Building an autonomous financial analyst with o1 and o3-mini

# Building an autonomous financial analyst with o1 and o3-mini Source: [https://openai.com/index/endex/](https://openai.com/index/endex/) [Endex⁠\(opens in a new window\)](https://endex.ai/), an AI platform for financial firms, is developing an AI Analyst that retrieves, synthesizes, and reasons through complex financial data\. They work closely with OpenAI to reinvent the workflow of investment professionals\. "Finance professionals don’t just need search results; they need structured thinking and deep analysis," says Tarun Amasa, CEO at Endex\. "We envision a future where every firm has access to teams of digital analysts, seamlessly augmenting time\-intensive workflows\.” By evaluating and integrating a range of OpenAI’s models, including GPT‑4o, o1‑mini, o1‑preview, and o3‑mini, Endex connects firms' internal data, public disclosures, and trusted financial sources to enhance research and augment key workflows\. Precision is paramount in finance, but many AI\-powered tools can miss critical details that impact decision\-making\. For instance, a missed adjustment in EBITDA reconciliation or an overlooked “change in control provision” clause can alter the financial outlook of an investment\. Endex takes a different approach\. Instead of relying on retrieval\-augmented generation \(RAG\), its agents use OpenAI’s reasoning models to retrieve and reflect data like a financial analyst, pulling facts, identifying inconsistencies, and contextualizing metrics\. Endex’s agents autonomously process financial reports, market data, and firm\-specific knowledge to complete tasks including: - Precedent transaction overviews - Earnings performance summaries - Investment committee \(IC\) memo preparation - Data room due diligence "OpenAI’s o\-series long\-context windows, prompt adherence, and reasoning capabilities make a significant difference," says Amasa\. "Instead of just finding a needle in a haystack, the models can analyze and detect discrepancies in financial data the way a seasoned analyst would\." > "Instead of just finding a needle in a haystack, the models can analyze and detect discrepancies in financial data the way a seasoned analyst would\." Tarun Amasa, CEO at Endex Agents can deliver outputs as emails, documents, Excel models, or slide decks, and enable analysts to confidently trace agent conclusions back to their sources, acting as a true extension of an investment team\. Endex has seen breakthrough performance using OpenAI’s reasoning models, enabling precise financial workflows: - **Multistep financial reasoning**: Previously, Endex relied on complex prompting, chained completions, and multiple verification steps\. With OpenAI o1, they’ve simplified this process without sacrificing accuracy\. - **Faster, more efficient AI**: Using OpenAI o3‑mini, Endex has improved intelligence at one\-third the latency per turn, enabling detailed multi\-step workflows such as analyzing confidential information packages \(CIPs\) and automating financial model reconciliation\. - **AI\-powered cross\-checking**: Endex agents can now identify discrepancies in financial data, flagging restatements in footnotes and surfacing inconsistencies with targeted citations\. This allows analysts to focus on decision\-making rather than manual verification\. - **Multimodal analysis**: Endex’s Finance Agent Retrieval \(FAR\) benchmark measures context usage on tabular and chart data \- key sources for financial professionals\. OpenAI’s o1 vision capabilities allow Endex to process investor presentations, internal decks, Excel models, and 8\-Ks with leading FAR scores\. - **Deliverable automation**: Endex generates detailed reports, reducing the manual work traditionally required for financial analysis\. This automation allows professionals to focus on high\-value strategy rather than data formatting\. "Finance professionals require structured, referenceable reasoning with attention to detail, and traditional LLMs struggled with this level of coherence," notes Amasa\. "OpenAI’s reasoning series models were the first to consistently meet this quality bar\." At the core of Endex’s platform is a rigorous evaluation and testing framework, developed in close collaboration with OpenAI\. This system allows the Endex engineering team to trace model outputs directly within the OpenAI platform, tracking key metrics like response latency, first\-token generation time, and reasoning depth\. Professionals can then assess outputs in real\-world scenarios\. Through blind user testing, financial experts preferred responses generated by OpenAI’s o1 model**70% of the time**over non\-reasoning models\. Endex also tracks performance metrics like response latency and first\-token generation time to continuously refine the system\. Fine\-tuning models like GPT‑4o mini and OpenAI’s o1 series through[Reinforcement Learning⁠](https://openai.com/form/rft-research-program/)– critical for precedent transaction analysis and other research\-intensive tasks – improved entity extraction and query intent mapping as well\. “Our collaboration with OpenAI has unlocked tools to tailor model behavior to the reasoning and output style professionals expect,” says Pratham Soni, co\-founder at Endex\. “We’ve deployed reinforcement fine\-tuned models to convert our custom datasets into targeted reasoning improvements\.” OpenAI is integral to Endex’s growth \- not just as a model provider, but as a strategic collaborator in refining evaluation techniques and pushing the boundaries of AI\-powered financial analysis\. “What excites me most about this collaboration is our shared vision for vertical\-specific AI\. Our work goes beyond APIs – it’s about building the agent\-user interfaces that will change how financial analysts do work,” says Amasa\. The Endex team sees a significant opportunity to scale\. They plan to refine their autonomous analyst capabilities, ensuring that the technology can handle increasingly complex financial tasks for their customers\. "Our clients will be pioneers in a world where AI systems are not just a tool, but a true co\-worker," says Amasa\. "With OpenAI’s models, we’re building systems that can plan, reason, and execute financial analysis\. It simply seemed like science fiction before\." > "Our clients will be pioneers in a world where AI systems are not just a tool, but a true co\-worker\." Tarun Amasa, CEO at Endex