Quick Verdict
Kensho is S&P Global’s AI engineering and product organization, not a self-service retail investment terminal. Its current public offering has two main areas. The Kensho LLM-ready API gives LLMs and agents traceable access to licensed S&P Global datasets. The Core AI Capabilities product family prepares unstructured enterprise content through Extract, Link, Scribe, and NERD, covering PDF extraction, company-identifier matching, financial speech transcription, and named-entity recognition and disambiguation.
The LLM-ready API emphasizes deterministic retrieval: the retrieval service does not perform AI reasoning or generation, and the same question is designed to return the same data answer. Results include sources and can link to an original filing in S&P Capital IQ Pro where available. Customers retain control of the LLM, orchestration, tools, and final interpretation. This reduces integration work, but it does not grant unlicensed access to every S&P dataset.
Dataset coverage, history, update frequency, latency, fields, and source-document links depend on the customer’s contract and entitlements. PDF extraction and transcription quality also depend on the actual input. This page, retrieved data, and downstream model output are not investment, trading, or financial advice. Data latency, provider revisions, contracted coverage, and historical-data limitations can change conclusions; historical performance, backtests, and matching scores do not guarantee future returns or performance.
Best For
Kensho is best for S&P Global data customers, banks, investment firms, and enterprise data teams that need to connect licensed information to internal AI applications. It also fits organizations processing large volumes of financial PDFs, cleaning company master data, transcribing earnings calls, or adding S&P-linked entities to text archives.
Individuals who need charts and screens should consider Koyfin. A team that wants to integrate several providers and private databases in an open framework can compare OpenBB. Developers testing a lower-cost market-data API can evaluate Alpha Vantage.
Kensho is not a strong fit for buyers that require public retail pricing, instant self-service activation, or ready-made buy and sell signals. Procurement commonly includes sales discussions, data contracts, security review, proof-of-concept testing, implementation, and a definition of support responsibilities.
Key Features
- LLM-ready API: natural-language and API access to licensed datasets including S&P Capital IQ Financials, Earnings Call Transcripts, Company Intelligence, and Private Company Financials.
- Deterministic retrieval: the retrieval layer avoids generative reasoning and returns traceable sources. A downstream LLM can still summarize the evidence incorrectly.
- MCP, Python, and SQL tools: remote or local MCP servers, a Python library, and simplified endpoints mapped to traceable function calls support several application patterns.
- Kensho Extract: transforms complex financial PDFs into machine-readable text and structured content for search, RAG, and downstream workflows.
- Kensho Link: matches incomplete or messy company records to S&P Global IDs and other identifiers, returning a score that helps users assess match quality.
- Kensho Scribe: speech-to-text designed for business and financial audio such as earnings calls and investor presentations, with attention to accents, audio quality, and jargon.
- Kensho NERD: identifies companies, people, events, and other entities in text and links them to S&P Global knowledge bases.
These capabilities should not be treated as a single mandatory stack. A buyer can have a retrieval problem, a document problem, an entity-resolution problem, or an audio problem. Each requires its own acceptance dataset and quality threshold. Public descriptions of speed or scale should not be extrapolated to scanned Chinese filings, unusual tables, low-quality calls, or a customer’s private entity catalog without testing.
Use Cases
The LLM-ready API supports company comparisons, financial-statement retrieval, transcript research, historical pricing and valuation work, M&A analysis, and pitch-book preparation. Extract can turn filing tables into structured review queues. Link can reconcile CRM or transaction records to a corporate identifier master. Scribe can make earnings calls searchable, while NERD can enrich news and documents with linked entities.
Every workflow needs an “as of” policy. Financial statements, estimates, transactions, private-company information, and entity mappings can change after new disclosures or provider corrections. Extracted documents and transcripts can also change after a model or review update. Store source files, timestamps, product versions, confidence scores, and human corrections.
Backtests must use point-in-time data and the universe that existed at each historical date. Today’s Capital IQ coverage cannot stand in for a historical universe. Include delisted companies, retain historical identifier mappings, and distinguish original values from later restatements. Otherwise survivorship and revision bias can contaminate results.
Latency varies by dataset rather than by the Kensho brand as a whole. A filing, consensus estimate, transcript, market price, and private-company record can have different publication and processing schedules. Ask for dataset-specific service definitions and verify provider and exchange entitlements before relying on a field operationally.
Pricing
Kensho does not publish a standard retail monthly price list. The LLM-ready API is presented through S&P Global Marketplace and directs prospective customers to a sales conversation. Extract, Link, Scribe, and NERD have documentation or sign-in surfaces, but access and commercial terms are generally configured by enterprise agreement.
Total cost can include the Kensho product, specific S&P Global datasets, API usage, seats, implementation, support, cloud infrastructure, and the customer’s chosen LLM. A demonstration may contain datasets or fields not included in the final contract.
Before purchase, request written definitions for dataset names, regions, entities, history, refresh schedules, latency, API quotas, concurrency, source-document links, retention, use in model training, generated-output rights, redistribution, service levels, and data handling after termination. Exchange or upstream-provider entitlements must be verified separately where applicable.
Pros
- Tight alignment with S&P Global datasets, identifiers, and enterprise workflows.
- Deterministic retrieval, citations, and source links support auditability.
- Extract, Link, Scribe, and NERD address distinct document, master-data, audio, and text problems.
- MCP, Python, and SQL-oriented tools support different enterprise application architectures.
- Customers retain control of the downstream LLM and orchestration.
- The product boundary avoids presenting generative reasoning as the authoritative data source.
Cons
- Enterprise sales rather than transparent self-service retail pricing.
- Product access does not automatically include every S&P Global dataset.
- Performance on non-English, low-quality, unusual-layout, or private-domain inputs needs independent testing.
- Extraction, transcription, and entity-linking errors can propagate into downstream AI systems.
- Financial and entity data can be revised, requiring snapshots and version controls.
- Data, security, privacy, model-hosting, and redistribution contracts can lengthen implementation.
Alternatives
| Tool | Better for | Main difference from Kensho |
|---|---|---|
| OpenBB | Building a multi-provider data layer and custom workspace | More open, while the user and providers own data quality and licensing |
| Koyfin | Self-service charts, screens, portfolios, and advisor reports | A packaged SaaS terminal rather than enterprise document and audio processing |
| Alpha Vantage | Rapid market-data API prototypes | Lower entry barrier with a different coverage model and fewer enterprise content tools |
| Seeking Alpha | Investment commentary, ratings, and earnings content | A content service rather than S&P enterprise retrieval infrastructure |
| TipRanks | Analyst, insider, and quantitative-rating tracking | A finished research interface rather than extraction and entity-resolution products |
FAQ
Is Kensho part of S&P Global?
Yes. Kensho describes itself as S&P Global’s innovation and AI engine, combining S&P intelligence with AI and engineering products.
Can an individual sign up for a complete free product?
The current site does not present a comprehensive retail free plan. It directs prospects to S&P Global Marketplace or the sales team.
Does the LLM-ready API generate investment advice?
No. The official description says the retrieval layer does not perform AI-driven reasoning or generation. Downstream applications still require human review and are not investment advice.
Does API access include every S&P dataset?
No such assumption is safe. Datasets, fields, history, and source links depend on the signed contract and entitlement.
Are Extract and Scribe perfectly accurate?
No. Layout, scan quality, overlapping speakers, accents, and specialized vocabulary can cause errors. Critical fields need sampling and human review.
How should survivorship and revision bias be controlled?
License point-in-time data, preserve query snapshots, include delisted entities and historical mappings, and record provider revision dates. Today’s latest response is not necessarily what was known historically.
Bottom Line
Kensho is valuable when an enterprise needs licensed S&P Global data or unstructured financial content inside an auditable AI workflow. Buyers should first identify whether they need retrieval, PDF extraction, entity matching, transcription, or text enrichment, then test that product with representative material. Contracts must specify coverage, latency, history, revisions, rights, and redistribution. No model output or backtest guarantees future performance, and final decisions remain the responsibility of qualified professionals.