Commercial credit decisions have long moved at the speed of the analyst: a new scorecard takes weeks to design, months to validate and another cycle to clear regulatory sign-off. Dun & Bradstreet says agentic AI can compress that timeline. The company announced July 6 that its D&B Commercial Graph is now available through Databricks Marketplace and OpenSharing, powering three agentic workflows for credit origination, policy optimization and portfolio risk monitoring. In one anonymized portfolio example, the workflows raised the bad capture rate from 30% to about 38%, avoiding more than $6 million in bad debt, the company said.
“Finance leaders are under pressure to move faster, support growth and manage risk with greater precision,” said Scott Spencer, general manager for finance and credit at Dun & Bradstreet. “With the help of agentic AI, tasks that would take analysts days or weeks are completed in seconds.” That speed runs on decades-old infrastructure: the workflows anchor on the D-U-N-S Number, the global standard for identifying commercial entities that D&B originated in 1963.
Commercial Graph Gives AI Agents a Verified Starting Point
The Commercial Graph holds more than 650 million trade lines worldwide, spanning the geographies D&B owns directly and its network of partners in Eastern Europe and Asia, Spencer said. “Our trade database is designed to take into account all of those trade lines and build a complete commercial profile of that small business,” Spencer said, “so that when that small business applies for a loan product, a lease or a small business credit card, they’re able to do so with the maximum set of choices out in the market.”
That scale is what makes identity resolution possible, and it’s where AI has delivered the earliest gains. A lender evaluating a small business applicant may encounter variants in the company name, address and owner’s name across different data sources. The system targets a 98% confidence threshold for matching a business to its proprietor before any credit decision proceeds, Spencer said. “AI does that stuff extraordinarily well,” he said. “Identity resolution is kind of the start of the process, and then we’re able to cross all the way through to credit issuance.”
Once a business’s identity is confirmed, the workflows move to policy. Traditional scorecard development involves internal modeling, governance reviews and regulatory approval before a revised policy reaches the portfolio. Spencer said the Databricks integration compresses the design phase. “The part around design and optimization can happen in minutes, not weeks,” he said. “You could see a world where scorecards [were] only getting rolled out once or twice a year. You could be rolling out one once a month.” Early results support that claim. “We’re seeing things like 20% KS lift, reduction in bads in the millions of dollars, with the ability to approve more customers,” Spencer said. “You’re able to say yes to folks, see your losses come down, which increases both revenue while decreasing bad debt write-offs.”
Portfolio Risk Is Where the Proactive Case Is Strongest
Portfolio monitoring today relies on threshold alerts: when a borrower crosses a defined risk parameter, the system flags the account. Spencer said the new workflow changes that dynamic. “We wanna actually push based on what we’re seeing on a variety of different signal data, including some of our supplier data and shipping data, to be more proactive with these accounts,” he said. Spencer said that same proactive posture could eventually extend beyond monitoring existing borrowers to vetting new ones. “If you’re a home builder and you’ve got to go source windows for a new complex, you can go tell your agent to go out and get you six proposals,” Spencer said, “and have that agent be able to represent your identity and then do some basic KYC and creditworthiness checks to bring you back qualified vendors.”
