Variance

Variance, based in San Francisco, builds AI investigative agents that autonomously run risk and compliance work, replacing the manual investigations that normally take teams of analysts days to complete. Founded in 2023 by chief executive Karine Mellata and Michael Lin, who worked together on fraud engineering and machine learning at Apple and Discord, the company began as the trust and safety tooling startup Intrinsic before focusing on financial crime, and raised a twenty one and a half million dollar Series A led by Ten Eleven Ventures, bringing total funding to twenty six million dollars. Its premise is that the industry is shifting from scoring risk to investigating it, because a classifier returns a probability while a real decision requires evidence.

Rather than producing a risk score, Variance's agents conduct full investigations across know your customer, know your business, anti money laundering, transaction monitoring, and fraud. When a review is triggered, an agent pulls entity registration documents, traces ultimate beneficial ownership across jurisdictions and corporate layers, checks sanctions lists and adverse media, and returns a cited, auditable decision within minutes, all following the customer's own compliance playbook, while a second agent monitors data continuously to surface suspicious relationships before they trip predefined rules. Three pieces make this work: a proprietary context engine that maps each customer's entire universe of entities, events, and relationships into one ontology for multi hop reasoning, a standard operating procedure enforcement layer that encodes compliance steps in plain English so agents execute them consistently with a full audit trail, and a marketplace of plug and play investigative tools spanning corporate registries across more than fifty jurisdictions, sanctions, court records, adverse media, and device signals, drawing on over one hundred fifty data sources. The platform processes more than seventy million context signals a day and executes hundreds of thousands of automated enforcement actions.

Variance is built for the strict realities of banks, fintechs, marketplaces, and Fortune 500 companies, and it stands out on deployment and governance: it supports fully on premise and air gapped installations with customer managed encryption keys, field level controls, and confidential computing, so sensitive data never leaves the institution and is inaccessible even to Variance. It uses its own reasoning approach rather than a choice of model and acts through data rather than driving a browser. For a compliance or fraud team that wants agents to run complete, defensible investigations at machine speed under its own procedures, including in sovereign or air gapped environments, Variance is a strong and well aligned fit; a team wanting a lightweight risk score, a general purpose agent builder, or customer model choice will find it purpose built for investigative rigor.