Inside Rogo's Series B: how a $50M raise wired a finance-native AI agent into 25+ investment firms

In June 2024, Rogo, a two-year-old startup building an AI analyst for investment workflows, announced a $50M Series B led by Thrive Capital, with Tiger Global, Khosla Ventures, and AI Grant participating. Alongside the round, the company disclosed adoption at more than 25 financial institutions, including Tiger Global, Two Sigma, Lazard, and Point72. The raise landed at a moment when the buy-side and sell-side were still triaging generic large language models against actual research workflows, and it stands as one of the first venture-scale validations of a vertical finance agent rather than a horizontal chat product.

Background: the finance-specific gap in generic LLMs

By 2024, most investment firms had run at least one internal pilot with a general-purpose LLM. The pattern was consistent across seat types. An analyst would paste a 10-K into a chat window, ask for a segment breakdown, and get back prose that read fluent but missed the finance-specific reasoning that a first-year associate does without thinking: reconciling segment reporting across restatements, tying management commentary to the numbers, or flagging that a non-GAAP margin excludes a line item the company just added.

Three constraints kept horizontal chat products from graduating out of those pilots. First, weak finance-specific reasoning: a model trained on the open internet has read a lot of Wikipedia summaries of DCF, but relatively little of the actual work product of an investment analyst. Second, no citation grounding: an answer without a page-anchored source is not usable in a memo that has to survive an investment committee. Third, no clean path to internal data: the interesting corpus at a hedge fund is not the public filings, it is the internal notes, expert-call transcripts, and prior deal memos that live behind the firm's own walls.

That gap defined the opening. A product that could reason over financial documents in the way an analyst does, cite what it used, and connect to internal systems had a defensible wedge that a horizontal assistant did not.

Approach: vertical training, named-customer land-and-expand

Rogo was founded in 2022 by Gabriel Stengel and John Willett and positioned from the start as a finance-native AI analyst rather than a general chat product. The Series B was announced by the company on June 18, 2024, with $50M led by Thrive Capital and participation from Tiger Global, Khosla Ventures, and AI Grant. TechCrunch's coverage of the raise framed it as a bet on a finance-specialized agent at a moment when horizontal AI startups were raising on much broader theses.

The approach had three visible components.

Domain training, not just prompt engineering

Rogo disclosed that it was hiring ex-bankers and ex-investors as domain trainers, and doing technical work to fine-tune models on financial reasoning tasks. The distinction matters. A prompt library sits on top of a generic model and coaches it into finance register; a fine-tuned system pushes the finance reasoning into the weights. The Business Wire release announcing the round described the product as an AI financial analyst trained on the workflows of the job, positioning that read as a direct rebuttal to the horizontal-copilot framing.

Named-customer land-and-expand

Rather than pursuing broad self-serve distribution, Rogo disclosed adoption at 25+ financial institutions at the time of the raise, and named Tiger Global, Two Sigma, Lazard, and Point72 among them. Named-logo disclosure at Series B is a deliberate choice: it trades some competitive quiet for peer-effect distribution inside a small buyer universe. When a fundamental hedge fund analyst sees Two Sigma and Point72 on the list, and an M&A associate sees Lazard, the reference class is doing part of the sales work.

The use cases Rogo disclosed mapped to the actual seat-level workflows those firms run: company research, market analysis, memo drafting, and meeting preparation. Each is a recognizable analyst task with a clear input surface (filings, transcripts, internal documents) and a clear output surface (a memo, a briefing note, a comp table). Building an agent that produces those outputs is a narrower problem than building a general research assistant, which is part of why a vertical agent could ship into production faster than a horizontal one.

An open-standards data layer

The research agent problem is ultimately a data problem: an agent is only as good as the sources it can reach. In 2025, Rogo disclosed an MCP-based integration with Daloopa, plugging structured financial data into the agent through the Model Context Protocol. Using MCP rather than a bespoke connector matters because it signals a bet on open standards for agent-to-data plumbing, and because it lets the same integration pattern extend to other structured and unstructured sources over time.

What happened: production usage and downstream integrations

The Series B is the natural checkpoint for this case, but the interesting part is what the disclosed customer roster implies about the state of adoption. Twenty-five-plus financial institutions is not a pilot count, it is a book of business that includes multiple seat archetypes: quantitative hedge funds (Two Sigma), multi-manager platforms (Point72), a crossover growth fund (Tiger Global), and an investment bank's advisory practice (Lazard). Each of those seat types values different things. A quant shop cares about programmatic access and audit trails; a multi-manager cares about pod-level customization and information barriers; a growth investor cares about private-company research and market maps; an advisory banker cares about pitchbook prep and comps.

Rogo shipping into all four archetypes at once suggests the product surface was general enough to cover them and specific enough to earn a paid seat in each, a combination that has historically been hard for horizontal copilots to achieve inside financial institutions.

Public follow-on milestones through 2025 ground the trajectory:

Each downstream milestone reinforces a piece of the original thesis. The Jefferies rollout extends the sell-side footprint that Lazard signaled at the Series B. The OpenAI partnership sits alongside the Claude-and-E2B stack disclosed later, indicating a multi-model posture rather than dependence on a single foundation model provider. The Daloopa MCP integration validates the data-layer bet. And the 400-job New York expansion, supported by New York State tax credits announced by the governor's office, is the operational tell that the customer roster has moved from disclosed logos to a revenue base that can support that kind of headcount.

What it signals for the industry

Three readings sit on top of the sourced facts.

First, the vertical-agent versus horizontal-copilot split is now a live commercial question, not a theoretical one. A finance-native product reached 25+ financial institutions at the Series B stage, in a buyer universe that is famously slow to adopt outside vendors, and did so while horizontal AI assistants were still building out finance capabilities as one of many verticals. That does not settle the debate, but it moves the burden of proof.

Second, open standards for agent-to-data plumbing are starting to matter. The Daloopa MCP integration is a small technical disclosure with a large architectural implication: if the interesting data at a financial institution ends up exposed through MCP servers rather than through vendor-specific APIs, the composition layer (the agent) and the source layer (the data provider) can evolve independently. That is a healthier shape for the ecosystem than the alternative, in which each agent has its own bespoke connectors.

Third, named-logo disclosure at the Series B stage is a distribution strategy in a small buyer universe. It works when the reference class is credible and the seat-level workflows are recognizable, and it is difficult to replicate without a genuine production footprint.