Half the errors, half the correction rounds.

The trap of 100+ pages of documents

Bottleneck: An average mortgage case takes hundreds of data fields and a mass of documents. Checking by hand, branch advisors often could not control all of it, so incomplete packages reached credit assessment, causing four frustrating correction rounds per customer on average.

Market pressure: With the Otthon Start home-ownership programme ramping up and the strict digital transformation expectations of the central bank (MNB), the bank’s capacity reached its limit. Instead of hiring, radically automating the existing process was the only way to grow.

Goal: Cut checking time by 80%, reach machine accuracy above 99% and halve the correction rounds, so customers are happier.

A tailored AI architecture in the lending chain

Not boxed software: We built a bespoke AI solution that fits the bank’s core systems and analyses 11 critical document types instantly, from the land registry extract to the sales contract.

Visual and language intelligence: The AI does not only read the data, it understands how it connects (a RAG framework), flags errors at field level and generates a correction checklist for the advisor, who can always override it.

Fast to scale: Thanks to the iterative rollout, the first measurable results arrived after 14 days.

80% faster processing, business potential set free

A sub-20-minute automated check replaced the one-to-two-hour manual document audit in retail mortgage lending, and halved the correction rounds per case from four to two.

The gain frees 270 working hours a month for network advisors, who can spend them on customers and new sales instead of administration. Full field-level transparency and auditability have become a cornerstone of the bank’s risk management and compliance.