How Kensho Uses LLMs and Machine Learning to Power Financial Data Workflows
With more financial firms eager to put AI in their employees’ hands, Kensho saw the need for a solution that would enable them to do just that — without the risk of hallucinations.
That’s where the Kensho LLM-ready API comes in. The tool integrates with large language models such as Claude, ChatGPT and Gemini, enabling customers to use natural language to query a range of S&P Global datasets, including transactions, earnings call transcripts and company financials.
“We’re using our years of accumulated domain knowledge to help users power their financial workflows in an era of rapidly accelerating AI adoption,” said Special Projects Machine Learning Engineer Abhishek Chaudhary.
Over the past year, a team of Kensho technologists including Chaudhary have been busy enhancing the LLM-ready API’s capabilities, making it even easier for customers to reap the benefits of S&P Global data. Chaudhary works closely with S&P Global strategic customers to understand what datasets and features are most valuable for them, and surfaces this valuable feedback to Kensho product teams.
About Kensho
Headquartered in Cambridge, Massachusetts, with an office in New York City, Kensho is the AI innovation and transformation engine of S&P Global. Combining S&P Global’s essential intelligence with deep AI and engineering expertise, Kensho builds trusted solutions that shape global markets.
How Kensho’s Strategic Collaboration with Anthropic Accelerated the LLM-ready API Momentum
Forming industry alliances is key to Kensho’s LLM-ready API’s success. The company recently collaborated with major LLM tool-provider Anthropic to deliver its data to customers, and has also announced integrations with platforms including ChatGPT, Amazon Quick Suite, and Databricks, powered by a cutting-edge Model Context Protocol (MCP) server built for the solution.
According to Software Engineering Manager Stu Harvey, the company was featured in Anthropic’s Claude for Financial Services announcement, highlighting Kensho as one of the first companies to provide a financial data retrieval tool through Claude.
“After the Claude for Financial Services announcement, we saw a tremendous influx of customer interest in utilizing our tools in their workflows,” Harvey said.
Harvey leads a lean 10-person team, so meeting the demands of a growing customer base required extensive collaboration with the company’s site reliability engineering, commercial and product teams, in addition to their colleagues at S&P Global.
“At one point, it felt like everybody at Kensho had helped out in some way, whether covering a customer call, fixing a bug in the system or testing out the UX of a new feature,” Harvey said. “As engineering manager, I wore a lot of hats but primarily helped coordinate efforts across teams and made sure all of our customers or prospective customers had a positive experience.”
How the Team Expanded S&P Global Data Coverage in the Kensho LLM-ready API
To further enhance the LLM-ready API, Kensho has worked to expand API support across S&P Global’s vast data estate. Financial data coverage has expanded to include datasets including Private Company Financials, S&P Capital IQ Estimates and more, while enhancing the solution with more detailed financial comparison capabilities and deeper auditability for public company financials data.
Software Engineer Keith Page explained that access to private company data is a big deal for clients, considering private companies outnumber public ones by a massive margin, therefore increasing data coverage.
Chaudhary added that these recent enhancements address problems that even experienced financial professionals often face. For instance, Apple’s 2024 fiscal year reporting period doesn’t align with Microsoft’s 2024 fiscal year, so comparing the two could be misleading.
To address this issue, Chaudhary said his team used the company’s financial domain knowledge to implement a system that detects ambiguous multi-company questions and informs users that they can use calendar period data to compare multiple company financials across the same time period.
“This is the ‘LLM-ready’ part of our API,” he said. “We’re not just wrapping data endpoints; we’re designing the entire experience to help users avoid pitfalls they might not know to be wary of.”
“We’re not just wrapping data endpoints; we’re designing the entire experience to help users avoid pitfalls they might not know to be wary of.”
To add this feature, Chaudhary’s team worked alongside the company’s infrastructure, security, machine learning and data teams, along with site reliability and the S&P Global Market Intelligence division’s sales teams. They also leveraged insight from S&P Global subject-matter experts with backgrounds in investment banking, hedge funds and other financial domains.
“It’s one thing to technically define tools wrapping S&P Global’s data, but it’s another to understand how professionals actually phrase questions,” Chaudhary said. “Our SMEs emulate real-world usage patterns we couldn’t anticipate from data alone.”
When his team encountered an issue where tools would time out for certain users, he said they first analyzed it with a hedge fund customer, and when that didn’t help, they went to the team at Anthropic. To troubleshoot the issue, Chaudhary and his teammates asked a technical user to send logs and bypassed the failed permission popup by allowing Claude to run tools without explicit approval.
And it worked.
“We shared the solution with both customers and Anthropic, who released a patch fixing the bug shortly after,” Chaudhary said. “That cross-collaborative problem-solving defines how we operate.”
Kensho’s ongoing collaboration with Anthropic has only grown more fruitful as the companies continue to co-innovate to push the boundaries of agentic AI. The Kensho team has since built an S&P Global Plugin for agents like Claude Cowork, releasing a suite of financial skills for AI agents and large language models.These skills are tailor-made for S&P Global customers to seamlessly execute agentic workflows, from building a company tearsheet to generating industry deal flow digests and earnings call previews.
How the Kensho LLM-ready API Improved Scalability, Self-Service and Customer Trust
The work that went into building and enhancing the LLM-ready API hasn’t just benefitted clients; it’s also unlocked scalability.
“By building self-service capabilities and refining our support processes, we’ve freed our engineering team from reactive troubleshooting to focus on building,” Chaudhary said.
The project also brought teams closer and made teammates feel more comfortable asking questions, leveraging each other’s expertise and efficiently routing issues to the right people. Meanwhile, customers have witnessed Kensho’s willingness to help them when questions arise, which has compelled them to continue using the product long-term.
“Other divisions are now considering how they can expose their specific data to users in similar ways,” Chaudhary said. “That ripple effect across S&P Global is significant.”
Harvey added that he’s proud of the team’s willingness to make use of frontier technologies in a way that counteracts the approach taken by many other organizations.
“We’re always evolving at Kensho, but when it comes to Model Context Protocol and GenAI, we’re really on the bleeding edge of tooling,” he said. “Everybody is building the ship while it’s at sail, but we’ve managed to build a useful and trusted product with rapidly evolving tools.”
This impact is amplified by Kensho’s global expansion. Chaudhary said that working with customers outside of the United States has given the team new perspectives, as diverse regions rely on different tech stacks, regulatory constraints and LLM providers.
“That diversity of input strengthens the product,” he said.
What It’s Like to Work at Kensho: An Intentional, Unselfish Engineering Culture
At its core, Page describes Kensho as a very “unselfish” company.
“Across our divisions, there is no culture of hoarding information or expertise, and that culture shows its worth in the software that we sell,” he said.
“Across our divisions, there is no culture of hoarding information or expertise, and that culture shows its worth in the software that we sell.”
The LLM-ready API is a reflection of the knowledge-sharing and teamwork that upholds Kensho’s culture. It’s an environment that Chaudhary said is inherently scientific in nature, rooted in data-driven priorities backed by human judgment and domain expertise.
“People should learn we’re a team combining deep financial domain expertise with cutting-edge ML engineering, unafraid to iterate with customers and partners to get things right,” Chaudhary said.
Employees at Kensho are driven by a “Go Team” attitude, Harvey added. Whether they’re resolving an issue with a customer or brainstorming new features, everyone is motivated to win — together.
“We don’t strictly adhere to our titles and responsibilities, but instead jump in to help out whenever we get the opportunity,” he said.
Frequently Asked Questions
What is Kensho’s LLM-Ready API?
Kensho’s LLM-ready API is a tool that integrates with large language models like Claude, ChatGPT and Gemini to let customers query S&P Global datasets using natural language. It enables access to data such as transactions, earnings call transcripts and company financials while helping users avoid common financial data pitfalls.
What is it like to be an engineer at Kensho?
Engineers at Kensho work on AI and machine learning systems that make S&P Global data more accessible and useful for customers. The work involves collaborating across infrastructure, security, data and product teams while building solutions that power financial intelligence workflows.
What is it like to work on AI at Kensho?
Working on AI at Kensho involves building tools that combine machine learning, financial domain expertise and large language models to power financial workflows. Teams experiment with emerging technologies such as generative AI, Model Context Protocol and agent-based tools while iterating with customers and partners.
