NS&I: Customer Chatbot
A chatbot used by customers of NS&I to obtain answers to product and service queries.
Tier 1 Information
1 - Name
NS&I Customer Chatbot
2 - Description
This tool is used by customers of NS&I to obtain answers to product and service queries that they may have.
The tool enables NS&I’s online customers to get rapid responses to queries without needing contact centre agents to intervene. This provides customers with quick and accurate answers and enables contact centre staff to focus on more complex queries.
3 - Website URL
Access to the tool can be found by the bottom right chat icon on the NS&I Website found here: https://www.nsandi.com/
Further information about the tool: https://www.genesys.com/en-sg/company/newsroom/announcements/genesys-launches-genesys-dx-the-next-chapter-in-customer-engagement
4 - Contact email
Tier 2 - Owner and Responsibility
1.1 - Organisation or department
National Savings and Investments (NS&))
1.2 - Team
Enterprise Service Management
1.3 - Senior responsible owner
Head of Service Operations
1.4 - External supplier involvement
Yes
1.4.1 - External supplier
Genesys Atos
1.4.2 - Companies House Number
Genesys: 02979911 Atos: 03290446
1.4.3 - External supplier role
Genesys are the vendor of the tool and provided further support and consultancy. Atos were the implementation supplier who provide staffing, support and maintenance.
1.4.4 - Procurement procedure type
Specified and procured by our incumbent technology and operational outsourcer, Atos.
1.4.5 - Data access terms
N/A
Tier 2 - Description and Rationale
2.1 - Detailed description
Genesys DX uses Natural Language Processing (NLP) to understand customer queries. Key elements include: Intent Recognition: Identifies what the customer wants to achieve by mapping their input to predefined intents. Context Maintenance: Tracks the conversation history to ensure continuity in multi-turn conversations. - Powered by ML algorithms to improve over time: Supervised Learning: Initially trained on labelled data to classify intents and responses. Reinforcement Learning: Learns from customer interactions to refine performance. Semantic Matching: Maps customer queries to relevant knowledge base entries using vector embeddings. - Rule-Based Systems: For straightforward workflows, it applies decision trees or if-else conditions. AI-Augmented Routing: Directs complex issues to human agents based on topic, sentiment, or urgency analysis.
2.2 - Scope
The NS&I Chatbot using Genesys DX is used for scenarios including: Product and Service Information Support: Answering common general non-account specific customer queries via chatbots or live agents.
Scenarios the tool is NOT Designed For: Highly specialised support: Tasks requiring detailed product and service expertise require direct human intervention. Financial advice: The chatbot cannot offer NS&I customers advice on their financial circumstances or savings needs.
2.3 - Benefit
The Genesys DX chatbot allows NS&I too analyse customer behaviour in real time, proactively address queries, and provide targeted assistance through chatbots, which seamlessly transition to human agents for complex issues. This ensures consistent, streamlined experiences while reducing response times and increasing customer satisfaction. Additionally, the platform consolidates customer data to give agents a 360-degree view, enhancing their ability to deliver informed and empathetic service
2.4 - Previous process
There was not a chatbot in place prior to the implement of this Genesys DX Chatbot. Atos identified a need to increase their contact centre productivity and improve NS&I’s customer experience. Retail customer service best practice was followed in the implementation of a chatbot
2.5 - Alternatives considered
Atos is NS&I’s incumbent partner for contact centre technology and operations. Their technology strategy is to buy, not build. Industry leading solutions were evaluated, resulting in the selection of Genesys DX as best fit
Tier 2 - Decision making Process
3.1 - Process integration
The tool opens up allowing customers to select pre-set questions about frequently asked questions whereby the tool will enable the customer to answer their questions in a preset way. The tool provides customers with information to help them identify which NS&I products best meet their needs, or how to fulfil their customer objectives when interacting with NS&I. The tool also provides the user with the ability to write a bespoke message at any point to the chatbot, this includes restarting the chatbot, speaking to an advisor. Once the user is satisfied that their question has been answered or that they have obtained the information they are seeking, they can either ask another question, close the application or speak to a customer service representative.
3.2 - Provided information
The tool provides the user with text responses and links guiding customers to information on the website, or to self-service journeys the customer can follow, having first authenticated their identity by logging in.
3.3 - Frequency and scale of usage
The tool is available 24/7 and handles 100k+ interactions per month
3.4 - Human decisions and review
The tool has undergone user testing to ensure that its outputs correspond to its training materials, and to ensure consistent tone of voice. Customers are responsible for ensuring the information provided back from the chatbot has answered their questions sufficiently otherwise they can speak to a human operator.
3.5 - Required training
The model was trained on NS&I’s public domain product and service information. It was further configured through conversation scripts, developed by Atos and Genesys. Contact centre agents are trained to intercede when requested by a customer or where required by the script. NS&I takes an assurance role in ensuring the process treats customers fairly and is compliant. The chatbot is prefaced by intro text outlining that it can only respond to generic queries and how to contact NS&I for specific help
3.6 - Appeals and review
The tool either prompts the customer to make a decision based on the information provided by leading them to a self-service journey which the customer makes a decision if their question is now answered or they initiate a subsequent offline operational process. There is no specific guidance on who to contact in the event of errors but it is assumed the customer would do this by the other contact channels.
Tier 2 - Tool Specification
4.1.1 - System architecture
https://help.mypurecloud.com/articles/key-concepts-of-dialog-engine-bot-flows-and-digital-bot-flows/
The Genesys Chatbot uses the Dialog Engine and uses natural language understanding (NLU)
Key components of the architecture include:
Data Ingestion: Data ingestion of the data held on the NS&I public facing website into the Cloud environment is conducted using the Genesys Architect tool.
The retrieved documents are processed using Genesys’ Digital Engine Bot Flows. Genesys Digital Engine Bot summarises the retrieved information and generates a user-friendly response, which is then presented to the user.
4.1.2 - Phase
Production
4.1.3 - Maintenance
The Tool supplier’s recommended updates are deployed and regularly updated. The NS&I contract with the service provider requires them to maintain a N-1 patching strategy (maintaining a software version that is one release behind the latest version) .
4.1.4 - Models
- Rule-Based Models Decision Trees; : Used for routing queries or determining next steps in automated workflows based on predefined rules. If-Else Logic: Supports basic chatbot functionality and customer interaction workflows. Session State Management: Tracks multi-turn conversations by adhering to predetermined conversational flows.
- Machine Learning Models Natural Language Processing (NLP) Models: Intent Recognition: Identifies user intent in messages by training on conversational datasets. Entity Recognition: Extracts structured data like names, dates, and account details from unstructured input. Sentiment Analysis: Determines customer sentiment to adjust tone or escalate interactions to human agents. Predictive Engagement: Analyses user behaviour on a website or app to predict and proactively address their needs. Reinforcement Learning Models: Continuously improves the bot’s responses based on customer feedback and success rates.
- Statistical/Mathematical Models Routing Optimization Models: Allocates interactions to the most suitable human agent or department based on efficiency metrics like wait time and skill matching. Customer Journey Analytics: Uses statistical techniques to map and analyse touchpoints across the user journey. Real-Time Analytics Models: Tracks metrics like average handling time, customer satisfaction, and query resolution rate to optimize performance.
- Hybrid Models AI-Augmented Human Assistance: Combines rule-based systems with NLP and machine learning for live agent support (e.g., suggesting responses or retrieving relevant knowledge base articles). Conversational AI: A hybrid of ML-driven NLP and rule-based fallback mechanisms for cases where intent is unclear.
Tier 2 - Model Specification
4.2.1 - Model name
Genesys Dialog Engine Bot
4.2.2 - Model version
01-2025
4.2.3 - Model task
Genesys Dialog Engine Bot Flows and Genesys Digital Bot Flows uses a natural language understanding (NLU) engine that can interpret and process information the customer provides as input, to provide a output answer including guidance were possible.
4.2.4 - Model input
Text-based customer questions in English
4.2.5 - Model output
Relevant guidance based on NS&I’s public domain product and services information and terms of service.
4.2.6 - Model architecture
Natural Language processing (NLP) Dialog Engine uses the following key actions Intents: The intention of the user. What is the user trying to do? Utterances: What would the user say to convey their intent? Slots and Slot Types: What values can the bot infer from the user’s utterance. What is the specific piece of information that the bot would map to an entity? Confirmations: A message sent by the bot to confirm that it understands the user’s intent.
4.2.7 - Model performance
Prior to release - User acceptance testing for accuracy by subject matter experts. Customer acceptance testing for usability and tone of voice.
4.2.8 - Datasets
NS&I’s public domain product and services information and terms of service
4.2.9 - Dataset purposes
Training the model required ingestion of the documentation in suitable formats. On completion, the model provides relevant responses to NS&I customer queries
Tier 2 - Data Specification
4.3.1 - Source data name
NS&I’s public domain product and services information and terms of service
4.3.2 - Data modality
Text
4.3.3 - Data description
Webpage name, webpage content. PDF name; PDF content.
4.3.4 - Data quantities
A knowledgebase containing circa 200 web pages and document links.
4.3.5 - Sensitive attributes
N/A - all information is public domain.
4.3.6 - Data completeness and representativeness
The data includes all NS&I’s public domain product and service information which is factual and correct. Questions not answerable by the chatbot from these sources may prompt a human agent intervention.
4.3.7 - Source data URL
4.3.8 - Data collection
NS&I gathered together all of its NS&I’s public domain product and service information. As this data is another form of publishing the same information to the same audience, the purpose is identical.
4.3.9 - Data cleaning
N/A - Not required.
4.3.10 - Data sharing agreements
N/A - the data is public domain.
4.3.11 - Data access and storage
NS&I and Atos staff have access to this dataset alongside NS&I customers and the general public due to the information being publicly available. As mandated by NS&I’s knowledge management and data retention policies, information assets are subject to regular review. Additionally, there are ad-hoc reviews of documentation as required when product and service terms change.
Tier 2 - Risks, Mitigations and Impact Assessments
5.1 - Impact assessment
A DPIA assessment was undertaken and confirmed that this tool contains and collects no Personally Identifiable Information. The tool requests no personal information from the customer, other than first name should they wish to speak to an advisor
5.2 - Risks and mitigations
The main risks are:
- data becoming out of date, and
- inaccurate outputs.
How are we mitigating against these risks: - User and customer testing to assure chats are delivered in accordance with the scripts and sources - As mandated by NS&I’s knowledge management and data retention policies, information assets are subject to regular review. Additionally, there are ad-hoc reviews as required when product and service terms change.