Demonstrating the potential of multi-agent AI to transform a critical lending control 

Industry

Technology

Introduction

UTB asked NashTech to deliver a Proof of Concept (PoC) to test whether a multi-agent AI system could automate this analysis in a reliable and auditable way. 

Overview

United Trust Bank (UTB) is a UK specialist bank operating across property development finance, bridging and specialist mortgages. Within its Property Development division, Quantity Surveyor (QS) reports are essential for assessing construction progress and determining whether funding drawdowns should be approved. These reports contain complex financial and project data that must be reviewed accurately and consistently to maintain strong lending governance. 

Business challenge

UTB’s QS team managed a high-volume, manual workflow where each report required careful extraction, validation and reconciliation of key financial and compliance elements. A typical review could take 30 to 60 minutes. This created several operational and risk challenges: 

• High processing effort and variable turnaround times 
• Inconsistencies due to diverse QS report formats 
• Risk that critical issues could be overlooked in manual review 
• Growing pressure to demonstrate credible, safe adoption of agentic AI 

UTB asked NashTech to deliver a Proof of Concept (PoC) to test whether a multi-agent AI system could automate this analysis in a reliable and auditable way. 

The solution

NashTech delivered a five-week agentic AI PoC. The core build was completed within four weeks, with the fifth week dedicated to addressing UTB’s queries and incorporating feedback. 

The PoC automated the full lifecycle of QS report analysis, from ingestion through extraction, validation, exception analysis and summary generation, using a coordinated multi-agent AI architecture. 

How it worked 

Automated ingestion:  

QS PDFs uploaded to SharePoint automatically triggered the workflow via n8n. 

Intelligent extraction: 

Gemini AI was used for high-accuracy PDF understanding after traditional extraction libraries proved inconsistent for UTB’s varied formats. 

AI-driven validation and reconciliation: 

Azure OpenAI agents applied UTB’s rules, performed reasoning and validated extracted fields against mocked financial data to simulate internal checks. 

Exception and risk detection: 

The system highlighted anomalies, discrepancies and risk indicators that influence drawdown assessments, often identifying items that manual reviews can miss. 

Automated report generation: 

A structured summary report was generated and stored in SharePoint, providing a consistent output for QS review. 

Technology stack used 

• Gemini AI – PDF intelligence 
• Azure OpenAI (GPT-4) – reasoning, validation, summarisation 
• n8n – workflow orchestration 
• SharePoint Online – input/output storage 
• MongoDB (mocked) – simulated reconciliation data 
• Azure container – runtime for the multi-agent workflow 

Results and achievements

Significant reduction in processing time 

QS report analysis reduced from 30–60 minutes to 5–6 minutes, achieving an 80–90 per cent improvement. This demonstrated the transformative potential of agentic AI to accelerate a key stage in UTB’s drawdown process. 

Improved accuracy and stronger risk control 

The PoC consistently identified discrepancies and exceptions that can be overlooked during manual checks. This shows how automation can support a stronger control environment by providing consistent, rule-based validation. 

Standardised and auditable outputs 

The PoC produced a uniform summary for every QS report, ensuring consistency across assessments and enabling a clearer audit trail for future decision-making. 

High stakeholder confidence 

Weekly demos and fast incorporation of UTB’s feedback built strong confidence in both the solution and NashTech’s delivery capability. The PoC also gave UTB a tangible example of how agentic AI can be applied responsibly in a regulated banking context, a key expectation from leadership and investors. 

Validated feasibility for future adoption 

The PoC confirmed both the technical feasibility and the operational value of scaling this solution further. UTB has requested a follow-on proposal to explore a full production implementation. 

Leanne Sweeney, Head of Business Transformation at UTB said,

“The proof of concept demonstrated just how transformative agentic AI can be. By engaging with NashTech, we were able to take requirements directly from our business experts, rapidly evaluate AI tools and language models and see tangible early results. At scale, we expect this capability to reduce a process that currently takes an hour to just a few minutes, freeing our experienced team to focus on exceptions and higher value work rather than routine data evaluation.”

Why NashTech 

NashTech brought together AI engineering, workflow automation and enterprise software expertise to deliver a sophisticated multi-agent PoC in just five weeks. Regular engagement and transparent progress updates ensured alignment with UTB’s expectations. 

The approach also demonstrated potential for future reuse or adaptation across similar lending workflows at other financial institutions. 

James Loveridge, Client Director at NashTech concluded; 

“There’s a lot of excitement, but also a lot of understandable scepticism around Agentic AI, especially in regulated environments like banking. This Proof of Concept was about removing that. In a controlled setting, we showed how agentic AI can be applied in a practical, transparent way to deliver genuine business value, without introducing unnecessary risk. By focusing on a real QS process and measurable outcomes, we helped UTB move from theory to genuine confidence, proving that agentic AI can support efficiency and control when it’s designed and governed properly”.