Acrodigital launches VLM-based AI-OCR platform 'Papers GO AI', targeting insurance companies
The AMW Read
Novelty is 1: VLM-based OCR is an existing trend; Acrodigital is a known but not top-tier player in document AI. Significance is 1: the launch is incremental for the sub-segment, with no cross-substrate capital or compute signal.
Acrodigital launches VLM-based AI-OCR platform 'Papers GO AI', targeting insurance companies
South Korean document AI firm Acrodigital has released 'Papers GO AI', a Vision Language Model (VLM)-based AI-OCR platform that converts unstructured documents—including handwriting, fax scans, receipts, and invoices—into structured digital data. Unlike conventional AI-OCR, which requires retraining for each new document type, Papers GO AI generates extraction templates automatically from sample uploads. The company reports a document classification accuracy of 99% and information extraction accuracy of 98%, and is currently being deployed by major domestic insurers for claims and underwriting workflows.
Why it matters: This launch exemplifies a recurring pattern in enterprise document intelligence: the shift from template-bound OCR to VLM-driven zero-shot extraction. Acrodigital, which has completed over 350 document digitization projects and claims a proprietary foundation model with 5 trillion parameters, is directly targeting the insurance vertical—a segment with high document heterogeneity and strict accuracy requirements. The product also reinforces the 'fastest-ARR-ramp' pattern within the AI-OCR substrate, as the elimination of manual template coding and model retraining lowers deployment friction from weeks to same-day integration, a key selling point for risk-averse enterprise buyers.
Grounding in the AI market: Acrodigital is a lesser-known player in the global document AI space, which is dominated by giants like ABBYY, Hyperscience, and emerging LLM-native tools from Microsoft (Azure AI Document Intelligence) and Google (Document AI). The company's claimed 99/98 accuracy benchmarks, while unverified by a third party, are competitive with the upper end of the market. However, the 5 trillion-parameter claim for a vertical OCR model is unusual—most foundation models at that scale are general-purpose LLMs. This may signal either an aggregated or loosely defined parameter count, or a deliberate marketing anchor to signal technical depth. The insurance focus is smart: it’s a niche with high willingness to pay due to compliance and error-cost implications. Still, the closed, on-premises or API-only deployment model typical of Korean enterprise SaaS may limit global scalability.
#AI-OCR #DocumentIntelligence #InsuranceTech #VLMModel #SouthKoreaAI #EnterpriseAI