![](https://emergencyservicestimes.com/wp-content/uploads/2025/05/ESS-Button-resized.png")
There are many ways to contact the NHS and in Wales they are using AI to enhance access to the non-emergency 111 service. Welsh Ambulance Service Digital chief Jonny Sammut talks to me about AlBot, an AI agent that is assisting thousands of people across Wales find the information they need in a trusted way.
If you go to the NHS 111 Wales website, you’ll be greeted by a pop-up box and Albot, a digital assistant, will ask you how it can help. Type in your question and Albot will consult its knowledge base to find the answer to your question. It’s accessing a walled garden of trusted, clinically accurate information provided by the Welsh Ambulance Service.
Jonny explains why WAST has decided to use AI agents in this way.
“111 is a high volume, high impact entry point to the organisation where people contact us by phone and online. We decided to trial artificial intelligence agents – bots – using the 111 offering because we wanted to start small, and more importantly, start safely.”
He focuses on trust and confidence. He talks about the information being clinically assured with escalation routes to humans where required. “For us, this is a controlled environment, where we can start to build confidence with agentic AI.” He is keen to emphasise safety here, he knows that AI agents are not technically tricky or innovative but it’s the use of clinically assured health information that means the user can trust what it says.
“We wanted to make sure what we did was safe, to do this incrementally, and make sure that we’ve got the right governance and ethics built in right from the beginning.” Working with two tech firms to deliver this project, RoboticsAI and Druid, Jonny says they are keen to have a collaborative approach to build the right patient journey with this technology.
“Agentic AI is definitely in the hype phase at the minute,” says Jonny. “We want to take it, use it meaningfully, and then start to build on it.” They are starting in a good place as the information – the health A-Z as he calls it – that underpins the NHS 111 Wales website is up to date, well-structured and didn’t require a lot of ‘cleaning’ to form the knowledge base for AlBot to consult. He is excited to look at other data sources for the bot to refer to but recognises that each one presents different challenges in terms of structure, decision making and safety.
And there’s that word again: safety. I ask Jonny what happens if queries are on sensitive topics – perhaps about mental health or harm. What would AlBot do with that and what trigger points would see a human intervene in the interaction?
“The first thing we did was we set some really good, hard safety boundaries,” he explains. “This wasn’t an afterthought. AlBot doesn’t provide direct clinical advice, it won’t provide diagnosis, but for those sensitive topics, like mental health conversations or in an emergency, it already has deliberate, pre-designed reference points to call 111 and 999 in those scenarios.”
The Trust ran a competition with its staff to name the bot and that’s how it came to be called AlBot. “We were clear from the outset that we didn’t want to pretend that the bot is or should be a human. It isn’t. So, what mattered to us, though, is the responses. They had to feel empathetic, calm and be non-judgmental.” The project team tested the tone of the responses, the use of language and carried out extensive testing before releasing this for public use including not using gendered language.
Like a human, it is learning. Jonny uses an example to explain.
“You might type in ‘I’ve got a headache’ and there are lots of different ways of saying the same thing. Each time AlBot receives a query, it refines the way it can then translate it and match it to the knowledge base. If I was to sit and have a conversation with AlBot, the more I’m engaging with it, the more it will learn my tone of language, my style of language, what I’m trying to get to do: so it will learn and adapt in the moment.”
The Trust launched AlBot in August and so far, it has had 30,000 conversations – that’s about 5% of the total calls to NHS 111 Wales during that time. User feedback is helping the team to understand what’s working and where improvements can be made. Speed of response is one concern and coverage is another – both of which are under consideration.
Jonny is pleased with the results so far. “Considering this implementation cost the equivalent to two family cars to get to where we are, it’s been an incredible use of money for public purse.” Although he says it’s not about reducing the cost of running 111, he says there’s no equivalence or way to show that users are shifting from one channel to another.
“But what I would say is that volume of the chat bot has sustained now. If it wasn’t successful, what you’d have seen is a spike of the usage in August, September, and then a plateauing; actually, we’ve seen a steady increase of the usage month on month.” This is impressive given there has been no accompanying marketing campaign and he says that users are predominantly coming from Wales with small amounts of non-UK usage.
Throughout our conversation Jonny has emphasised safety and starting small, building incrementally but that doesn’t mean he lacks ambition when it comes to wider use of AI in the Trust. “It’s balancing hype with delivery,” he adds, setting out what could follow.
“The next step for us would be using things like a virtual avatar. Currently someone can talk to AlBot while typing and texting if they are accessing it on WhatsApp. The next stage would be to create a voice version of our bot so you can have a conversation with it. I guess the future is wide open with this.”
For now, AlBot is at work only in Wales, but Jonny is talking to his peers in ambulance services across the UK, sharing the learning and developing concepts to move the use of AI forward not only 111 but all areas of urgent and emergency care.
