AI will restore our customer confidence, says Natwest
Posted: 3 July 2017 | By Charlie Moloney
An artificial intelligence (AI) system, which can record and analyse customer conversations, will be trialled by Natwest in order to ensure they provide better quality information and advice, a Royal Bank of Scotland (RBS) press release announced last week.
Natwest has partnered with Recordsure, a London-based fintech company, whose AI, with customer consent, will record face to face and over the phone customer interactions, which will be analysed by human agents and the AI itself.
According to Natwest’s press release, the Recordsure AI can identify distinct parts of a conversation, so it will know, for example, when a customer is being given an initial disclosure, financial advice, and general conversation. It should be able to inform staff if certain topics or issues have not been discussed.
Conversations are immediately encrypted and uploaded to a secure cloud, before being archived at former UK and US military facilities, originally built as nuclear bunkers
Security is said to be a top priority in the introduction of the AI, and conversations are immediately encrypted and uploaded to a secure cloud, before being archived at former UK and US military facilities, originally built as nuclear bunkers.
“We want to restore trust and give customers complete confidence in us”, said Les Matheson, CEO of Personal and Business Banking at RBS, parent company of the Natwest brand, which reported a £7 billion loss in February 2017 after being rocked by misselling and conduct charges.
Today we spoke to Kit Ruparel, CTO at Recordsure, who answered an information request from Access-AI and explained some of the finer points of the Recordsure system:
How does the Recordsure AI deal with things like slang, accents, and informal speech when it is listening to conversations? How will customers be sure that the AI will accurately record what they say?
Spoken English is very different to written English, so the Recordsure AI is trained on the bank’s real audio conversations that include slang, accents and informal speech. In addition, we supplement this AI training with a ‘Knowledge Model’ that comprises millions of hand-curated examples and variations of spoken English from pronunciations, through idioms, colloquialisms, and dialogue patterns.
Regards accurate recording, the recorded audio itself is retained as the authentic conversation record for the period required by the regulators, and whenever any review or judgement is to be made by RBS staff regards the conversation they are always referred to the relevant section of the audio itself for review.
Is the Recordsure system a machine learning (ML) system? I.e. will it continue to learn and grow through continued use?
It is a Machine Learning system and as well as performing regular manual training (learning) exercises to provide proven, measurable improvements, we also use techniques such as ‘Reinforcement Learning’ and ‘Unsupervised Learning’ to ensure ongoing continuous improvement of the solution.
How has the Recordsure AI been built? Is big data involved?
A lot of big data is involved! Essentially four key data sources are used:
- A manually curated Knowledge Model consisting of millions of pieces of information and what they mean, ranging from examples of things that can be said, through to financial concepts such as bank accounts, mortgages, pensions, investment and borrowing methods, etc.
- Web crawled data from sites such as the FCA, as well as banks’ own websites
- Real audio conversations from across the industry, using an AI method known as ‘Transfer Learning’ so that there is no actual data transfer between Recordsure clients, but instead, only the ‘middle layers’ of neural nets are re-used.
- Finally, and most importantly, RBS group’s own conversations with clients
Will Recordsure have any access to records of the conversations that Natwest customers have with the AI?
Recordsure staff, such as our scientists, who are directly involved in training the system, as well as necessary technical operation staff that need to support the system, have access to the records. These staff follow similar employment vetting and security training procedures regards data-handling as the bank’s own policies, and this has been verified by RBS.
Records are only made available to relevant Recordsure staff on a need-to-know basis during the system training windows, all access to records is automatically logged and the audit trail is kept indefinitely.
The locations, processes and systems used to access the records are highly secure and have been audited by the bank, with the systems and methods very much aligned to the bank’s own procedures.