RBS looks to improve customer satisfaction by unlocking its web chat data in Hadoop, eyes voice data from phone calls next

The Royal Bank of Scotland is working with the 'data wrangling' company Trifacta to gain insight into its customer service web chat logs, as more and more customers turn to this channel when it wants to interact online with the bank.

RBS stores approximately 250,000 such chat logs and the associated metadata per month. The bank stores this unstructured data in Hadoop, however before turning to Trifacta this was simply a huge - and previously untapped - source of information about its user base.

RBS was only analysing small samples of these chats for quality assurance and training for its 300 web chat agents. “Of those 250,000 [monthly] chats we analysed 200 for quality assurance, that’s no good,” said RBS’ head of big data and innovation Dan Jermyn, speaking at the Strata+Hadoop conference in London today.

Data Wrangling

Connor Carreras, a customer success manager from Trifacta said companies like RBS aren’t using the data they have “because it is messy, non-tabular”, and the staff analysing that data are “used to using tools which can’t handle unstructured data”.

Read next: Analytics and business intelligence tools for enterprises: What are the best alternative BI products?

Silicon Valley-based Trifacta specialises in taking the unstructured data in a Hadoop data lake, where traditional schemas and BI tools aren't fit for purpose, and ordering it for analysis within the enterprise.

Carreras did a demo of the platform, taking the chat logs and highlighting certain parameters. For example: credit_card_question, mortgage_question and current_account_question. The tool learns patterns as you start to train it and bunch these together, giving the data some structure, with no coding required.

Once the data is in order RBS could, in theory, start to map paramaters like a customer service agent’s ID number to the time stamp for a customer’s response wait time. This would effectively give them a leaderboard of agent's average response time to customer queries.

Carreras said that ‘wrangling’ the RBS web chat log data took them “a few weeks for the wrangling piece and end-to-end one to two months”. Why the delay? “We had to build out the Hadoop infrastructure and go into production,” said Carreras.

Tripping over gold bars

By getting their Hadoop data lake in order RBS was able to stop “tripping over gold bars” as Jermyn’s boss is fond of saying. “Customers are actually telling us what is going on and we can start to use that,” said RBS’ Jermyn.

RBS found there was a disconnect between how well a customer service agent performed and how well they were marked on the follow-up survey. This was because if the service the customer wanted wasn’t available online they would mark a zero for unsatisfied, regardless of the agent’s best efforts to help.

Now that his team can analyse the chat logs in full they can start to look at the context of the conversations, “we can use sentiment analysis to understand if they are doing a good or bad job and give a fairer assessment,” said Jermyn.

Read next: RBS says it's 'twelve months ahead' of other UK banks with Luvo AI customer assistant

RBS can also tie in other proprietary data sources and layer on data visualisation tools from SAS to provide insight such as uptake of certain products from customers and geographic analysis.

For example, Jermyn’s team saw that they had poor customer satisfaction in Australia and couldn’t work out why. They soon realised that expats and holidaymakers weren’t able to access key online products while they were down for maintenance, which is generally early in the morning UK time. RBS discovered that it “had a UK centric view of the world”, said Jermyn.

Jermyn is now able to proliferate this sort of capability across the thousand-strong analytics team within RBS. “We want to be agile, we don’t want there to be any technology blockers. If they have an idea to improve stuff for customers they should be able to,” he said.

What next?

It looks like this is just the first step for Trifacta and RBS. Jermyn sees a huge opportunity to go one step further if his team were able to analyse telephony data in the same way.

He said: “There is voice and text stuff we are looking at to apply to the work we have done already. [Voice] is a massive source of what customers are saying that we want access to. When you look at web chat it is 250,000 chats per month but telephony is twenty million calls a year.”

He says RBS is looking at translation services as the first step for this project already.


Copyright © 2016 IDG Communications, Inc.

Shop Tech Products at Amazon