I think we're ready to begin. Hello everyone. I'm Omar Aftab, VP of Conversational AI here at Microsoft, and today we've all heard a lot today about Copilot experiences at Build. Underneath all those experiences is Copilot Studio, your single stop, end to end tool that provides a complete life cycle for creating, publishing and managing copilots from a single pane of glass. With Copilot Studio, you can build your very own custom copilots or extend and customize all of the Copilot experiences that you've been seeing at Build Today. Because really those are the two key asks we hear from customers. I want to build my own new custom Copilot for my users on this specific channel. Say you want to build an HR Copilot and embed it on your HR website for employees to interact with. Or I want to enable the use of my own business data, my own processes, my own enterprise specific scenarios into these existing Microsoft Copilot experiences so I can configure and tailor them for my own users. Say you want to surface those same HR experiences into M365 Copilot itself so folks can get HR answers right there in the flow of their work. With Copilot Studio, you can do both of these through one SAS based graphical tool. And underneath all of this, Copilot Studio brings together and integrates with a variety of AI technology pieces from across the company, from Azure Openai Studio to Azure Cognitive Services to Azure Bot Service and more. Let's have a look at the quick overview of the product. With generative AI and a powerful dialogue builder, Microsoft Copilot Studio lets you create your own conversational experiences and make them available wherever users are, across multiple channels. Click to create a copilot and describe what you wanted to support, providing instructions about behavior and tone, and pointed to a website to further infuse it with data. That's it. In just these simple steps your Copilot is instantly capable of rich multi turn conversations. Since Copilot Studio is a full end to end SAS with no infrastructure to deploy or maintain, you can easily publish this to your support site or other channels like Microsoft Teams so your customers or employees can use it quickly. Or you might want to further infuse your Copilot with additional knowledge by easily adding a number of different data sources from the Knowledge page, including Documents, SharePoint sites, Dataverse, and data from your CRM and ERP systems. Copilot authors no longer need to build and curate hundreds of dialogues for every question a user may ask. Just add your knowledge, your actions, then your topics and your Copilot will do the rest. Copilot Studio can also extend and customize Microsoft Copilots like Copilot for Microsoft 365, Copilot for Sales, and Copilot in Dynamics 365 customer service, enabling personalized company specific use cases with access to back end systems that Microsoft Copilots don't have existing access to. Extending Microsoft Copilot really is that simple, and with Copilot Connectors you can once again leverage Power Platform connectors, Power Automate flows or prompts. Once you've published a Microsoft 365, you've now unlocked additional ways Microsoft Copilot can support you and your organization. Microsoft Copilot Studio gives you the power to take actions on your data. Back with the security you've come to expect from Microsoft copying. The building has changed forever with unlimited ways to extend and enhance copilots across a myriad of use cases thanks to Microsoft Copilot Studio. Try it now at AKA dot Ms. slash Dry Copilot Studio. Great. So as you saw in that little vignette, Copilot Studio covers the full life cycle of Copilot building, from creating your Copilot in publishing it live, to analyzing its performance and continually improving it. When you try this out, and and you should, you'll see that we've made creating copilots incredibly simple. You can now get started simply by chatting with Copilot Studio and describing what you want to build. You could say, build me a Copilot to handle customer service for laptops. Oh, and you know what? It really shouldn't answer any pricing questions. Or you might say I'd like a copilot that provides HR support for my employees. Please use the information on my HR SharePoint and this employee handbook file and voila, you're copilot is ready grounded with those capabilities and specifications. Next, you can add enterprise knowledge to give it all the information specific to your company, your processes, your products, from your company websites to your files, to your internal SharePoint sites, to now all of your company structured data in database, in fabric, in graph, as well as all your data in various third party sources. Simply point to those sources and with the clicker to your copilot is now an expert over them, ready to answer a host of user questions and chat over all of that knowledge. So the generative part of your copilot is ready, but companies often also want to have tight control. It's step by step over some business critical topics like say returns or purchases. You can easily supplement the generative AI conversations and specify exactly what you want the copilot to do when specific topics come up. And you can do this with a few clicks using the built in powerful graphical studio. If a user asks about disability leave, for example, you can trigger a custom workflow that informs them of their legal rights, asks a set of questions in exact order, looks up their eligibility, and completes and submits their leave application. So you you have both generative and custom dialogues in one system and it will simply orchestrate them for you. You can also build plugins and call actions, from calling your own custom enterprise APIs to connecting to your back ends with the thousands of prebuilt connectors that come with Copilot Studio. And that's it, you're done. You just publish your Copilot to the channel of your choice, from Teams to your public website to social media or whatever you wanted. Or you can choose to add the experiences you've just built to Microsoft 365 Copilot or one of other first party Copilot experiences. There's no infrastructure to deploy, no models are trained, nothing. Your copilot is instantly live and ready to use. In fact, we often build copilots for customers during the course of a meeting and publish them live for folks to interact with before the meeting ends. Let's see a demo of this arc in action. Copilot is incredibly simple with Copilot Studio. Let's say I want to create a copilot to serve as an HR copilot for Contoso. I start by simply describing my copilot and what it will do in natural language. I also asked her to be friendly but remain professional at all times. I'll also point my copilot to Contoso's website so employees can ask questions about the company, with the answers generated live based on the content, alongside all of the preexisting knowledge baked right into the copilot. Finally, I can even specify things I don't want the copilot to talk about. Great, now I can instantly test my copilot right here and iterate on my instructions to further refine my copilot as needed. Let's go ahead and ask to employees receive an award after five years at the company. As you can see in my copilot was able to find the information on the website and generated a contextual answer to my question. I can even ask follow up questions like how long will it take for it to arrive and my copilot is smart enough to know the context of what I'm asking about. OK, this looks great. And just like that my new copilot has been created. I can see the instructions that have been generated based on my earlier answers, and I can easily make changes and iterate on these anytime I need to. The first thing I want to do is give my copilot some more knowledge. As well as the public website I added earlier. I can add SharePoint sites, upload files and point it to structured data stored in Dataverse or the Microsoft Graph. I can also leverage thousands of connectors to pull in data from third party systems. For example, I have information related to training courses held in Dataverse, and I can easily add this to my Copilot knowledge by using the new Data versus Knowledge source. I simply choose whichever tables I want to add, and in this case, I'll choose training course information. That's it. I've now equipped my copilot with more information and I can test it right away. Let's ask what new training courses have been added in the last month. I can see from the citations that this answer came from my Dataverse tables and the copilot pulled the answer from just the right relevant rows in my table. Now I can go ahead and just publish this to get it into the hands of employees right away. After just a few seconds, my new Copilot is published and I can enable it on a wide variety of channels like a website, a mobile app or in this case Microsoft Teams, which I can turn on with just the click. Now over in Teams, users can chat and get their questions answered instantly with my new Copilot. Great. So now that our Copilot is live and users are interacting with it, you can get conversational insights into how it's working right there in Copilot Studio. You can see what the generative system is doing, where it's successfully answering questions, and where it doesn't yet have the necessary knowledge or tools to respond. For example, you could see the system is struggling to answer questions about your return policy, and you can easily provide a link to the policy and give it that knowledge. And now that particular problem is solved. You could see that many people seem to end up having to talk to a human agent when they try to change their payment method for their orders. And you can quickly create a topic to handle that. With a few clicks. You can even easily connect to all of Microsoft's other conversational AI services and tools. For instance, you could build a custom fine tune model in Azure I Studio and easily use that model inside of your copilot. So let's continue our demo with this analyze and improve part of our ARC. To users, it's vital to understand how it's being used, what is working well, and most importantly, where should you focus your valuable time on enhancing your Copilot? Thanks to the expanded built in analytics and Copilot Studio, it's easy for me to get actionable insights. As a maker, I'm now able to understand the top themes of questions users are asking and review the top themes of user queries that might Copilot could not answer. This helps me identify major gaps and take immediate action to improve the performance. For example, as you can see in the new CSAT tab, there are two themes of user queries where users are dissatisfied. One when they're asking about the new updated vacation policy, and the second when they try to apply for short term disability leave. I'm able to drill into those two themes and see that for the most frequently asked topic, there is a high escalation rate leading to dissatisfaction. I can even see content moderation metric and easily see exactly where you don't have the necessary knowledge to answer the question. You now have full transparency and understanding of how you can improve these capabilities and actionable insights the answers generated off of your existing data. I can see where my copilot has blocked harmful content, jailbreak attempts, and ungrounded content, giving me confidence that my copilot is secure. So let's get my copilot some additional capabilities. For example, I would like the copilot to check vacation balances and submit for time off approvals. To do this, I'll start by adding an additional file that has the vacation and time off policies for Contoso. This document includes various tables and rich media. Fortunately, with our new multimodal capabilities which use GPT 4V copilots can now interpret images, diagrams, and tables embedded with documents. This enables copilots to extract and relay crucial visual data, along with scanning the text components of that knowledge source. Employees can now ask how many vacation days do I get a year? And the copilot can use and understand the table within this policy document to relay that they have 15 days since they've been with the company for less than five years. Continuing my efforts to address the escalation themes around vacation, I also want to add an action to retrieve the vacation balance for employees so employees can easily find out how many days they have available. Vacation data is held in an external system, so I will add a custom connector to connect to that back end. As you can see, I can customize and configure the description and all of the inputs and outputs that are required. Finally, I want to build a topic to allow employees to submit a new disability leave request. Contoso has a step by step process that such requests need to go through for legal reasons. For scenarios where more control is needed and repeatable workflows need to be carried out, I can create a specific workflow for the copilot to use when the user brings up a specific topic. Creating this is simple, I'll just subscribe the topic I want to create, Create me a short term disability request, ask for the start date. How many days off the employee would like approval for and if they've spoken with their manager about the request yet? As you can see, the topic has been created with everything generated for me. I can continue to iterate on this topic and expand it further. Let's go ahead and add a flow to look U the employees manager and another action to send the request via e-mail for approval. Now I can go ahead and test this topic and see each step of the dialogue successfully being carried out. I see all the information I need to get a time off approval sent to my manager. At this point, the Copilot can both provide generative answers over a set of knowledge and can trigger specific workflows for specific need, which is a common enterprise ask all within one system. And again, I can simply publish these updates and see them immediately reflected in the live Copilot deployed in Teams. So there you have it, a complete end to end lifecycle for creating, managing, publishing, analyzing, iterating copilots from one single unified SAS based experience. Now I'd like to pause for a second and just call out the sheer amount of functionality that's packed into what we just saw. What would you do if you wanted to build a copilot without using Copilot Studio? Well, you'd create a subscription in open the Eye and configure your LM model. And then you'd set up a vector database, an index for all of your knowledge, and create a vector search over it. You're build some custom workflows and dialogues and other system, and then you train an animal model for triggering those custom workflows. And then you build an orchestrator to arbitrate between the generative and the dialogue systems. And then you'd host your generative model and services. And then you'd build custom connections to all of your back ends. And then you build a custom analytics and reporting pipeline and more and more. Or you could use Copilot Studio and be up and running in minutes, not months. So we've seen how to create our own enterprise specific Copilot experiences with Copilot Studio. Let's see how simple it is to publish those same experiences into M365 or any of the other Microsoft Copilot experiences. That's how easy it was to create a custom Copilot and deploy it on Teams, my own website, or the channel of my choice. But I'm also able to extend Copilot for Microsoft 365 with the copilots I make as well, so users can interact with my enterprise scenarios right inside of Microsoft Copilot. From the library, you'll see I can select a Copilot I've previously made and publish it right to Copilot for Microsoft 365. Once it's published, I'll head over to Microsoft 365 Copilot and start chatting with it right there. I can also add the same Copilot to Teams. I can speak to it one-on-one and see prompts right there that I could get started with. I can also add it to group chats so all members of the chat can interact with this copilot together. So easy. The same building experience, the same custom Copilot experiences that we just walked through. You can choose to publish them and infuse them into any of the Microsoft first party Copilot experiences. And the same extension pattern that you just saw holds for all Microsoft Copilot experiences. Copilot Studio extends them all. It's like the one ring here is a SharePoint copilot for example. Microsoft Copilot that can be extended with Copilot Studio. Let's look at another example of how you can extend a Copilot within SharePoint. This SharePoint site has all the event details for a major event, Relacom 2024. The attendees can quickly get information from the copilot. However, employees looking for food vendor information couldn't get answers from the Copilot due to the information not being available on SharePoint. To fix it, the Copilot can easily be customized via Copilot Studio to connect to external sources. With a simple click, the user is now able to continue editing their Copilot and Copilot Studios full authoring experience. Relicanth catering vendor information is available as records in Dataverse from Copilot Studio. It can easily be added as additional knowledge sources and to give attendees even more menu details. They can also add the catering vendors public website as another knowledge source. This way the Copilot can answer any additional meal questions from the attendees. Providing information is great, but the Copilot can take it further by allowing attendees to submit their meal choices ahead of the event. This can be easily done by adding an action in Copilot Studio. In this case, the action updates employees meal choice as part of the attendee records in Dataverse. There you have it. Now the Copilot will take the action automatically without needing additional configuration. When they're satisfied with the Copilot's additional capabilities, they can easily publish their changes to become available on the Recon 2024 SharePoint site. Now, employees can get all the event information they need when they chat with the Recon 2024 Copilot in SharePoint. With the enhancement from Copilot Studio, they can easily get catering vendor info and ask follow up questions on the meal menu from sources outside of SharePoint. When they're ready, the Copilot can use actions to collect their meal choice conversationally before attending Relacom 2024 with Microsoft Copilot Studio, The options to extend are limitless. You can extend Copilot for sales, Copilot for Service, Copilot for Insights, Viva Copilot and Power Apps, Copilot for finance, and more. Great. We already have 10s of thousands of enterprises using Copilot Studio today. From copilots that provide external support, to copilots that provide employee facing experiences to copilots that help with partners. All built with Copilot Studio, and customers love it. Cineplex, for example, is finding that their Copilot is saving an average of 30 minutes every single time it's used. And Central, which is a global children's charity is using some of our Co pilots as agents that can handle complex long running tasks around child welfare. And I'll talk a little bit more about these new agentic experiences later in our talk. And one of the things that customers love the most is the speed of deployment and the quick ROI. BYU, for example, was able to start saving 150 hours in the first week with their experiences, and they're rapidly expanding their copilot footprint. And Rabobank is even using copilots in their digital and telephony channels. They love the simplicity in control they have with Copilot Studio. So let's actually, I think the, the, the main difference with Copilot Studio is the quick ROI and the, the breadth of customizability that it provides. And I think we should we let's hear from one of our customers ensure who's actually using Copilot Studio for the insurance sector. Digital marketplace for our customers to be able to compare home and auto insurance online. We saw how complicated and outdated the process is. Every single insurance carrier have different. Rules. So let's try to do something to change the customer experience. Everything was manual. We had to start looking at what solutions are there through custom AI. When Microsoft came out with their Neural Voice product, we decided to utilize Copilot in order to create a friendly John. Right now, Friendly John can handle around 60% of customer questions and on our next release, we'll be going up to 90%. The quoting process changed completely the time string from days to 30 minutes on average. It was a game changer. With Power Platform and Copilot Studio, we can innovate and implement anything we want. The quality of work, the customer journey is just more scalable, repeatable and we can just do things much faster and smoother than. Before, if we didn't have these tools, if we didn't have these systems, building such complex areas as quoting as voice copilots would take a traditional development team years. We're able to do it thanks to all of the tools within months. Great. SO now let's change gears a bit. I want to share some of the few, some of the powerful new capabilities we're lighting up in Copilot study. And this is just a subset of a few things that I thought this audience would find interesting. The first one is an obvious little daylighter. How often have we taken a picture of something to share with someone that you're interacting with? Contractor and insurance agent, a user can now just upload a file or an image to Copilot and it will understand it and use it to understand it, to basically understand the content of the file and answer your questions. So hey Copilot, here's my bill. What's this RTA tax line mean? This unlocks a variety of interesting end user scenarios. Here's a photo of this error. Here's an invoice. Here's a copy of my insurance policy. Am I eligible for this benefit? A major thing I want to talk about, and you'll have seen in some of the keynotes today, is that copilots can now act as independent agents, ones that can be triggered by events, not just conversation, and can automate and orchestrate complex, long running business processes with more autonomy and less human intervention. For instance, a copilot can now react behind the scenes when an e-mail arrives and use generative AI to trigger the appropriate chain of actions in response. From understanding the intent of the e-mail, to looking up the sender's details and account, seeing their previous communications, checking inventory, responding to the sender, asking for their preferences, and then taking the appropriate actions to close the ticket, orchestrating and shepherding that entire process over days, let's have a quick glimpse of this. Co pilots can now react to more than chats to events, actions, schedules, and more. For example, we can set a trigger on our new IT help desk copilot, activating it when a new employee is hired. This triggers a complex multi day workflow, from sending an e-mail asking the employee to pick their preferences, to getting any exceptions approved, to updating the hiring manager. What's more, you can view how the copilot is handling these processes and how many times a human has had to get involved. You can even teach it to do something else. For example here I see that more than 24% of new hardware requests are now requesting an exception, and you can simply tell the copilot to limit exceptions to only designer roles that may need non standard hardware. These powerful agentic capabilities transform copilots from working with you to working for you. Now in limited preview. So another interesting capability is memories. Now, one of the frustrating things about sport has traditionally been having to repeat information who you are, what your policy number is, what your address is. There's there's absolutely no continuity of conversation. But now Co pilots, just like human agents, will learn from previous conversations and utilize this knowledge to continually personalize their interactions. So Copilot may not need to ask you for your laptop model or your address when you call again for the same issue. Conversations will become long running, contextual, and deeply personalized. Just more human. When I call my insurance agent, for example, he knows who I am, he knows my policies, he knows what we last talked about, and the conversation picks up from there. So it will be with Copilot. Now one of the most common questions my team and I get around security. So Copilot Studio is a globally secure SAS. It's fully HIPAA high trust, PCI compliant. Admins have complete control and governance options, from audit logs to data loss prevention to granular control over features and the AI behavior. It's secure by default. It automatically tailors its responses based on who it is talking to, what permissions and data access they have, and most importantly, it's built on Microsoft's Responsible AI platform that protects your users and systems from harm, provides content control, and is continually improved every single day. Now, as couples become more pervasive, how you interact with them is also changing. You can now talk to to Copilot is pretty much the same way that you would talk to any other colleague and teams. They're simply pull them in into a chat. They'll be able to read and understand the chat history. You can add mention them and ask them whatever you want. You can see this in the example on the screen. Mahesh invites the Contoso Finance Copilot, which is built in Copilot Studio and which has knowledge of Contoso's budget and policies, into a chat. The copilot can glance through the history of what was being discussed in the group chat and then use its knowledge to respond. Yes, based on this year's FY and current spend, we do have money for this extra purchase. Think about what this means. I can build a finance copilot, give it the knowledge for my enterprise, and people can just pull it into any conversation no matter what time of day. It can mean multiple conversations at the same time, answering questions from multiple different users about my specific enterprise budget and financial issues. And to the user, it's the same as talking to anyone else on Teams. So get started right now. AKA dot Ms. slash try Copilot Studio. Build your own Copilot Today & up to get an early preview access to some of these latest capabilities. We have a number of other sessions and hands on labs today drop by our booth. Our Copilot Studio team will be there to our bill to talk to you and thank you. Yeah, We have a couple of minutes for questions. We actually had a number of questions being submitted online. So for those who can stay, let's take at least maybe one or two questions here. Our first question is, is Copilot Studio included with M365 Copilot licenses? So Copilot Studio is included within 265 seat licenses. If you buy M365 copilot, then Copilot studio which is the tool to configure and customize and extend it is also included in the box. And of course if you want to completely standalone copilot, say you want to build a customer facing bot, you can also buy Copilot studios separately in just paper message. Great. Thank you, Omar. Next question we have is, do I have to train a custom natural language model for Copilot Studio? No. Copilot Studio comes in with a built in large language model that orchestrates over all of your topics in generative responses. There's no model training requirement. If you do need a custom model for some esoteric reason, then Copilot Studio integrates with Azure Cognitive Services, CLUE and Luis with no cliffs extensibility. All right, I think we have time for one more here. OK. And I think we're getting signal to leave, so let's just do one more. One more OK last one, do you have a limit on the size of files that you can upload as knowledge? Yeah, it's currently 3 megabytes per file, but it'll actually grow to 10X that over just the next month or two. With the new multimedia features we talked about where the doc can have tables and images and all that stuff in it, you really need a larger file and that will be addressed in the bottom one or two. And we're around if anyone wants to just talk to talk to us about questions informally. Yeah. Thank you.
