NASUWT: Notice Period Chatbot
Using AI-powered Chatbots to provide teachers with information about their contractual obligations.
NASUWT is the Teachers’ Union that supports 250,000 members across the UK. For more than 100 years they have provided expert advice, support and guidance to teachers, and campaigned to protect and improve teacher’s pay, working conditions and professional status.
The union wanted to test using chatbot technology to allow their members to self-serve and access reliable information quickly 24/7. To keep the scope manageable, and prove the value of this technology before rolling it out across more topics, we decided to train the chatbot on just one topic and only launch it on the Notice Periods web page.
Implementation
We started by reviewing the information available to members on the union’s Notice Period page and working with the NASUWT team to identify the queries that the chatbots should be trained to answer. We determined that to answer a user’s query correctly, the chatbot must know which country and type of school they teach in. We trained the chatbot to answer queries from members who teach in publicly funded schools in England or Wales, as these teachers make up the majority of the union’s members and most of the information published by the union relates to these teachers. However, we still wanted to provide general information and contact details for members who teach in Scotland, Northern Ireland, or independent schools. To achieve this we used a feature within EBM’s chat widget to enable or disable free text input at different points in the chat and therefore have more control over the conversation flows.
At the start of a conversation, the chatbot introduces itself, stating in bold that it is a chatbot and not a human and that it can answer questions about notice periods. This welcome message clearly defines what it is, its purpose and the topic it can help with. It also provides a link to alternative contact details if the user wants help with another topic or to speak to a human.
We built a start flow that asks users to select their country and school type. To make collecting the required information easier, free text input is disabled in the start flow. Users can select from a list of quick reply buttons to proceed through the flow swiftly. These quick reply buttons are non-persistent so that later in the conversation users can’t scroll up and try to change their selection out of context. Instead, during the flow they are given the option to ‘go back’ a step to correct their selection and later in the conversation, they can select ‘restart’ and get back to the start flow.
After the start flow, if a user teaches in a publicly-funded school in England or Wales free text input is enabled. The chatbot provides key information about notice periods for these teachers and then asks the user to type their query or select from a list of popular topic buttons. These popular topic buttons allow users to quickly navigate to common FAQs and act almost like a menu, as users are given the option to ‘see popular topics’ again after the chatbot responds to their query. The chatbot uses Google Cloud’s Dialogflow for Natural Language Understanding to answer free text queries. It remembers if the user teaches in England or Wales, and will tailor its responses to that country accordingly.
If the user teaches in Scotland, Northern Ireland, or an independent school, then the chatbot also provides key information about notice periods for these teachers. However, free text is still disabled and users are instructed to contact their member advice team if they have further queries.
For certain queries, it is more appropriate for the user to speak to the Union’s member advice team, so the chatbot directs them to the relevant contact details for their country. To contact the England member advice team, members are advised to complete a Contact Us form on the website. Rather than just point users to this form, we used the in-built web form feature in EBM’s chat widget to replicate this form within the chatbot. The web form opens within the chat widget and when a user completes this Contact Us form, the chatbot emails their form submission with a copy of their chat transcript to the member advice team who can then follow up with the user directly. The user also receives a confirmation email from the chatbot to reassure them that their query has been passed on. To send these emails, the chatbot is integrated with a third-party email service called SendGrid, part of the Twilio suite of tools.
Results & future plans
Since launching in March 2024, the chatbot has responded to 99% of queries correctly. Just 1% of queries have required retraining or needed a new response added to the chatbot. With our support, NASUWT are using our chatbot management platform to monitor, report and improve the chatbot. There was a concern that adding a new channel for members to contact the member advice team might increase their workload, however, the majority of queries are handled solely by the chatbot and less than 3% of users have completed the chatbot Contact Us form.
With the success of their first chatbot, NASUWT would like to expand the content of their chatbot as well as develop and roll out more chatbots across their website to provide their members with 24/7 support for a multitude of topics.