LGSCO: LGSCO Chatbot
AI enabled chatbot widget linked to our website, which provides information about the role of the LGSCO.
1. Summary
1 - Name
LGSCO Chatbot
2 - Description
AI enabled chatbot widget accessible from our website. Users can ask questions about the role of the LGSCO and obtain guidance about how to make a complaint to our organisation and bodies in jurisdiction.
3 - Website URL
www.lgo.org.uk
4 - Contact email
Tier 2 - Owner and Responsibility
1.1 - Organisation or department
Local Government and Social Care Ombudsman
1.2 - Team
New Technology Group
1.3 - Senior responsible owner
Customer Service Manager
1.4 - Third party involvement
Yes
1.4.1 - Third party
Talkative
1.4.2 - Companies House Number
Company number 09660808
1.4.3 - Third party role
Provision and customisation of an off-the-shelf solution.
1.4.4 - Procurement procedure type
Framework Agreement Call-Off
1.4.5 - Third party data access terms
Access is not readily available. Access is granted by LGSCO only if technical issues arise.
Tier 2 - Description and Rationale
2.1 - Detailed description
The tool is an AI enabled chatbot linked to our website. The chatbot is available 24/7 and assists the public to obtain guidance on our jurisdiction and role; guide them to submit complaints, get support with online account issues and provide signposting to other organisations. The AI knowledge base collates information solely from our website and any internal documentation we add to the knowledge base.
2.2 - Benefits
The primary objectives of the chatbot are to:
-Provide 24/7 customer support.
-Provide guidance and manage expectations about our role. Including:
-How to make a complaint to both us and bodies in jurisdiction.
-To provide information on specific complaint categories, such as the information contained in our factsheets about our jurisdiction.
-Enhance customer satisfaction by increasing availability and responsiveness.
-Improve accessibility to our service by providing another route in line with modern expectations
-Access harder to reach demographics. For example, younger people and people who don’t have English as a first language.
The key business benefits derived from this tool are to: -Alleviate workload by reducing the number of interactions that do not necessarily require advisor intervention. -Reduce the burden on our Operational Support team,by helping customers to resolve their own technical issues.
The main success measures involve the chatbot handling enquiries without the input of an advisor. This will be achieved by an overall reduction in the volume of: -Signposts to other organisations. -Premature complaints -Oral referrals to make a complaint online -Calls to frontline staff regarding our Assessment and Investigation Teams -Helpdesk calls regarding online account issues
The data and overall output is reported monthy.
2.3 - Previous process
N/A
2.4 - Alternatives considered
This tool was complimentary to our already established front line service. Whereby users can already contact us via phone, online and post.
Tier 2 - Deployment Context
3.1 - Integration into broader operational process
The chatbot provides guidance and support about the role of the LGSCO. The information provided is already available on our website. The chatbot acts to find, collate and summarise that information in an easier and quicker way for the website user. It is designed to manage user expecations about our role. Users can then make an informed decision as to how and when to complain to ourselves and bodies in jursidction. The chatbot uses information already available on our public website to provide advice and guidance to users. This may include whether we can investigate their complaint. It does not make decisions and users can still make a complaint notwithstanding advice from the chatbot, which will then be subject to human review.
3.2 - Human review
The bot is monitored daily. The knowledge base provides details of when the bot was unable to answer questions and also make suggestions where answers could be improved. These are manually validated (or dismissed) weekly. Our New Technology Group reviews the overall impact of this work and the group meets monthly.
3.3 - Frequency and scale of usage
The tool is currently available for use on our website 24/7 up to 31st March 2026. To date, estimates are around between 1K and 2K interactions a month.
3.4 - Required training
No training is required to use the chatbot
3.5 - Appeals and review
N/A. The chatbot does not make decisions. if the users disagrees with the advice other avenues to contact the LGSCO are available.
Tier 2 - Tool Specification
4.1.1 - System architecture
A cloud-based AI chatbot, using a custom LLM and drawing on externally facing knowledge articles about our service. Bot is delivered as an app on our existing website and is available 24x7x365
4.1.2 - System-level input
Knowledge database is driven by URLs associated with www.lgo.org.uk and any documentation we input.
4.1.3 - System-level output
Information collated from www.lgo.org.uk
4.1.4 - Maintenance
As per the 3rd party tool owner’s schedule and adhoc where required. See also 2.3.2 Human Review.
4.1.5 - Models
Large Language Model
Tier 2 - Model Specification
4.2.1. - Model name
OpenAI GPT4.1 (unless changed in the customer dropdown). Talkative supports models from OpenAI, Google, and AWS Bedrock (Anthropic+LLAMA)
4.2.2 - Model version
OpenAI GPT4.1 (unless changed in the customer dropdown). Talkative supports models from OpenAI, Google, and AWS Bedrock (Anthropic+LLAMA)
4.2.3 - Model task
The model takes the customers query, parses the data from the knowledgebase and provides a response if a match is found.
4.2.4 - Model input
The model accepts natural language user queries alongside a “Knowledge Base” of customer-provided data, which can include website URLs, file uploads (PDF/Docx), or manual text entries.
4.2.5 - Model output
The model generates text-based responses strictly limited to the information within the Knowledge Base, returning a standard “I don’t know” response if the query cannot be answered by the provided data.
4.2.6 - Model architecture
This system uses a Large Language Model (LLM), specifically implemented as a Retrieval-Augmented Generation (RAG) chatbot. The model is optimized for strict grounding, meaning it is instructed to ignore external world knowledge and only utilize the specific data variables and content passed through the knowledge base. More info on the model can be found here: https://developers.openai.com/api/docs/models/gpt-4.1
4.2.7 - Model performance
Deployment readiness was evaluated through functional testing to ensure the ‘I don’t know’ logic-gate triggers reliably when information is absent from the Knowledge Base, preventing hallucinations. The primary safety and bias mitigation strategy is structural: by strictly grounding the model via RAG, the system is constrained to only surface information explicitly approved and provided by the deploying organization. Furthermore, the system includes a comprehensive test suite allowing LGSCO to test responses en masse, ensuring outputs remain appropriate, unbiased, and aligned with organisational guidelines prior to and during deployment
4.2.8 - Datasets and their purposes
The foundational model’s training data is proprietary to the API provider (e.g., OpenAI). However, for model refinement and task execution (RAG), the system relies exclusively on the ‘Knowledge Base’ dataset. This dataset is curated and maintained by LGSCO and consists of verified public website pages, policy PDFs, and FAQ documents to ensure the model’s outputs are grounded in approved LGSCO information
2.4.3. Development Data
4.3.1 - Development data description
This section is not applciable due to this being an off the shelf solution. More info can be found here: https://developers.openai.com/api/docs/models/gpt-4.1 Talkative does not use any customer data sets for developing the tool. The data of our clients is only used for the knowledge base feature that determines the responses in a given customer environment. Data is never used to train models or shared with third parties at Talkative.
4.3.2 - Data modality
Talkative is currently just using the text modality, version 4.1 was trained on many modalities
4.3.3 - Data quantities
The Talkative knowledge base can be as small or as large as LGSCO requires, it is currently at ~5MB
4.3.4 - Sensitive attributes
Only information in chat messages will be processed by the API; the LGSCO environment is set up such that users are instructed not enter personal information.
4.3.5 - Data completeness and representativeness
N/A
4.3.6 - Data cleaning
N/A
4.3.7 - Data collection
Data is collected purely for the process of responding to questions through the chat system. All data is encrypted at transit and at rest. Once the chat session has completed, the data is held within the targeted AWS cloud environment only for as long as LGSCO requires for the purposes and monitoring, but it can be deleted at any time and the data retention schedule can also be altered to be as long or as short as LGSCO requires
4.3.8 - Data access and storage
Data in the Talkative environment is never used for training or developing the model. The data is held in a UK AWS instance that is only accessible to the CEO and CTO of Talkative. LG SCO staff are able to view the data (e.g. chat transcripts) by logging into the Talkative portal
4.3.9 - Data sharing agreements
No data-sharing agreements are in place. We have a data processing agreement with LG SEO. We also have DPAs with our sub-processors that process the data
Tier 2 - Operational Data Specification
4.4.1 - Data sources
www.lgo.org.uk internal documentation Chat box on the website for user input.
4.4.2 - Sensitive attributes
Users are not required or requested to enter any personal details into the chat.
See also risks
4.4.3 - Data processing methods
N/A
4.4.4 - Data access and storage
Chats are stored within the software for 1 month before being automatically deleted. The Customer Service Manager, IT Manager and Intake Managers (x2) have access to the chat transcripts. LGSCO can grant temporary access to Talkative for maintenance purposes.
4.4.5 - Data sharing agreements
No data sharing agreements are in place. Data gathered from the chatbot relates to volume, subject matter and key themes. No personal data is gathered or shared.
Tier 2 - Risks, Mitigations and Impact Assessments
5.1 - Impact assessments
Privacy Impact Screening. Completed 19/09/2025. The chatbot does not compel users to provide personal or sensitive data. Therefore, a full PIA was not required. If a user enters personal details, the information is not redacted by the bot, but the data is contained within a closed, password protected, system that is only viewable by 4 people with LGSCO account holder and supervisor access rights. The interactions cannot be seen by the software supplier and they are automatically deleted after 1 month.
Equality Impact Assessment. Completed 09/12/2025. The chatbot is designed to provide an additional contact route and to increase accessibilty and expanding operating hours. We are not closing down any other contact routes and therefore all demographics can still contact us via their preferred method.
5.2 - Risks and mitigations
We do not consider there is a risk to particular individuals or groups. We are aiming to increase overall accessibility and we are not removing any other routes. The chatbot does not make decisions. Users are not required to enter personal data. The risks are around not chatbot not providing the right answers. This is mitigated by regular quality checks and by maintaining an up to date website.