Dina is a product manager on the Google Assistant team. She leads feature discovery across Google Assistant surfaces. Prior to Google, Dina’s experience varied between management consulting, product management and startups.Dina has MBA from Harvard Business School. She earned her bachelor's degree in Electronics Engineering from the American University in Cairo.View the profile
Tanya Gupta is a product manager focusing on developer experience and App discovery within the Google Assistant. Prior to this, Tanya was an engineering manager responsible for scaling the Assistant. Tanya is currently an MBA candidate at Haas School of Business at UC Berkeley. She earned a bachelor's degree in Computer Science from Carnegie Mellon University where she graduated with the highest honors.View the profile
About the talk
For every developer, finding and retaining users is critical to success. This session will be a pragmatic guide to all of the discovery tools and channels the Google Assistant offers, and an overview of the best development practices to drive retention.
Hello everyone. Thank you all for coming it is so exciting to see you here by name is Sonya Gupta and I am a product manager on the Google Assistant focusing on natural language understanding and discovery of third-party actions. And this is Dina. I had dad's I'm also a product manager and I lead Discovery and engagement. Where are here today to talk to you about how to make your action more discoverable and build a strong and consistent user base. This faster than be a surprise to you at this point that
distance is a large and ever-growing ecosystem. We have over 1 million actions across more than 500 million devices in more than 25 local. Now the question is, how can you stand out? How can you get the share of that user base? Many of you may recognize this and see the similarity to a marketing funnel and a lot of building a user base is creating a brand that users recognized immediately. We've broken this funnel into three stages the first and biggest stage is all about creating
awareness for your action. So that users know that it exists and what they can use it for Once they know about your action, the second stage is all about creating an engaging experience. So that users are immediately hooked. And the last page is about retaining user so they come back day after day week after week year after year. For all three of these stages, we'll talk about different techniques and different tools that you can use to improve your action. Let's Jump Right In and talk about some of the ways that you can build awareness around your action
and make it more discoverable to users. There are various Discovery opportunity than Google Assistant starting from assistant directory to in conversation suggestions to a knob Discovery to Belton until you can even ask Google Assistant. What can you do and it will respond back with relevant actions to try now, let's jump and focus on the assistant directory also known as Explorer Explorer as a one-stop-shop for all assistant actions. Is available on web at assistant. Google.com
Explorer and the assistant app on both Android and iOS, the directory is interesting because we see that more than 30% of our third party actions being discovered and used directly from there that are supported and explore and how you can optimize each of them. One of the primary use cases for Explorer is browsing when we designed the landing page for Explorer. We made sure that the most relevant and personal recommendations are showing at the top of
the page would last relevant recommendation shown as you scroll down. At the high level. There are three key areas to highlight in the landing page. First we have fun sexual suggestions. These are relevant recommendations based on the user's contact such as time and location. For example, I can see here. That's where my morning I can play my favorite playlist or listen to my daily brief. Next comes Dynamic categories to date we have four of these what's new which highlights new high-quality action was popular with highlights
overall popular attractions. What's trending which highlights newly popular actions and you might like which recommends action that the user may like based on their history with assistant. We have also just lost editorial fixed. This is a new feature highlighting high-quality and seasonal actions through Visual campaign. An interesting facts about the landing page is that we see that actions that got promoted in that page specifically and dynamic categories are ten times more likely to be used on that day. If you ask me, how can I ensure that my action gets
featured in the landing page? I would say that the Main Ingredients here as quality and quality can mean a variety of things. It means solid ratings positive reviews said they have signals among other things. The Second Use case that Explorer support is categorical browsing and search here. We're trying to expose the user to the variety of options available within a category of Entry while also allowing the user to easily find their favorite actions through a quick search. Let's say for example, I'm interested in food and drink by going to that
category. I can see all the relevant actions group my head. Categories. For example, I can see here that I can order food from Dunkin Donuts and Domino's Pizza. Whereas I can find recipes from Food Network. I swear search on how to complete the idea is to expose the user to the variety of capabilities supported by an action of Interest while also providing predicted suggestions to the user based on their search key term now if you want to make sure that's your option is showing and the rights category and it's being surface for the right search key terms you
want to ensure that's your display name your description and Sample invocation are reflective of what reaction does and actually include useful keywords. Finally and most importantly let's move to the action page. The action page is a welcome mat your action. It should reflect your brand and it should invite the user to try your action out at a glance action phase consists of two main sections. The top section should provide the user with a quick summary of what's your action does through your imagery description and Sanford and vacations
that idea about your action before trying it out. The second section covers more details. The user here can see your ratings and our newly launched reviews. This allows the user to assess the quality of Your Action before you give it a try. Know that you walk through the main user Flows In explore, how can you optimize your page and really shine in the directory? Let's go to an example. I'm a French Bulldog lover and I wanted to publish an action about French bulldogs. Unannounced I chose a short and concise
name French Bulldog. Trivia it explains exactly what my action does and it fits nicely on the screen on your right eye fixed a long display name that got got cut off and I also use redundant key words such as the best. Under left for my imagery. I created Graphics based on my dog their eye contact and they look great on a large screen. On the right I use the distorted picture of my dog's the undeniably cute. It doesn't look great on a large screen and it doesn't reflect well on my brand.
For my description on the left, I wrote a short and glanceable description, but I made sure that's all the important key words to me. Are there a word such as Frenchy's French Bulldog lovers dog lovers dog and fox This is great as its would potentially allow my action to be Surfers 4D search key terms on the right. I have a long description that got him and I didn't use any use for keywords. Finally for my sample invocation on the left. I maximize the number of sample invitations on the screen. I also made sure that they were
descriptive instead of just limiting myself to talk to French Bulldog. Trivia. I use more descriptive sample invocation such as ask French Bulldog trivia about grooming. So in summary, if you want your auction page to really shine in the directory you want to do to do to do this for things you want to have a short and unique display name you want to have a descriptive Yabba glanceable description. You want to high-five Stamford invocations that are descriptive and you want to use high quality Graphics encouraged to you that happens keywords to your action, but don't
overdo it and definitely don't include keywords that are not relevant to your actions say otherwise, you may be penalized in a bit and talk about in conversation Discovery through Belton and tan So another important aspect to building awareness. It's a surface your action when it's most relevant to the user. One such opportunity is when the user is interacting directly with the assistant, but natural language understanding is hard in addition to the multitude of languages. There's dialects their swing and there's cultural variation
the increase the number of ways that a user can express the same thing. So let's pick an example of some of the ways that are user can express that they're stressed out and they want to come down. These examples here are just some of the ways users would say it in American English. It doesn't include other variations of English such as British English or Indian English. Then when you start thinking about other languages, you can imagine just how long this list would be. Well, I work with a team of expert linguist to Define intent that capture
all of these natural language variations into one definition. Has Brad mentioned in the developers keynote be built in intent or a common taxonomy to describe actions across Google services? As a developer, you can integrate your action with irrelevant built-in in time and this allows Google to handle the natural language understanding and the semantic parsing for you. Google will extract the relevant parts of the Prairies and pass it to your action in a structured format. So let's take a few of these queries and see what they look like as a
built-in in 10. On the right here, you can see that we've defined the intent and called it start calming activity. We've also identify three arguments. The first is the activity the user wants to partake in second is how long they want to partake in that activity and third is a reason why they're stressed out. On the left you can see how we've Matt the queries to the built-in intent. So in that second example, we've extracted breathing exercises and mapped it to the calming activity and extracted the my interview and. Became the stressor.
Building in time to make it so that developers don't need to worry about the natural language variation. You just need to make sure that you can fulfill all of the arguments specified by the belt women time. Another benefit to built-in intense is that egg is Google information about the type of queries that your actions can handle and how good it is in responding to those queries Google can then use that info to apply Ai and surfacer action when it's most relevant to the user. On the screen here. You see some
Explorations on how built-in intense can be used to make actions more discoverable. Example on the left there. The user has asked to check their Leo horoscope and is presented with for high-quality action that Google is confident can fulfill that request. Building intent are not just limited to the assistant and previous sessions. You may have heard about our actions and how it is also built on Dothan and 10. What that means is that when you integrate your action with a built-in and time, not only are you creating opportunities for Discovery
inside the assistant but across a variety of Google services such as Google Play and Google search. So now I know all of your itching to find out how to integrate with a built-in intent. So let's do a quick demo. Can we switch to the laptop, please? But here you can see that we signed into dialogflow and I've already selected my action here called breathing, but though I have a series of intense and I've already defined and to integrate with a built-in intent. I can create a new intense. Let's call this intense demo.
Inside of the events section. I can see all of the building in tents that are available for me to integrate with. Given that my action is about meditating and breathing exercises. I think start calming activity probably makes the most sense. But here I can see a few example queries that would map to this building in time as well as all of the arguments. Know the proper way to fulfill. This intent would be to create a webhook r i could parse the argument values and respond to the user accordingly, but for the sake of time, I'm just going
to create a simple response year. I can save my intent and try it out in the simulator. Now as usual I could choose to speak directly to my acting but let's try one of the query that I know Master this building in 10. All right, as you can see here. I'm presented with a list of four actions that I can use to meditate for 1 hour. Let's switch over to the phone and try it on on the real assistant. I want to meditate for 1 hour. Alright, which one would you like to try and you can see the same action with my action being the first one there? I switch back to the
lights, please. We are excited to announce built-in intense as part of Google IO. These are just some of the intense that were working on this year. And the ones are the stars next to them are already available as a developer preview encourage all of you to sign into dialogflow. Try them out familiarize ourselves with them and see how they can fit into your action. the word that we talked about two different ways that you can build awareness reaction first, but
optimizing the assistant directory page and second by integrating with built-in in now that your action is discoverable. Let's jump to the next stage which focuses on creating an engaging experience. So that users are immediately hooked. In this instance, we access our project through the lens of critical user Journeys. We pinch Google Assistant to help our users throughout their day, but based on our research and from analyzing our usage Behavior. We have seen that we truly offer values and CD key moments in the morning
during the commute and in the evening. In the morning, we had part users from the minute they get out of bed till they had out of the door from setting alarms to checking the weather to listening to the news that assistant is always there. Next comes to come you Journey imagine you are driving to work and you have a million things on your mind. The assistant can make that time productive. It's good to help you with navigation listening to your favorite podcasts or even rescheduling your next meeting. Finally comes the evening Journey the time to distress
still there for you. It could help you with things like control your smart home devices play relaxing music or how do you cook a meal as a developer? What are you focus on morning productivity media on the go or smart home by focusing on one of these critical user Journeys, you'd be optimizing for moments where we truly believe we can offer assistance to our users. This will automatically lead to higher engagement and would potentially lead to more organic Discovery opportunities. Another key source for engagement is our
newly launched action length. Action links are hyperlinks that point directly to your action. They make your action discoverable from any platform or Surface. They allow users to engage directly with your actions are prospective. Oh how they found this link weather on the website and a social media campaign or even on a billboard. I am personally considering having a billboard for my action. Isn't this awesome? On the assistant side. We're going to promote these links everywhere. You're going to start seeing these links
on your action page in the directory. This will allow users to try your action easily and engage with your directly if they had a good experience. They're also going to be able to share it with their friends. The poor thing action links is very easy. All you need to do is go to the actions consoling over there. You can publish a URL for your action and choose a title for your leg. Once you do. You can instantly get the link that you can start sharing once you have it published link. You're encouraged to promote it everywhere. You might have seen this in the
developers keynote, but that's face has been some great experiences on Google Assistant helping users meditate. They also happen to have a reading and Gage social following now every time they have a new updates since they since they have an action link. They can direct their users directly to that update they can do so through a blog post like they have done. So on Tuesday or do earthquakes week. So just to recap so far we have talked about awareness through the assistant directory and built-in intense engagement through critical user Journeys and action links.
Now, let's move to the last piece of the funnel retention. Retention is all about creating a loyal base of users that derive value for their action. Not only do these users bring consistent traffic, but they also eventually become your evangelist and they feed the funnel by creating awareness for action. Loyal users the best way is a lot of what Dina talked about with by creating a high-quality action wide Dynamic and unique content that users find meaningful. Once you're confident that you're actually does this there are few tools that you can leverage to
re-engage and maintain a relationship with a user. The first feature allows you to send an update on a daily basis to the user at a time specified by them. Let me give you a personal example of how I like to use this feature every day at 2 p.m. I like to grab a coffee and just escaped from work for a few minutes and Destroy. I'm also an avid Story Reader and reading stories really relaxes me. First set of the future so that every day as I'm getting my coffee. I receive an update with a story to read.
Another feature that helps of re-engaging users is a push notification feature as a named announced this feature allows you to push a notification to the user whenever your action deemed it appropriate. Let's say for example, you're action track cryptocurrency prices and there is a fluctuation in the price your action can automatically push an update to the user so that the user can take the appropriate response. Both of these features can be enabled for your action inside of the actions on Google console. Once you've enabled it, you can start prompting users
to see if they'd like to receive these updates or these notifications from your action. The last tool that I want to talk about bills on the Wii user Journeys that Dina talked about as part of the engagement section. We really believe that these are some of the best time that use the assistant can provide the most benefit to the user. For this reason we built the routines assistance feature which allows the user to customize the information that they received when they trigger routine. So freaked out when I say Hey Google good morning.
I've customized my phone to take itself off of Silent to tell me about my commute the weather as well as my first meeting and then kick it into my favorite music. I need developer. You can also enable your action to be incorporated as part of a routine. Following the same model as a daily update. Once you enable this for your action. You can start prompting users to see if they'd like to incorporate it as part of one of their routines. After the user. Do the next time they trigger routine your actual automatically get
called. Another important aspect to retaining users is to make sure that users can have a smooth and free-flowing conversation with your action. Have you ever had that situation where you walked up to someone and you ask them a question and he just didn't get it. So you rephrase it a little bit hoping that they'd be able to understand you this time and they shrug their shoulders One Last Time by third time hoping that they'd be able to answer and tell you what you're looking for and they still could not help you
frustrating was that experience? In the same way you want to make sure that your action understand your user and is able to give it the information that is looking for. We have a few tools that you can use to ensure that. The first is a session flow feature inside of dialogflow, which allows you to analyze the conversation flow users are having with your action. What should the suction flow of the mobile Legends action as an example? This action was built a little over a month ago
and quickly Rose to become a top 10 action globally. So it makes for a great case study. Reaction for those of you who aren't familiar with it is based on the mobile Legends game and it tells users about the different characters from the game. So you can see from the session flow hear that about 97% of users that start a conversation do so in the default welcome in times where they're presented a list of characters to choose from let's dig a little bit deeper and see where the conversation goes from here.
Are you can see sessions low allows you to break down the conversation on a person tent basis. I was able to click on the default welcome intent and see what the next set of intense that trigger come there. In this case about 57% of users clicked on one of the characters to learn about it. I want to dig one level deeper and see what happens from there. 41% of users that selected the character read about the character and then 32% dropped off from there as we can see from that red line coming out of the intense session flow is a powerful tool
for you to identify what the most popular intense in your action are this is great. If you're looking to expand your action and want to see how users are already leveraging it. The second you think that I would recommend using session flow for would be to use it to identify where the large drop offs are maybe there's certain times where it just not understanding the user or it's not providing the user the information they're seeking. Identifying these drop-offs is a great way to ensure that you're not losing users when they interact with your action.
And the last thing I'd encourage you to use this procedure for would be to identify the intense. I just aren't getting any love and that might mean restructuring your action. So that users can find those intense much more easily. Another tool I want to quickly touch upon is the history feature inside of dialogflow. This feature allows you to see the most popular queries that users are issuing to your action. It's a great way of ensuring that your action response in the appropriate way and is giving the user what they're looking for. You can also see
a few of these queries have an! Next to him that's never a good thing these exclamation marks mean that either your actions could not be reached or that it David default fall back in time, or worst of all it just quit. In all of these cases that is a poor user experience and the users less likely to engage again with your action. The first is talking about creating a smooth conversation and the second was all about reading the metric right It's always important to keep a lookout on the
metrics to ensure that your action or keep that your users are having a good experience and their action is performing consistently. So let's talk about a few metrics that are particularly important. The first is a star rating that users are giving your action B star ratings are good signal as to whether users enjoy your action and are likely to come back. Ideally, you're the star rating should be always either be five star or 4 star and if you're seeing a lot of one star into stars, that means something is going wrong and it needs to be fixed. Are the ratings to keep a
lookout on are things like active users average conversation length messages and abort rate. All of these should be what you expect them to be. You want your heart rate to be low you want your active users to be high. You want your average conversation went to be what it is if you Any deviations that means against something has gone wrong with your action. Different things in this session. Let's quickly recap. What they were we broke the funnel into three parts. The first was we talked about awareness where we talked about two different to
optimizing for the essence assistant directory page and seconds integrating with built-in intense. The second was about creating and engaging experience and this could be done by tailoring for the key user Journeys where the assistant can provide the most benefit. And the last part of the funnel was about retaining user so that they can come to come back repeatedly and here we talked about creating a compelling conversation and using metrics to identify pit balls. This concludes our talk for today. Thank you so much guys.
Buy this talk
Access to all the recordings of the event
Buy this video
With ConferenceCast.tv, you get access to our library of the world's best conference talks.