Kensho

Kensho is the artificial intelligence division of S&P Global. It started in 2013 as a Harvard-born startup doing natural language search over financial markets, and S&P Global bought it in 2018 for roughly $550 million, then one of the largest AI acquisitions in finance. Since the acquisition it has settled into a clear role: the machine learning engine that makes S&P Global's mountain of financial data usable.

That framing matters if you found this page hoping for a stock-picking app. Kensho isn't a retail product and never has been. Its customers are banks, asset managers, and engineering teams that work with financial data at scale, plus S&P Global itself, whose products like Capital IQ Pro run Kensho technology under the hood.

What Kensho actually does

Kensho ships a family of machine learning services, each aimed at a specific messy-data problem in finance:

• Scribe transcribes audio with models trained on financial speech, so earnings calls come back with tickers, acronyms, and numbers intact instead of mangled.
• Extract pulls tables and figures out of PDFs and filings and turns them into structured data.
• NERD (Named Entity Recognition and Disambiguation) spots companies, people, and organizations in text and ties them to unique identifiers, so Apple the company never gets confused with the fruit.
• Link matches messy internal company records to S&P Capital IQ identifiers, a job that otherwise eats weeks of an analyst's time.
• Classify tags documents and passages against financial taxonomies so they can be searched and filtered.

The LLM-ready API

Kensho's most visible recent work is the LLM-ready API, launched in beta in late 2024. It lets large language models query S&P Global datasets in plain English: Capital IQ financials, market data, earnings call transcripts, business relationships, and M&A transactions. Ask an assistant what a CFO said about margins last quarter and it can retrieve the actual transcript passage instead of guessing. Kensho also runs a Model Context Protocol (MCP) server for the same data, with integrations announced for ChatGPT, Databricks, and Amazon Quick Suite, and a collaboration with Anthropic to bring the data to Claude.

For research teams experimenting with AI assistants, this is the piece worth watching. It turns a data licensing relationship into something a chatbot can use without inventing numbers.

Who it's for

Kensho makes sense for three groups: financial institutions that already license S&P Global data and want to query it through AI tools, engineering teams that need finance-tuned transcription or entity linking as building blocks, and data teams stuck reconciling company identifiers across systems. Individual investors and small teams without an S&P relationship will find little here to buy.

Pricing

There is no public price list. Kensho sells through enterprise agreements, usually tied to S&P Global data licensing, so budget for a sales conversation rather than a checkout page. Some tools have offered trials in the past, but plan on a procurement process, not a credit card signup.

Our take

We rate Kensho highly for the problems it picks. Entity linking, finance-aware transcription, and clean extraction from filings are unglamorous, genuinely hard problems, and Kensho's versions are among the best because they train on S&P Global's own labeled data. The LLM-ready API is a smart bet on where research workflows are heading, and the MCP server means it plugs into tools teams already use.

The honest downside is access. Everything worthwhile sits behind enterprise sales and an S&P Global relationship, pricing is opaque, and there's no meaningful way for a solo developer to kick the tires the way you can with an open API. If you don't specifically need S&P's data, open-source models plus a cheaper data vendor will get you most of the way there.