AI to ROI: How Dun & Bradstreet automated supplier risk evaluation

Dun & Bradstreet (D&B) is a 185-year-old data and analytics firm with more than 590 million business records in its global database. In 2026, the company partnered with IBM to build D&B Ask Procurement, an AI assistant designed to automate supplier risk evaluation for enterprise procurement teams.

What the system does

The assistant runs on IBM WatsonX Orchestrate, integrating Meta Llama 3 and Mistral LLMs with D&B’s proprietary data to handle tasks that previously required manual research across multiple systems: supplier risk scoring, fraud detection, compliance checks, and supplier onboarding workflows. The critical integration step was connecting the AI layer to existing ERP and procurement systems so that outputs fed directly into decision workflows, rather than requiring a separate human review step to transfer results.

Measured results

The outcomes were concrete. Multi-step supplier analyses that previously took hours were reduced to seconds. Time spent on procurement tasks decreased by 10–20%. Risk assessments became more consistent and defensible, with outputs derived from structured data rather than analyst judgment alone, which made the results auditable and easier to justify internally.

Three lessons for product managers

Authors Ray Rike and Peter Buchanan from the AI to ROI newsletter draw out three lessons from the D&B implementation.

The first is that proprietary data is the actual competitive moat. D&B’s 590 million records made the assistant useful in a way that a generic LLM applied to the same tasks could not replicate. The AI layer derived its value from the data asset, not from the model itself.

The second is that workflow integration is not optional. A tool that produces accurate output but requires a manual transfer step to reach decision-makers does not change how long work takes. The efficiency gains at D&B materialized because outputs fed into existing ERP systems without an intermediary step.

The third is that each efficiency gain was measured at the task level — hours saved per analysis, percentage reduction in task time — which is what made the ROI calculation defensible to leadership. Abstract claims about capability do not survive budget conversations; time saved per task does.

For product managers at B2B software companies, this case study offers a template for positioning AI features: identify the proprietary data asset, connect output directly to existing workflows, and measure efficiency in task-level time savings rather than in capability claims.