Bradford Council: Supporting the ASC front door with AI digital assistants

ASC information and advice was identified as an area with strong potential for an AI solution, due to high demand for signposting services - over 4,500 calls a month – and the fact that no personal information needed to be processed. Often people arrive at the ASC Front Door at a time of crisis and their situation is always highly personal. Websites can be hard to navigate to find relevant information to complex situations and there are multiple ASC services/websites available which can be overwhelming.

This case study focuses on Bradford’s Digital Assistant, Annie, which is accessed online and through WhatsApp. 

Annie offers a new 24/7 access channel, helping Bradford to comply with Section 4 of the 2014 Care Act, which requires councils to ensure that people have access to the right information and advice at the right time. Annie strengthens Bradford’s ASC prevention offer by supporting immediate queries and signposting resources blended from 100s of trusted services and websites, helping users access the support and solutions they need. This includes vital out of hours support for carers, ensuring they can access help and reassurance when Bradford’s ASC front door team are unavailable outside standard 9–5 hours.

This new front door system improves accessibility to services and advice in multiple ways:

• Locked box: Digital assistants operate from a ‘locked box’, meaning that they are specifically trained on trusted, approved information and advice, including content from national, regional, and other expert sources. They do not draw on unapproved content from the internet or other sources. The content in the locked box is curated by the council, and Aida has an automated content update process to ensure all information is current. 
• Intelligent, informed and personal conversation: The ability to use the wealth of content in the locked box, allows Annie to blend information from different sources to be as comprehensive as possible. Annie can also ask clarification questions to ensure the information she is sharing is as useful as possible. As the conversation progresses, Annie references the situation the person is in, as such the responses and conversation becomee highly personalised and relevant.
• Live directory information: Annie can interface with local directories (e.g. connect to Support in Bradford, a community or Local Offer directory) and pulls in in local services as required.
• Channel agnostic: Annie can be deployed across platforms and online, allowing for flexibility based on council systems and user preference.
• WhatsApp: Whilst Annie is available from Bradford’s website to use online, Bradford are actively focusing on WhatsApp as the preferred communication channel. With over 90 per cent of all people of all ages in the UK using WhatsApp, it is a simple, easy and effective way in which to deliver the ASC information, advice and guidance. WhatsApp allows for two way messaging meaning Bradford can proactively send messages and updates to people. It’s secure, free to use, means that people don’t need to navigate complicated websites and they can return to their conversations and the links shared at any time. 
• Empathetic, supportive and kind: Annie is trained to deliver responses with empathy, kindness and a supportive tone. The training process was more involved than just giving Annie access to the right data. Responses were designed in a granular way, with things like tone and sentence length iteratively adjusted in the training process.
• Expandable resource: Alongside being available 24/7, Annie is able to manage multiple conversations at one time.
• Linking conversations into the backend: Built into Annie is the concept that should a person ask more than four questions, they probably have a complex situation and need to talk to a human. At that point they are directed to the front door (tel & email) and given a PIN code to use, so that colleagues can pick up the original conversation and progress without repetition.
• Organisational dashboards: All conversations with the digital assistants are fully audited via a cloud-based dashboard. Conversations in different languages are automatically translated, conversations are summarised, highlighting key themes and points. A wide range of metrics is also captured, including time of day, conversation length, and information sources used.

Bradford engaged with a wide range of stakeholder groups throughout the project, these included the colleagues from the Independence Advice Hub and wider Adult Social Care Departments, voluntary community sector partners and the Co-Production Board – Individuals with lived experience of using Adult Social Care services.  Testing of Annie took place across internal and external client groups, it was iterative and refinements were made after each round of user testing to ensure Annie was communicating in the most natural and effective way.

Conversations about accessing social care can be sensitive and may need to be followed up on by a human. Behind Annie, there is an active monitoring layer. If any conversations highlight a safeguarding issue, these are flagged. All conversations are fully audited and auto summarised, allowing the team to ensure that nothing is missed. This also enables online interactions to be connected with in-person service delivery, creating a more joined-up service journey.  
The project team worked in collaboration with partners in Information Governance,  to ensure that Annie was underpinned by comprehensive information governance and cyber security measures, which is now available for reuse by other councils. 

Bradford Council also initially completed a Data Protection Impact Assessment (DPIA) in January 2025 to identify potential risks associated with implementing the AI digital assistant. 

During this process, concerns were raised about the possibility of citizens sharing personal information despite being advised not to. To mitigate this, the team introduced a clear banner at the start of every interaction reminding users not to enter personal details. A privacy notice, which users must accept before beginning a chat, was also created to reinforce this message. If any personal data is shared such as DOB, bank details, this is identified by the active monitoring layer and redacted in the backend. 

It was also a recognised risk of children accessing the AI service. As the tool is specifically designed for Adult Social Care, the system directs any child-related queries to the appropriate Children’s Services pathways. Any safeguarding concerns identified are escalated accordingly. 

IT security risks were also identified. To address these, an initial PEN test was carried out, and annual PEN testing is now scheduled to ensure ongoing cyber-security resilience and compliance. The technical partner is also ISO 42001 accredited and the platform has been developed in accordance with NCSC Secure AI System Development principles.

Whilst there have been flagged messages, to date none of the conversations Annie has had have resulted in a safeguarding concern. Rather the active monitoring has identified vulnerability around people not having food or electricity, and the council has been able to provide food parcels to support them. 

Bradford Council has a dedicated AI Strategy board which has been newly developed to support any AI projects going forward. 
Having worked with Bradford, Norfolk and West Northamptonshire on Aida (and further councils using the underlying platform for public health services), Aida has a comprehensive Cyber Security & IG pack (available on request) which includes:

• Acceptable Use Policy
• Data Protection Policy
• Equipment Security Policy
• Forensic Readiness Plan
• Information Security Policy
• Information Sharing Policy
• Product Roles and Responsibilities 
• Records Management Policy 
• Removable Media Policy 
• Security Incident Management Plan. 

Since its launch in October 2025, Annie has had around 50 conversations per week – equivalent to one full time advisor. 20 per cent of usage is out of contact centre hours and it has been used in languages including Hindi, Italian, Serbian, Urdu and Welsh. Proactive feedback which is scheduled to follow up with people confirmed that Annie has a 74 per cent satisfaction rate. Where there is dissatisfaction this is usually around the person’s situation being too complex or services not being available to them.

Colleagues in Bradford’s ASC front door team actively use Annie to support calls in the contact centre. Whilst the contact centre team has a depth of knowledge, Annie often provides them with additional resources or ideas to discuss with people. Colleagues can use Annie both on the web and mobile, which means as social workers are visiting people, Annie works as a valuable, always available resource to give people immediate support. 

To build on the success of the pilot, the team are looking to shift 20 per cent of all information and advice queries towards Annie over the next 12 months. Marketing and communications work to promote this channel shift and increase trust in this new service is ongoing. 

Based on the approach and impact of Annie other local authorities are in the process of developing their own versions to support at the ASC Front Door.

Bradford is using the conversations that Annie is having to improve the content on their website, as well as feed back to commissioners about services that might be needed in the community. The conversations Annie is having are never standard, they are often complex and always highly personal. Annie is able to blend trusted sources of information, advice and guidance to empathically and effectively support people in a variety of situations, giving people easier access to early help.

There were some key learnings and takeaways from the project, including:

• Senior buy in: having corporate buy in at an early stage ensured that the project could maintain momentum.
• Coproduction: working with others, particularly IT colleagues from day one, made sure that technical feasibility was at the centre, whilst designing a product that was useful for residents. 
• Preconceptions around AI: being open minded about how colleagues will engage with AI supported the implementation process

Jane Hancer

Email: [email protected]