"With all these visitors trying to navigate such a large library of content that is not always relevant to them, I was looking for ways to make the website more responsive to the member's personal needs."
NASUWT’s website is a massive hub of knowledge for the teachers they represent. There’s extensive written advice and guidance, across multiple policy areas. However, the structure of the advice is particularly complex, due to legislative differences across the four UK jurisdictions and contractual variations between different workplace settings.
This complexity made life hard for members who were trying to find relevant information. They had to navigate through broad policy categories and sift through large amounts of content that might not apply to their specific situation.
Michelle Graham, Digital Campaigns and Communications Manager, recognised that whilst the internal architecture was sound, the front-end user interface for guiding members to specific information they needed was "a bit clunky."
The issue was particularly acute for the popular "Teachers' Notice Periods and Resigning From Your Job" page, where members with varying contract types struggled to find the specific resignation guidance relevant to their circumstances.
The solution
NASUWT worked with a chatbot development partner, EBM.ai, to develop a chatbot that can be installed on those pages and can guide members through the content on the page in a way that was personalised and much easier-to-understand. The chatbot could even contact the member support team, summarising the conversation for them, when the member needed more advice.
EBM.ai developed the AI chatbot using Google's Dialogflow platform, which uses natural language understanding (NLU) to interpret user input and respond naturally.
The chatbot was designed to handle conversations similar to a human adviser, using 'intent detection' to match user questions with the correct responses from a pre-determined knowledge base.
The solution takes a triage approach, starting with a series of Q&A buttons to determine the member's location and type of school, then filtering advice accordingly. Members in Northern Ireland and Scotland have slightly more limited options due to the nature of available advice, but most users can then use free text input alongside additional button options for commonly asked questions.
NASUWT was careful to choose a phased approach, first implementing a single-page chatbot rather than a site-wide generative AI solution. They selected a popular web page with comprehensive advice and relatively definitive answers, managed by a policy colleague who would be open to the change.
The chatbot includes several important features:
Integration with existing contact forms that go directly to the advice inbox with conversation transcripts
Fail-safe 'situation-specific' responses for niche cases
Built-in safeguarding alarms
White-labelled branding to appear as an integral part of the website
Key benefits
Improved member experience: Members can get answers within 3-4 clicks that might otherwise take considerable time to find in page content
Better insights: The web team receives direct insight into what members want to know, allowing content to be aligned with actual member questions
Cost savings: Reduced need for other user feedback tools previously used on the website
Staff efficiency: Minimal contact forms submitted through the bot indicate most users find satisfactory answers quickly
High accuracy: The bot consistently performs with over 98% confidence in its responses
Reduced workload: As one policy lead noted, the chatbot "produced benefits for colleagues in terms of a reduction in the number of queries that the MSA Team deal with. In doing so, they are freed up to look at other more complex issues".
Common questions
How difficult is it to implement?
Michelle emphasises the importance of taking a cautious, phased approach. EBM provided comprehensive support including building the dialog flow, bot interface and widget script, plus a short testing phase during the proof of concept stage.
What skills are needed?
The team went through basic onboarding to learn the EBM dashboard but didn't need to manage or retrain the bot independently. Monthly support reviews with EBM handle monitoring and any retraining required.
What are the main challenges?
The biggest challenge was managing colleague concerns. Staff had fears at both ends of a spectrum - some worried about increased workload pressure, others about AI potentially replacing human advice staff. Michelle addressed these by:
Emphasising the project was never about replacing staff
Explaining the bot sits between the website and phone lines as an additional tool
Demonstrating how it handles "low-hanging fruit" questions, freeing staff for more meaningful work
How do you ensure quality and accuracy?
The chatbot achieved over 98% confidence in responses during monthly reviews. Complex questions that could tip over into casework are diverted to contact details, and there are built-in safeguarding alarms for sensitive situations.
What about data protection?
The EBM dashboard retains identifiable conversation data for a limited period, but members' details are not retained when submitted via forms. There's an analytics dashboard showing overview, intents, and feedback where given.
Getting started
Choose the right use case: Start with a popular page that has comprehensive advice and relatively definitive answers
Get colleagues on board: Work with someone who's open to change and can provide subject matter expertise
Take a phased approach: Begin with a single-page bot rather than site-wide implementation
Address staff concerns early: Have robust conversations about values and approach - reassure colleagues this is about enhancing, not replacing, human work
Plan for testing: Bring in advice team members whose frontline experience can help build the knowledge base
Be transparent: Always tell people when AI is being used
Common pitfalls to avoid:
Don't rush into site-wide generative AI without proving the concept first
Don't ignore staff concerns about job security - address them head-on
Don't assume the bot will work perfectly from launch - plan for iterative improvement
Looking ahead
NASUWT is now moving to the next phase, upscaling to generative AI and covering multiple web pages. When a large corporation approached them with a competing solution focused on "saving money" and "cutting jobs," Michelle used ChatGPT to compare proposals and demonstrated that EBM had better understanding of NASUWT's values and superior credentials for working with non-profits.
The next phase will use generative AI to scrape the notice periods page plus additional advice and information pages. Michelle notes: "I have every confidence in the bot and I'm hoping the groundwork we've done internally will yield results when we get to the testing phase again."
As Policy Lead, Paul Watkins, reflected: "I would certainly consider how this could be used for the benefit of the Union and members in other areas."
AI transparency statement: This case study was developed based on written interview responses from NASUWT staff members. Claude.ai was used in reformatting these to fit a case study template.
