Document Processing · Risk Management
Deutsche Bank needed to review third-party vendors' documentary evidence against its control framework and regulatory expectations faster without removing expert judgement, because onboarding and periodic reviews required specialists to examine multiple evidence files and make documented risk decisions manually.
Reported impact
The platform reduced document review times, accelerated vendor onboarding and processing timelines, delivered cited recommended outcomes at around 90% accuracy versus human validation, and gives Deutsche Bank a scalable basis for handling higher assessment volumes with cost-avoidance potential while keeping trained humans in control.
Deutsche Bank built TPRM AI in-house and deployed it in Global Procurement & Vendor Management in December 2025 to accelerate third-party risk reviews while preserving human decision rights. The cloud-based platform ingests documentary evidence supplied by vendors, including business continuity submissions that may arrive as four to five separate files, and processes the material through a sequence of specialized AI agents. One agent retrieves the relevant Deutsche Bank control questions, a second drafts answers from the evidence set, and a third proposes an assessment outcome with citations to exact source passages. Those suggested answers and cited outcomes are then served to human assessors, who validate the recommendations and make the final approval decision. The source does not name storage services, but it does describe a scalable human-in-the-loop document analysis workflow designed for transparent verification and reuse across additional risk assessment areas.
Third-party vendors submit documentary evidence to Deutsche Bank, sometimes as four to five separate files. TPRM AI ingests those files, retrieves the relevant internal control questions, drafts answers from the evidence, and generates a suggested assessment outcome with citations to exact passages. Human assessors then review the cited recommendation against the control framework and make the final onboarding or review decision.
"A manual review would typically take around 30 minutes per evidence document."
"Because it’s cloud-based, the solution is designed to scale as assessment volumes grow and requirements evolve."
"In cases like this, a manual review can take around three hours. With AI-supported document review followed by human validation, the work can be completed in about 30 minutes."
"TPRM AI not only provides us with speed of processing, but also achieves around 90% accuracy in recommended outcomes, as benchmarked against human validation."
"TPRM AI now analyses up to five documents in under two minutes and provides suggested outcomes with citations."
"TPRM AI, an agentic AI platform developed by our colleagues in the Technology, Data & Innovation team and implemented in Global Procurement & Vendor Management in December 2025."
"The first one retrieves the right Deutsche Bank control questions, the next one drafts answers based on the vendors’ evidence, and the third proposes an outcome with precise citations, while human experts stay firmly in control to make the final decision."
"This solution uses multiple AI ‘agents’ to speed up the assessment of vendor documents and suggest an outcome."
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
In-house Sample Oct 20, 2025 BNY BNY wanted to scale AI adoption across the company without fragmented point solutions, while maintaining strong governance, risk management, compliance controls, and practical support for workflows such as payments,… 75% more AI solutions in production compared to the prior quarter Contract Review Assistant · Eliza · Eliza agents In-house Sample May 26, 2025 BNP Paribas BNP Paribas must design and distribute investment products tailored to each retail customer while handling very large volumes of information, complying with stringent regulation, and giving advisers and customers clear… several billion possible combinations algorithm · automated preparation of performance reports · internal conversational assistants In-house Sample Mar 5, 2025 BNP Paribas BNP Paribas needed to reduce stress, delays, and manual errors in mortgage processing, where advisors had to enter, review, and verify large document files and the bank needed an average of 11 days to issue a digital… 60% of data is automatically matched and validated by AI. Doc Factory · Intelligent Document Processing 