S&P Global Market Intelligence launched Adaptive Retrieval last month, promising to give clients greater flexibility in accessing and using the ratings giant’s data via their large language models (LLMs) and agents.
Complementing S&P Global’s existing Deterministic Retrieval tool, which is built on the Kensho LLM-ready API, the latest innovation enables users to query and make requests of the data using natural language.
Data Management Insight spoke to Bhavesh Dayalji, head of Kensho Data & Intelligence at S&P Global Market Intelligence, to find out more about Adaptive Retrieval.
Data Management Insight: What specific pain points is Adaptive Retrieval capable of solving?
Bhavesh Dayalji: AI agents and LLMs need data that’s cited, verifiable and auditable. Historically, getting high-quality data into those systems has been hard work. You have to find the right sources, validate them, then build all the logic to pull them accurately across enormous datasets.
That takes deep domain expertise that most teams just don’t have and it slows everything down. What we’ve done with Adaptive Retrieval is take that burden off the customer.
You submit a question in natural language, the solution searches and routes across our data sources for you and comes back with a cited response. It’s built for complex and open-ended work, and particularly for agentic workflows and multi-step analyses where you don’t always know upfront exactly what data you’re going to need.
DMI: How does it complement S&P Global’s existing data retrieval capabilities?
BD: There’s no single way customers want to access our data within their AI applications and workflows. Sometimes, you need to ask a question that’s more precise and targeted. You know exactly what you need, and you want full control over how it’s retrieved.
That’s Deterministic Retrieval, the Kensho LLM-ready API that more than 500 active and trialing customers are already using in their AI workflows today for structured queries and predictable results. Other queries are more complex, requiring reasoning across a range of data sources. That’s where Adaptive Retrieval comes in, handling the routing and the decomposition automatically. Customers can use one or both retrieval methods depending on their needs. Together they make up the S&P Global AI Data Portal, which gives them a single point of entry to all of our trusted data.
DMI: How can clients use and access it?
BD: We’ve built our retrieval solutions to not only meet customers at every point on the spectrum of AI and agentic workflows but to adapt as their architectures evolve. These solutions connect to any AI application via a dedicated MCP server and we’ve already pre-integrated the solutions into leading AI applications so customers can get started right away. Adaptive Retrieval is available today in Claude and Gemini Enterprise, and Deterministic Retrieval is natively integrated with platforms including Claude, ChatGPT, Copilot and Cohere, with more on the way.
How you actually use it is quite simple. Once you’re connected, a user or an agent asks a question in natural language, and the answer comes back with citations that link right back to the underlying S&P Global data.
DMI: How does it safeguard data? BD: Trust and accuracy are foundational to how we’ve built this. With Deterministic and Adaptive Retrieval, answers are cited and trace back to the actual S&P Global data behind it, so customers can verify what they’re getting and avoid the risk of hallucinated, unattributable outputs you often see in generative AI systems. We take the same approach to safeguarding our customer queries. When customers query our data through their own AI applications, we simply retrieve the data they asked for and return it.
Subscribe to our newsletter
