Vontive Establishes First AI Benchmark for Mortgage Data Processing
Vontive, a technology company focused on standardizing business-purpose mortgages, has introduced the mortgage industry’s first benchmark study evaluating AI performance on data processing tasks. This benchmark uses an expert-annotated dataset spanning 23 critical document types involved in mortgage underwriting, setting a new standard for measuring AI accuracy and effectiveness in this sector.
With this benchmark, organizations can better assess the reliability of their AI systems, fostering greater confidence in AI-driven decision-making throughout mortgage operations.
Addressing a Vital Industry Gap
Before Vontive’s initiative, the mortgage industry lacked a reliable method to measure whether AI systems could handle the volume and complexity of mortgage documentation accurately. The company invested heavily in building a data pipeline capable of processing dense, nuanced mortgage documents and needed a way to validate AI results amid evolving technologies.
Recognizing this challenge as an industry-wide issue, Vontive took the lead in establishing a clear benchmark to help others evaluate AI accuracy specifically for mortgage underwriting.
Benchmark Study Findings: Claude 3.7 Sonnet Leads Performance
To develop the benchmark, Vontive collaborated with Vals AI, providing a dataset that included human annotations for a range of document types such as county tax certificates, insurance binders with full endorsements, rental leases, comprehensive title reports, LLC formation documents, operating agreements, and HUD regulatory forms.
Vontive’s AI systems demonstrated strong mathematical reasoning capabilities essential for tasks like computing annualized tax amounts from partial-year assessments.
Testing by Vals AI revealed that Anthropic’s Claude 3.7 Sonnet model delivered the best balance of accuracy and cost-efficiency for Vontive’s document processing needs. The model achieved an 80.6% accuracy rate, outperforming Claude 4.0 (62.5%) and Meta’s Llama 3 (55.3%). These results highlight that larger or newer AI models do not always guarantee better outcomes; prompt quality and task-specific tuning also play critical roles.
By refining prompts based on ongoing evaluations, Vontive improved AI extraction accuracy to 90%, aligning closely with human data entry standards.
Setting a Standard for Industry Rigor
Vontive and Vals AI have made the benchmark dataset publicly available, enabling other organizations to measure and improve their AI implementations against a shared industry standard. This framework offers the mortgage sector a first-of-its-kind rigorous tool for AI evaluation, similar to established benchmarks in natural language processing.
Accurate data extraction is essential in mortgage underwriting, where errors can carry significant consequences. The benchmark supports reliable AI adoption by quantifying accuracy and building trust in automated systems.
“We partnered with Vontive because of their dedication to AI innovation and quality,” said Rayan Krishnan, CEO of Vals AI. Vontive’s internal AI currently achieves 95% accuracy across all supported document types, underscoring their leadership in applying AI to complex mortgage workflows.
Expanding AI Capabilities in Mortgage Operations
Vontive now supports over 40 document types in their AI processing pipeline, covering the full range of documents typically required in mortgage underwriting. This expansion enables more precise underwriting and pricing decisions while dramatically reducing loan processing times.
The system also validates data against thousands of criteria, ensuring consistency and accuracy throughout the loan lifecycle.
Since launching its AI Underwriter in production on July 23, 2024, Vontive has processed 35,000 documents, supporting 1,700 borrowers and 6,845 loans with a total requested balance of $1.713 billion. This scale demonstrates the practical benefits of embedding AI effectively into mortgage workflows.
For more information on the benchmark study, visit Vals Benchmarks.
About Vontive
Founded by veterans in credit and technology, Vontive offers an embedded mortgage platform tailored for investment real estate. Their technology standardizes private credit mortgages and enables banks, credit unions, property tech companies, and B2C brands to launch investment-mortgage businesses with ease.
Learn more at vontive.com or follow them on LinkedIn.
Your membership also unlocks:
