RavenPack
RavenPack is one of the oldest language analytics businesses in finance, founded in 2003 and headquartered in Malaga, Spain with a New York presence, and it now trades under two live names. The company name carries the original business, which reads unstructured financial text at scale and turns it into structured signal: entity resolved sentiment and event detection across more than 40,000 news sources, filings and earnings transcripts, negative news monitoring across more than 7 million companies, executive sentiment drawn from earnings calls, and packaged quantitative factors sold to hedge funds, banks and asset managers. The platform name is Bigdata.com, launched on 22 October 2024 and presented as Bigdata.com by RavenPack, which is the current delivery surface and the one a new buyer meets first. Both names belong to the same business.
Everything the company sells rests on that one capability, so there is no second business to set apart. What did change is the architecture. Bigdata.com is a retrieval augmented generation platform combining vector search, run on Vespa Cloud at billion document scale, with the entity knowledge graph the company spent two decades building, exposed through an API, a research assistant, desktop and mobile applications and autonomous research agents that run tasks and produce daily pre market notes. That is a re platforming of delivery on top of a continuous data and entity resolution asset, rather than a legacy vendor bolting on artificial intelligence.
The content estate is the other half of the business and is unusually well disclosed. More than 170 content providers are licensed, with the Financial Times, the Economist Intelligence Unit and Preqin named individually, alongside a podcast corpus of roughly 20,000 shows and 5 million episodes, fund holdings, jobs data, environmental and governance scores, corporate fundamentals and regulatory filings. In July 2026 the company launched what it calls the tokenization of content, a marketplace in which agents retrieve and pay for licensed premium content per token rather than per document, priced publicly.
Funding includes a $20 million round led by GP Bullhound with participation from the European Investment Bank Group in 2024, and a subsequent investment from FT Ventures alongside the content agreement. Vendor material states more than 100 global financial institutions use the platform, without naming them.
Capability Axes
Capability grades
15 of 15 axes rated · 8 graded A or B
Every output comes from language processing, from event detection across more than 40,000 news sources to a retrieval layer that answers questions over billions of documents. What carried over from two decades of operation is the entity resolution and knowledge graph that makes retrieval land on the right company.
Inline citations tie every claim to its source, which makes retrieval checkable, but confidence isn't shown and nothing says what happens when no supporting passage turns up. Autonomous agents that write daily pre market notes have no stated review before their output reaches an analyst.
The architecture is described openly, pairing vector search with an entity knowledge graph, and context per query is said to shrink by up to 100 times. No retrieval precision, hallucination rate or independent evaluation backs any of it.
The organizations named are content suppliers and infrastructure partners rather than customers with results. More than 100 financial institutions reportedly use the platform, none named, and the tenfold research efficiency claim has no baseline.
A portfolio manager's query reveals live interest in a name, competing funds use the same platform, and per token billing logs every retrieval. Whether query text is kept, used for ranking or separated between clients isn't stated.
Companies and markets are the subjects, not consumers, so European data protection law and the content licenses govern rather than US financial privacy rules. Executive sentiment work does involve named individuals, though the platform offers no data processing terms or retention schedule.
A trust center sits on the company's own subdomain, and a service organization controls examination and the ISO information security standard appear as feature bullets on the pricing page, but neither the report type nor the audit period is stated.
Selling analytics and licensed content isn't itself regulated, yet a Spanish provider falls under European AI rules, and news signals delivered to trading desks sit close to market abuse rules. None of that is addressed.
Coverage based signals favor heavily covered companies, so a clean negative news screen on a thinly covered firm may mean the coverage was missing, not the conduct. No density or detection rate by market, language or company size is published.
Output feeds investment memos and trading signals, where a misstated fact costs money before anyone notices, and no warranty, accuracy commitment or remedy is published. Citations let a customer check each claim, which helps but isn't recourse.
More than 170 content providers are licensed, the Financial Times and Preqin among them, and a public catalog shows what sits behind each answer. The language models doing the generating, and where they run, aren't named.
Public API documentation, self serve keys and distribution through a major cloud data platform make the data easy to build on. No portfolio, risk or order management system is named, which matters for signal customers working inside quant pipelines.
Data residency appears only as a two word feature on the enterprise tier, with no region list and no word on where query text and retrieval logs are stored.
Usage is billed per token with a public rate card by content class, where regulatory filings cost about a fortieth of premium news, and seven worked examples run from about $0.57 to $1.82 each. The enterprise tier and the older analytics products, where institutional buyers usually land, are quoted rather than priced.
Hedge funds, banks and asset managers are the documented buyers, plus developers building agents, so for all its breadth of content the platform serves one corner of finance.
Product updates
Material product, regulatory, evidence and commercial changes at RavenPack, each verified against a live source and tagged to the capability axis it bears on. Funding rounds and awards are not product changes and are not logged.
RavenPack made an official Bigdata.com connector, built on MCP, available inside Microsoft 365 Copilot Chat and Copilot Cowork. It brings company, market and macro data into the conversation, covering news, filings, transcripts, tearsheets, media sentiment and a customer's own uploaded content, with a source attached to each answer.
The Bigdata.com MCP server added two tools: find_entities, which resolves organizations, people, places and products to entity IDs by name, filters or both, and list_entity_filter_values, which lists the accepted values for a filter such as a person's employer or a place's region. Companies, ETFs, funds and other securities remain under find_securities.
RavenPack launched what it calls the tokenization of content on Bigdata.com, a marketplace in which AI agents retrieve, license and pay for premium content per token rather than per document, drawing on more than 170 licensed providers spanning market data, newswires, research, transcription, expert networks and global media. The vendor states that precision retrieval sends a model only the excerpts that carry the answer, reducing context consumed per query by up to 100 times, and that existing provider agreements are honored so a firm bringing its own content licenses pays only for retrieval. Access runs through one connection via MCP or API with entity resolution from the company's knowledge graph built in.
Compared With
Most editorial comparisons pair two vendors the index assesses as direct competitors for the same buyer. Some pair vendors that are adjacent rather than rival, where the useful question is where one ends and the other begins. Each carries a verdict, the buyer conditions that favor each vendor, and a graded side by side.
Alternatives to RavenPack
Vendors in the same categories as RavenPack whose profiles look most alike, each with a line on what the product does. Similar is not the same: check that each one does the job you need. No vendor pays for placement.
FinTech Studios runs an AI intelligence engine that turns news, filings, regulatory updates and market data into cited research for financial professionals.
Termina sells quantitative due diligence on demand to venture capital, private equity, corporate development and M&A teams, and to founders.
Terminal X, built by Project Pluto Inc, is an agentic research platform for institutional investors, founded in New York in August 2022 by chief executive Hyun Hong, who spent a decade in…
Onthos, formerly Theia Insights, sells AI that classifies companies by what they actually do.
Accelex automates the extraction of investment data from private markets fund documents for asset owners, allocators, institutional investors and the asset servicers acting for them.
Binocs produces the analysis a deal process needs, using agents organized like an investment team.
The index publishes no overall score or ranking. See what we assess and the methodology.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | Data Protection Terms | Implementation | Source |
|---|---|---|---|---|
|
Free tier with $25 of credits, no card required; thereafter usage based from fractions of a cent per token, with published worked examples at roughly $0.57 to $1.82 per research run
$0 baseline
|
Usage based per token of content consumed, with an optional flat subscription per content set and a quote based enterprise tier. | Not published. Data residency, single sign on and audit controls are named as enterprise tier features without terms. | None stated for self serve. Enterprise onboarding is described as including a named team and an agreed service level, without figures. | Vendor Published |
On the platform side, pricing is unusually open. Charge is per token of content consumed, published as starting from fractions of a cent, with no seats and no subscription required to begin. A public rate card breaks cost by content class, and the spread is itself informative: regulatory filings sit roughly forty times below premium news, with earnings transcripts, corporate fundamentals, environmental and governance scores, jobs data and fund holdings in between. Seven named worked examples are published with per run costs from about $0.57 to $1.82, each labeled with the artifact produced and its approximate output size, and an interactive calculator lets a buyer model monthly cost against expected run volume before speaking to anyone.
An alternative model is published rather than hidden: a heavily consumed content set can be switched to a flat subscription, which removes its per token content charge and leaves only the lower search and retrieval tokens. Committed use and volume discounts, custom dataset licensing and tailored deployment are named as available without figures.
Two gaps are worth a buyer's attention. The enterprise path, which is where an institutional buyer of the legacy analytics and factor products will land, is quote only, and those legacy products carry no published price anywhere. And the token unit means cost scales with how much content an agent retrieves rather than with team size, so a poorly scoped agent that retrieves broadly is the main route to an unexpected bill. The published calculator mitigates that considerably.