NatWest
genAI digital assistant (Cora+)
Cora, NatWest's digital assistant launched in 2017, went from over 5 million queries in 2019 to 10.8 million in 2023; in June 2024, NatWest added a generative AI layer, Cora+, built with IBM watsonx.
Key points
- Cora digital assistant with a genAI layer (Cora+) for complex financial questions.
- Built with IBM watsonx Assistant on IBM Cloud, after a twelve-week pilot.
- Volume grew from over 5 million queries in 2019 to 10.8 million in 2023.
- Evidence B, confirmed status.
Objective
Have the digital assistant absorb most of the everyday financial-management questions, then move upmarket with GenAI (Cora+) to hold more complex and more human exchanges without systematically routing to an advisor.
The deployment
Cora is NatWest's digital assistant, launched in 2017, available on the website, the app, and social media. Its volume climbed from over 5 million queries in 2019 to 10.8 million in 2023. In June 2024, NatWest launched Cora+, a generative AI layer built with IBM (watsonx Assistant on IBM Cloud) after a twelve-week pilot: natural-language responses, understanding of complex questions (credit, mortgage), proactive assistance based on history, and conversation summaries for handoff to an agent.
Results Proof B
Volumes published by NatWest in its official communications (T1) and the Cora+ launch documented with partner IBM (T2). Concordant volume figures over several years, but operational and not audited in financial results, hence B.
How it works
Documented architectureThe stack in detail
- plateforme IBM watsonx Assistant Conversational foundation of Cora+: understanding of complex questions and natural-language responses.
- llm LLM generatif watsonx Generative layer for the responses and conversation summaries passed to agents; the exact model is not named in the sources.
- infra IBM Cloud Hosting of the solution.
- integrateur IBM Client Engineering Support for the twelve-week pilot and then the launch of Cora+.
How it runs, concretely
For ops teams-
1Customer question customer
The customer asks Cora on the website, the app, or social media (balance, payment, management, product).
-
2Conversational response AI
Cora+ answers in natural language, including on more complex topics such as credit or mortgage, drawing on the customer's history.
-
3Handoff to an advisor AI / customer service
If needed, Cora+ passes a conversation summary to a human agent to continue without starting over.
-
4Monitoring and training data team
The team tracks volumes, satisfaction, and the share handled without a human, and enriches the knowledge base.
The volume of conversations resolved without an advisor and satisfaction. If the share handled without a human drops, the service cost rises again on the contact center side.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- Structured, up-to-date banking knowledge base
- Conversation history for training
- Access to customer history for proactive assistance
Org prerequisites
- Choice of an assistant platform (watsonx or equivalent)
- A handoff-with-summary rule to an advisor
- Knowledge base governance
Possible stack
- IBM watsonx Assistant
- Alternative: assistant platform + generative LLM
- Multichannel integration (web, app, social)
The plan, step by step
- Step 1Choose a pilot scope on the existing assistant (high-volume topics) and define the handoff-with-summary rules to an advisor.Deliverable: Validated pilot scope and success criteria.
- Step 2Build the GenAI layer on the knowledge base: natural-language responses, conversation summaries, access to customer history.Deliverable: Working GenAI assistant in pre-production.
- Step 3Pass the banking compliance, tone, and escalation tests, with adversarial question sets.Deliverable: Risk and compliance validation file.
- Step 4Launch the pilot in production on a fraction of traffic and measure resolution without a human and satisfaction.Deliverable: Readout of the twelve-week pilot.
- Step 5Broaden the topics covered and generalize to full traffic, continuously enriching the knowledge base.Deliverable: Generalized GenAI assistant with a volumes/satisfaction dashboard.
First step: Start from an already-deployed assistant and add a GenAI layer on a pilot scope, with an automatic summary for the handoff.
The same pattern, proven in other industries
ADT
genAI customer agent
Air India
genAI customer agent (automated customer service)
AT&T
Governed enterprise genAI assistant: an LLM connected to internal knowledge, open to tens of thousands of employees and called at scale by automated systems
Sources
- S1 NatWest launches Cora+, the latest generative AI upgrade to the bank's digital assistant Primary archive pending
- S2 We're giving our digital assistant Cora, a generative AI upgrade Primary archive pending
- S3 NatWest engages IBM to revamp chatbot service with generative AI Secondary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.