Ian Crosby’s Synthetic Raises $10M Seed to Build Agentic Bookkeeping AI

Ian Crosby, who previously founded the accounting software Bench and the payroll tool Teal, launched Synthetic in San Francisco last year. The startup’s five‑person team is developing an AI agent that connects to a company’s banking feeds, payroll systems, billing software and email inboxes.

When the agent encounters a transaction it cannot classify confidently, it poses a targeted question to a human bookkeeper for clarification. This approach is described as agentic bookkeeping, meaning the software can act autonomously on routine tasks while deferring uncertain cases to people.

The seed round totals $10 million, led by Khosla Ventures. Angel investors include Tobi Lütke, CEO of Shopify, and Basis Set Ventures.

Crosby told BetaKit, “I don’t know if this is possible yet,” underscoring his uncertainty about the feasibility of the technology at this stage. The product is currently in a “crash test phase,” during which the system still makes mistakes that require human correction. BetaKit notes that accounting users demand near‑100 percent accuracy to trust automated bookkeeping outputs.

For investors, the round reflects confidence in Crosby’s entrepreneurial track record despite the steep technical barriers inherent to agentic bookkeeping, which requires deterministic reconciliation, robust data connectors, and auditable trails to satisfy regulators. For engineers, the primary challenges involve minimizing model hallucination, ensuring transaction‑level verification, and building reliable human‑in‑the‑loop query generation without slowing down workflows. For prospective customers, the near‑term signals to monitor are enterprise pilot programs, published accuracy or reconciliation metrics on real‑world data, the release of connectors to major payroll and banking platforms, and evidence of SOC 2‑type controls or third‑party audits that demonstrate compliance and reliability.

Watch for those pilot results and accuracy data in the coming months to gauge whether Synthetic can move beyond the crash test phase and deliver a bookkeeping agent that meets the stringent accuracy demands of the accounting profession.