- Machine learning will not answer all your prayers, says Samik Chandarana, head of data analytics, applied artificial intelligence and machine learning for JPMorgan's corporate and investment bank.
- Chandarana helped design a three-layer approach to evaluating, testing and incorporating artificial intelligence.
- The approach translates theory into code and useful business cases. JPMorgan tracks ideas so employees don't waste time on projects the bank has already rejected.
- This story is part of a BI series on how AI innovation is playing out across industries like healthcare, retail, consumer and enterprise technology, transportation, advertising, and media. You can check it out here.
The first thing Samik Chandarana needs you to understand is that machine learning will not answer all your prayers.
That might be unwelcome news to the thousands of executives who have seized on the buzzword and its cousin, artificial intelligence, to make their old companies sound new again. And it certainly sounds odd coming from someone like Chandarana, who has worked in JPMorgan's corporate and investment bank since 2017 as head of data analytics, applied artificial intelligence and machine learning.
But it's a key lesson when one considers the increasingly lofty expectations for new tech. Wall Street saved $41.1 billion using AI in 2018, according to an April report from IHS Markit, and AI's business value is seen reaching $300 billion globally by 2030.
So how exactly does a company deploy the technology?
Support system
Chandarana - along with fellow JPMorgan executives Lidia Mangu and Manuela Veloso - has taken a measured approach. After careful consideration, Chandarana decided to position the tech as a support system for business lines within JPMorgan's investment bank as opposed to dictating how, when, and where it should be implemented.
One example is DeepX, a market-making algorithm previously named LOXM that uses machine learning techniques to decide when to execute orders and in what size, depending on market liquidity. The project was originally conceived by the equities trading unitCK, and pulled in expertise from Chandarana's operation as needed. The technology went live in 2017.
"They had the main expertise, they knew the business." Chandarana told Business Insider. "My day-to-day contribution in terms of what they were already doing is quite low aside from making sure that the centralized resources and resource pool that I'm bringing together is there to help them accelerate."
That's not always the case. Sometimes senior executives will hear about something new and force subordinates to find ways of working with it, and the technology becomes more of a marketing tool than something of use to the business.
Just look at blockchain, the decentralized ledger technology that's been used for breeding and collecting digital cats and tracking lettuce.
A lot to lose
With a reported tech budget of $11.4 billion in 2019, JPMorgan has Wall Street's biggest warchest to potentially invest in AI and ML. That also means it's got a lot of money to lose if it does not develop and apply the technology wisely.
"You can't build a utility for the sake of a utility," Chandarana said. "You build it up use case by use case and make sure it has some form of commercial impact at each point."
Chandarana has developed a three-layer approach to evaluating, testing and incorporating artificial intelligence techniques. He oversees one layer, while Mangu, JPMorgan's head of machine learning center of excellence, and Veloso, the bank's head of AI research, sit atop the others.
Chandarana runs the bottom layer, which pairs data scientists with front- or back-office colleagues working directly with internal or external clients. From markets and payments all the way through to operations, data science teams work with specific business groups to understand what they do and how artificial intelligence might help them.
Chandarana wanted to avoid creating parallel teams for problems that would be better tackled together.
"You're guiding people who have relationships and deep-seated knowledge about the business to try and advance the agenda and co-design the solution with the businesses they serve," Chandarana said.
Mangu, who spent 17 years at IBM's Watson Research Center before joining JPMorgan, leads the middle layer, which consists of technique experts in various disciplines of artificial intelligence. Whether it's natural language processing, speech-to-text, or deep learning, the team supports the first layer by translating white papers or research into useable computer code that can be deployed across the bank.
While Mangu's team stays on top of the most current themes, Chandarana said the bank won't use a technology simply just because it's trendy.
"A lot of people use the words cutting-edge or bleeding-edge technology," Chandarana said. "We owe it to our customers to deploy technology in a controlled manner. Therefore it's about appropriate technology not always classified as cutting or bleeding edge. The word 'appropriate' is important."
The top layer is the most theoretical. It is made up of the research team, led by Veloso, a top academic who was hired by the bank in 2018. Her team tackles issues like the ability to explain how algorithms actually came to a solution, known as interpretability, and discussions around ethics and fairness.
Veloso's team also pitches in on projects in the first two layers where deeper research is needed.
'Just a little bit better educated'
And there's further collaboration as well, Chandarana added. There is Rob Casper, the JPMorgan's chief data officer, whose team plays a critical role by cleaning, collecting and collating the necessary data. Then there is Apoorv Saxena, JPMorgan's global head of artificial intelligence and machine learning services, whose group is focused on increasing the uptake of AI applications across the firm by helping develop common platforms, reusable services and solutions.
The goal of the strategy is to successfully implement AI technology, but it can be just as important to identify where the tech shouldn't be deployed.
JPMorgan has set up a system of logging each project. Whether it's a matter of the data not existing in the proper form - or at all - or a project not being worth pursuing at all, everything needs to be written down so the same mistakes aren't repeated. And employees need to be able to find it.
"Making that discoverable means that we're just a little bit better educated than we were yesterday," Chandarana said. "Maybe that means people can collaborate and come up with new ideas about how to attack problems. Or maybe that means they know not to attack it at all because it's already been done."