Blog: Wells Fargo Team of Business, IT and Data Scientists Selects AI Projects
By John P. Desmond, AI Trends Editor
Wells Fargo has developed a system for prioritizing AI projects from the 80 business units of the 3,000-person company, two speakers explained in a session at 2018 AI World in Boston.
“We have tried to figure out how to organize around AI,” said Krish Swamy, Senior VP, Enterprise Analytics and Data Science at Wells Fargo.
Co-presenter Brian Pearce, Senior VP, Enterprise AI at Wells, described three approaches to how large organizations tend to organize for AI. The first is that “IT owns AI” and is focused on acquiring needed software and services. A second approach is that “business owns AI” and takes a consultant strategy towards AI, building relationships with business unit managers to try and focus AI efforts. A third approach is “data science owns AI,” an approach often used today and favored by data scientists, who Pearce said “like their sandboxes.” Wells Fargo chose to balance data science, technology and business, forming a “virtual team” which combined these three legs.
That decision was made in mid-2016. Since then Wells has developed a set of AI platforms and capabilities that can be applied to different business problems.
“We partner with business users to evaluate new opportunities to use AI to transform experience or streamline operations,” Pearce said. “We are learning from industry best practices and onboarding new use cases.”
Business Unit Managers, IT and Data Scientists Form a Team
Wells formed a team of business unit managers, IT and data scientists to review proposals from business units, for how to apply AI within Wells. “Our AI solution consultants translate business challenges into use cases that can be addressed with current and future AI capabilities,” Pearce said.
The team created a ranking model to guide the decision-making. Important considerations included the value being delivered to the overall business, the readiness of the business unit team to embark on the AI project, and support from business unit management for the project.
This effort helped to focus AI efforts at Wells, to reduce the risk of failed projects. “The number of things we can do very well is inversely proportional to the company’s size,” Swamy said of the company’s AI development efforts.
Pearce said business units request help for many types of projects. “The first thing we ask is, where is your data and is it usable?”
Pearce said his success is measured on the value of the models delivered. “An AI model itself has no value. It needs to be used to influence a business process,” he said. “And we don’t want to build a model and then have it sit on the shelf.”
Challenges include working with the legacy IT infrastructure. “Introducing AI into that can be challenging,” Swamy said. “We could be better at working across data science and technology.”
Issues around data privacy can also be contentious. “In regulated organizations like banks, data access and data privacy can pull in opposite directions,” Pearce said. “That balance has to be struck.”
Data governance and model governance come into play. And the suggestions from AI models need to be explainable. “Credit models cannot be a black box,” Pearce said.
The team in 18 months has identified 500 application projects to work on, he said.
During questions at the end, an attendee in the audience asked if the AI capability could have been applied to help Wells address issues of fraud that have recently plagued the company. The questioner referred to it as “the elephant in the room.” Disclosed in 2016, millions of fraudulent checking and savings accounts had been created on behalf of Wells customers without their knowledge.
Wells now has models in place to help identify “operational risk” that would flag such activity, Pearce said.
Learn more at Wells Fargo.
Source: AI Trends