What if you could turn your useful documents, FAQ’s and other sources of information into an easy to use AI powered Chatbot. With Microsoft latest tool you can and it’s easy!
QnA Maker is a new preview service offered by Microsoft. It creates a Chatbot that allows users to interact with useful pieces of information in a simple chat form. The interesting thing about this is how easy it is to create, it doesn’t rely on expert knowledge of AI or Bots or coding, its accessible to business users.
We created the bot below using these 5 simple steps and it took no more than 10 minutes to get our Chatbot live.
- Identify the information you want to enter into your knowledge bank. Right now it supports FAQ’s and product information. The information you are looking to upload has to be in what they call pairs, where there a clear answer to any questions or headers. This makes it easy for the tool to pick them up.
- You then go to https://qnamaker.ai/, create new service, name the service and upload your information, this can be a url or a file up to 20mb. You can also choose to create it from multiple sources such as existing FAQ webpages, .tsv, .pdf, .doc, .docx, pdf and .xlsx, or you can even start from scratch and enter your pairs manually.
- The QnA Maker then extracts what information it can and presents it in a Q+A format, showing you what answers it will give to each questions. It automatically generates a welcome message to start the conversation. You can now start to widen the scope, you can enter different welcome messages for different users. You can also choose to add multiple ways of wording different questions to help improve the chances of the correct answer appearing. The more you add the more user friendly it is.
- It’s time to train your bot, go to test. Now you will see a demo of your bot. You have to go through a fairly rigorous round of testing. This involves asking questions in a way the user of the bot will, using a range of natural language and seeing if the answer it displays is correct. If it displays the wrong answer you can tell it what it should have displayed and it will retrain. The bot will learn every time you do this and provide more accurate results.
- Finally you publish your bot. This puts it out in the open and gives you an url endpoint service that you can connect to a Chat based application. One way of doing this is through Azure. All you need to do is sign up to your azure portal, create a new bot using the QnA template and enter the knowledgebase id and subscription key you get when publishing your QnA service. Click save, then select the channels you want to make your bot available on. The bot is ready to be used by anyone you want. You can publish the bot in a range of ways including Skype, Skype for business, web chat, Microsoft teams, slack, yammer, and Facebook messenger.
After going live it is easy to update and retrain your knowledge bank at any time. You can add more information and additional options to keep your knowledge bank up to date. The outcome of this is a user friendly, natural language interface that has the ability to answer questions for the users.
The potential applications for this across most organisations is huge. It could be applied to:
- Provide internal knowledge services to customer facing roles
- Deliver interactive training and support
- Help new starts learn about the organisation
- Provide customers with a better FAQ interface.
One of the best things about QnA Maker is that it its currently free. It is limited by number of transactions (messages) per minute or month but there is currently no charge for the service until you cross this monthly threshold. We expect that Microsoft will develop this into a full service, building on Azure cognitive services, depending on interest.
Interested in more on bots, AI or data? Get in touch to discuss how we can help you.
This Chatbot is based on the Microsoft FAQ’s for office 365 and Windows as an example. Ask it questions about Windows, product keys, installation, errors, different versions and a range of other questions.