Daloopa — AI Financial Data Infrastructure for Analysts

AI-powered financial data infrastructure that automates fundamental data extraction from SEC filings for equity analysts and financial institutions.

Overview

Daloopa is an AI-powered financial data infrastructure company founded in 2019 by Thomas Li, a former Point72 analyst, alongside Jeremy Huang (ex-Airbnb and Meta) and Daniel Chen (ex-Microsoft). The New York-based company has raised $52.5 million total across multiple rounds, including an $18 million Series B led by Touring Capital with Morgan Stanley participation in August 2025 and a $13 million strategic investment from Pavilion Capital announced July 31, 2025. Backers also include Nexus Venture Partners and Uncorrelated Ventures.

The platform serves over 150 major financial institutions, covering nearly 5,000 public companies globally with claimed accuracy exceeding 99% and full source attribution to regulatory filings. Daloopa delivers up to 10x more data points per company than unnamed competing providers; no independent comparison has verified that figure.

Daloopa's public positioning has shifted from financial data provider toward agentic finance infrastructure. In July 2025, the company launched its Model Context Protocol (MCP) with a connector for Anthropic's Claude for Financial Services, then expanded to OpenAI's ChatGPT in February 2026, treating MCP as its primary go-to-market channel for AI-native buyers rather than building a standalone interface. A Slashdot directory listing confirms Daloopa is named alongside S&P Global as a core data source within Anthropic's Claude for Financial Services, though no primary Anthropic product page has confirmed partnership terms or launch date. Daloopa also offers a custom GPT accessible through OpenAI's GPT directory.

In February 2026, Daloopa published the FinRetrieval benchmark: 500 real-world financial questions tested across three frontier agent systems (OpenAI Agents SDK with GPT-5.2, Anthropic Agent SDK with Claude Opus 4.5, and Google ADK with Gemini 3 Pro). All three reached approximately 90% accuracy on Daloopa's structured database versus roughly 19-20% from public web sources, a gap of up to 71 percentage points, confirmed by independent coverage. The benchmark is vendor-commissioned and has not been independently replicated; treat figures as vendor-reported. CEO Thomas Li framed the finding directly: "Accuracy in AI-driven finance isn't just a model problem, it's a data access problem."

The benchmark's secondary finding carries more competitive weight than the headline number: closing the remaining gap from 90% to 99%+ is blocked by data infrastructure failures, specifically fiscal calendar misalignment for non-US companies and entity naming inconsistencies, not model failures. All three agent systems performed better on US companies than non-US companies because US companies predominantly use December year-ends while non-US companies often do not. By naming these problems publicly, Daloopa is staking a claim on the solution space and signaling that raw model improvements from Anthropic or OpenAI won't close the gap without better underlying data.

How Daloopa processes documents

Daloopa's extraction pipeline targets SEC filings, earnings transcripts, and supplemental disclosures, normalizing outputs into structured datasets with every data point hyperlinked to its source filing for auditability. The core products operate as a layered stack.

The Excel Add-in provides one-click model updates with hyperlinked source verification inside existing spreadsheets. Within it, the Retrofitter links existing Excel models to source documents and audits historical inputs, while the Updater automatically refreshes models with the latest disclosures while maintaining native formatting, with a claimed error rate below 1%.

Scout is an AI-powered model-building tool that generates complete financial models from natural language prompts with transparent reasoning. A vendor-published workflow demonstrates Scout's MCP integration inside Claude retrieving six years of standardized depreciation data across Microsoft, Alphabet, Amazon, Meta, and Oracle in a single conversation, generating an interactive dashboard with seven chart views and five data tables. The vendor claims five-minute completion; no third party has verified this timing.

The API platform offers 8 endpoints with JSON and CSV output, webhook notifications, and keyword search. The MCP layer sits above this, providing a standardized client-server architecture that replaces custom AI integrations with connectors for both Claude and ChatGPT-based workflows, with containerized deployment support. GA and beta status for individual MCP components are not publicly disclosed.

The FinRetrieval benchmark identified two infrastructure gaps that cap accuracy below 99%: non-December fiscal year-ends for non-US companies cause calendar misalignment, and inconsistent entity naming conventions or tickers cause retrieval failures. These are the stated next product frontiers.

Use cases

Equity research and financial modeling

Analysts save an average of 2 hours per ticker each earnings season, with a 70% reduction in time for new model builds. The platform enables simultaneous updates across all coverage models rather than forcing prioritization of select companies, a structural advantage during earnings season when multiple companies report within days of each other.

The FinRetrieval benchmark demonstrated the agentic use case concretely: hedge funds can use MCP-connected agents to identify quarter-over-quarter inflections, run scenario simulations, and generate equity research with full source traceability. On structured Daloopa data, agents achieve approximately 90% accuracy versus roughly 19-20% on public web sources. The benchmark does not specify which of the three tested agent frameworks achieved the full 71-point gain versus smaller gains.

Due diligence and private markets research

Private equity and hedge funds access 13 years of historical data including company guidance, KPIs, segment breakdowns, and geographic data for comparable company analysis. The claim of 10x data density compared to unnamed competing providers is vendor-reported without independent verification. Alkymi addresses a comparable need in financial services, focusing on extracting and transforming unstructured documents into standardized datasets for investment workflows. Cognaize takes a neuro-symbolic approach to the same financial document extraction problem, targeting structured data extraction from complex filings for institutional clients.

Compliance and audit

Every data point is hyperlinked to its original SEC filing or earnings transcript, enabling compliance teams to trace extractions to source for regulatory requirements. As Li stated in the February 2026 benchmark release: "We're entering the era where AI is no longer optional in finance, but accuracy and auditability are non-negotiable." This positioning distinguishes Daloopa from general-purpose large language model tools that generate outputs without source grounding. Acuity Knowledge Partners takes a services-led approach to the same auditability requirement, combining AI-powered document processing with human analyst oversight for 800+ financial institutions. For teams evaluating agentic document processing more broadly, the MCP architecture Daloopa has adopted represents one implementation pattern for connecting structured financial data to autonomous agent workflows.

Technical specifications

Resources

• Daloopa homepage
• Daloopa products
• Daloopa API
• FinRetrieval benchmark report (PR Newswire, February 2026)
• Scout MCP workflow demonstration
• TechCrunch: Daloopa trains AI to automate financial analysts' workflows
• completeaitraining.com: FinRetrieval benchmark coverage
• CB Insights: Daloopa company profile

Company information

Headquarters: New York City, NY Website: https://daloopa.com/ Founded: 2019 Employees: ~300