[MUSIC] OMAR AFTAB:
Hello everyone. I'm Omar Aftab, VP of Conversational
AI here at Microsoft. We're going to talk a little bit today about Copilot Studio, which Satya announced in
his keynote this morning, and how you can use it to build your very own Copilot or
customize Microsoft Copilot. I think we all believe that
the future is conversational. There was a Gartner report
recently that said, in just three years, a third of all work experiences
will be conversational. That's 80 percent of company to customer interactions
will be with a digital concierge
or a virtual agent. As you're aware from all of our other announcements
today at Ignite, there's an entire
Copilot ecosystem that's emerging around
Microsoft Copilot. Our first Copilot solution
was GitHub Copilot. That's been GA for
more than a year now. We're all familiar with it and the amazing leap in productivity that it
has already brought. We have a set of Copilot
experiences designed for core business functions
like Copilot for sales, Copilot for service, and more. We're enabling anyone to create low-code and no-code solutions with Copilot for Power apps, Power pages, and more. Just a few weeks ago, we announced that
Copilot for M365 was now GA. Underneath this emerging
ecosystem is Copilot Studio, which is a new product that
allows users to extend and customize first party Copilots or build their own
custom Copilots. Because those are
the two key asks we're hearing from customers. I really want to
enable to use of my own business data,
my own processes, my own enterprise
specific scenarios into these Microsoft
Copilot experiences, so I can configure
and tailor them for my users and
my own employees. Or I want to build my own new custom Copilots
and publish it to the channel of my choice for
my users to interact with. With Copilot Studio,
you can do both of these things through one
SAS based graphical tool. Within the studio, you have a variety of conversational
tools at your fingertips. You can enable
generative AI over your enterprise
knowledge, your websites, your SharePoints, your files, and instantly get rich multi-turn GPT
conversations over them. For key scenarios, you
can build manual topics where you control the flow
of turn by turn dialogue, or you can call actions. Call APIs, build
AI plug-ins with hundreds of pre-built connectors to common back-end systems. You will even be
able to build and use custom Copilot GPTs, similar to what you saw in the recent open AI announcement. Then you take all these powerful
capabilities and you create your very own custom Copilots to deploy your own site or
channel of your choice. Or you can choose to extend Microsoft first party Copilots with your own
enterprise scenarios. For example, you could enable your employees
interacting with M365 Copiloting
teams to ask about their benefits information or look up sales status or more. Now, Copilot Studio will be included with the
M365 Copilot skew. You get the Microsoft Copilot, and along with it
you get the tool to configure it and customize it. Underneath all of this, Copilot Studio
brings together and integrates with a variety
of technology pieces from Power Virtual Agents to
Power Platform AI Builder, to Azure Open AI Studio to
Azure Cognitive-Services, Azure Bot-Service,
and much more. Let me double click on the first of those two use cases the
customers are asking for. Building your very
own custom Copilot. Copilot Studio exposes
a full end to end life cycle for Copilots within
a single pane of glass. You can build, deploy, analyze, manage Copilots all from within
the same web experience. Since it's a SAS service, everything you build
is live instantly. You can simply sign up and start creating your Copilot
in the graphical studio. And the first thing you do is enable generative AI over
your knowledge base. You simply point to
your knowledge sources, the ones you want your
Copilot to understand, your websites, your SharePoints, your files, your back-end APIs. The Copilot will be able
to instantly deliver rich, multi-turn generative AI powered
conversations over them. You could point to, for example, your HR policy documents, your HR SharePoints, your benefit brochures, your
PDFs or files and then ask, what is the leave policy and get answers relevant
for your company. Then, fine, you've got your Copilot up and running with generative answers
over your knowledge. You'll want to
supplement it with any additional specific
topics you want it to handle. Customers often tell us that in addition to
generative answers, they want to have tight
control over what the Copilot says for
business critical topics. Say you want to
process a return or take an order and you don't
want to send that to the LLM. That too can be easily
done right within Copilot Studio.You can use the
graphical studio to build custom-manual topics where you control the flow of turn
by turn dialogue explicitly. Now your Copilot can answer
questions both using generative AI and your own
custom-built responses. But beyond just
answering questions, you'll also want
your Copilot to call APIs and take action
on the user's behalf. To do this, you can easily integrate the Copilot
with your back-end APIs, with hundreds of
pre-built connectors to common systems that come
with Copilot Studio. You could call, for example, your SAP back-end or any of
your custom enterprise APIs. You can also create generative
AI plug-in actions and Copilot will
automatically chat with the user and gather the information
necessary to use them. Think about this for a second. Say you want to call
an order status API and it requires an
e-mail address. All you need to do is
expose that API as a plug-in and the system will
automatically figure out, this needs an e-mail address. It'll ask the user, hey, what's your e-mail address? With that response, it'll call
the plug-in and invoke it. Now your Copilot can answer
questions and take actions. Once you're ready to get
your Copilot up and running, interacting with your
users and employees, you can simply deploy it with a few clicks to the
channel of your choice. Since Copilot Studio
is a SAS service, it'll be live and ready
to use immediately. There's no infrastructure
to deploy or manage. Then as users interact
with your Copilot, you can see how it's working
with built-in analytics. You can monitor your
Copilot's performance, get insights into
what's working well, what new topics or knowledge you might want
to add into it next. Let's see all of this
in action with a demo. I'm going to ask
Gary on my team to come up and build us a Copilot. (applause) GARY PRETTY:
Thank you. (applause) Awesome. Thank you so much, Omar. All of the customers
that you can see on the screen right now are already using Copilot Studio to build
their own custom Copilots. Using the power of generative AI to bring
value to their users. But today we're going to focus
on one specific customer, which is Holland America. Holland America have
been recognized for 150 years as leaders
in cruising. They operate 11 ships, more than 500 sailings a year
to all seven continents. But booking a
cruise can be hard, and Holland America know that there's a lot
to think about. The type of state
room that you'd like to book, the location. Which cruise do you want? Which excursions will
you have along the way? Holland America are using Copilot Studio to build a
custom Copilot that can act as a virtual concierge on hollandamerica.com and be with their users throughout
their digital journey. Over the course of the
next 10 minutes or so, I'm going to show
you how simple it is to get started
with Copilot Studio. How you can make your Copilot instantly useful by
grounding it in data. How you have control to add those specific use cases
using our authoring canvas. Then how we can scale that
out with plug-in actions. Let's switch over to the demo machine and
we'll take a look. Awesome. Here, I am on the creation screen
for my Copilot, and all I need to do is provide
a name like I have done here and I've pointed it to
the Holland America website. Now from here, all I
do is click "Create", and within a few seconds, it will have created a
Copilot just like this one. Now, instantly because I'm pointing to hollandamerica.com, I can start asking questions. Here I say, do I need a
passport for my cruise? We can't necessarily anticipate all the questions that
customers are going to ask, but using the data
on the website, we search across that and
we use the power of GPT and our generative
answers capability to summarize an answer, rich and with citations so that your users have full confidence as to where that data came from. Even better, I've heard that I need six month validity
on my passport. I ask why do I need
six month validity? But I don't have to say why do I need six month validity on my passport because the
generative answers capability uses the conversation
history and the context of the conversation
so that I can have a natural and fluid
engagement with my Copilot. Now that's pretty incredible. But beyond the use
of generative AI to answer these probably
thousands of questions, any questions that
your users have, what most companies need
is control to build more specific use cases and
scenarios for their Copilots. For that we use topics. I'm going to head over
to the topics page. Now a topic is an
individual module of conversation that you can design to determine what
questions will your Copilot ask. What messages will it send? I've started building
one of these in our state of the art
dialogue builder. You can see here, it's actually relatively simple
for me to get going. I provide some phrases
that are examples of what your users might say in order to trigger that
particular topic. Then I can add
nodes to my canvas. Here I can add a
question node to ask them if they have
an existing booking. I can branch my
conversational logic. If they do have an existing
booking, I can say, great, now I can get you
personalized recommendations based on your existing spend, based on your
existing engagement, your recent engagement
with Holland America. Adding more nodes to this
canvas is incredibly simple. There's a wide range of
things that you can do. But what I want
to do is just add an extra message node just to
really introduce the topic. Sure, I'm happy to help
you find that information. Then instantly, we can say what excursions are
available for my crews? You can see now that it's executing in the test chat
on the right hand side, I even have full
visibility of exactly where I am within my
topic at all times. The testing experience is really right there
along with you, so that if you want
to jump out and make some changes and then carry on with your conversation,
then you can do that. Here, I can say, yeah, I've got an existing booking, they ask me for my
booking number. We're now at that it's gone down the right
branch for the condition, and it's asking me the question. Here I'll give it
my booking number. It makes a call out
to Power Automate. We have integration with lots of external systems
including Power Automate, and right there
it's brought back personalized recommendations
for potential excursions for me on my cruise. That's awesome. Let's just stop there for a second and
just really highlight. We had the power of
generative answers, the ability to ground my Copilot in my data in order
to answer questions. We also though, have full
control of the conversation. When you're building the topic, you are entirely in control. When you need it, that
control is there. Now we're going to move on
and we're going to talk about plug-in actions which
Omar was mentioning before. Because it's fantastic
that I can build out my specific scenarios
in my Copilot. But how can I then scale it
to make sure that my Copilot really has a rich plethora of capabilities available to it? Adding the plug-in action
is incredibly simple. I simply open the
wizard and I can search across the available
thousands of connectors. I can search across my own
pre-built connectors which wrap my own enterprise APIs. I can also use Power Automate flows and the new flows for Copilot that
have been announced. Once I select one of these, it takes me just three clicks
to add it to my Copilot. I've already added one here. This is one of Holland
America's APIs for getting cruise itinerary
and booking details. There's nothing really
complicated here. It's just going to call
that API endpoint and return a JSON payload with
a lot of information. But where this gets
really magical, is when I head over to
our generative AI page and I enable generative actions. Generative actions
will enable your bot, using the power of
GPT and Azure Open AI to pick the correct plug-in, depending on the user's query, execute that plug-in, and then provide an
appropriate response without any authoring needed. Let's take a look at
what that's like. I'm going to restart
my conversation. Now that generative
actions is enabled, I'm going to ask, what date
does my cruise depart? I'm also going to open a
new dynamic chaining log, which will show me the
plug-ins that have been picked for me
by the Copilot. As soon as I ask that question, it's going to select hopefully the Holland America plug-in. It's done that but it has also recognized as Omar said before, there's a missing piece of
information that I just don't have in order to be
able to execute this API. It has asked me that question, it has generated the
question for me. I didn't have to
author anything. Let's go ahead and put
in our booking number. Just like that, it will
populate the booking number, call the API and generate
a response for me, a very succinct response that's contextual and relevant
to my question. What happens if I ask something
more open ended though? Where does it stop? That should select
the same plug-in, pick the same plug-in for me
and then provide an answer. Now, think about that. That was a more open ended question. I could have meant anything. I didn't provide
the context that I was still talking
about a cruise. I didn't even provide
my booking number. It used the whole context of the conversation that I've been having over
the last few terms, which was entirely generated for me to select that plug-in, execute it, and provide
me with this rich answer. All of this was based
of three clicks. Adding the plug-in,
in three clicks is all I had to do to
enable this experience. (applause) Let's just quickly summarize. We saw at the beginning
how straightforward and quick it is getting started building your own Copilot, how quickly you can make it useful by grounding
it in your data. Today we saw a public website, but you can also point your
Copilot to SharePoint sites, OneDrive sites, and even bring your own data with
Azure OpenAI Studio. Then we saw how you're always in control
where you need it. Using the state of the
art Dialogue Builder, and the rich intuitive
experience that you have there. Then finally, we saw how you
can scale your Copilot by simply adding your
own custom APIs using a thousands of
pre-built connectors. We watched as your Copilot generated turn after turn,
after turn of the conversation, with only three clicks
worth of effort from me. With that, I'm going
to hand back to Omar, but thank you so much. (applause) OMAR AFTAB:
As we just saw, you can use Copilot Studio to easily create your
very own Copilot like Holland America has. But the other thing
that customers have been asking for
as we discussed, is the ability to extend Microsoft Copilot
experiences with their own enterprise scenarios. Now let's have a look at
how you can do that with all the same powerful conversational capabilities
that we just saw. Say all your employees are using Copilot for Microsoft
365, for example. Say you'd like to allow them to ask questions
about medical leave, or simply query one of your back-end systems right from where
they are right now, right from within Copilot
for Microsoft 365. Well, with Copilot Studio, that's incredibly simple to do. You start from Copilot
in Microsoft 365, and you find that it can't
do something you want. You discover and
launch Copilot Studio. Then you use the same
intuitive graphical interface that Gary just showed with all of those
capabilities to build new topics that you want
Copilot to be able to handle. You can look up enterprise data, expose generative answers, call your back-end APIs, all of the things we
just saw and more. Then you can publish those back to Copilot
with just a click. Now, when you ask the same
question in Microsoft 365, it can answer your request. I'm going to ask Kendra
and my team to come up and show us two demos. The first is a quick
video that shows extending Microsoft
365 Copilot with an AI plugin to look up and return some enterprise
back-end information. The second will show extending Microsoft 365 Copilot with a full conversational
topic. Kendra. KENDRA SPRINGER:
Thanks, Omar. OMAR AFTAB:
Thank you. (applause) KENDRA SPRINGER:
Awesome. For this first demo, I'm going to touch on the two different
ways that Omar just highlighted that's possible for extending your
Microsoft Copilots. As you can see, you can now easily access
both ways to extend directly from the left hand
nav of the Copilot studio. Let's first start by taking
a look at AI plugins. The new AI plugin
experience allows you to take a single plugin action, whether it's a connector, a power automate flow, or one of the new
dynamic prompts. It allows you to wrap
these plugins in order to extend your
Microsoft 365 Copilot. Soon you will also be
able to use these across multiple Copilots created
within the Copilot studio. I'm going to now show
you a quick video of the end to end
experience from creation to extension within
Microsoft 365 Copilot. SPEAKER 1:
Kim, a Customer Service
Representative, is preparing for an
upcoming meeting with City Power and Lights. KIM:
Show me the case details for City Power and Lights. SPEAKER 1:
Copilot retrieves the case details from Dynamics 365. But Kim wants to know
the case resolution, and so she continues. KIM:
Summarize case resolution. SPEAKER 1:
Copilot says, Dynamics plugin doesn't have any information
about the case resolution. Kim asks Copilot how
to create a plugin. Copilot informs Kim to use Microsoft Copilot Studio to
create and manage plugins. Kim navigates to
Microsoft Copilot Studio, and create a new prompt Plugin. She names the plugin as Summarize case resolutions
by case number, and adds necessary fields
from Dynamics 365. She tests the prompt
with case number, and confirms the output. She saves the prompt. Now Kim comes to Copilot
from Microsoft 365, and enables the prompt plugin
from the Plugin fly out. She retries the question, and this time Copilot returns the case resolution details. This shows how easy
for one to create new plugins in Microsoft
Copilot Studio, and use it in Copilot
for Microsoft 365. KENDRA SPRINGER:
I hope you can see how incredibly easy that was to do, and how incredibly powerful that is for extending your
Microsoft 365 Copilot. Now we're going to
go back to what was announced earlier in
Rajesh's Keynote. While wrapping in action
such as a connector, or a power automate flow as
a plug-in is a simple way to light this up as an
action within Microsoft 365. However, sometimes
you might need to access data from
multiple sources, or even manipulate that
data that you retrieve. That's where
conversational plugins in Copilot Studio really come in, built within the intuitive
dialogue builder that you saw earlier. Better yet, you now benefit from all of the life
cycle management features that you
get when building topics right within
Copilot Studio. Let's go ahead and take a
look at this in action. SPEAKER 2: Microsoft
Copilot Studio allows customers to create their own
conversational experiences, and make them available to users wherever they are across
multiple channels. Microsoft Copilot Studio
can also be used to extend and customize
Microsoft Copilot across surfaces like Microsoft
365 and Windows enabling personalized company
specific use cases for your customers
and employees. Let's take a look
at how this works. Here I'm using Copilot
in Microsoft 365, and I've tried to
ask a question about the current outstanding
repair orders. But as we can see, Copilot
doesn't know how to handle this enterprise
scenario natively. With Microsoft Copilot Studio, lighting up this scenario can be done in just a few minutes. I can create a topic using
the new Copilot trigger, and provide a description of
what my new topic will do. This ensures Microsoft
Copilot understands its purpose and knows when
to invoke it for a user. The information that's needed
to respond to the query about repair orders is going to come from
multiple places. The great news is
that I can easily connect to external
systems by adding plugins based on thousands of prebuilt connectors and
my own Enterprise APIs. I've already added
an SAP plugin to fetch the latest repair
order information, and a custom connector
plugin to pull a list of product recalls from one
of our enterprise APIs. Calling these
plugins is as simple as adding a node to my Canvas. I also want to ensure
that we don't provide information about
completed repair orders. Thanks to the power of
the Dial-Up builder, I can filter this information out with a simple
power effects formula. Finally, I'll use the built in prompt node to easily
use GPT to combine and summarize this
disparate datasets into the response that I
want to provide to my user. With my topic finished, I can choose to publish it to
Microsoft 365 with a click. Now, back in Microsoft
Copilot asking the same question
instantly yields the response that I wanted drawing data from multiple
enterprise data sources, and harnessing the power of GPT. Microsoft Copilot
Studio allows for limitless opportunities
to extend and enhance Copilots
for companies and users across a
myriad of use cases. This is only the beginning
with the support for even more powerful capabilities
coming after Ignite. KENDRA SPRINGER:
Now we'll switch to our last example of how you can extend Copilots using
Copilot Studio. So far we've really tapped into how you can extend
Microsoft 365 Copilot. But there are Copilots across the entire Microsoft ecosystem, and you can also use Copilot
Studio to extend those. Like I said, this pattern
isn't limited to just M365. Here's a demo of the very same extension
pattern being used in yet another Copilot. In this case, we're
going to look at Power Apps Copilot Control. Do I have to go to play
this one? Here we go. Adding the Copilot control to any Canvas app
takes just seconds. Out of the box, you're instantly able to
answer questions from public websites and
data verse tables. But sometimes you want to extend your Copilot with
specific use cases, and for that you can extend
with Copilot Studio. Here you can build topics using the rich authoring Canvas
that we saw earlier. Adding questions and messages, and using conditions to
control your conversation. When you're done again, you simply hit "Publish", and your Copilot will
show up now within your Copilot control with all of the changes
that you just made. Now back in Power Apps, you can test your new Copilot to make sure everything is working. Thanks for bearing with
me on all the screens, and all the videos.
Appreciate it. To wrap this up, hopefully in this session, what you will take away is how simple Copilot Studio makes it to build your own Copilot, or extend Microsoft Copilots. Don't just take our word for it. Try it today for yourself
at aka.ms/TryCopilotStudio. (applause)
