cases, contexts, and things that could
go wrong? How do you know if you’re on the right path—or the wrong path? By the
time you’re usability testing, it’s too late (and too expensive) to ask these questions.
For additional advice on conducting voice interaction research:
Download our Best Practices in Voice Interface Research which includes:
• An Iterative approach to designing for voice, where we identify and explain what
research methodologies we recommend at each stage of the design process.
• Recommendations for successful voice interface research, helpful tips for
researchers who are new to this platform.

19
3. Be a good host
(i.e., setting
expectations)
Device: Welcome to Twitter Reader. I can tell you what is happening, can read your
latest mentions, the last tweets you liked, or were posted by you. I can also search
for trends by city. When I am reading out a list of tweets or trends, you can ask me
to tell you more about the tweet or trend by number. Now, how can I help you?
Heather: Wait, what can you do?
Device: Here are the latest tweets from your timeline...
It helps to think of voice applications a little like a host would: welcome people,
make them feel comfortable, and get them where they’re trying to go as elegantly as
possible. In designing tech experiences, making users feel comfortable should
include many of the same practices that go into web and software design:
consistency, clarity, and comprehensibility; users should know where they are and
what they can do; they should have as much information as they need and only as
much information as they need, and they should have it when they need it.

20
3.Be a goodhost
(i.e.,setting
expectations)
“If I have to go online to
learn how to use [a voice
application], I’m already
online—I’ll just use that to
do what I wanted.”
-Bilee, participant
That said, while it’s important to be unambiguous about what users can do, don’t be
too prescriptive in explaining how to accomplish these tasks. Instead of listing
everything they can say, give some suggestions and design for all likely variations.
Also, avoid “onboarding.” Users just want to do what they came to do. This is even
more important in voice applications, because the ability to “x out” of an onboarding
experience may not exist or may not be obvious. Instead, welcome users quickly, get
them into their task efficiently, and ensure contextual help is always available should
they have questions.
To ensure your users feel comfortable:
• Set expectations at every decision point. Namely, what users can do and
suggestions for things they can say.
• Look for opportunities to provide context-aware self-help within the voice
application. Where possible, avoid sending users off to look at the device app or
your own website for help.
• In multi-step processes, let the user know where they are in the process.
• Follow our advice for writing effective commands, including conducting research
to identify all the various ways users might choose to issue commands.

21
4. Design for
context and
continuity
Chris: Alexa, what’s the yellow light mean?
Echo: The yellow light means you have a new message
or a notification. You can say ‘play my messages’ or
‘read my notifications.’
Chris: Alexa, play my messages.
Echo: No messages from today. You have one
notification. You can say ‘read my notifications.’
Chris: Read my notifications.
Echo: [silent]
Chris: Alexa, read my notifications.
Echo: [reads notifications]
In this case, Alexa does a great job of being a good host and using natural commands. The user didn’t even have to think about
what to do when he saw the yellow light, he just asked. The problem is Alexa’s failure to maintain continuity. Instead of, “You have
one notification. You can say ‘read my notifications’,” why not simply ask, “would you like to hear it?” and allow the user to say yes
or no?
Examples like this abound in our research. Voice technology is still in its early stages. At the time of our research, one couldn’t ask
Alexa the address of a location and then ask “how far is that from home?” Technical limitations exist, but nevertheless, as much as
the technology will allow, good voice interaction design pays attention to context, taking into account where the user is likely to be,
both physically and in the process of task completion. Further, good design is consistent. Users should not have to wonder if
different commands and responses will produce different results.
To design for context and consistency:
• Look at every transition in your task flows to see what can be combined, simplified, or removed. Don’t assume the same flow for
web or software should be used for voice.
• If there is a next logical step to the process, anticipate that step, ask the user if they would like to take that step, and listen.
• Observe the physical contexts in which your voice application is likely to be used and design for those.
• Review all your commands, responses, and confirmations to ensure consistency among words and actions.

22
5. Use natural
commands
“That’s the other problem I
have. Sometimes even the
ones that I like, I can’t
remember the commands to
get them to work, and it’s
[sigh], oh god, now I have to
go look it up.”
–David, participant
Many participants in our research said they had a hard time remembering the right
commands. In traditional web design, good design focuses on recognition over
recall, but with voice interactions, this gets flipped. Since there’s no visual stimuli to
drive recognition, the user is forced to recall important aspects of the interaction. The
less we ask of users in this regard, the better the experience. Users shouldn't have
to learn a new language to interact with your voice application. In this way voice
interfaces can add to the user’s cognitive load instead of making her life easier.
To make things easier for users:
• Commands should be natural and easy to remember.
• Commands should be consistent across applications.
• Allow multiple commands for the same action.
• Commands should be sufficiently unique to help with recall and to avoid errors of
misinterpretation.
Spend time listening to people talk to learn how they think and speak about the tasks
you're trying to enable. Depending on where you are in the development process
and the investment you're able to make, you might do this through a “wants & needs”
focus group or an open card sort, or by simply observing people in their natural
environment like their homes or offices. As much as possible, voice interface
commands should match your users’ thought processes and vocabulary. We
recommend conducting usability testing with a range of users prior to launch in order
to make sure you’ve captured the full range of expressions a person might naturally
use when interacting with your voice application.

23
6. Make the user
feel heard (i.e.,
confirmations)
Device: Do you want to log into your profile for a faster checkout experience?
Brian: No.
Device: Sounds good!
[Brian laughs. To interviewer:] It’s just very inhuman. It’s like, OK, I’m glad we
agree. Just do what I asked, you robot. It’s trying too hard.
We all want to be understood, and
when we ask for something, we want a
response. When talking with one
another, we often confirm we heard
what someone said not with verbal
confirmations but with a change in eye
contact or a nod of the head. Smart
speakers can’t to do this.
Confirmations can take many forms and
practitioners should consider which
confirmation is most suited to the
interaction in question. Our overall
guidance here is that confirmations
should be kept to as few as necessary to
reassure the user and as brief as
possible in order to keep the
conversation moving. (Keep in mind that
error messages are also confirmations,
even if what they’re confirming is
something that didn’t or can’t occur.)

24
6. Make the user
feel heard (i.e.,
confirmations)
Confirmation types and when to use them:
Explicit confirmation: For example, “I heard you say [X]. Is that correct?” For
interactions where a mistake would be significant, it is critical to make sure the
user’s command was understood. Placing an order that would result in the user
being billed or calling someone from their contacts list are important to get
right. This can also be used when the system’s natural-language processing
isn’t 100% confident it understood a command.
Implicit confirmation: In implicit confirmations, the question is implied in the
response. For example, “What’s my commute look like?” “Traffic is heavy and
your commute is estimated to take approximately 42 minutes.”
Sometimes, instead of a confirmation, an acknowledgement will do:
Nonspecific confirmation: For example, “Okay.” A general confirmation is
most effective when the command is simple and straightforward and a mistake
would not be critical.
Nonverbal confirmation: For example, a device’s lights turn red and it plays a
dissonant chime. Although smart speakers have a very limited repertoire of
nonverbal communication, they do have lights and chimes. Think of how much
R2-D2 and BB-8 conveyed with the same vocabulary! Use a nonverbal
confirmation when the fail state is not critical and the acknowledgement may be
conveyed unobtrusively.
And sometimes, a confirmation isn’t needed at all:
No confirmation: Occasionally a confirmation isn’t necessary because
confirmation is conveyed through other means. For example, when asking a
smart speaker to turn on a light or to pause music that’s currently playing.

25
7. Have some
personality
“I love Dom [the voice
persona for Domino’s Pizza].
I already have a relationship
with Dom [through the app
and website] and Dom has
not done me wrong. I
already have a high opinion
of it, so it messing up, I know
it’s going to get it right
eventually. I’ve never used
the Progressive thing so it
messing up on the first time
makes me think it’s a miss. If
it was obviously Flo that
would’ve been cute, that
would’ve upped my opinion
of it from the start. If it had
used a voice I recognized.”
–Amanda, participant
Web content and ad content are often referred to as having a “voice” based on the
tone of the writing—voice interfaces literally have a voice, and many of the same
rules apply. But unlike “voice” used to describe writing, users look to voice
interactions to be more entertaining than screen-based interfaces, perhaps to fill in
for the lack of visual stimuli, perhaps because an audible voice is so personal that
many anthropomorphize the voice-powered assistant. For this reason, so many of
our participants talked about the voice interface’s “personality.”
You don’t need to have a recognizable voice persona, as Domino’s does. And if you
don’t have one, don’t force it. As technology writer Cennydd Bowles* points out,
“Marketers can’t resist an opportunity to force a damn relationship on you. Truth is,
I don’t want to talk to most of my products. They’re dumb utilities. Close and forget.
I want a spade, not the experience of digging.” But our research showed clearly that
users do respond to the personality of voice interfaces, and your voice application
will have a personality whether by design or default.
*Bowles, Cennydd. "What happens next with Conversational UIs – Cennydd Bowles – Medium." Medium. February 19, 2016. Accessed October 07,
2017. https://medium.com/@cennydd/what-happens-next-with-conversational-uis-b9e4699541d5

26
7. Have some
personality
Almost all our participants identified humor as the hallmark of personality. But humor
can be tricky for brands. We recommend exploring the use of humor (if it’s consistent
with your brand), but proceed with caution. As mentioned previously, humor works
best when you can safely assume the user is in a low stress situation and engaged
in a low risk task. They will be far more receptive to a dash of humor if they’re asking
about surf conditions or current movies than they will if they’re checking a flight’s
status or tracking an important package. With humor as with error handling,
practitioners should conduct ideation exercises around ‘what could go wrong’ and
test designs before sending them into the market.
When designing your voice interactions:
• If you haven’t defined your brand personality, do so.
• Listen to how your customers communicate with you and match their tone. For
example, are they casual or formal?
• Ensure your voice application’s tone, word choice, etc., are in line with your
“voice” across other touchpoints.
• Look for ways to differentiate from your competitors’ voices.
One interviewee, when comparing different voice interfaces she used, said
“willingness to help,” in addition to humor, defined the personality of her preferred
assistant. How do you convey “willingness to help” with a computer program? By
using clarifying questions and suggestions instead of error messages. And that
brings us full circle to where we started this section on best practices: handling
errors with grace.

27
What’s next for
Smart Speakers?
We’ve entered an exciting time for
voice technology. Dramatic
advances in natural language
processing and machine learning
have ushered in the start of true
consumer-facing voice
interactions. Significant
investments by some of tech’s
most monied companies have
propelled explosive growth in the
category, both in diversity of
offerings and units sold. With over
70% of the market, Amazon is
clearly the one to beat. Amazon’s
product strategy for Echo seems to
be to release products almost like
betas, trying out new form factors,
new interaction models, and new
functionalities. But will this serve
as a valuable public laboratory or
will it undermine confidence in their
brand? Perhaps both. Further,
when competing against cash- and
talent-rich companies like Google
and Apple (and maybe
Facebook?), first-mover advantage
is not a barrier to entry.
We recommend device
makers work with
strategic insights
partners to develop best
practices that can be
used by those building
in this uncharted space.
To learn more about
research to develop
best practices, read
how Google partnered
with AnswerLab and
developed design
principles to guide their
mobile advertising
clients. Case study
Our research provides a snapshot
of this changing landscape. We
believe many of the sentiments
we’ve uncovered—and certainly the
best practices we recommend—will
continue to be relevant for some
time. Smart digital leaders, product
managers, designers, and
marketers will certainly want to
keep abreast of this new platform
as it develops.

28
What’s next for
Smart Speakers?
Based on our research, here are some of the developments we’re continuing to watch:
Changes to the business ecosystem
Monetization: When Amazon, Google, and/or others provide a path to profitability for developers, we expect the quality of voice
applications to improve materially. Will your voice application be ready to compete?
New Entrants: Will Apple’s long-anticipated HomePod be able to carve out its own space? Is Facebook really preparing their own
entry? Samsung? Microsoft? What will the effect be of these and other entrants into the space?
New functionality
Biometric authentication: Amazon just announced that they can do what Google has been doing, recognize which person is
speaking and adjust the experience to that individual. But none of the smart speakers yet have true biometric authentication, and
given the security concerns expressed by those we surveyed and interviewed, this will be critical for “v-commerce” and other
interactions with sensitive information. Will you be positioned to capitalize on secure transactions in voice?
Visual feedback: Amazon recently launched the Echo Spot and before that the Echo Show, both smart speakers with a screen as
part of the interface. Google is rumored to have a similar product in development. Expect to see more experimentation in this space
with multi-modal interactions. Will you be prepared for all the usability challenges that come with that?
Always on and everywhere
Always on: Participants in our research voiced few concerns about the technology listening to their conversations. (Where they
had concerns was with sensitive information being shared with voice applications.) It was clear from our interviews that
convenience was more important to them than privacy concerns. We expect smart speakers will eventually reach a point where
they are listening for more than just their “wake word” and they’ll remember more about their users. This will open up new
possibilities for contextually-aware, continuous interactions. What will this mean for voice applications? Will you be ready?
New contexts: Smart speakers’ voice assistants are already making inroads in the automotive space and that will continue. They
are already used to manage smart appliances, and we expect they will soon become the hub of the smart home, with microphones
throughout the house. The voice assistants in our smart speakers are already on our phones, and while those devices are not yet
delivering seamless experiences across devices, they soon will. They are not yet in most computers or mixed reality (VR/AR)
devices, but we anticipate they will be. As voice technology expands into new contexts and permeates our lives, those companies
that support users throughout their day will be poised to become equally as ubiquitous. Will your company be one of them?

29
Your customers are talking –
make sure you’re listening
The landscape will continue to shift as smart speakers gain functionality
and conversational interfaces improve. Along with these improvements,
your customers’ needs and expectations will also increase.
Do you have questions about designing for voice interactions or want help
navigating your voice interaction strategy?
HOW TO WORK WITH US
AnswerLab can support your voice experience efforts in the following ways:
• Bring our Smart Speaker team on site for a Q&A as you begin exploring your strategy for
designing voice interactions.
• Gain ongoing smart speaker insights—stay tuned for our industry-specific findings
and recommendations.
• Engage with AnswerLab workshops to help you define your digital strategy and plan for the
user insights you’ll need.
Interested in working with us? Contact us at answerlab.com/contact-us
About AnswerLab
AnswerLab delivers insights and advice to create exceptional
digital experiences. The world’s most innovative brands rely on
our research to improve user engagement, reduce development
costs, and increase conversion rates. We partner at each stage
of the product development cycle, helping digital leaders envision
new experiences, optimize existing ones, and measure
their impact.

30
We ran an online survey of 1,000 smart speaker owners
throughout the United States with a panel provided by the market
research company Lucid. Participants for this segment of our
research included people representing:
• A range of device ownership including the Google Home
device and Amazon’s Echo, Dot, Show, and Tap devices
• Ages 21 to 75
• Smart speaker ownership ranging from one month to two years
About the
research
We carried out in-depth in-person interviews with 10 smart
speaker owners in Sacramento. Why Sacramento? It’s one
of the 15 most demographically representative cities in the
United States. Participants for this segment of our research
included people representing:
• A range of device ownership including the Google Home
device and Amazon’s Echo and Echo Dot devices
• Ages 24 to 60
• Mix of household incomes ($20k - $200k)
• Mix of education levels (high school graduate or higher)
Participants submitted seven days of diary entries,
providing a snapshot of a typical week of activity. Following
the diary submissions, in-home interviews allowed us to
observe and investigate smart speaker usage in context.
The interviews were 90-minute sessions that included both
general exploratory questions about participants’ current
and desired use of smart speakers as well as usability tests
of several representative voice applications.
We reviewed the current state of the smart speaker user experience
through a multi-method research study including both qualitative and
quantitative methods.
Qualitative Research Quantitative Research

31
Smart Speaker
Research Team
Chris Geison is a UX Researcher at AnswerLab where he leads research to help
fortune 500 clients identify and prioritize insights that improve their business results.
Drawing from his experience in digital strategy at Charles Schwab and a fascination
with motivation and change honed working in behavioral health, Chris has led
studies ranging from the role of emerging technologies in the future of banking to the
behavioral effectiveness of workspaces, with a particular focus on
conversational interfaces.
Chris Geison
UX Researcher,
smart speaker
research and
insights lead
For more than a dozen years, Ryan has been advising the world's top digital
properties on how to improve their user experience. He leads research across a
variety of industries, including telecommunications, financial services, automotive,
retail, and design agencies. Ryan specializes in customized, quantitative and
behavioral research methodologies across mobile, tablet, ad desktop. Before joining
AnswerLab, Ryan was a Senior UX Consultant at Keynote Systems, Inc., a leading
provider of UX research software. Early in his career, he worked at Accenture as a
developer, team lead, and technology manager for complex backend
system integrations.
Ryan Haupt
Principal UX
Researcher,
quantitative
survey lead

32
Lin Nie
UX Researcher
Lin Nie is a UX Researcher
with a PhD in Experimental
Psychology. Before
AnswerLab, she consulted
for Amazon’s most profitable
product, and led foundational
user research for startups.
Her research in cognitive
science and artificial
intelligence has appeared in
Wired and Slate.
Smart Speaker
Research Team
Beth Devine
Project Manager
Beth manages research
project logistics for Fortune
500 clients representing a
variety of industries,
including e-commerce, retail,
financial and pharmaceutical.
With a background as an
organizational psychologist,
she has a decade of
experience in management
research and theory. Beth is
also skilled in qualitative and
quantitative research
methods and data analysis.
Amy Buckner
Chowdhry
AnswerLab CEO
Amy Buckner Chowdhry
founded AnswerLab over a
decade ago to help the
world’s leading brands build
better digital products. Under
her watch, AnswerLab has
grown to become a trusted
UX insights partner to
companies like Google,
Facebook, Amazon and
more.

Answer lab best practices in research and design for voice user interfaces

More Related Content

What's hot (13)

Similar to Answer lab best practices in research and design for voice user interfaces (20)

More from Isidore Gotto (13)

Recently uploaded (20)

Answer lab best practices in research and design for voice user interfaces