What Nanonets changed

• Shifted core messaging from "data extraction" to "AI agents that run business processes."
• Introduced "Trail," a context graph that connects policies, ERP data, and tribal knowledge.
• Added a new "Self-learning Loops" feature where agents learn from human corrections.
• Launched an agent platform connector to integrate with Claude, ChatGPT, and Salesforce Agentforce.
• Added specific use-case modules for Reconciliation, Order Management, and Claims Processing.
• Updated the offer to include $50 in free credits with no credit card required.

Nanonets is moving beyond simple data extraction to full process automation. They now position themselves as a "system of action" that can complete entire workflows, like claims processing and supplier onboarding, using a new "context graph" called Trail to handle complex business rules.

• Replaced the specific "Invoice processing" section with a broader "Turn documents into usable data" section.
• Added a prominent badge for being "#1 on the IDP Leaderboard."
• Expanded document examples to include POs and bills of lading instead of just invoices.
• Updated the call-to-action link from "accounts payable" to "data extraction."
• Added a clear "Start free. No credit card" offer at the bottom of the page.

The product is moving away from being just an "accounts payable" tool. By highlighting their #1 rank on the IDP Leaderboard, they are positioning themselves as a top-tier AI engine for any document type, not just finance.

• New headline emphasizes that edge cases ruin ROI and cause AI deployments to fail.
• Added a specific example of a complex problem involving five different systems.
• Updated the automation claim to a specific "94% of complex processes" instead of general "millions of pages."
• Added "GL coding" and "2-way matching" to the invoice processing feature list.
• Removed the "Agent Builder" section and detailed security badge descriptions (SOC 2, GDPR, etc.) from the main flow.
• Added "ComplexConstraints" to the list of leaderboards where their model ranks #1.

The product is moving away from general automation talk to focus on why AI projects usually fail: the hard "edge cases" that humans usually have to fix.