RavenPack: EDGE analytics, financial factors and Bigdata.com research tools

RavenPack combines analytics derived from news and other textual information with research tools for financial institutions. Its catalogue brings together EDGE datasets, precomputed Factors and Bigdata.com. A broker or investment team can use these at different points in a workflow: structured event records for models and monitoring, aggregated indicators for dashboards, or retrieved evidence for analyst and AI research. These are related products in one provider family, rather than interchangeable licences.[1]

News Analytics attaches entity and event tags to stories and provides sentiment, relevance, novelty and temporal information. These fields answer different questions: which organisation a story concerns, what happened, whether it adds new information and when an event occurs. Historical and real-time material includes premium news and web sources. A sentiment value is therefore only one part of the record; entity resolution, event relevance and duplicate or repeated coverage also affect how a research model interprets the information.[3]

RavenPack’s technology description positions EDGE as the current generation of its natural-language-processing infrastructure, producing both historical and live analytics. It describes sentence-level sentiment derived from models trained on its curated archive and an expanded event taxonomy. This is a data-production process; the resulting analytics can feed a customer’s models without requiring that customer to use Bigdata.com’s conversational research interface.[6]

Factors are available through CSV files, Snowflake and Python-based queries. The library includes Company News, Earnings Intelligence, Controversy News and Credit News, plus macro indicators such as C-Level Economic Sentiment and Forex Sentiment. Aggregation frequency and field content vary by family: the catalogue, for example, describes multiple daily Company News snapshots alongside daily macro factors. A factor feed should be mapped to its own production schedule and definition rather than assuming one common update frequency.[4]

Bigdata.com’s retrieval architecture ingests web, premium and private material, identifies entities and events, resolves them to a knowledge graph, and indexes passages for semantic retrieval. Company relationships include subsidiaries and locations, allowing a query to follow a corporate relationship as well as match text. The returned passages carry source references. That provides a basis for checking the evidence used in a research answer, while the consuming application determines how those references are presented to its users.[7]

The Search quickstart documents REST requests authenticated with an X-API-KEY header and a Python SDK route. Search filters can constrain time, entity and sentiment; the maximum number of returned chunks limits the material retrieved. These controls let an implementation separate its investment universe and observation period from a broad natural-language query. The guide demonstrates the interface rather than promising that every query returns a complete market record.[9]

With MCP, Bigdata retrieves authorised context and passes it to the chosen AI host. The MCP terms distinguish that delivery from the host’s generation of the final answer: RavenPack does not control the host model’s reasoning or settings. Connecting a model therefore creates both a data-access relationship and a separate model-provider relationship.[15]

Bigdata.com’s public page offers Starter, Pro and Max plans and an enterprise enquiry route. It displays a 20% yearly option alongside regular and reduced monthly figures. The figures below preserve that public presentation; the selected billing cycle determines the actual upfront charge.[11]

The pricing documentation describes token consumption for retrieved content, model use and stored private content. Enterprise pricing can use an annual committed spend drawn down through usage, with negotiated volume discounts, contracting, monitoring and invoicing. Its enterprise package also lists service-level commitments and a dedicated account manager. These commercial arrangements differ from assuming a fixed price for every search or a single licence covering all datasets.[12]

The usage guide identifies the actual rates for an API key through the subscription-quota endpoint: each billing unit has a per-token price in US cents. Cost is the sum of the consumed tokens multiplied by the corresponding rates. Account balances are also reported in cents. This permits a customer to reconcile a call against its own rate card; the documentation’s example balances and token counts are illustrations, not universal prices.[13]

Prepaid pay-as-you-go credits expire after 12 months. Subscription allowances roll over during the subscription unless the plan says otherwise; exhaustion blocks further use until refresh or top-up. Subscriptions renew automatically. Cancellation stops future renewal and normally does not refund the current period. Fees exclude taxes.[14]

A separate subscription to a heavily used content set can remove its per-token content charge, while standard search and retrieval charges remain. That distinction affects the overall cost even when access to the chosen content set is described as unrestricted.[11]

Private-content ingestion has a two-step upload flow: obtain a single-use upload URL and document identifier, then upload the file. Enrichment extracts and annotates the content; an optional status check shows when processing is complete. Indexing and long-term storage occur by default, with no_index and no_store options available to change that behaviour. The implementation must therefore distinguish uploading a file from making it searchable or retaining it for later use.[10]

The privacy documentation states that user data is not used to train large language models. Uploaded documents are logically separated for the uploading organisation, with sharing inside that organisation controlled by the user. It also describes need-based employee access for service delivery, compliance and resolving reported issues. These statements concern Bigdata’s handling of customer data and should be assessed separately from the retention settings of an external model connected to it.[16]

The general service terms require connected models to support and operate with Zero Data Retention; customers must verify and maintain that configuration.[14]

RavenPack publishes separate sales and technical-support contacts, including regional telephone numbers. The public support page does not specify a common service window or numerical response target. That page is a contact directory; the service agreement determines any incident priority, availability or resolution commitment for a licensed dataset.[17]

Bigdata’s developer support page directs individuals to general product support, managed evaluations to a trial-support address, and enterprise customers to a priority route supplied by their account manager. This distinguishes a self-service account from an organised trial and a contracted enterprise deployment. It does not publish one response-time promise applicable to all three.[18]

Bigdata’s standard Grounding Right permits a single inference and retention of the resulting derivative output, including after account termination. It does not permit retaining or redistributing the underlying source content, or using it to train, fine-tune or evaluate a model. A subscription or API key therefore does not establish a permanent source archive or general model-development licence.[14]

The separate MCP terms require deletion of source material, extracts and cached copies on termination. They allow limited retention of lawfully generated responses for regulatory compliance, internal records, audits or routine backups, without further reuse to recreate or commercially exploit the service. These are MCP-specific exit conditions; the applicable master agreement and order need to establish continuing rights.[15]

MCP terms describe the service as informational, exclude investment, tax and legal advice, and assign users responsibility for checking output accuracy. A source reference makes evidence inspectable; it does not establish that the host’s conclusion is accurate or authorise a trade or other action.[15]

Bigdata.com’s service agreement names RavenPack International S.L.U. as the operator.[14]

RavenPack’s website terms use RavenPack International S.L. and govern the public website. Those website conditions are separate from Bigdata’s service terms and from a commercial dataset licence, which must specify permitted access, distribution and use.[19]

RavenPack dates its financial-data analytics activity to 2003. Its current company page identifies Armando Gonzalez as CEO and co-founder and describes customers including banks, asset managers and hedge funds. The development of Bigdata.com sits within that existing analytics business rather than identifying a separate unrelated provider.[20]

The company identifies Marbella, Spain, as its global headquarters and New York as its North American headquarters. These are the published operating locations; a customer’s actual supplier entity and contract jurisdiction remain those in the relevant agreement.[21]

On 1 December 2025, RavenPack announced an FT Ventures investment and a Financial Times content-licensing agreement. The announcement describes FT news and archive access within its analytics products and Bigdata.com and names Bullhound Capital and Molten Ventures as existing backers. It describes an investment and content partnership, not a purchase of the entire RavenPack business by the Financial Times. The article does not disclose the ownership percentage.[22]

Operator

RavenPack group; Bigdata.com is operated by RavenPack International S.L.U.

Service scope

Structured news and alternative-data analytics, financial factors, research search and AI data services.

Customer types

Banks; Hedge funds; Asset managers; Quantitative researchers; Financial analysts; Enterprise developers

Delivery

SaaS; API; Hosted

Pricing publication

Public and quoted pricing

Headquarters

Marbella, Spain

Platforms

RavenPack EDGE; News Analytics; RavenPack Factors; Forex Sentiment; Bigdata.com; Research Agent; Workflows API; Search Service; Content API

Integrations

REST APIs; Python SDK; MCP; CSV; Snowflake; SharePoint; Email ingestion

Pricing scope

Public Bigdata.com plans and token consumption; enterprise and dataset scope require commercial agreement.

Implementation

Entity filters and API integration, with configurable private-content ingestion.

Support

RavenPack sales and technical contacts; Bigdata.com individual, managed-trial and enterprise support routes.

Permissions and regulatory scope

Data and research technology; MCP terms describe informational services and retain user responsibility for output assessment.

Provider metadata checked 21 Oct 2026Source 1Source 2Source 3Source 4Source 5Source 8Source 10Source 12Source 14Source 15Source 17Source 18Source 19Source 21Source 9

Forex Sentiment contains two different constructions: macro-news sentiment supports trend-following factors, while currency media attention supports mean-reverting factors, with smoothing periods affecting turnover. The published predefined currency list is AUD, CAD, CHF, EUR, GBP, JPY and NZD. RavenPack supplies daily data with history from 2004 and delivery through RavenPack or Snowflake. The performance charts on the product page are backtests of daily-rebalanced strategies over 2004–2023; they are not presented here as live customer returns.[5]

The 15 December 2025 S&P Capital IQ announcement describes filings and earnings-call transcripts as live on Bigdata.com for S&P Global Market Intelligence licensed users. Activation or a suitable licence is required. Catalogue inclusion therefore does not by itself show that every account has the same source entitlements.[23]

The developer introduction labels the Monitors API as coming soon while documenting Research Agent, Workflows, Search and Content services. The planned recurring-search service should therefore be distinguished from currently documented research and template workflows, even though both involve repeated analysis.[8]

The usage guide retains documentation for older quota-based subscriptions, including API Query Units, alongside the current token-credit model. It instructs customers to identify the account’s subscription type rather than apply the token calculation to every legacy account. API and MCP usage can expose content-token counts, although the guide notes an exception for fetch/search tools constrained by the OpenAI output schema.[13]