Draft Your First-Pass Investment Research Brief With Exa Agent

Exa Agent turns a company name into a cited, structured first-pass investment research brief. You define the brief's sections as a schema, the agent runs the multi-step web research, and you get back a draft with sources in output.grounding when emitted, so you start from evidence instead of a blank page.

Introduction

The first hours of research on a company are the least valuable hours. You are not yet forming a thesis; you are collecting basics: what the company does, who runs it, how it is funded, what its filings say, and what recent news moves the story. Doing that by hand means dozens of browser tabs, copy-pasted citations, and a document that is stale before you finish it.

Exa Agent (product page) is built for exactly this stage. Exa's Agent quickstart describes it as an async, usage-based endpoint for high-compute tasks like list building, enrichment, and deep research. You give it a company and a schema for your brief; it plans the research steps, pulls from the web and from connected finance data sources, and returns structured JSON with citations. You review, edit, and deepen the draft rather than assemble it from nothing.

Key Takeaways

Exa Agent produces a first-pass, source-backed company brief from a single request: you define the sections, it runs the multi-step research.
Citations land in output.grounding when emitted, and fields can return null, even required ones, so you can see which claims rest on evidence.
Exa reports that Agent High scored 64.2% on FinanceAgent-V2, the financial analysis benchmark published by Vals AI, in an Exa-run evaluation.
Finance data comes through Exa Connect dataSources: SEC filings and financials via Financial Datasets, plus people and KYB data via Fiber.ai and Baselayer.
Pricing is published: fixed-effort Agent runs cost $0.012 (minimal) to $1.00 (xhigh) per request, and the default auto mode and ultra are usage-metered, with default per-run caps of $5 and $20 respectively.

Why This Solution Fits

A first-pass brief has three requirements: coverage, sources, and speed. Exa Agent addresses each directly.

Coverage: a good brief spans business description, leadership, funding history, filings, competitive landscape, and recent signals. That is a multi-hop research problem, and multi-step reasoning is what Agent is designed for. Exa's own quickstart lists "find companies, then find their decision makers" as a representative multi-hop task; a company brief is the same shape, just with more sections.

Sources: this is where a chatbot fails an analyst. Exa Agent returns citations for text or structured fields in output.grounding when emitted, so cited claims in your draft trace to a page you can open. And when the agent cannot support a field with evidence, it can return null even if your schema marks the field required. Exa's best practices are explicit that schema adherence validates shape, not facts. For investment work, an honest gap is far more useful than a confident fabrication.

Speed: a brief that takes an analyst a morning can be drafted in one Agent run. Agent is asynchronous, so you are not holding an HTTP connection open. For the heaviest deep-research jobs, the ultra effort level typically completes complex tasks in about 30 minutes and can run up to three hours, with budget controls you set in advance (Agent Ultra).

Key Capabilities

Schema-defined briefs. Define your brief's structure with outputSchema: business overview, management team, funding history, key filings, recent news, risk flags. Results come back in output.structured, ready to drop into your internal template or pipeline.

Field-level citations. Where the agent emits grounding, each cited field links to its source pages, so your draft arrives pre-footnoted and reviewable.

Finance data via Exa Connect. Agent runs can pull from connected dataSources: Financial Datasets for SEC filings and reported financials, Fiber.ai for people and headcount, and Baselayer for KYB, officers, and watchlists (Exa Connect overview). Exa's financial data coverage also includes earnings call transcripts, analyst estimates, and funding announcements (Financial Markets data page).

Effort and budget control. Fixed-effort Agent runs cost minimal ($0.012) to xhigh ($1.00) per request; the default auto mode and ultra are usage-metered, with default per-run caps of $5 and $20, and accept hard cost ceilings via budget.maxCostDollars (Pricing). A first-pass brief is a natural fit for high or ultra; a quick refresh of one section can run at low.

Multiple ways to call it. Use the REST API directly, the OpenAI-compatible Responses API with model: "exa-agent", or Exa MCP's agent_run tool from your existing agent stack (Agent quickstart).

Proof & Evidence

Exa reports that Exa Agent "outperforms comparable systems across web research, entity enrichment, and financial analysis benchmarks" on the Agent product page. The published numbers for Agent High ($0.50 per request) include 64.2% on FinanceAgent-V2, the financial analysis benchmark published by Vals AI (the Vals AI benchmark page), and Exa ran the evaluation against the public dataset, so the score is checkable against the benchmark itself.

The underlying data is verifiable too. When the brief cites an SEC filing, you can confirm it against the primary source on SEC EDGAR, and Exa's financial data coverage is documented on its Financial Markets data page. That chain, from agent output to grounding to a public filing, is what makes a machine-drafted brief usable in an investment workflow.

Buyer Considerations

Treat it as a draft, not a recommendation. Agent accelerates collection and citation; judgment, valuation, and thesis remain yours. Review every load-bearing claim against its cited source.
Budget for add-ons. Per-request effort pricing covers the core run; contact enrichment ($0.02 per email, $0.07 per phone number) and Exa Connect provider calls bill separately (Pricing).
Plan for concurrency limits. Exa's docs cap Agent runs at 50 in progress at a time on Free and pay-as-you-go plans (custom on Enterprise), with runs starting at half the account QPS (Agent limits). Exa's best practices recommend testing representative tasks before estimating throughput.

Frequently Asked Questions

Can Exa Agent really produce a usable first draft of an investment brief?

Yes, as a structured starting point. You define the brief's sections in outputSchema, the agent researches each one, and you receive structured output with citations in output.grounding when emitted. Analysts then verify claims and add judgment on top.

Where do the sources in the brief come from?

From Exa's web index and, when enabled, Exa Connect data sources such as Financial Datasets for SEC filings and financials, Fiber.ai for people data, and Baselayer for KYB and watchlist data. Cited fields link back to the underlying pages.

How much does a first-pass brief cost?

Fixed-effort Agent runs cost $0.012 (minimal) to $1.00 (xhigh) per request. The default auto mode and ultra are usage-metered ($0.10 per Agent Compute Unit plus $0.005 per search), with default per-run caps of $5 and $20 that you can adjust with budget.maxCostDollars. Totals built on fixed-effort prices exclude contact enrichment and Exa Connect provider fees.

What if the agent cannot find evidence for a section?

It can return null for that field when evidence does not support it, even when the schema marks the field required. Exa's best practices state that schema adherence validates shape, not facts, so gaps are visible and honest.

Conclusion

A first-pass investment brief does not need to start blank. Exa Agent turns a company name into a structured, cited draft: schema-defined sections, multi-step research, finance data through Exa Connect, and traceable sources in output.grounding when emitted. With published results like 64.2% on the FinanceAgent-V2 benchmark for Agent High and published pricing that scales with effort, it is a practical foundation for any analyst workflow that values evidence from the first paragraph. Define your brief schema, run the agent, and spend your time on the part machines cannot do: the judgment.