This is a guest post from Dr. Iris Yuster, Senior Business Development Manager, Microsoft Educator Community.
The Microsoft Educator Community recently launched a new bot built with the Bot Framework to provide educators with a personal human-like assistant that can direct them quickly to relevant content. One of the key roles of the bot is to help educators better engage with Office 365 and develop their teaching skills with Microsoft’s Office products.
Since the bot was added to the site in January, engagement with the bot has been high, with over 40,000 messages from educators being exchanged. Educators have found the bot to be human-like, valuable, and easy to use. They shared positive praise for its assistance and helpfulness. In fact, we’re beginning to see a trend towards educators preferring to start their engagement on the site with the bot. Educators who use the bot during their visit to the site have 3-times longer session duration and 2-times higher number of page views per session.
Driving Business KPI and Democratizing AI in Three Steps
Define users’ intents and flows
The Educator Community bot combines pre-defined scenarios to define users’ intents and flows. The scenarios support business KPIs flows, as well as free text dialog messages for over 60 intents.
Pre-set Dialogs Flows
The bot’s first message includes a personal introduction and set of expectations, describes what the user and the bot can do, and gives an option to start the interaction with one of the pre-defined scenarios.
For example, when the educator selects “Courses,” he gets a set of featured courses to select and start consuming Office 365 content.
It is important to direct the user through the pre-defined dialog to other options and enable continuous interactions, such as “browse more courses”, “find another product”, “go back to all activities”, or write a free text message.
Intents and Free Text Messages
In human-to-human conversations, the participants usually don’t have to predict each other’s intents or plan their answers. In human-to-bot conversations, an intelligent and effective bot should be able to anticipate user intent and answer it with the correct per-defined dialog script.
In this example, the user couldn’t find an interesting course and typed “Sway.” The bot then searched and presented relevant Sway items, and offered the user some options of how to continue.
Personalized Educator-Bot Interactions
One of the most important capabilities of the Educator Community bot is to provide a personalized experience based on individual interests, previous interactions with the bot, and user’s activities on the site. The site machine learning (ML) modules were tailored to the bot along with some more personalized elements:
- User preferences – Just for you ML for user activities on the site
- Related content – More like this ML Collaborative Filtering module
- Previous User-Bot interactions – The bot remembers which content the user selected and presents new relevant content
- Continue Chat – same as a natural conversation starting from the last point that it was ended
To maximize business KPIs with the bot, users have to get only the pre-defined scenarios that are relevant to their based on their profiles, behaviors, and interests.
Train the Bot to be smarter
Training the bot is a critical and on-going process to make it smarter and maximize its effectiveness. This process includes both business and technical development.
On the business side, the engagement with the bot is processed through an analytics platform which enables measurement of the bot’s contribution to the business and offers an opportunity to optimize it. Analysis of the bot’s interactions enables us to learn what is missing, what should be improved and how to support the educator to achieve his goals. This often leads to defining new and missing intents and rephrasing pre-defined dialogs.
On the technical side, the Educator Community Bot was built using the Microsoft Bot Framework platform and includes Microsoft’s Language Understanding Intelligent Services (LUIS), an active learning system that can understand users’ messages and interpret them into the predefined intents. LUIS is part of Microsoft’s Cognitive Services, a collection of 25 intelligent APIs that allows machines to see, hear, speak, understand and interpret our needs using natural methods of communication Almost every intent includes several utterances, for example: Intent – “Badges”, Utterances – I want a badge, where is my Badge, how can I get a Badge, and so on. The business team works closely with the developers to direct them with new intents that LUIS should be trained on.
The bot can play a central role for business only if it’s smart, personalized, and brings added value that doesn’t exist in any of the other channels. This kind of bot should know how to provide a fun and easy experience that helps the customer achieve specific goals. The human-like dialog is most useful if it’s relevant and effective, and not just a limited loop or QA list. The combination of LUIS, Microsoft Bot Framework, and machine learning enables us to optimize the bot experience. We believe every business should consider the value a bot may add to their customer experience and try creating one. Customers are already waiting to start the dialog, willing to share their interests in the context of a chat platform, and helping to develop a stronger brand.