BANKING

How banks can use the latest AI text analytics to improve customer loyalty

How banks can use the latest AI text analytics to improve customer loyalty

By Dan Somers, CEO at Warwick Analytics (www.warwickanalytics.com)

A major UK bank was looking to improve its customer loyalty. It was already using the latest analytical tools including social listening, sentiment analysis and a large data science team but they were experiencing limitations and not making enough progress.

The bank was keen to find more opportunities to improve customer loyalty and reduce their operational costs by gaining more customer insight. They were also interested to see what online feedback their main competitors were receiving.

The bank invited Warwick Analytics, specialists in AI text analytics, to carry out analysis using their automated text analytics platform PrediCX.

Some key outcomes from the analysis included:

  • £200m per annum of churn mitigation was identified through operational opportunities
  • £240m per annum of churn mitigation was identified through CX opportunities
  • Customer Service savings opportunities of £20m per annum through automation
  • Ongoing tactical opportunities identified for marketing effectiveness and propensity

The analysis

PrediCX analysed Tweets associated with the bank, and 23 of their competitors, over a 3 month period.

Sentiment Analysis can present issues with accuracy, context and multiple sentiments. The technology overcomes these by using the latest in machine learning and AI to classify by ‘concepts’ instead of ‘key words’. The result is a highly tuned model, built in a single day, with thousands of classes that pick up signals of issues,customer sentiment and more importantly customer intents.

Automation is optimised with ‘human-in-the-loop’ functionality which means it only asks for human input when it needs validation and it then trains itself behind the scenes.

Key findings

The automated analysis found several key opportunities for improvement and savings:

Firstly, the key topics of conversation were classified and then broken down into many classes of detailed, statistically-relevant signals. Service and Customer Service emerged as the most common topics for the majority of the banks.

The AI then classified the detailed intent of the customers to get a deeper understanding than existing sentiment analysis was able to provide. The larger banks displayed the highest churn with challengers and mid-tiers performing better. The supermarket banks generally had low churn but very high negative advocacy, representing a customer base who are all not 100% happy, but much more loyal. When breaking down the root causes of customer churn, it was evident that customer service, branch and ATM closing and interest rates all caused the greatest churn in the market.

The amount of effort expended by customers at different banks to set up new products was also analysed. Current accounts, business accounts and mortgages are associated with the most proportional churn whilst current accounts and mortgages were the most challenging to set up (ancillary services such as share accounts, notary services etc. are also highly negative but very small).

For the customer services teams, there seemed to be distinct patterns of churn and comments based on time of the day and day of the week, indicating opportunities for customer service resourcing.

Finally, the bank was able to get a cross market analysis of who and when customers were churning with their competitors. This highlights when different banks are strong or vulnerable and could present tactical marketing opportunities i.e. targeting potential customers and increasing switching propensity.

Conclusion

 A number of key recommendations for the bank were identified just by this analysis alone:

  • Estimated 20% of churn is caused by incidents and this represents maximum opportunity.
  • 10% improvement is c. £200m pa revenue compounded yearly
  • Comparable best-in-class churn e.g. Nationwide is 25% lower.
  • Potentially a further 10% i.e. c. £200m opportunity with improvements
  • Online and mobile banking is a key issue, and is causing direct churn
  • Sentiment is middling. It does not appear to correlate to churn for the market
  • Drivers of churn are mostly customer service, branch closures, marketing offers,
  • interest rates and vulnerability issues
  • Early warning can help predict churn tactically and intercept likely churners
  • 28% of Tweets and potentially all non-voice queries can be automated. This could be £20m pa saving
  • Business banking, current accounts and ancillary services have the highest churn,
  • and insurance the highest negative advocacy
  • Mortgages, current accounts, savings and overdrafts cause the most attritional set-up.
  • Opportunity to improve the journey
  • There are distinct patterns and opportunities to change customer services planning
  • over the week and day to reduce churn and costs

It’s clear to see how the adoption of the latest machine learning for text analytics can present far more insight, a much deeper level, than traditional sentiment analysis and text analytics. More significantly it is able to correlate these signals to operational issues to reduce costs and improve customer loyalty.

With the latest technology, this level of insight can be set up in a matter of days, delivered in near real time and without the need for a data scientist to maintain the model.

To Top