Now Reading
Trust Is Central To Banking And CustomerXPs Helps Them Leverage It, Says Rivi Varghese


Trust Is Central To Banking And CustomerXPs Helps Them Leverage It, Says Rivi Varghese


Founded by fintech product experts, CustomerXPs believes that like a human soul, a customer too has a soul – the sum total of what a customer is. “Among all the industries, banking is the only one that captures the customer’s soul," said Rivi Varghese, CEO, in a candid conversation with Analytics India Magazine. He shared some interesting insights about how their company leverages artificial intelligence, machine learning and analytics, to provide real-time enterprise solutions to global banks, their growth plans and more.



Varghese entered the world of AI and real-time analytics about 18 years ago. A course on statistics during his MBA at IIM Bangalore exposed him to methodologies for massive data crunching which encouraged him to conceive a startup that focused on the same.

“In 2006, we started creating solutions using fuzzy logic and eventually advanced to tools in AI," he says. They then decided to sharp focus on developing extreme real-time solutions for banks using AI, ML and analytics.

"Among all the industries, banking is the only one that captures the customer’s soul."

-Rivi Varghese
Analytics India Magazine: Why did you begin with banking as your area of interest?

Rivi Varghese: When we started about 10 years ago, we had already interacted with more than 200 banks in over 20 countries. That’s when I realised that banking is that one unique sector where the entire life of the customer goes through the bank. While every other industry has a fragmented, unidimensional view of its customers, a bank knows everything about you — how much you earn, your marital status, where you travel to, whether you live on rent, even how much fuel your vehicle consumes.

But we found it odd that banks were saying they were in the relationship business, but in reality, they were not quite aware how exactly. We started CustomerXPs to help banks actually discover their customers’ soul, and realised that AI and ML are the way forward. We started developing our product using computational technologies and fuzzy logic engines. We are talking about chatbots now, but our customers went live with it eight years ago.

AIM: How is your approach of providing AI to banks different?

RV: We believe in putting the end state in perspective and then work towards it. We put substantial effort into understanding a client from scratch before developing the AI solution. AI for us is the means to an end and not the end in itself. Almost every bank has made an investment in some way or the other in AI and ML.

The usual practice is to buy costly software licenses, put hundreds of ML engineers and spend millions of dollars buying huge hardware. After years of such costly spends, we still find that banks are saddled with “know why” and the science, but still haven’t figured out the “know how” – as to what is the exact problem they are trying to solve. We decided to tilt this practice.

We sit down with banks to understand their problems in depth before developing solutions that work for them. Mere models and fancy tech jargons will not solve problems - we need a different, real, smarter way of getting things done.

AIM: Does that mean that you offer customised AI solutions to banks?

RV: Let's first look at the three top hurdles preventing large banks from deploying large-scale AI.

First they don’t have access to absolute real-time data. This is because data warehouse is more or less a failed concept, and everywhere the data is late by hours, days or even months. The data is not structured properly and even if you feed it into the ML models, they don’t work the way they should.

Second if you have non-siloed data, though you could create phenomenal models which help in fraud detection and protection, they still lack the last mile connectivity to influence a transaction is in flight, the flitting moment of truth. Not being able to do this, reduces your AI efforts to just yet another report.  The best way to do this is with an enterprise-wide anti-fraud system, which is connected in real-time to all the decision points in a bank.

Third the sheer complexity of managing huge software. Conventional products are dated and most of the time, with origins from the statistical background. If you really want to do AI, you need to look at what the world is doing today, now and be on the bleeding edge and not have a dated approach.

So, what we have instead is an appliance model. We put all the information into the appliance which contains all the latest software including GPU. GPU does massive volumes of parallel processing and we encapsulate all of this into a box. While the data is incoming real-time, the massive processing can be done easily.

In the appliance model, a bank doesn’t incur huge capital expenditure, as it is a monthly subscription, and there is no need for licenses – think of software and hardware being free. This keeps the vendors on their toes to deliver every month, instead of taking 100% of the licenses up front and leaving the bank to figure out what it intends to do.

"While every other industry has a fragmented, unidimensional view of its customers, a bank knows everything about you."

-Rivi Varghese
AIM: Interesting. And how you are dealing with the issue of money laundering? Where does analytics fit in?

RV: Banking is about customers’ trust and the moment you lose it, it is gone. Our core product is in enterprise fraud management and financial management which consists of fraud, anti-money laundering and compliance. We use AI and ML to ensure that all of these works better.

When we say AI and ML for money laundering, it means a lot of things. Dated AML systems, generate a lot of alerts. In a differentiated approach, we need to mimic human behaviour and understand what they’re trying to do. The engine learns from it so that the human effort with respect to intervention, alert resolution times, and better hits are addressed. From an investigator’s perspective, instead of looking at many alerts, there are fewer alerts in the first place, so lesser time is spent per alert, and the overall accuracy is increased.

See Also

Say the investigator has to enter a phone number search from millions of accounts for interlinkages. With traditional analytics, it takes anywhere from a couple of hours to a few days to process unstructured data, but in our case, our AI engines accelerate this and get this intelligence in seconds. What we are doing is applying ML to work in sync with our core belief and not because AI is trending.

AIM: Tell us about the analytics and AI tools you use

RV: We have a core product which is a real-time decision engine, to analyse, detect and stop fraud within the transaction window itself. The other is a thinking engine in our appliance model. In both cases, we use open source tools and technology such as dashboarding, enhanced in-memory capabilities to constantly improve our system.

AIM: Could you highlight some of the use cases in banks?

RV: Our use cases are centred around fraud detection, compliance, anti-money laundering and generating alerts. For instance, in fraud, it is about the quantum of money saved for the bank.

AI and ML at its core, operate on probability. There type A and type B errors. Making the error rate close to zero but not zero requires substantial levels of optimisation. The other models work on information asymmetry in banking systems such as locating the user based on their place of carrying out a transaction. This metadata is also important for ML parameters.

Every bank is now thinking about AI within its own realm and when we look at banks there are only a few labelled cases. Say out of 100 transactions, only 1 is fraudulent. And, they land up extrapolating it to 99 times. So, if you have a 10% error in your basic model, the error rates get multiplied by 99 times and the whole thing becomes unpredictable.

When you go deeper into serious, large-scale enterprise deployment for banks, these are the hard problems that the banks are trying to crack – and we crack these innovatively.

AIM: How big is your team?

RV: We are now close to 140 FinTech experts who work on building, customising and implementing large-scale banking enterprise systems. We know it is vital to first define each problem precisely and then decide on the right ingredients to solve it. Once we have these sorted, we then decide on hiring. We work with leading System Integration partners to give us the scale.

AIM: What is the roadmap for 2019?

RV: Every day the world wakes up, we will make available to our banks ‘whatever the world has learnt till then’ global intelligence to be used for every decision related to “trust” in a bank.  This is the goal that we are working towards, we expect all our key customers signing up for this vision of ours.



Register for our upcoming events:


Enjoyed this story? Join our Telegram group. And be part of an engaging community.


Our annual ranking of Artificial Intelligence Programs in India for 2019 is out. Check here.

Provide your comments below

comments

What's Your Reaction?
Excited
0
Happy
0
In Love
0
Not Sure
0
Silly
0
Scroll To Top