OpenAI Unveils GPT-5.5 Financial Features to Automate Wall Street Models

OpenAI

Deep research tools and direct Excel integrations compress multi-day financial modeling and presentation workflows into minutes while maintaining full auditability.

AI-assisted summary of Multiplying workforce impact: Stephanie Anani, Solutions Engineer, OpenAI — original source. Check the original for full context.

Source published: 2026-07-03

Detailed notes

## Background Published by OpenAI on YouTube on June 8, 2026, this video features Solutions Engineer Stephanie Anani discussing deployment strategy and organizational transformation. According to the page description, it focuses on how financial institutions can partner with OpenAI to embed AI into daily workflows, accelerate customer-facing innovation, and scale trusted AI systems across regulated environments.

## TL;DR OpenAI Solutions Engineer Stephanie Anani presents how financial services organizations can leverage AI to eliminate administrative overhead and elevate baseline workforce capabilities across five key functional dimensions. Powered by the GPT-5.5 frontier model with financial context built into its core intelligence, the platform features deep research capabilities, custom financial app connectors, and an Excel integration that generates audit-ready dynamic workbooks. Through a live demonstration as a Blossom Bank investment analyst preparing for an Investment Committee meeting, Anani demonstrates how ChatGPT compresses complex workflows—from deep research and three-statement modeling to creating polished, audit-ready slide decks—from hours or days into minutes. Ultimately, this integration offloads routine operational overhead so human professionals can focus on high-value business decisions and human judgment.

## Atmosphere & style The presentation carries a professional, authoritative, yet highly practical and modern corporate tone typical of an enterprise tech demonstration. Anani delivers her walkthrough with structured clarity and realistic roleplay, creating an engaging and action-oriented experience that bridges high-level AI capabilities with day-to-day financial workflows. The arc moves smoothly from organizational challenges and benchmark data to an energetic, step-by-step practical showcase, leaving the audience with a sense of efficiency, clarity, and empowerment.

## Key takeaways - [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) AI integration aims to elevate employees' baseline capabilities and operating levels rather than simply speeding up routine work. - [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) A recent OpenAI survey indicates that 75% of ChatGPT users can complete tasks they were previously unable to accomplish. - [01:29](https://www.youtube.com/watch?v=m2TV8slGQKc&t=89s) ChatGPT integrates directly with key financial data providers like Dow Jones, LSEG, and S&P, as well as native Excel workflows. - [01:29](https://www.youtube.com/watch?v=m2TV8slGQKc&t=89s) OpenAI introduced GPT-5.5, which embeds native financial context directly into its core model intelligence. - [01:29](https://www.youtube.com/watch?v=m2TV8slGQKc&t=89s) GPT-5.5 demonstrates state-of-the-art performance on financial analysis based on the 'OpenAI Banker Bench' benchmark. - [02:54](https://www.youtube.com/watch?v=m2TV8slGQKc&t=174s) ChatGPT's deep research capability creates executable research plans and queries trusted sources like SEC filings and earnings transcripts. - [03:41](https://www.youtube.com/watch?v=m2TV8slGQKc&t=221s) Enterprise prompt presets and custom skills allow institutions to standardize modeling and presentation formatting across teams. - [03:41](https://www.youtube.com/watch?v=m2TV8slGQKc&t=221s) ChatGPT generates interactive Excel workbooks with explicit assumption notes and fully dynamic, auditable formulas. - [03:41](https://www.youtube.com/watch?v=m2TV8slGQKc&t=221s) AI assistance compresses multi-day financial modeling and scenario planning workflows into mere minutes. - [06:21](https://www.youtube.com/watch?v=m2TV8slGQKc&t=381s) The 'extended thinking' setting enables deeper AI reasoning cycles to generate audit-ready slide decks complete with presenter notes. - [06:21](https://www.youtube.com/watch?v=m2TV8slGQKc&t=381s) AI systems are designed to support and augment human judgment rather than replace professional decision-making.

## Detailed walkthrough ### [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) Empowering Financial Services Workforces with AI

Stephanie Anani, a Solutions Engineer at OpenAI, introduces her work assisting OpenAI's largest enterprise clients in extracting maximum value from AI technology. Building on Lee's previous discussion of workspace agents, Anani shifts the focus to how individual employees can dramatically elevate their output. She addresses a recurring challenge shared by organizational leaders: employees spend excessive time on necessary administrative tasks that do not actually require human judgment. AI bridges this gap by offloading operational overhead, enabling personnel to concentrate on high-value business drivers.

- A recent OpenAI survey revealed that 75% of ChatGPT users were able to complete tasks they previously could not accomplish. - The primary goal of AI integration is not merely speeding up existing work, but elevating the overall operating level and baseline capabilities of employees. - Unlocking these operational gains requires providing workforce support across five key functional dimensions.

---

[01:29](https://www.youtube.com/watch?v=m2TV8slGQKc&t=89s) Frontier Models Built for Financial Context and Workflows

Anani frames ChatGPT as an ideal enterprise platform tailored for financial services, starting with the critical requirement of context. Enterprise users need reliable access to trusted, up-to-date data sources to execute high-stakes analysis. To support this, OpenAI has established app connectors to major financial data providers, including Dow Jones, LSEG (London Stock Exchange Group), and S&P. Recognizing that extensive financial analysis occurs in spreadsheets, OpenAI built ChatGPT directly into Excel, enabling professionals to manipulate and query workbooks using natural language.

- Underpinning these custom financial capabilities is the GPT-5.5 frontier model. - Financial processes are directly embedded into GPT-5.5's core model intelligence rather than layer-on top. - OpenAI developed the "OpenAI Banker Bench" benchmark to evaluate model performance on complex financial tasks, demonstrating GPT-5.5's state-of-the-art performance against competing frontier models.

---

[02:54](https://www.youtube.com/watch?v=m2TV8slGQKc&t=174s) Deep Research for Investment Analysis

Transitioning to a live demonstration, Anani roleplays as an investment analyst at "Blossom Bank" who receives an urgent request from her manager: prepare a fresh investment view on QXO for an upcoming Investment Committee (IC) meeting tomorrow. To build a solid foundation for the decision, she launches ChatGPT's deep research capability. She configures the deep research prompt to query trusted financial connectors and reputable websites, prioritizing SEC filings, earnings transcripts, and investor presentations to produce a comprehensive dossier.

- ChatGPT automatically constructs an executable research plan before generating the report, allowing the user to inspect or adjust the approach beforehand. - The generated deep research dossier surface key qualitative and structural insights, such as QXO acquiring Kodiak—a crucial factor impacting the investment thesis.

---

[03:41](https://www.youtube.com/watch?v=m2TV8slGQKc&t=221s) Converting Context into Financial Models via Excel Integration

To turn qualitative research into actionable numbers, the analyst leverages ChatGPT to construct a full Excel workbook. Using a specialized enterprise prompt preset called the "Blossom Bank three-statement model" skill, ChatGPT automatically applies the firm's exact modeling conventions. This eliminates the need for long, repetitive prompts by embedding corporate modeling standards directly within the custom skill.

- ChatGPT outputs an interactive Excel workbook rather than a static, black-box summary. - The model embeds explicit comments detailing assumptions and data sources directly within the sheet. - All output values retain full dynamic formulas, enabling analysts to audit the underlying calculations manually. - When evaluating bear, base, and bull valuation scenarios, ChatGPT processes the context, updates the model, and provides a clear summary of changes, key numbers, and strategic insights. - Workflows that previously took hours or days are compressed into minutes, allowing financial teams to move at market speed.

---

[06:21](https://www.youtube.com/watch?v=m2TV8slGQKc&t=381s) Delivering Audit-Ready Decks for the Investment Committee

The final step requires distilling complex quantitative outputs into a presentation suited for senior leadership. Utilizing the "Blossom Bank deck build" skill, the analyst transfers the deep research dossier and valuation model into a slide deck structured according to corporate presentation standards. She enables the "extended thinking" setting, allowing ChatGPT to execute deeper reasoning cycles to ensure high-quality slide generation.

- The generated deck opens with a clear strategic recommendation for the Investment Committee: "approve with conditions." - The presentation automatically renders visual charts to illustrate key decision points and valuation outcomes clearly. - Comprehensive presenter notes are populated with explanatory comments detailing why specific conclusions were reached, establishing end-to-end auditability. - Anani emphasizes that AI does not replace human judgment; rather, it frees up time so professionals can focus on making well-informed decisions.

## Notable quotes > [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) 75% of ChatGPT users were able to complete tasks they previously could not accomplish. — Stephanie Anani

> [01:29](https://www.youtube.com/watch?v=m2TV8slGQKc&t=89s) OpenAI Banker Bench — Stephanie Anani

> [06:21](https://www.youtube.com/watch?v=m2TV8slGQKc&t=381s) approve with conditions — Stephanie Anani

## Mentioned Stephanie Anani, OpenAI, ChatGPT, Dow Jones, LSEG, London Stock Exchange Group, S&P, Excel, GPT-5.5, OpenAI Banker Bench, Blossom Bank, QXO, Investment Committee, Kodiak, SEC

[Source](https://www.youtube.com/watch?v=m2TV8slGQKc)

中文简报

OpenAI推出GPT-5.5金融功能，实现华尔街建模自动化深度研究工具与直接Excel集成在保持完全可审计性的同时，将耗时数天的财务建模和演示汇报流程压缩至数分钟。 OpenAI解决方案工程师Stephanie Anani展示了金融服务机构如何利用AI消除行政管理开销，并在五个关键业务维度上提升员工的基础能力。该平台由核心智能中内置金融背景知识的前沿模型GPT-5.5驱动，具备深度研究能力、定制金融应用连接器，以及可生成审计就绪动态工作簿的Excel集成功能。通过模拟Blossom Bank投资分析师筹备投资委员会会议的现场演示，Anani展示了ChatGPT如何将涵盖深度研究、三表建模以及制作精美可审计演示文稿的复杂工作流，从数小时或数天压缩至数分钟。最终，这种集成减轻了日常运营负担，使专业人士能够专注于高价值的商业决策与人类判断。 AI集成旨在提升员工的基础能力和业务水平，而不仅仅是加快常规工作速度。 (0:06) OpenAI近期的一项调查显示，75%的ChatGPT用户能够完成他们此前无法完成的任务。 (0:06) ChatGPT直接集成了Dow Jones、LSEG和S&P等主要金融数据提供商，并支持原生Excel工作流。 (1:29) OpenAI推出了GPT-5.5，该模型将原生金融背景知识直接嵌入其核心模型智能中。 (1:29) 基于“OpenAI Banker Bench”基准测试，GPT-5.5在金融分析方面展现出了最先进的性能表现。 (1:29)

Reading summary

OpenAI Solutions Engineer Stephanie Anani asserts that embedding financial context and enterprise tools into AI models allows financial institutions to eliminate operational overhead and raise employee baseline capabilities without displacing human judgment. Embedding Financial Context and Core Intelligence (1:29) Enterprise adoption in high-stakes sectors like financial services depends on model access to verifiable, current information. Anani notes that OpenAI addresses this by integrating connectors to major data providers—including Dow Jones, London Stock Exchange Group (LSEG), and S&P—and embedding ChatGPT directly into Excel to enable natural language workbook manipulation. According to Anani, these workflows are powered by the GPT-5.5 model, which incorporates financial processes directly into its base intelligence rather than as an external layer. OpenAI benchmarked this capability using its proprietary 'OpenAI Banker Bench' to evaluate performance against competing frontier models on complex financial operations. Autonomous Deep Research for Investment Theses (2:54) In a simulated investment workflow for 'Blossom Bank,' Anani demonstrated using ChatGPT's deep research capability to evaluate a company, QXO, ahead of an Investment Committee meeting. The system prioritizes primary regulatory and corporate sources, including SEC filings, earnings transcripts, and investor presentations retrieved via trusted data connectors. Before compiling the final dossier, the tool generates an inspectable research plan, allowing analysts to review and modify search vectors. In the demonstration, the research dossier identified pivotal qualitative developments, such as QXO's acquisition of Kodiak, that directly inform the strategic investment thesis. Dynamic Spreadsheet Modeling and Auditability (3:41) To convert qualitative findings into financial projections, the workflow deploys custom enterprise prompts, such as the 'Blossom Bank three-statement model' skill. This skill encodes internal modeling standards and accounting conventions directly into the prompt layer, removing the need for manual, repetitive configuration. Rather than outputting static calculations, the model generates dynamic Excel workbooks with functioning mathematical formulas, explicit assumption notes, and sourced citations. This structure permits manual auditability across bear, base, and bull valuation scenarios, compressing financial modeling tasks that historically required hours or days into minutes. Synthesizing Findings into Decision-Ready Presentations (6:21) The final stage translates financial models and qualitative research into an Investment Committee slide deck via a dedicated 'deck build' skill. By activating an 'extended thinking' setting, the system runs deeper reasoning cycles to align slide structures with institutional standards. The resulting presentation includes data visualizations, an explicit committee recommendation ('approve with conditions'), and comprehensive presenter notes explaining the analytical rationale behind each conclusion to ensure end-to-end accountability. Anani concludes that while an OpenAI survey indicates 75% of ChatGPT users complete tasks they previously could not execute, the technology is designed to eliminate non-judgmental administrative burden rather than replace human oversight in final decision-making.

精炼阅读摘要

OpenAI解决方案工程师Stephanie Anani指出，AI赋能金融服务的核心在于接管耗费时间且无需人类判断的运营与行政负担，从而提高员工的基础生产能力并聚焦于高价值业务驱动因素；OpenAI调查显示，75%的ChatGPT用户得以完成此前无法独立承担的任务。金融上下文对接与底层模型构建 (1:29) Anani表示，金融行业的高风险决策必须依赖可靠且时效性强的数据基础。为此，OpenAI开发了针对道琼斯（Dow Jones）、伦敦证券交易所集团（LSEG）以及标普（S&P）等主要金融数据提供商的应用连接器，并直接在Excel内嵌入ChatGPT，使专业人员能够运用自然语言查询并操作工作簿。在底层架构上，相关能力由GPT-5.5前沿模型支撑。OpenAI将金融工作流程直接植入模型核心智能而非外挂处理，并通过其构建的“OpenAI Banker Bench”基准评估，验证了GPT-5.5在复杂金融任务上优于同类前沿模型的表现。深度研究功能支持投资背景调研 (2:54) 在模拟“Blossom Bank”分析师为投资委员会紧急评估QXO项目的演示中，ChatGPT的深度研究功能被用于自动建立全面的档案。在正式生成报告前，系统会自动拟定一份可执行的研究方案，允许分析师在执行前核查并调整方向。该研究工具优先调取SEC文件、业绩电话会纪要与投资者演示文稿等高信度来源。生成的深度研究档案能够梳理关键结构性信息，例如QXO对Kodiak的收购动向，为后续的投资论证提供定性依据。 Excel集成下的动态三表模型推演 (3:41) 为将调研信息量化，分析师调用了预设的“Blossom Bank三表模型”技能。该企业预设技能直接固化了机构内部的建模标准，避免了冗长且重复的提示词编写，将过去耗时数小时或数天的流程缩减至数分钟内完成。 ChatGPT生成的Excel工作表具备完整交互能力而非静态摘要，单元格内附有详尽的假设说明与数据来源注释，并保留了所有底层动态计算公式以供人工审计。当进行悲观、基准及乐观三种情景测算时，模型能自动响应上下文更新报表，并清晰呈现变动摘要与核心洞见。投委会汇报生成与全流程可审计性 (6:21) 在决策交付阶段，分析师使用“Blossom Bank幻灯片构建”技能，将深度研究档案与估值模型整合为符合企业汇报规范的演示文稿。通过启用“拓展思考”（extended thinking）设置，ChatGPT展开了更深层次的推理循环，以保证材料的逻辑与生成质量。系统生成的演示材料包含附带条件的批准建议以及关键决策与估值图表，并在演讲者备注中完整记录了得出结论的支撑逻辑。这一设计确保了从原始数据检索、财务建模到最终建议的端到端可审计性。 Anani总结认为，AI技术的介入并非替代人类的专业判断，其核心价值在于分担日常繁复流程，让金融从业人员能够将精力投入到高质量的事实研判与审慎决策之中。

Detailed notes — zh

## Background 本视频由 OpenAI 于 2026 年 7 月 3 日发布在 YouTube 官方频道。内容聚焦于金融机构的落地部署策略与组织转型，深入探讨金融企业如何与 OpenAI 携手将 AI 嵌入日常工作流，在强监管环境中加速创新并规模化落地受信任的 AI 系统。 ## TL;DR OpenAI 解决方案工程师 Stephanie Anani 阐述了如何利用基于 GPT-5.5 与行业标准数据连接器（Dow Jones、LSEG、S&P）的 AI 基础设施，将金融员工从机械繁琐的无判断力劳动中解放出来。在现场演示中，她以虚构机构“Blossom Bank”评估 QXO 并购案为例，展示了分析师如何借助 Deep Research 快速挖掘核心定性事实，并利用 Excel 原生 ChatGPT 与自定义技能自动化构建包含透明公式与批注的三张表财务模型。随后，系统在数分钟内完成悲观、中性、乐观三套情景推演，并借助深度思考模式（Extended Thinking）一键输出附带完整证据推理链的投委会决策汇报材料。整套工作流展示了受信任上下文与企业技能如何打通投资研判全闭环，印证了 AI 旨在释放时间以强化而非取代人类的专业审慎判断。 ## Atmosphere & style 整个分享呈现出极其专业、严谨且务实的叙事基调，兼具前沿科技感与现代投资银行的工作节奏。Stephanie 的讲解从容自信、逻辑缜密，没有空洞浮夸的技术概念兜售，而是直接依托真实的 Excel 与投研场景展开一气呵成的实机演示。观看体验兼具高信息密度与流程上的顺滑感，真实传达出尖端 AI 将专业分析师从繁重建模内卷中解救出来的松弛与掌控力。 ## Bulletin - [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) OpenAI 调查显示 75% 的 ChatGPT 用户借助 AI 实现了此前无法独立完成的工作能力突破，技术核心价值在于全面拉升组织的运营基准水平。 - [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) OpenAI 推出集成 Dow Jones、LSEG 及 S&P 的金融数据连接器，依托 GPT-5.5 原生金融智能并在 BankerBench 基准评测中树立行业新标杆。 - [02:44](https://www.youtube.com/watch?v=m2TV8slGQKc&t=164s) 通过 Deep Research 挖掘并购情报与自定义技能建模，投研团队可将原本耗费数天的财务建模与多情景推演压缩至数分钟内完成。 - [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) 方案实现 Excel 公式级透明审计与投委会汇报决策链回溯，强调 AI 并非取代人类判断，而是释放精力以支持审慎决策。 ## Key takeaways - [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) 企业痛点在于员工被大量不需要人类专业判断的机械流程捆绑，AI 的使命是接管此类任务以释放核心生产力。 - [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) 调查表明 75% 的 ChatGPT 用户能完成此前无法独立实现的工作，AI 正从单纯的提效工具转变为重构组织能力底座的关键。 - [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) 针对金融业务对权威时效数据的要求，OpenAI 上线对接 Dow Jones、LSEG、S&P 等顶级金融数据源的专属连接器。 - [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) GPT-5.5 在底层直接内嵌金融流程推理能力，并在专有评测基准“OpenAI BankerBench”中刷新了行业领先表现。 - [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) 深度嵌入 Excel 的 ChatGPT 允许分析师直接使用自然语言与复杂工作簿无缝交互，极大降低了高级数据处理的操作门槛。 - [02:44](https://www.youtube.com/watch?v=m2TV8slGQKc&t=164s) Deep Research 功能通过自主生成并执行检索计划，能从 SEC 申报与财报会议中敏锐捕捉到改变估值逻辑的关键并购线索。 - [02:44](https://www.youtube.com/watch?v=m2TV8slGQKc&t=164s) 通过自定义银行技能（Skill）固化机构专属的建模规范与会计惯例，分析师无需繁琐 Prompt 即可一键生成合规的三张表模型。 - [02:44](https://www.youtube.com/watch?v=m2TV8slGQKc&t=164s) AI 生成的 Excel 模型拒绝黑盒化，所有推导均保留标准公式，并在关键单元格附带假设依据与数据源批注以满足严苛合规审计。 - [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) 在悲观、中性、乐观三种情景假设推演中，Excel 端模型可将传统需耗时数小时至数天的交叉核算大幅压缩至数分钟完成。 - [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) 借助深度思考模式（Extended Thinking），模型能够调用充足推理算力自动构建逻辑缜密的投委会决策汇报演示文稿。 - [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) 演示材料在给出明确结论的同时，通过演讲者备注完整保留底层逻辑与取舍原因，实现决策链条自底向上的完整可追溯性。 - [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) 端到端打通“受信任上下文 → 深度调研 → 财务建模 → 商业决策点”闭环，AI 的定位是赋能而非替代人类专家的终局审慎判断。 ## Detailed walkthrough ### [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) 金融服务的 AI 赋能与核心底座在企业日常运营中，员工往往耗费大量时间处理繁琐但缺乏实质判断力的机械任务。OpenAI 解决方案工程师 Stephanie Anani 指出，各行业客户反复反馈同一痛点：员工正被那些“必须完成、却完全不需要人类专业判断的工作”牢牢捆绑。AI 的核心价值正是接管这类流程，从而释放员工的精力，让他们专注于能够真正驱动商业增长的高价值决策。OpenAI 最近的一项调查显示，“75% 的 ChatGPT 用户能够做到他们以前无法做到的事情”；这绝非单纯加快工作效率，而是本质上“提升了整个组织的运营水平”。为了使金融机构充分发挥这一潜能，OpenAI 针对金融服务量身打造了多维度的系统性支持。金融业务的基石在于权威且具有时效性的上下文数据。为此，OpenAI 推出了对接 Dow Jones（道琼斯）、LSEG（伦敦证券交易所集团）以及 S&P（标普）等顶级金融数据源的应用连接器（App Connectors），使模型能实时吸纳受信任的市场情报。同时，针对金融行业对电子表格的高度依赖，OpenAI 构建了深度集成于 Excel 的 ChatGPT，允许分析师直接使用自然语言与复杂工作簿无缝交互。支撑这些前沿能力的是新一代 GPT-5.5 模型，团队直接将金融服务业务流程内嵌至模型的原生智能中，并通过专门构建的评测基准“OpenAI BankerBench”验证其在各类专业金融任务中的顶级表现。 - **工作赋能统计**：75% 的 ChatGPT 用户借助 AI 实现了此前无法独立完成的工作能力突破。 - **数据源生态**：首批上线包括 Dow Jones、LSEG、S&P 等在内的行业标准金融数据连接器，保障底层事实的权威度。 - **评测与模型基准**：由 GPT-5.5 提供原生推理支撑，并在衡量前沿模型金融任务表现的“OpenAI BankerBench”测试中刷新行业标杆。 --- ### [02:44](https://www.youtube.com/watch?v=m2TV8slGQKc&t=164s) 投资分析实战：深度调研与三张表财务建模 ```mermaid flowchart LR accTitle: 投资研究端到端工作流 accDescr: 展示从外部权威数据源检索，经由深度研究与 Excel 建模，最终生成决策汇报材料的完整流程。 subgraph S1["数据与调研"] A["权威数据源 (SEC文件/财报会/S&P等)"] --> B["Deep Research 生成投资档案"] end subgraph S2["财务建模"] B --> C["Blossom Bank 技能库 三张表财务模型"] C --> D["Excel 工作簿 含透明公式与假设批注"] end subgraph S3["决策交付"] D --> E["投委会 PPT (含图表与演讲备注)"] end ``` 在现场演示中，Stephanie 模拟了一位虚构机构“Blossom Bank”的投资分析师。面对主管紧急发来的即时需求——为次日召开的投资委员会（IC）会议准备一份关于标的公司 QXO 的最新分析视角，分析师首先调用 ChatGPT 的 Deep Research（深度研究）能力构建全面事实画像。分析师通过配置信任站点并联动专属连接器，将调研重点锚定在 SEC 申报文件、财报电话会议纪要以及投资者演示文稿上。ChatGPT 会先自主梳理出一套完整的检索计划供分析师审核确认，随后迅速整合出长篇投资档案。在此过程中，报告精准捕捉到了一个具有决定性影响的关键线索：QXO 正在推进对 Kodiak 的收购交易，这一重大并购将根本性地改变其估值逻辑。调研完成后，挑战转变为如何将海量定性信息快速量化为专业财务模型。通过调用银行内置的“Blossom Bank 三张表模型”自定义技能（Skill），分析师无需撰写繁琐冗长、反复调试格式的 Prompt，因为机构特有的建模规范、会计惯例已固化在该技能之中。用户只需输入一句简要指令，ChatGPT 即可在 Excel 中自动化构建标准财务工作簿。Stephanie 特别强调，生成的结果绝不是不可追溯的“黑盒子”，表格中每一项推导均保留了标准计算公式，模型还在关键单元格中附带了详细的假设依据与数据源批注，确保整个财务分析过程具备严密的合规审计能力。 --- ### [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) 情景分析与投委会汇报：构建可审计的决策链在与投资主管进一步沟通后，团队明确了投委会的核心诉求：全面量化评估 QXO 在悲观（Bear）、中性（Base）和乐观（Bull）三种情景下的业务表现。分析师将管理层的假设直接输入 Excel 端的 ChatGPT，模型迅速理解情景差异并完成交叉核算，即刻输出变动摘要、关键财务指标以及核心洞察。这项以往通常需要分析团队耗费数小时乃至数天反复搭建校验的情景推演，现在“短短数分钟内即可完成”，使投研节奏能够近乎实时地跟上瞬息万变的市场步伐。为了向投资委员会清晰传达核心信号，下一步工作是将复杂的数字模型转化为精简的决策汇报材料。分析师激活了“Blossom Bank 演示文稿生成技能”，并开启了深度思考模式（Extended Thinking），让模型调用充足的推理算力反复推敲结构。结合此前的 Deep Research 调研底稿与最新估值结论，ChatGPT 最终生成了一份版式精美的投委会决策 Deck：材料首页给出了清晰明了的结论建议——“附条件批准”（approve with conditions），内页则由系统自动绘制了直观反映估值区间与决策节点的图表。尤为关键的是，演示文稿的演讲者备注（Presenter Notes）中完整记录了每一项结论背后的推理逻辑与取舍原因，使得投委会成员能够沿着证据链自底向上回溯整个决策过程。Stephanie 总结道，这套流程完整跑通了“受信任上下文 → 深度调研 → 财务建模 → 商业决策点”的闭环。她强调：“AI 并不是要取代人类的专业判断，而是为了把一天的时间解放出来，让大家能有充分的时间做出这些审慎的判断。” ## Notable quotes > [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) 必须完成、却完全不需要人类专业判断的工作 — Stephanie Anani > [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) 75% 的 ChatGPT 用户能够做到他们以前无法做到的事情 — Stephanie Anani > [00:06](https://www.youtube.com/watch?v=m2TV8slGQKc&t=6s) 提升了整个组织的运营水平 — Stephanie Anani > [02:44](https://www.youtube.com/watch?v=m2TV8slGQKc&t=164s) 黑盒子 — Stephanie Anani > [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) 短短数分钟内即可完成 — Stephanie Anani > [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) 附条件批准 — Stephanie Anani > [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) 受信任上下文 → 深度调研 → 财务建模 → 商业决策点 — Stephanie Anani > [05:33](https://www.youtube.com/watch?v=m2TV8slGQKc&t=333s) AI 并不是要取代人类的专业判断，而是为了把一天的时间解放出来，让大家能有充分的时间做出这些审慎的判断。 — Stephanie Anani ## Mentioned Stephanie Anani, OpenAI, ChatGPT, GPT-5.5, OpenAI BankerBench, Dow Jones, LSEG, S&P, Excel, Deep Research, Blossom Bank, QXO, Kodiak, SEC ## Video description Deployment strategy and organizational transformation: how financial institutions can partner with OpenAI to embed AI into daily workflows, accelerate customer-facing innovation, and scale trusted AI systems across regulated environments. [Source](https://www.youtube.com/watch?v=m2TV8slGQKc)

Source quotes

[0:06] Stephanie Anani: 必须完成、却完全不需要人类专业判断的工作 [0:06] Stephanie Anani: 75% 的 ChatGPT 用户能够做到他们以前无法做到的事情 [0:06] Stephanie Anani: 提升了整个组织的运营水平 [2:44] Stephanie Anani: 黑盒子 [5:33] Stephanie Anani: 短短数分钟内即可完成 [5:33] Stephanie Anani: 附条件批准 [5:33] Stephanie Anani: 受信任上下文 → 深度调研 → 财务建模 → 商业决策点 [5:33] Stephanie Anani: AI 并不是要取代人类的专业判断，而是为了把一天的时间解放出来，让大家能有充分的时间做出这些审慎的判断。

Source quotes

[0:06] Stephanie Anani: 75% of ChatGPT users were able to complete tasks they previously could not accomplish. [1:29] Stephanie Anani: OpenAI Banker Bench [6:21] Stephanie Anani: approve with conditions

Original transcript

[0:01] [music] [0:06] >> Hi everyone. I'm Stephanie, a solution [0:08] engineer here at OpenAI. [0:11] I work with some of our largest [0:12] customers to show them how they can get [0:14] value from this technology. [0:17] So, Lee has spoken to you about how [0:19] workspace agents can really transform [0:21] your team. [0:22] In this section, I want to get a bit of [0:24] a deep dive into how your individuals [0:27] can get lots of value from AI. [0:31] You tell us the same thing over and over [0:33] again. [0:34] Your employees are spending time doing [0:37] work that needs to happen, but work that [0:39] doesn't actually require human judgment. [0:42] This is where we think AI comes in. [0:45] AI can free up their time so that they [0:48] can work on the things that actually [0:50] drive value in your business. [0:54] A recent survey we did found out that [0:56] 75% [0:58] of ChatGPT users were able to do things [1:01] that they couldn't do before. [1:04] That's not just doing your job faster. [1:06] That's increasing the operating level, [1:08] and we think that your employees should [1:10] be working at this increased capability. [1:14] In order to do that, we think they need [1:16] support on five key dimensions. [1:20] And today, I really want to specifically [1:22] share why ChatGPT is the platform for [1:25] financial services. [1:29] First is context. [1:31] To get your jobs done, you need access [1:33] to trusted sources with up-to-date [1:36] information. [1:37] We've worked hard to make sure that [1:40] we've connected to some of the most [1:42] trusted pieces of information in [1:44] financial services. So, for example, we [1:46] have app connectors to Dow Jones, LSEG, [1:49] and S&P. This is just the start of what [1:52] we're doing, and we're continuing to [1:53] work on this. [1:56] Additionally, we know that a lot of [1:58] financial services work happens in [2:01] Excel. So, we built chat GPT in Excel [2:04] that you can interact with your [2:06] workbooks in natural language. [2:11] What is underpinning these capabilities [2:13] is GPT 5.5. [2:16] We went a step further this time, and we [2:18] decided to embed financial services [2:21] processes directly in the intelligence [2:24] of the model. [2:26] In order to test this over time, we [2:28] developed OpenAI Banker Bench. This [2:31] gives you the ability to see how [2:33] frontier models perform on financial [2:35] services tasks. You can see that GPT 5.5 [2:39] is state-of-the-art above all the other [2:41] frontier models. [2:43] Now, I love telling you about all these [2:45] advancements, but I actually want to [2:47] show you. So, let's head over to the [2:48] demo. [2:54] So, I am an investment analyst at [2:57] Blossom Bank. My manager has just sent [2:59] me this message. We need a fresh view on [3:01] QXO for an IC meeting tomorrow. [3:07] So, I go on to chat GPT, and the first [3:10] thing I know, if we ever want to have an [3:13] effective investment decision, is [3:15] gathering as much context as possible. [3:19] You all know and love deep research. So, [3:22] we're going to use this capability to [3:23] gain more information around QXO. [3:27] We're also going to have the ability to [3:29] connect to some of these apps so that we [3:31] ensure that our trusted sources are used [3:33] in this deep research. [3:36] Lastly, we can connect it to sites that [3:38] we trust and we respect. [3:41] So, here I have my prompt. We're going [3:43] to create an investment dossier deep [3:45] dive, and I I focus on things like SEC [3:48] filings, earning transcripts, and [3:50] investment presentations. [3:53] So, what happens behind the scenes is [3:55] that ChatGPT creates a plan. [3:58] At this point, you have the opportunity [4:00] to go in and change anything. [4:02] But, I'm pretty happy with what I see [4:03] here. So, let's start the report. [4:08] Once the dossier is created, we can go [4:10] through it and try and find key [4:12] information that's necessary. [4:15] So, one thing I realized is that QXO is [4:17] acquiring Kodiak. That is going to have [4:20] a massive difference in whether we [4:22] should invest in them or not. [4:26] These reports are great, but how do we [4:28] actually turn this into numbers? [4:31] This is where we use ChatGPT to create [4:35] an Excel workbook for our investment [4:38] statement. So, here we have a specific [4:41] skill that we're using, the Blossom [4:42] Bank's three-statement model. [4:45] In this skill, we've decided exactly how [4:48] we like our three-statement models done. [4:51] And so, instead of me having to write a [4:53] detailed prompt telling ChatGPT exactly [4:56] how to do things, all of that knowledge [4:58] is within the skill, and then I can [4:59] write a brief prompt saying specifically [5:01] what I'd like to do. [5:06] ChatGPT is able to create an Excel [5:08] workbook. [5:10] Now, what I love about this is that it's [5:13] not just a black box. We can see that [5:16] ChatGPT has left comments about [5:18] assumptions and sources. [5:21] Another really great capability is that [5:24] all of the numbers actually have [5:26] formulas. So, you can go in and actually [5:29] audit these capabilities yourself. [5:33] So, I've had a meeting with my manager, [5:35] and we've decided that we're going to [5:37] take a very specific approach with this [5:39] IC. [5:41] We want to be able to measure the bear, [5:44] base, and bull case for QXO. [5:47] So, I've dumped all of that context into [5:50] um ChatGPT and Excel. [5:53] We see that it reads the context, it [5:54] decides what needs to be done, and then [5:57] I get a result. [5:59] I get a summary of what changed, key [6:02] numbers, and insights to glean from. [6:05] And this is incredible. [6:07] Before, this might have taken hours or [6:10] days. Now, I'm able to do this in [6:13] minutes with ChatGPT and Excel. So, now [6:16] we can move almost as fast as the [6:18] markets. [6:21] This is great, but for an IC, we know [6:24] that we really need to take out all of [6:26] the important signal and present it in a [6:28] way that can just inform the decisions [6:31] that are going to be made. [6:34] Our next step on this journey is [6:35] actually transforming this into a [6:37] presentation that's ready for the IC. [6:42] The first thing we use is a Blossom Bank [6:45] deck build skill. [6:47] So, this gives us the ability to put in [6:49] all the know-how as to how Blossom Bank [6:52] likes to create decks. [6:55] Then, I also put on extended thinking [6:57] because I want ChatGPT to really take [6:59] its time to make sure that we're getting [7:01] the output that we want. [7:05] Lastly, I make sure I add all of that [7:06] context. So, the deep research report [7:08] and also the valuation that we've done. [7:15] ChatGPT is able to create this deck. So, [7:18] let's have a look at at what's been [7:20] created. [7:22] Here, we can see there's a clear [7:24] decision, approve with conditions. [7:28] What I also love is that we've also [7:31] created charts in the presentation. So, [7:34] you can see clearly what the decision [7:37] points are and the valuation outcomes. [7:41] Another great capability is that in the [7:44] presenter notes, there actually comments [7:46] around why decisions were made. So, you [7:48] can actually audit this process end to [7:51] end. [7:53] So, what we've been able to see is not [7:56] just a feature run-through. It's going [7:59] from trusted contacts to deep research, [8:02] going from deep research to a financial [8:05] model, and going from a financial model [8:08] to decision points that actually move [8:11] the business forward. [8:14] AI does not replace the need for human [8:16] judgment. It just frees up the day so [8:19] that you can use the time to make those [8:21] judgments. [8:25] >> [applause]