Morgan Stanley has been investing in artificial intelligence R&D and earlier this year launched a pilot program where some of its financial advisers are assisted by AI in everything from helping wealthy clients decide what to do with their money to suggesting what to write in emails to those clients.
Company researchers have also been thinking about using AI to make employees more productive, Pegah Ebrahimi, COO of Global Technology Banking at Morgan Stanley, said from stage at the AI Summit in San Francisco this week.
Another big application for AI in the financial services industry is cybersecurity and fraud detection, she said. It’s not feasible for a human to examine every single transaction, but a sophisticated AI model may surface anomalies across a vast pool of transaction data.
Ebrahimi shared some of the top challenges Morgan Stanley has been up against as it continues exploring various ways AI technologies can be applied to its business:
1) Finding Useful Data
One of the biggest problems is scarcity of data. Yes, you read that correctly. While companies have access to more data than they’ve ever had before, datasets that are useful for AI applications in the financial sector are few and far between. Those would be datasets where the data is labeled.
“Everyone knows the best AI we’ve had is around supervised learning, and that all requires labeled data, and there’s real lack of labeled data in financials,” Ebrahimi said.
Labeling data means organizing it to make it ingestible for machines. While labeling pictures of dogs as dogs when training a deep neural network to recognize dogs in images can be done by any unskilled worker, labeling diverse and complex financial data requires a level of expertise that may make it cost-prohibitive to use human labelers.
2) Finding the Right Talent
The second big problem is scarcity of talent for AI model development. “AI talent is really scarce,” Ebrahimi said.
It is doubly challenging to find people who not only understand AI technology but also have domain expertise specific to what Morgan Stanley does – a necessary combination of skillsets that’s difficult to come by.
3) Selecting the Right Use Case
The third challenge is selecting the right use cases. Morgan Stanley is currently running a pilot program, augmenting wealth-management advisor with AI algorithms to help them make better decisions on behalf of their clients.
For the next use case, everyone has a ton of ideas, but there are some major considerations that go into committing to one of them. The biggest one would be the likelihood of the final product meeting user expectations.
Since this is a new technology, companies still have a lot of work ahead of them in building trust, and if an organization commits to a use case that will not deliver accurate results, that trust will be more difficult to build.
This is why Ebrahimi’s philosophy is to walk instead of running with new AI applications. “You have to build confidence in that,” she said.
4) Open Source vs Proprietary
The vast pool of open source technology counterintuitively presents another big challenge: choice. There’s so much open source AI technology out there that it’s difficult to decide when to use open source and when to develop proprietary tech.
Open source models may be useful but businesses need to think about ways to stand out from the pack. “The real value is not in the generic models but in the domain expertise, proprietary models,” Ebrahimi said.
5) Tough Compliance Standards
Finally, and obviously, whenever a financial services firm invests in a new technology, it has to answer some tough questions around compliance. It took a while before the cloud services market matured enough to meet the financial industry’s compliance requirements, and she expects to see the same dynamic with AI.
Financial compliance standards don’t make AI research impossible for companies like Morgan Stanley, but they do make innovation either slower or more expensive than in lighter-regulated industries, she said.