Numos is building an AI platform aimed at the enterprise CFO's office, with the explicit goal of being a finance AI that finance leaders can actually trust. Rather than forcing teams onto a new system of record, Numos layers on top of the existing finance stack, accounting systems, billing tools, data warehouses, and the inevitable web of spreadsheets, and continuously analyzes transactions flowing through them. From that vantage point its AI agents can automate the repetitive work that consumes finance teams while surfacing faster, more strategic insight.
The platform spans several core workflows. For strategic FP&A, it automates variance analysis and dynamic reporting. For month-end close, it streamlines bank reconciliations and produces audit-ready documentation. It also handles accounts payable, invoice routing and payment workflows, and accounts receivable, invoicing, collections, and revenue recognition, so teams can run quote-to-cash as one continuous process. The recurring theme is reducing manual data gathering so finance professionals can spend time on decisions rather than spreadsheet plumbing.
Numos was founded by Parijat Sarkar and Mitul Tiwari, who bring deep finance, data, and enterprise-AI backgrounds; Sarkar was previously a senior vice president at Zenefits, and Tiwari served as ServiceNow's head of natural language AI and director of engineering. That combination of finance-operations and applied-AI experience underpins the company's pitch to risk-averse enterprise buyers.
The company announced a $4.25 million seed round co-led by General Catalyst, with participation from Operator Collective. Customers cited in the launch see roughly 80% faster FP&A reporting cycles and close their books in less than half the usual time. By emphasizing trust, audit-readiness, learning a company's specific accounting policies, and non-disruptive integration, Numos is positioning itself for finance teams that want AI leverage without sacrificing control or accuracy. The bet is that enterprise finance will adopt AI fastest when it augments existing systems and produces explainable, defensible outputs.
