Rox Raises $100M Series B for Enterprise Revenue Agents

Ishan Mukherjee wants revenue agents to do work whose commercial payoff can take months to appear. At Rox, that puts a research assignment inside the sales pitch: improve the software while customers are still working toward the outcome that will tell them whether it helped.

The San Francisco company announced a $100M Series B at a $1.3B valuation on November 11, 2026. General Catalyst and March Capital co-led the financing, with Sequoia Capital, Factory, Databricks Ventures and Prosperity7 Ventures participating. Rox builds AI agents for enterprise revenue teams, connecting company data with prospect research, outreach, deal work and account expansion.

The financing will support AI research, custom models, evaluation infrastructure and enterprise expansion. For buyers, the consequential question sits inside that research agenda: how does an agent learn to handle a customer relationship before enough time has passed to measure the sale?

What the $100M Series B funds

Rox describes its target customers as the Global 2000 and plans to deepen its work across healthcare, industrial businesses, financial services, energy and technology. Selling into those organizations gives the company an assignment that extends beyond generating messages. The agent needs to use the right account information, understand the requested task and fit the systems the customer's team already operates. That places Rox within the enterprise-agent market covered in DevCuration's reporting on AI agent governance, where adopting software also means deciding what it may do.

The round follows $50M across Seed and Series A announced in November 2024. The new announcement identifies Sequoia Capital and Factory as existing investors, and Databricks Ventures and Prosperity7 Ventures as new participants. It reports the $1.3B valuation without specifying its pre-money or post-money basis, so the figure should be read as the announced valuation rather than a reconstructed capitalization table.

Rox reports more than 10x annual recurring revenue growth in its first year in market. Absolute ARR is undisclosed in the announcement, leaving the growth rate without a public revenue base. That limits comparisons with companies publishing revenue totals, even as the financing provides a concrete indication of investor commitment.

For a revenue leader evaluating the platform, the useful connection is between the proposed research spending and the work an agent will be trusted to perform. Better account research can support a seller's preparation; permission to update records or orchestrate outreach adds a different operating responsibility. The buyer needs to understand which actions require review and where responsibility stays with the sales team.

Sales feedback arrives on the customer's schedule

Rox's October research on its Outbound Reward Model describes a difficulty familiar to anyone studying commercial behavior: replies arrive late and reflect several influences at once. Timing, prospect selection, contact accuracy and the sender's reputation can all affect the result. A message's outcome therefore offers an imperfect explanation of the agent's own contribution.

The company describes using the changes sales development representatives make to drafts as a more immediate learning signal. An edit records a person's judgment while the message is being prepared, giving researchers something to examine before the recipient responds. Rox says it developed the model with hiddenweights and tested whether its scores related to subsequent replies.

Those are company-reported research findings, with the strengths and limits of that evidence. Their commercial relevance lies in the feedback mechanism: a seller's intervention can become useful development material instead of disappearing after the email is sent. The quality of that signal still depends on the people, accounts and situations represented in the underlying work.

This creates a product question that procurement can investigate directly. A buyer can ask how the system handles an unfamiliar audience, whether a human can correct its work and how that correction affects later behavior. A demonstration becomes more informative when it includes the customer's difficult cases, especially the ones that make an experienced seller stop before pressing send.

Rox's founders bring sales and data infrastructure together

Rox's current company team page identifies co-founder and CEO Ishan Mukherjee, co-founder and CTO Shriram Sridharan, co-founder and CGO Diogo Ribeiro, and research co-founder Christopher Ré. The mix puts commercial leadership, infrastructure engineering and research inside the founding group. Those responsibilities meet in the account record the agent needs to understand.

In its November 2024 investment account, Sequoia described Mukherjee's earlier work building Pixie and then leading growth functions at New Relic. The investor also described the team's experience across infrastructure and go-to-market systems. That history helps explain the company's focus on complex enterprise accounts without establishing that any particular customer will achieve the same result.

Sequoia's early description emphasized a warehouse-native system integrating with established enterprise platforms. Rox's current product materials likewise describe work inside existing customer systems. Read together, the sources suggest a practical route into an incumbent software estate: make more of the information already held there useful in the seller's day.

That route also places demands on implementation. A warehouse can contain several versions of an account's history, and the newest meeting may have less usable information than an older one. The organization supplying the data has a continuing role in deciding which records are current, who can access them and what the agent is permitted to do with them.

Engineering the follow-up before it reaches a customer

Rox's October 14 Living Evals engineering account describes a production failure in which the agent used an older call's transcript when the latest meeting lacked one. The company turned the case into a reproducible evaluation. For a revenue team, the error is easy to recognize: fluent writing can carry the wrong conversation into the next customer interaction.

The engineering process uses production traces in Braintrust and Devin to construct isolated test cases, with engineers reviewing the resulting pull requests. Rox describes checking data-changing actions through code and evaluating other responses against written criteria. The process supplies evidence of how the team investigates behavior; it is not an independent assurance that every deployment will behave correctly. It offers a company-specific view of the evaluation problem also examined in DevCuration's coverage of Plurai's agent simulation.

Rox's security materials also state that automated features operate within configured parameters and require ongoing oversight. That matters to the same buyer choosing where an agent belongs in the sales process. Review, access controls and escalation need to follow the work as it moves from gathering information toward acting on it.

The Series B gives Mukherjee's team resources to pursue that development across more industries and account histories. Each additional deployment brings another version of the problem the transcript example exposed: the customer's next conversation depends on the software understanding which previous conversation matters. The seller will still be there when the reply arrives.