The crucial link between data, use cases and training models

4 min read


Presented by Envestnet | Yodlee

For the financial services and tech industry, successful AI and analytics strategies require expertise in the complex world of data and modeling. In this VB Spotlight event, learn why it’s critical to partner with experienced data and AI orgs to develop and launch AI initiatives.

Register to watch free on-demand!

AI initiatives can make or break a financial services company in today’s market, and it all comes down to data. “It’s crucial to take a systematic approach, not only to drive value-oriented insights that are applicable to the business, but to ensure you’re using the technology strategically,” says Om Deshmukh, head of data science and innovation at Envestnet | Yodlee.

He spoke with Joe DeCosmo, CTO and CAO of Enova, and Nicole Harper, director of corporate strategy at Jack Henry & Associates, during a VB Spotlight event, exploring the critical connection between determining strategic AI use cases and choosing the right data — a more complex undertaking than most organizations realize.

Solving the right problems

“Any and every problem where there is an availability of data can be solved by using one ML technique or the other,” Deshmukh says. “But does that mean every problem should be solved? Absolutely not.”

There are two considerations, he adds. The first is identifying the problems where bringing data to bear can provide real insight. The second is ensuring the organization has access to data that’s reliable, generalizable and can be enriched to drive a particular insight.

For example, Envestnet | Yodlee has built scalable proprietary algorithms that analyze consumer financial transaction data, all the way down to micro-level clusters, such as how often they go out to eat or order food in. From there, it derives personalized insights that can enrich a customer’s engagement with a financial institution, in the form of financial advice and recommendations, and help the institution determine what their customers are looking for.

“We know that the opportunities are vast to apply AI and ML techniques to improve the experience, but a regulated financial institution is treading carefully and gaining learnings, and there is a lot of risk,” Harper says. “How we de-risk is by developing a way to prioritize use cases. Think of a value-based approach to the matrix and rate the different use cases. What is this business challenge?”

And if it’s a problem that can be solved by AI, it’s crucial to nail down the objective, whether it’s improving customer experience, driving revenue or improving efficiency.

Applying the right data

“We make sure that we have a well-defined business challenge and use case before we move forward with any type of data-driven solution,” DeCosmo says. “That then informs what data we gather, how we build the sample of the data that we’re going to use.” It’s critical as well to have an unbiased sample that offers a good representation of the behaviors the institution is trying to pin down.

“It’s the classic garbage-in, garbage-out, scenario,” says Deshmukh. “It’s a well-worn aphorism but is often overlooked.”

“A lot of times there is a lot of business pressure to just building models and showing some outputs,” he explains. “We go to great lengths to ensure that our data is sampled across multiple different stratified dimensions so that the insights that we derive are truly generalizable, not just along the dimensions that are of interest to us, but also along the dimensions which are not seen today, but which may become prominent, let’s say, a couple of months from now.”

Choosing the right data partner

At Enova, machine learning and automated decisioning has been part of their DNA for a decade, DeCosmo says, driving decision-making across every touchpoint. As a financial institution, it’s crucial that the data be reliable and relevant.

“We try to be very selective about both data that we incorporate and external data,” he explains. “There is an endless supply of data these days, and so we try to be very diligent in the data partners that we work with, because we’re also putting our trust in them, that they’ve provided and built a good data product for us.”

AI is a team sport, Harper adds, and requires an ecosystem approach with data, AI platform and fintech partners. Organizations need to select partners wisely, and select them for innovation, especially in a climate where funding is often an issue.

“When choosing fintechs that you want to partner with, they need to be viable, sustainable and have a good runway to be in business as they may face some headwinds,” she explains. “It also expands the importance of third-party due diligence and limiting and de-risking the selection of your partners; but there is a vast ecosystem.”

For an in-depth look at how data can make or break an algorithm, and how to identify the right data to increase the power of your AI solution, don’t miss this VB Spotlight event.

Register for free now!


  • How does your use case inform the data required for your AI training model?
  • How does data diversity and maturity affect your AI initiatives?
  • What kind of data enrichment is needed to ‘feed’ your AI applications?
  • How might large de-identified datasets help increase your AI solution’s predictive power?


  • Joe DeCosmo, CTO & CAO, Enova
  • Nicole Harper, Director of Corporate Strategy, Jack Henry & Associates
  • Om Deshmukh, Head of Data Science and Innovation, Envestnet | Yodlee
  • Michael Nuñez, Editorial Director, VentureBeat


Source link