# Wisedocs says insurance claims still break AI _Wisedocs' Denys Linkov told AI Engineer that fragmented, fax-heavy claim files remain AI's hardest data problem, even as its MLCR benchmark jumps to 75%._ **Published:** 2026-10-18 **Source:** https://www.startuphub.ai/ai-news/ai-research/2026/wisedocs-says-insurance-claims-still-break-ai --- Denys Linkov, head of AI program at Wisedocs, told [AI Engineer](https://www.youtube.com/watch?v=W4pnzOURwHQ) that insurance claims remain AI's messiest data problem because the files behind them are fragmented, duplicated and often still faxed. Linkov said Wisedocs helps companies process complex medical claims by taking unstructured claim files, medical records, legal forms and insurance documents and turning them into structured, reviewable data. He said duplicate records can run to 50% or higher, a single PDF claim can hit 8 gigabytes, and reconstructing a 30-year worker history means stitching name changes from marriage, city moves and provider switches. The host pressed on timelines and Linkov said response windows can be 7 or 21 days depending on the state, while adjudicating a complex claim can stretch from weeks to years, and the initial report is rarely sufficient. That last point shapes how the integration actually runs. Linkov said Wisedocs has shifted hard since 2024 and built MLCR, Medical Language Comprehension and Reasoning, with Artificial Analysis to test whether models can answer medical claim questions across reasoning tiers. The company released an open dataset for the first three tiers and keeps a private set with Artificial Analysis for higher reasoning levels, he said. When the eval launched the best high-reasoning accuracy was 20% according to Linkov, and a few months later it reached 75%. Linkov framed the plumbing underneath as enterprise drag. He said AI hit in 2023 and 2024 but most carriers had not finished cloud migrations, big data adoption or clean data lakes, and he put the global fax market at over $7 billion. He cited EHR adoption at 8% in 2008 rising to the 80s and 90s during COVID, and said workers compensation alone spans 50 different state forms. It is a useful contrast to the document AI market outside insurance, where Microsoft, [Google](https://www.startuphub.ai/ai-news/ai-research/2026/deepmind-s-scale-how-agents-run-at-google) and AWS sell cloud classifiers and APIs like Reducto and LlamaParse serve engineers, while Tennr, DataCare IDA and Wisedocs are built specifically for healthcare intake. Linkov was candid about failure modes and quiet on fixes. He argued that in litigated cases information is left out on purpose and that organizations without digital transformation cannot process cases fast enough to catch it, but he did not detail how Wisedocs validates completeness or resolves conflicts when records contradict. He also did not say how the 75% MLCR score translates to production accuracy on live, private claims where adjudication standards vary by line. The question he left open is the core one. If duplicate rates stay near half and the ground truth lives across faxes, state forms and decades of provider changes, what does it take to prove a model understood the medical story, not just the benchmark. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.