Citi US Personal Banking turns to AI to ‘delight’ customers with personalised services

Customer analytic records

Citi has developed a customer analytic record (CAR), which provides a view of each customer’s data, including their financial records, the financial products they use and their interactions with the bank through online banking, visits to branches, phone calls, or email.

The bank is linking its CAR data pool with automated decision-making software, supplied by US technology company Pegasystems, which it has branded the Omnichannel Decision Engine.

The decision engine uses Pega’s Customer Decision Hub software to analyse customers’ interactions in real time and recommend relevant services to offer customers at any given time.

“We’re not pushing things to our customer that they don’t necessarily find relevant, because they’ve indicated to us that they’re not interested, at least not right now,” said Dutta.

The decision engine is able to suggest offers or services that might genuinely interest the customer, flagging them up for bank staff to recommend or delivering them direct to the customer through text, mobile banking and other channels.

Bank staff have reported that it’s making a “tremendous world of difference” to their conversations with customers, Dutta added.

Untangling the IT web

Citi used a wide range of decision engines, including its own in-house rules engine and Pega’s Chordiant software, to manage its interactions with customers, but they were not connected together. This meant the bank’s mobile and internet banking decision engines operated independently.

Citi came under pressure to replace the software it used to communicate with customers on its website and mobile apps some four years ago, after Pega began to wind down technical support for Chordiant.

“We have to understand the customer sentiment to understand what their pain points are, or what’s working well. So unstructured data is a big area for us when it comes to improving customer experience”

Promiti Dutta, US Personal Banking, Citi

In its search for a replacement for Pega’s Chordiant decision engine, the bank issued several requests for proposals to technology suppliers, but each time reached an impasse. That changed in 2019. “We finally pulled the trigger on making the decision after doing a really quick review of the different tools that existed,” Dutta told Computer Weekly.

Citi reviewed proposals from 20 suppliers, including Pega, Salesforce and Adobe, before opting for Pega’s Customer Decision Hub.

The project required bank staff from multiple business departments, including analytics, technology, the fair lending and privacy teams, to work together.

Zero tolerance for failure

Although Citi had experience with Chordiant, its replacement, the Customer Decision Hub, had more advanced capabilities. And Citi needed to develop new business processes to take advantage of them.

As a regulated financial services company, there was zero tolerance for failures, said Dutta. “That was a huge challenge because you were having learnings and having to adapt to those learnings on the fly while making sure that you didn’t negatively affect anything,” she said.

Citi’s analytics team worked closely with teams across the bank to work out what business processes they were using, why they were using them, and how they would fit together.

The task was made more difficult during the pandemic because Covid meant that much of the work had to be carried out remotely. “There were many, many workshops, many, many hours spent on calls,” said Dutta.

The team’s strategy was to find enthusiastic volunteers from relevant parts of the business and invite them to work on the project. “Anyone who was ready for transformation came on the transformation first, and we never said no to anyone who said yes,” she said.

Dutta started with the marketing department, and began moving marketing tasks over to Pega’s Customer Decision Hub, case by case. Next came servicing and customer engagement.

“We figured out the paradigm and the protocol that works there. And now we’re applying it over and over again,” she said.

Explaining to other parts of the business how Pega’s Customer Decision Hub works and how to link it to the bank’s data was one of the biggest challenges faced by the project team. “Everyone thinks you can just plug it in and it works. That’s not the case. It’s not like your home Wi-Fi systems,” said Dutta.

Today, Citi has fully integrated its web and mobile channels into Pega’s Customer Decision Hub. It is well advanced at linking its network of branches and call centres.

Unstructured data

One of Dutta’s priorities is to find ways of using unstructured data, including notes taken by bank staff when a customer visits in person or contacts the call centre.

“We have to understand the customer sentiment to understand what their pain points are, or what’s working well,” she said. “So unstructured data is a big area for us as it comes to improving customer experience.”

The bank is looking at using AI technology to automatically produce transcripts and summaries of calls with customers and to identify the primary purpose of each call – a process that is not always straightforward when customers talk about several issues in one call.

Collecting this data will allow the bank to check whether it has solved the customer’s problem and to identify whether they are having to call back more than once with the same problem.

“It is important to figure out what’s causing our customers most grief – and how to quickly resolve that grief – so we’re able to better prioritise,” she said.

The bank is also looking at ways to capture the voice of customers in real time and to identify how they are feeling about their interactions with the bank. “Imagine if we’re able to capture the voice of the customer in real time to identify issues as they arise. You can get ahead of so many problems that other customers will not face because we’ve proactively remedied them,” said Dutta.

Self-learning technology

Citi also has plans to deploy the self-learning capabilities of Pega’s Customer Decision Hub. In the financial services industry, the use of self-learning technology is highly regulated.

“There are parameters that are predetermined by our model risk management platform, where we are allowed to tweak the weightings based on what’s happening to our customer and how they feel about it,” said Dutta. “So some of the parameters and models are consistently shifting and learning from what the customer says in their engagement.”

From Ford Focus to Tesla

One of the main challenges of the project was keeping multiple parts of the business involved in the project updated. If she were to run the project again, Dutta said she would be more proactive about keeping other parts of the business informed.

“One thing that I would do differently is figure out how to be far more communicative across all stakeholders, so people are more in the know about the different steps and challenges,” she said.

Dutta’s advice to other companies planning to move from a system like Chordiant to Pega’s updated Customer Decision Hub is to put strong change management processes in place across every aspect of the project.

“You’re going from the Ford Focus of engines recording to the Tesla self-driving engine. You don’t operate both in the same fashion, so the way you drive one doesn’t work in the other. So change management becomes the key,” she said.