3 min read

Optimize your chatbots with new AI features for Power Virtual Agents

Share

Microsoft Power Virtual Agents has been a key part of the overall industry transformation around conversational systems. Since the general availability launch in December of 2019, our focus remains on democratizing AI and empowering citizen developers to easily develop conversational AI agents with a simple, no-code maker studio. Throughout the month of March 2021, we’re launching in preview a suite of AI capabilities to help you optimize chatbots and improve the bots over time.

Keep reading to learn the details on how the newest release of AI capabilities will enable Power Virtual Agents to provide a more personalized and natural chatbot experience.

Topic overlap detection

With the amount of chatbot content growing, topic-overlap inevitably happens, resulting in the bot asking “Did you mean” clarifying questions more often.

Today, we are announcing an AI-powered topic-overlap detection tool which helps bot authors to identify and refine overlapping topics, reducing the need for the bot to ask clarifying questions.

An AI-powered topic overlap detection (preview) tool will be provided to help customers reduce topic overlaps. Using this tool, bot makers can quickly identify the trigger phrases that are currently confusing the bot and allow to adjust them accordingly.

Suggestions from chat transcripts to address gaps

Today we are also announcing a new topic suggestion feature that generates topic suggestions from chat transcripts. What this means is that as users chat with the bot, the system will analyze chat transcripts that don’t trigger a topic and suggest new topics that will help makers address those gaps. For example, when a lot of users are talking about “Thanksgiving hours,” the system will automatically suggest for this topic to be created.

As users chat with the bot, the system will Analyzes chat transcripts for end user queries that don’t trigger a topic, suggest new topics the maker should consider adding that will help you address those gaps. For example, when a lot of users are talking about “Thanksgiving hours”, you should create a topic for this.

Bots get better over time

With continuous learning in Power Virtual Agents, each conversation will make the next one better. Using signals from responses coming from the “Did you mean” questions, the bot will learn automatically, removing the need for the bot to have to clarify the same question over again. Thanks to this new AI capability, the more users interact with the bot, the better it will become when addressing questions.

Automatic triggering improvements gradually improves topic triggering based on end user behavior. When the bot asks a user to choose between topics, it will learn from their choices.

Personalized conversations

Through the course of conversations with a user, the bot will reuse information from Microsoft Graph and Microsoft Azure Active Directory, leveraging it to enhance and personalize future conversations. For example, if a user mentions their name, email, or zip code, these properties are stored and leveraged in later conversations without having to re-prompt the user.

In the near future, Power Virtual Agents will also be enabled to add User Graph personalization. Data that is stored in Microsoft Dataverse or populated by other apps (for example, Portals and Microsoft Dynamics 365 Customer Service) will be able to be leveraged as context variables for the bot to use.

Through the course of conversations with a user, the bot will reuse information from Microsoft Graph and Microsoft Azure Active Directory, leveraging it to enhance and personalize future conversations.

Cutting-edge NLU model boosts topic triggering performance

In Power Virtual Agents, conversational AI is infused in each step of the bot building journey, wrapped in an experience friendly to business users. All these are made possible by hosting multiple AI models and AI capabilities on a single service, the core of which is PVA’s transformed-based NLU model NLU model.

Traditionally, intent triggering is formalized as a multi-class classification problem, in which the model is highly associated with the known categories; any change to these categories will require building a new model.

Power Virtual Agents hosts multiple AI models and AI capabilities on a single service, the core of which is a transformer-based natural language understanding (NLU) model. Different from the traditional approach, Power Virtual Agents language understanding model uses the example-based approach powered by a deep neural model. Such large-scale model only needs to be trained once with massive amounts of data using AI supercomputing, and can be later used for specific tasks with just a few examples  and no further training. This is part of Microsoft’s AI at Scale initiative—which basically changes the way AI is developed. It provides an intuitive way for bot makers to work on the bot content simultaneously and confidently, without having to involve AI experts.

Delivering on our vision

Our mission is to democratize conversational AI, empower every individual and every organization on the planet to easily build, manage, and use intelligent bots—all without requiring any coding or AI expertise. To realize this vision, we will continue to introduce advanced AI capabilities in Power Virtual Agents, allowing bots to improve without any maker intervention.