Accend Sees AI Running 80% of Commercial Loan Decisions by 2027

The Underwriting Singularity: How AI Will Reshape Every Lending Decision by 2027 With Accend — Ai4

Ai4

Autonomous AI underwriting will compress commercial loan assessments from weeks to under an hour by 2027, turning turnaround velocity into the primary driver of market share while eliminating systemic accounting errors.

AI-assisted summary of The Underwriting Singularity: How AI Will Reshape Every Lending Decision by 2027 With Accend — original source. Check the original for full context.

Source published: 2025-08-28

Detailed notes

## Background Published on YouTube by Ai4 on August 28, 2025, this presentation features Pranjal Daga, Co-Founder of Accend. The session outlines 'The Underwriting Singularity,' explaining how AI-driven financial spreading and risk analysis will fundamentally transform commercial lending decisions by 2027 through massive operational speedups and risk mitigation.

## TL;DR Pranjal Daga illustrates how legacy underwriting software creates multi-million-dollar balance sheet distortions and explains how human-in-the-loop AI compresses two years of manual financial spreading into thirty days. Drawing on Jamie Dimon's 2008 risk oversight playbook, he outlines the three-stage shift from augmentation to full underwriting autonomy. Daga shows how sub-two-hour credit assessments create a compounding data flywheel that lets early adopters capture market share. He concludes with a concrete implementation roadmap and predicts that 80% of commercial lending decisions will be autonomous by 2027.

## Atmosphere & style The presentation carries an urgent, executive-level keynote tone that balances sobering cautionary tales of legacy banking errors with forward-looking optimism. Daga delivers his insights with sharp, strategic clarity, leveraging stark numerical contrasts—like a single $76 million sign-inversion error—to command attention. It feels like a high-stakes briefing tailored for risk officers and financial leaders, urging them to modernize before systemic velocity leaves their institutions behind.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) Human-in-the-loop AI processes 300,000 financial statement line items in under 30 days, a task requiring roughly two years of manual analyst effort. - [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) Legacy commercial spreading software produces severe data errors, exemplified by an inverted -$1M expense turned into a +$75M asset. - [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) Audits reveal that foundational errors affect over 10% of loan portfolios under legacy systems, distorting risk committee capital allocations. - [05:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=300s) Modern AI allows financial institutions to scale crisis-tested risk oversight from five manual daily reviews to 500 continuous automated reviews per day. - [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) Commercial lending transformation follows three stages: Augmentation (3x speedup), Automation (10x speedup), and Autonomy (100x speedup targeted for 2027). - [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) Autonomous underwriting allows institutions to perform preliminary evaluations of complex $50 million credit applications within two hours rather than weeks. - [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) AI creates a compounding flywheel where faster underwriting drives deal volume, feeding back richer data to continually calibrate risk models. - [15:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=900s) A pragmatic rollout begins by backtesting legacy error baselines and targeting high-volume, standardized portfolios before enterprise-wide deployment. - [15:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=900s) Underwriting velocity is shifting from a back-office cost-cutting measure into a primary driver of customer acquisition and win rates. - [15:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=900s) By 2027, an estimated 80% of commercial lending decisions will be executed by AI, reducing average decision turnaround to six hours or less.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) The Underwriting Singularity and the $76 Million Nightmare

AI has rapidly shifted from an optional experiment to an operational necessity in financial services. At Accend, the focus is helping fintechs and banks radically accelerate commercial underwriting. To illustrate the scale of this shift, processing 300,000 financial statement line items manually would take a team of human credit analysts roughly two years. Accend executes this same workload in under 30 days using a human-in-the-loop AI model. However, the core challenge in modern commercial lending extends beyond pure operational efficiency—it centers on catastrophic inaccuracies produced by legacy spreading and underwriting software.

During a backtest of five sample credit cases for a partner bank, Accend discovered a severe accounting distortion generated by an entrenched legacy vendor. The legacy system misclassified a negative $1 million expense item as a positive $75 million asset. This single sign-inversion error transformed a distressed financial profile into a phantom credit surplus, illustrating how automated underwriting without robust verification creates three-digit-million-dollar portfolio exposures. The bank immediately abandoned its three-year vendor contract. Subsequent audits across other financial institutions revealed that similar foundational errors affected more than 10% of audited loan portfolios, forcing risk committees to base major capital allocations on fiction rather than deep-rooted facts.

- **Workload Benchmark**: 300,000 financial statement line items processed in under 30 days with AI vs. 2 years of manual analyst time. - **The $76M Discrepancy**: A legacy system converted a -$1,000,000 expense into a +$75,000,000 credit, creating a $76 million accounting distortion. - **Portfolio-Wide Contagion**: Over 10% of portfolio files audited at another institution contained critical legacy data errors, exposing lenders to severe financial and reputational losses.

---

[05:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=300s) The Jamie Dimon Playbook and Scaling Risk Oversight

The blueprint for surviving modern systemic and credit shocks is rooted in the 2008 financial crisis. At an *Acquired* podcast live recording in New York featuring JPMorgan Chase CEO Jamie Dimon, Dimon highlighted the principles that allowed JPMorgan to emerge as a winner while peers like Lehman Brothers, Bear Stearns, and Washington Mutual collapsed. Central to Dimon's strategy was the concept of the "fortress balance sheet"—an operating posture focused on preparation rather than attempting to predict the exact timing of a macro downturn.

During the height of the 2008 crisis, JPMorgan replaced its standard five-week credit review cycle with risk committee meetings held five times every single day. In 2008, those daily reviews required exhausting, manual spreadsheet aggregation by dozens of analysts. Translating that strategy to modern AI capabilities changes the game: instead of five manual risk check-ins per day, institutions equipped with human-in-the-loop AI can conduct 500 continuous risk reviews daily. This allows real-time portfolio monitoring and risk mitigation at an unprecedented level of granularity.

```mermaid flowchart TD accTitle: The Credit Transformation Stages and Compounding Flywheel accDescr: Diagram illustrating the three stages of AI transformation and the compounding data advantage flywheel.

subgraph Stages["Stages of AI Transformation"] direction LR S1["Stage 1: Augmentation • 3x Speedup • Spreading assistance • Analyst baseline"] --> S2["Stage 2: Automation • 10x Speedup • End-to-end spreading • Smart money focus"] S2 --> S3["Stage 3: Autonomy (2027) • 100x Speedup • Instant assessments • Relationship focus"] end

subgraph Flywheel["The Compounding Advantage Flywheel"] direction TB F1["More Data & Intake"] --> F2["Faster Underwriting"] F2 --> F3["Higher Win Rate & Volume"] F3 --> F4["Smarter AI Decisions"] F4 --> F1 end ```

---

[10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) The Three Stages of AI Transformation and the Compounding Advantage

Commercial lending is progressing through three distinct evolutionary stages. The first stage, **Augmentation**, delivers a 3x speed improvement by assisting human analysts with basic financial statement spreading. While many lenders mistakenly treat augmentation as the finish line, it is merely the entry point. The second stage, **Automation**, unlocks a 10x improvement by fully automating end-to-end spreading workflows, which is where top-tier banks are currently allocating capital. The final stage, **Autonomy**, slated to arrive by 2027, will deliver a 100x improvement. In this autonomous paradigm, AI generates complete risk assessments in hours, allowing human credit officers to concentrate almost exclusively on borrower relationships, complex negotiations, and unusual edge cases.

To demonstrate this operational delta, modern AI-driven lenders can evaluate a complex $50 million commercial credit application between 2:00 PM and 4:00 PM on the same day, whereas traditional institutions still take weeks to complete initial spreading. This creates a compounding data flywheel: lenders with AI process more deals, capture more behavioral data, calibrate risk models faster, and win market share from slower incumbents.

- **Analyst Equivalence**: Processing 300,000 line items manually equals 600 continuous analyst days without sickness or vacation. - **Underwriting Velocity Example**: Complete preliminary assessment of a $50 million loan within a 2-hour window (2:00 PM to 4:00 PM). - **Compounding Flywheel**: Moving from 10 closed loans per month to 100+ loans without expanding back-office headcount.

---

[15:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=900s) The Monday Morning Action Plan and 2027 Predictions

Implementing an AI-first underwriting strategy does not require overhauling the entire bank overnight. Institutions must adopt a phased, pragmatic implementation roadmap starting immediately:

1. **Audit the Existing Error Baseline**: Backtest historical spreading data to uncover hidden calculation discrepancies and misclassified accounting items. 2. **Target High-Volume Portfolios**: Deploy AI automation on a single, standardized asset class or business segment rather than attempting an enterprise-wide rollout on day one. 3. **Benchmark Real Turnaround Speed**: Measure credit delivery timelines in hours rather than days or weeks to maintain competitive conversion rates. 4. **Pivot from Cost-Cutting to Market Acquisition**: Treat underwriting velocity as a primary customer acquisition lever to capture borrowers before competitors issue term sheets.

By 2027, the lending landscape will reach a definitive turning point: - **Credit Decision Velocity**: Industry average turnaround times will drop to six hours or less, with market leaders approving commercial credits in under one hour. - **Financial Statement Analysis**: Spreading time will fall from eight-plus hours down to under one hour, exceeding human expert accuracy benchmarks. - **Market Penetration**: Approximately 80% of all commercial lending decisions will be executed through AI-driven underwriting engines.

Lenders face a stark binary choice over the next two years: embrace autonomous underwriting to become the JPMorgan Chase of the modern era, or resist adoption and face the fate of Bear Stearns, Lehman Brothers, and Washington Mutual.

## Notable quotes > [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) By 2027, AI will fundamentally transform how every lending decision is made, creating what I call 'The Underwriting Singularity' – the point where AI becomes faster, more accurate, and more insightful than any human underwriter. — Pranjal Daga

> [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) Processing 300,000 financial statement line items manually would take a team of human credit analysts roughly two years. Accend executes this same workload in under 30 days using a human-in-the-loop AI model. — Pranjal Daga

> [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) The legacy system misclassified a negative $1 million expense item as a positive $75 million asset. This single sign-inversion error transformed a distressed financial profile into a phantom credit surplus. — Pranjal Daga

> [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) Subsequent audits across other financial institutions revealed that similar foundational errors affected more than 10% of audited loan portfolios, forcing risk committees to base major capital allocations on fiction rather than deep-rooted facts. — Pranjal Daga

> [05:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=300s) Central to Dimon's strategy was the concept of the 'fortress balance sheet'—an operating posture focused on preparation rather than attempting to predict the exact timing of a macro downturn. — Pranjal Daga

> [05:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=300s) Instead of five manual risk check-ins per day, institutions equipped with human-in-the-loop AI can conduct 500 continuous risk reviews daily. — Pranjal Daga

> [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) The first stage, Augmentation, delivers a 3x speed improvement... The second stage, Automation, unlocks a 10x improvement... The final stage, Autonomy, slated to arrive by 2027, will deliver a 100x improvement. — Pranjal Daga

> [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) Modern AI-driven lenders can evaluate a complex $50 million commercial credit application between 2:00 PM and 4:00 PM on the same day, whereas traditional institutions still take weeks to complete initial spreading. — Pranjal Daga

> [15:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=900s) By 2027, approximately 80% of all commercial lending decisions will be executed through AI-driven underwriting engines. — Pranjal Daga

> [15:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=900s) Lenders face a stark binary choice over the next two years: embrace autonomous underwriting to become the JPMorgan Chase of the modern era, or resist adoption and face the fate of Bear Stearns, Lehman Brothers, and Washington Mutual. — Pranjal Daga

## Mentioned Pranjal Daga, Accend, Jamie Dimon, JPMorgan Chase, Lehman Brothers, Bear Stearns, Washington Mutual, Acquired, Ai4

## Video description Pranjal Daga, Co-Founder, Accend

Abstracts

By 2027, AI will fundamentally transform how every lending decision is made, creating what I call "The Underwriting Singularity" – the point where AI becomes faster, more accurate, and more insightful than any human underwriter. The lending industry stands at a critical inflection point where early AI adopters are already seeing 90%+ time savings and unprecedented accuracy in financial statement analysis, while others risk being left behind in an increasingly competitive market. I will reveal the three key stages of AI transformation that leading institutions are implementing right now, share real-world data on ROI and risk reduction from AI underwriting deployments, and outline the specific technologies and strategies that will define the winners in the next two years. This talk will equip you with a clear roadmap to navigate your institution's AI transformation and position yourself as a leader in the new era of intelligent lending.

[Source](https://www.youtube.com/watch?v=zgdg_Rfgqq8)

中文简报

Accend预计到2027年80%的商业贷款决策将由AI完成自主AI核保将在2027年前将商业贷款评估时间从数周压缩至一小时以内，在消除系统性会计错误的同时，使周转速度成为市场份额的核心驱动力。 Pranjal Daga阐述了传统核保软件如何导致数百万美元的资产负债表失真，并解释了人机协同AI如何将耗时两年的手动财务报表拆分录入工作压缩至三十天。借鉴Jamie Dimon在2008年的风险监管策略，他概述了从辅助增强迈向完全自主核保的三阶段转变。Daga展示了耗时不足两小时的信贷评估如何构建复合数据飞轮，助力早期采用者抢占市场份额。他最后提出了具体的落地路线图，并预测到2027年将有80%的商业贷款决策实现自主化。传统信贷录入软件存在严重缺陷，曾将 100 万美元支出反转为 7500 万美元资产，且审计显示超 10% 的贷款档案受此类错误影响；而人机协同 AI 能在 30 天内完成人工需 2 年处理的 30 万行财报条目。 (0:00) 借鉴摩根大通在 2008 年金融危机中的风控机制，现代金融机构借助人机协同 AI，可将危机时期每天 5 次的高强度人工审查升级为每天 500 次的连续自动化全量监控。 (5:00) 商业信贷正经历“增强（提速 3 倍）”、“自动化（提速 10 倍）”与“自主化（提速 100 倍，面向 2027 年）”三阶段；自主化阶段可将 5000 万美元贷款的初审时间从数周缩减至 2 小时，构建数据与获客的复利飞轮。 (10:00) 机构推进 AI 转型应从回测历史错误基线与高周转标准化资产切入；预计到 2027 年，约 80% 的商业信贷决策将由 AI 驱动，行业平均信贷审批周期将降至 6 小时以内。 (15:00)

Reading summary

According to Accend, commercial lending is moving from manual, error-prone legacy spreading to autonomous AI underwriting, which the firm projects will compress decision turnaround times from weeks to hours and reshape institutional risk oversight by 2027. Legacy Spreading Inefficiencies and Critical Data Distortions (0:00) Manual financial statement spreading imposes severe operational bottlenecks. Accend notes that manually processing 300,000 financial statement line items requires roughly two years—equivalent to 600 uninterrupted analyst working days—whereas a human-in-the-loop AI model completes the same volume in under 30 days. Beyond labor constraints, legacy automated spreading software introduces significant accounting distortions. In a backtest of five credit cases for a partner bank, an established legacy vendor converted a negative $1 million expense into a positive $75 million asset, producing a phantom $76 million credit surplus that prompted the bank to terminate its three-year contract. Subsequent audits across financial institutions revealed that more than 10% of reviewed loan portfolios contained critical legacy data errors, causing credit committees to allocate capital against flawed financial baselines. Scaling Continuous Risk Reviews Using AI (5:00) The operational blueprint for managing credit shocks is modeled on JPMorgan Chase's response during the 2008 financial crisis. To maintain a "fortress balance sheet" focused on readiness rather than macro forecasting, CEO Jamie Dimon replaced JPMorgan's standard five-week credit review cycle with risk committee meetings held five times per day, relying on manual spreadsheet compilation by analyst teams. Accend argues that modern human-in-the-loop AI enables lenders to scale this defensive posture dramatically. Instead of five manual daily check-ins, automated systems can perform up to 500 continuous risk reviews per day across an entire portfolio, providing granular, real-time monitoring and early distress detection that were impossible under manual workflows. Three Evolution Stages and the Compounding Flywheel (10:00) Commercial underwriting is progressing through three structural phases. Stage 1, Augmentation, delivers a 3x speed improvement by assisting analysts with basic spreadsheet data extraction. Stage 2, Automation, achieves a 10x improvement via end-to-end spreading workflows—the current investment focus of leading institutions. Stage 3, Autonomy, projected for 2027, aims for a 100x speed increase, generating comprehensive risk assessments in hours so credit officers can focus primarily on borrower relationships, structured negotiations, and complex exceptions. This operational acceleration allows lenders to evaluate complex $50 million commercial loan applications within a two-hour window (such as 2:00 PM to 4:00 PM), rather than the multi-week timeline common among traditional lenders. Underwriting velocity triggers a compounding data flywheel: handling higher deal volumes yields more intake and behavioral data, which refines AI risk calibration and permits lenders to scale from 10 to over 100 closed loans monthly without adding back-office staff. Phased Implementation and 2027 Market Projections (15:00) Rather than attempting an immediate enterprise-wide system overhaul, Accend recommends a phased implementation plan: backtest historical spreading data to expose existing error rates, deploy AI automation across a single standardized asset class, benchmark credit delivery in hours, and treat underwriting velocity as a primary customer acquisition lever. By 2027, Accend forecasts that roughly 80% of all commercial lending decisions will be driven by AI engines. Industry-average credit decision turnaround is projected to drop to six hours or less, with leading institutions approving applications in under one hour. In addition, financial statement spreading times are expected to decline from over eight hours to less than one hour while exceeding human expert accuracy standards. Accend frames the next two years as a decisive turning point for commercial lenders, asserting that institutions adopting autonomous underwriting will build defensible market advantages, while those continuing to rely on legacy manual workflows risk severe competitive displacement.

精炼阅读摘要

商业核保科技公司 Accend 指出，商业信贷正从依赖传统表格工具与人工分析转向人机协同的人工智能驱动模式。该公司认为老旧系统存在系统性数据扭曲隐患，并预测到 2027 年全行业 80% 的商业借贷决策将由 AI 执行，审批周期将从数周压缩至数小时。传统信贷录入低效与千万级账目失真 (0:00) 商业信贷的传统财务报表录入（spreading）与核保正面临低效与高差错率的双重瓶颈。根据 Accend 的测算，人工团队手工处理 30 万条财务报表明细大约需要两整年时间，而借助人机协同（human-in-the-loop）AI 模型可在 30 天内完成同等工作量。然而，比效率低下更严重的是老旧财务录入软件中存在的灾难性错误。在对一家合作银行的 5 个贷款样本进行回测时，Accend 发现某主流传统供应商系统将一笔 100 万美元的负项费用误分类为 7500 万美元的正向资产，直接导致了 7600 万美元的账目失真，将一个财务处于困境的借款主体包装为拥有资金盈余的优质对象。该银行随后终止了与该供应商为期三年的合同。对其余金融机构的后续审计显示，超过 10% 的抽检信贷档案存在类似的底层数据缺陷，表明缺乏稳健验证的自动化工具正导致风控委员会依据虚假数据进行高额资本配置。借鉴堡垒资产负债表实现高频实时风控 (5:00) 应对系统性与信贷风险的防御逻辑可追溯至 2008 年金融危机。摩根大通（JPMorgan Chase）首席执行官 Jamie Dimon 曾在播客中强调，该行在贝尔斯登、雷曼兄弟和华盛顿互惠银行崩溃中得以幸存的核心在于构建“堡垒资产负债表”（fortress balance sheet），其关键在于立足全方位防御准备，而非试图预测宏观下行拐点的具体时点。在 2008 年危机高潮期，摩根大通将常规的 5 周信贷审查周期缩短为每日召开 5 次风险委员会会议，当时该机制依赖数十名分析师耗费大量精力手工汇总电子表格。Accend 提出，现代人机协同 AI 能够将该危机应对策略转化为日常机制：金融机构能够实现每日 500 次的连续风险审查，以更细的颗粒度对贷款资产组合实施全天候实时监控与风险缓释。信贷智能化三阶段与数据复利飞轮 (10:00) Accend 将商业信贷智能化划分为三个发展阶段：第一阶段为“增效”（Augmentation），通过辅助分析师进行财务报表数据提取带来 3 倍提速，这仅是当前行业的基线；第二阶段为“自动化”（Automation），实现端到端的完整报表录入，带来 10 倍提速，是当前头部银行重点投入的方向；第三阶段为预计 2027 年到来的“自主化”（Autonomy），将实现 100 倍的提速，AI 能在数小时内生成完整的风险评估，促使人类审批官转向客户关系维护、复杂谈判与极端边缘案例处理。以具体业务场景为例，现代 AI 驱动型机构可在当天下午 2:00 至 4:00 的两小时窗口内完成一项 5000 万美元商业信贷的初步评估，而传统机构仍需数周。审批效率的提升进一步构建了数据复利飞轮：机构处理的交易越多，沉淀的借款人行为数据越丰富，风险模型调校越快，从而在不增加后台人手的前提下将月放款量从 10 笔扩充至 100 笔以上，加速争夺传统同行的市场份额。渐进落地路径与2027年行业格局预测 (15:00) 部署 AI 核保系统无需在全行范围内一蹴而就，Accend 建议机构采取四步落地路径：首先回测历史财务录入数据以审计基线错误，排查隐藏的计算偏差与分类错误；其次选择单一大体量且标准化的资产类别作为突破口，而非全业务铺开；第三是以“小时”为单位重新衡量信贷交付周期；最后将核保速度视作核心获客杠杆，在竞品出具意向书之前锁定借款人。针对 2027 年的行业格局，Accend 作出预测：商业贷款平均决策周期将降至 6 小时以内，领先机构将在 1 小时内完成审批；财务报表录入耗时将从 8 小时以上降至 1 小时以内，且准确率超过人类专家水平；约 80% 的商业借贷决策将由 AI 引擎执行。Accend 宣称，未能跟进自主核保转型的机构将面临重蹈 2008 年危机落后机构覆辙的风险。 Accend 论证了商业信贷从人工录入迈向自主化核保的潜在效率与数据复利优势，但其给出的 2027 年核心指标（如 80% 决策渗透率与 100 倍加速）本质上仍属于技术供应商的前瞻性预测。金融机构在落地过程中，仍需面对跨周期模型有效性验证、极端边缘风险判定以及遗留系统替换等实际复杂性的检验。

Detailed notes — zh

## Background 该视频由频道 Ai4 于 2025 年 8 月 28 日在 YouTube 平台发布，主讲人为金融科技公司 Accend 联合创始人 Pranjal Daga。演讲围绕“2027 年核保奇点”展开，深入探讨了 AI 如何在商业信贷与核保决策中实现 90% 以上的时间节约与前所未有的报表分析准确率，并为金融机构梳理了迈向智能化信贷的三阶段转型路线图与落地策略。 ## TL;DR Pranjal Daga 首先揭示了传统遗留系统在财务数据提取中可能导致高达 7600 万美元单笔严重算错的隐蔽风险，并对比了人机协同 AI 将两年分析工作量压缩至 30 天的高效表现。演讲借鉴摩根大通 CEO 杰米·戴蒙在 2008 年危机中的风控哲学，指出 AI 能将人工每日 5 次的风险复核跃升至自动化 500 次巡检。此外，他系统阐述了信贷核保从辅助增强（3x）、流程自动化（10x）到 2027 年完全自主化（100x）的三阶段演进路径，并提出了周一早晨行动指南与 2027 年行业超 80% 决策由 AI 驱动的终局预测。 ## Atmosphere & style 整场演讲呈现出兼具危机警示感与战略前瞻性的专业基调，语调沉稳笃定且数据详实。开篇通过真实发生的巨额数据差错案例迅速建立紧迫感与戏剧张力，让在座高管意识到遗留系统的致命短板；随后无缝过渡至华尔街危机复盘与务实落地框架，展现出极强的商业洞察力和对金融工程进化的信心，为观众营造出一种站在行业范式大转移转折点上的振奋体验。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) 金融科技公司 Accend 借助人机协同（Human-in-the-loop）AI 架构，在 30 天内完成了传统人工分析团队需耗费两年（约 600 个工作日）才能处理的 30 万条财务报表分录。 - [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) 传统遗留核保系统存在严重的质量隐患，在实际案例中甚至将 -100 万美元的支出项误算为 +7500 万美元收益，造成 7600 万美元的惊天差错，且受测机构中超 10% 的信贷资产组合受此类错误影响。 - [05:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=300s) 汲取摩根大通 CEO 杰米·戴蒙在 2008 年危机中的管理策略，AI 技术可将当年依靠人工加班进行的每日 5 次风险复核，升级为全自动、高并发的每日 500 次实时系统级巡检。 - [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) 信贷核保 AI 变革经历三大阶段：第一阶段“辅助增强”（提效 3 倍）、第二阶段“自动化展开与分析”（提效 10 倍），以及预计 2027 年到来的第三阶段“自主化”（提效 100 倍）。 - [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) 在人机协同的自主化阶段，一笔 5000 万美元规模的大型信贷交易的前期完整财务评估可在短短 2 小时内完成，人类专家将精力主要转向客户关系维护与极端边缘案例审批。 - [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) 早期布局 AI 的机构正在形成数据与业务的复合优势飞轮：更多的底层财务数据带来更高交易吞吐和更优决策质量，反哺 AI 模型持续进化，推动信贷组合规模实现 10 到 100 倍扩张。 - [13:30](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=810s) 落地行动建议包括立即排查内部报表识别的符号与数据错误、选定业务量最大的单一资产板块先行试点，并建立以“小时”为单位的时效监测体系以争夺获客优势。 - [13:30](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=810s) 到 2027 年信贷决策平均审批时间将缩短至 6 小时内（领跑者达 1 小时内），财报深度分析压缩至 1 小时内且准确率超越顶尖人类分析师，AI 驱动决策将覆盖超 80% 的商业信贷交易。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) 传统系统的“7600万美元噩梦”与人机协同的突破在金融服务行业中，AI 已经从可有可无的创新尝试转变为关乎生存的必修课。金融科技公司 Accend 的联合创始人 Pranjal Daga 指出，公司致力于协助银行和金融科技机构加速企业客户的信贷审批与核保流程。仅在过去的一个月内，Accend 便处理了来自各类财务报表的超过 30 万条独立分录（line items）。如果让传统的人工分析师团队来完成同样的工作量，整整需要耗费两年时间，而通过人机协同（Human-in-the-loop）的 AI 架构，Accend 在不到 30 天的时间内就完成了全量处理。然而，相比于处理速度从两年骤降至 30 天的效率飞跃，更令金融机构高管震惊的是现有遗留系统（Legacy Providers）所带来的灾难性质量隐患——在实际案例中，遗留系统对净收入的严重算错直接导致了灾难性的信贷决策。在深入剖析这一被称为“7600 万美元噩梦”的真实事故时，Pranjal Daga 直言调查结果甚至“让我感到脊背发凉”（made my blood go very cold）。Accend 对一家业务量庞大的合作银行客户进行了回溯测试，调取了其此前合作的某传统知名技术服务商处理的 5 起案例。核查发现，该遗留服务商在处理报表时竟然将支出项错算成了正向收益：原本真实的数值应当是负 100 万美元（-$1M），系统却直接识别并录入了正 7500 万美元（+$75M），由此产生了整整 7600 万美元的惊天差错。该银行客户此前曾与这家传统供应商签署了为期三年的合同，但面对如此巨大的致命缺陷，银行即便支付了全额合同款项也果断选择弃用，转而寻求真正具备准确率保障的方案。这种荒谬的数据错漏绝非孤例。Accend 在与另一家金融机构的合作中发现，其信贷资产组合中竟然有超过 10% 的案例存在类似的严重失误。当信贷决策建立在虚构而非严谨的事实之上时，金融机构面临的不仅是高达数亿美元的潜在资金损失风险，更是足以摧毁整个投资组合的致命灾难。Pranjal Daga 强调，这类错误不仅会吞噬资本，更会引发不可挽回的声誉崩塌，最终瓦解市场与生态系统对金融机构的信任基础。 - **月度处理规模**：单月处理财务报表分录超过 300,000 条。 - **效率倍增对比**：人工分析师耗时 2 年（约 600 个工作日，且不计病假与休假），人机协同 AI 缩短至 30 天以内完成。 - **典型严重算错案例**：净收入本应为 -100 万美元，遗留系统误录为 +7500 万美元，单笔误差达 7600 万美元。 - **高风险比例**：在受测机构中，超过 10% 的信贷资产组合受此类数据识别错误影响。 --- ### [05:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=300s) 杰米·戴蒙战术手册：从危机防御到每日数百次风险复核围绕核保的未来，Pranjal Daga 将核心议题提炼为三大板块：首先是揭示为何“7600 万美元错误”正大概率隐秘地发生在各大银行自身的账目之中；其次是汲取摩根大通（JPMorgan Chase）CEO 杰米·戴蒙（Jamie Dimon）在 2008 年金融危机中沉淀的管理策略；最后则是展望 2027 年到来的“核保奇点”（Underwriting Singularity）。届时所有的放贷决策都将深度依赖 AI 协同，未能及时转型的机构最终只能沦为商学院里的失败案例反面教材。戴蒙的战术手册对当前的 AI 风险管理极具启发。Pranjal Daga 回忆起不久前在纽约现场聆听《Acquired》播客对杰米·戴蒙的专访。在那场对谈中，戴蒙详述了 2008 年次贷危机期间摩根大通如何在大批老牌金融机构中胜出。当时雷曼兄弟（Lehman Brothers）、贝尔斯登（Bear Stearns）以及华盛顿互惠银行（Washington Mutual）相继倒闭，而摩根大通不仅存活下来，更是全行业的赢家。戴蒙制胜的核心举措之一，是将通常每五周召开一次的信用与风险评估委员会会议，强行提频为“每天召开五次”；另一个举措则是构筑坚不可摧的“壁垒式资产负债表”（Fortress Balance Sheet）——其核心逻辑在于“做好万全准备远比精准预测危机更重要”。然而，戴蒙在当时未曾言明的是，2008 年那套支撑一天五次高管风险复核的底层机制，完全依赖于大量员工手工翻找电子表格（Excel）拼命加班，整个流程极其繁琐且难以拓展。Pranjal Daga 设想，如果将类似 2008 年的危机压力置于现代 AI 驱动的环境中，机构不再受限于人工制表的时间成本，每天进行的将不是 5 次风险复核，而是能够实现全自动、高并发的 500 次实时风险审查。这种指数级扩展的监控能力，将从根本上重塑金融机构穿越周期与应对极端风险的能力。 - **关键历史对比**：2008 年危机中摩根大通将信贷与风险会议频率从“每 5 周 1 次”提升至“每天 5 次”。 - **管理核心哲学**：杰米·戴蒙强调“防患于未然的准备（Preparation）远比预测（Prediction）更重要”。 - **AI 对风险管理的重构**：将人工时代每日 5 次的手动风险审查，直接跃升为每日 500 次的自动化系统级风险巡检。 --- ### [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) AI 变革的三阶段演进与飞轮复合优势金融机构在引入 AI 技术时必须清醒认识到，人机协同的意义不仅在于比人工节约 600 个工作日，更在于以超越竞争对手的速度抢占市场先机。Pranjal Daga 梳理了信贷核保向 AI 进化的三个必经阶段：第一阶段是“辅助增强”（Augmentation），通过工具协助分析财务报表，将处理速度提升约 3 倍。许多机构误以为走到这里就已经到达了终点线，但实际上这仅仅是变革的起跑线。第二阶段是“自动化”（Automation），通过全自动化完成整个财务数据报表展开（Spreading）和深度财务分析，带来 10 倍的效率提升——这正是目前聪明资本和头部顶尖银行重金布局的深水区。最终的第三阶段则是预计在 2027 年实现的“自主化”（Autonomy），它将带来 100 倍的效率质变。在自主化阶段，繁琐的数据提取与交叉验证由 AI 全面接管，人类专家只需集中精力维护客户关系并审批边缘极端案例（Edge Cases），委员会决策周期将被极度压缩。Pranjal Daga 列举了一个近期的突破性案例：一笔高达 5000 万美元的大型信贷交易，在人机协同系统的支持下，团队仅在下午 2 点到 4 点之间的短短两个小时内就完成了全套财务评估报告。虽然现阶段 5000 万美元规模的最终决策仍需人类高管最终拍板，但评估时效已经与需要数天甚至数周的传统核保形成了降维打击。 ```mermaid flowchart TD accTitle: 三阶段信贷核保演进与复合优势飞轮 accDescription: 展示从辅助增强、流程自动化到完全自主化的演进路线，以及数据与交易相互促进的正向循环飞轮。 subgraph Stages["核保 AI 变革的三大阶段"] S1["第一阶段：辅助增强 (Augmentation) • 报表录入与辅助分析 • 3倍处理提速（起跑线）"] S2["第二阶段：自动化 (Automation) • 完整财务报表展开分析 • 10倍效率提升（聪明资金聚集）"] S3["第三阶段：自主化 (Autonomy，2027年) • 人类聚焦客户关系与边缘案例 • 100倍综合提升（5000万贷款2小时完成评估）"] S1 --> S2 --> S3 end subgraph Flywheel["数据与业务复合优势飞轮"] D["更多底层财务数据"] --> T["处理更多交易规模"] T --> B["更高质量信贷决策"] B --> A["更强大的 AI 学习模型"] A --> D end Stages -.-> Flywheel ``` 在此技术演进之上，早期践行者正在构筑如同“四维国际象棋”般的复合优势飞轮：更多的业务数据输入哺育出更高的处理吞吐量，带来更高质、精准的信贷决策，进而反哺训练出能力更强的 AI 系统。如果一家传统机构每月仅能人工推进 10 笔贷款交易，引入人机协同的机构就能将其投资组合扩容 10 倍乃至 100 倍，通过数据与规模的复利效应拉开无可逾越的壁垒。 - **阶段一：辅助增强（Augmentation）**：3 倍效率提升，主要覆盖分录拆解与辅助处理，属于数字化转型门槛。 - **阶段二：自动化（Automation）**：10 倍效率提升，实现报表自动展开与标准财务比率分析，是领先银行当前的攻坚重点。 - **阶段三：自主化（Autonomy）**：100 倍效率提升，预计 2027 年全面成熟；复杂案例（如 5000 万美元贷款）在 2 小时内完成前期评估。 - **飞轮复利**：数据积累 $\rightarrow$ 交易吞吐增加 $\rightarrow$ 决策质量优化 $\rightarrow$ 算法与系统迭代，实现 10x-100x 的业务扩张。 --- ### [13:30](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=810s) 周一行动计划与 2027 年核保奇点预测针对参会的高管与信贷负责人，Pranjal Daga 给出了能够落地的“周一早晨行动计划”（Monday Morning Action Plan），建议切忌盲目大拆大建，而应采取分步推进的策略。首先，第一步是排查内部数据质量隐患，立即核查自身机构现存资产中是否存在类似“7600 万美元计算错误”的系统性偏差；第二步，找准机构内部业务量最大、最具代表性的资产组合细分板块（Portfolio Segment）先行试点，而非一次性全盘推倒重来；第三步，建立严密的时效监测指标，评估现有审批流程是以“小时”为单位还是仍以“天或周”为单位流转，若仍需数周则已严重落后于市场基准；最后，务必认识到提效的终极目标不是单纯省钱，而是为了在获客端建立压倒性优势，以闪电般的放款体验争夺更多市场份额。关于即将在 2027 年降临的行业格局，Pranjal Daga 给出了四项极为具体的量化预测：第一，全行业商业信贷决策的平均审批时间将缩短至 6 小时以内，而行业领跑者将在 1 小时内完成决策；第二，原本耗时超过 8 小时（复杂案例耗时更长）的标准财务报表深度分析，将普遍降至 1 小时以内；第三，AI 分析的准确率将全面超越人类顶尖专业分析师的平均水准；第四，届时市场上超过 80% 的商业信贷决策都将由 AI 系统深度参与或驱动，AI 将彻底变成信贷市场的“基础入场券”（Table Stakes）。在演讲的尾声，Pranjal Daga 直言摆在所有金融机构面前的只有两条截然不同的道路：要么选择拥抱 AI 技术成为银行业生态中的赢家，凭借敏捷决策与高质量风控不断扩大市场份额；要么固步自封，像 2008 年金融危机中的贝尔斯登、雷曼兄弟和华盛顿互惠银行那样，在行业智能化洗牌的浪潮中被颠覆淘汰，沦为商学院案例库里的警示教材。 - **行动步骤 1（排查风险）**：复核现有报表展开工具与外包服务，排查净收入等核心科目的结构性正负符号与识别错误。 - **行动步骤 2（切入高频场景）**：选择业务量最大、标准化程度最高的一个业务板块启动试点改造。 - **行动步骤 3（时效考核）**：以“小时级”响应取代“天/周级”流转，重塑风控与评审周转链条。 - **2027 年信贷决策时效预测**：行业平均决策缩短至 6 小时内，市场领跑者达至 1 小时内。 - **2027 年财报分析时效预测**：深度财报分析从目前的 8+ 小时压缩至 1 小时以内，准确度超越专业人类专家。 - **2027 年渗透率预测**：AI 驱动决策将覆盖超过 80% 的商业信贷交易，成为全行业标配基础能力。 ## Notable quotes > The Underwriting Singularity – the point where AI becomes faster, more accurate, and more insightful than any human underwriter. — Pranjal Daga > [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) made my blood go very cold — Pranjal Daga > [05:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=300s) 做好万全准备远比精准预测危机更重要 — Jamie Dimon > [00:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=0s) 7600万美元噩梦 — Pranjal Daga > [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) 第一阶段是‘辅助增强’（Augmentation）……许多机构误以为走到这里就已经到达了终点线，但实际上这仅仅是变革的起跑线。 — Pranjal Daga > [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) 第二阶段是‘自动化’（Automation）……这正是目前聪明资本和头部顶尖银行重金布局的深水区。 — Pranjal Daga > [10:00](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=600s) 第三阶段则是预计在 2027 年实现的‘自主化’（Autonomy），它将带来 100 倍的效率质变。 — Pranjal Daga > [13:30](https://www.youtube.com/watch?v=zgdg_Rfgqq8&t=810s) 要么选择拥抱 AI 技术成为银行业生态中的赢家……要么固步自封……沦为商学院案例库里的警示教材。 — Pranjal Daga ## Mentioned Pranjal Daga, Accend, Jamie Dimon, JPMorgan Chase, Acquired (Podcast), Lehman Brothers, Bear Stearns, Washington Mutual, Ai4 ## Video description Pranjal Daga, Co-Founder, Accend Abstracts By 2027, AI will fundamentally transform how every lending decision is made, creating what I call "The Underwriting Singularity" – the point where AI becomes faster, more accurate, and more insightful than any human underwriter. The lending industry stands at a critical inflection point where early AI adopters are already seeing 90%+ time savings and unprecedented accuracy in financial statement analysis, while others risk being left behind in an increasingly competitive market. I will reveal the three key stages of AI transformation that leading institutions are implementing right now, share real-world data on ROI and risk reduction from AI underwriting deployments, and outline the specific technologies and strategies that will define the winners in the next two years. This talk will equip you with a clear roadmap to navigate your institution's AI transformation and position yourself as a leader in the new era of intelligent lending. [Source](https://www.youtube.com/watch?v=zgdg_Rfgqq8)

Source quotes

Pranjal Daga: The Underwriting Singularity – the point where AI becomes faster, more accurate, and more insightful than any human underwriter. [0:00] Pranjal Daga: made my blood go very cold [5:00] Jamie Dimon: 做好万全准备远比精准预测危机更重要 [0:00] Pranjal Daga: 7600万美元噩梦 [10:00] Pranjal Daga: 第一阶段是‘辅助增强’（Augmentation）……许多机构误以为走到这里就已经到达了终点线，但实际上这仅仅是变革的起跑线。 [10:00] Pranjal Daga: 第二阶段是‘自动化’（Automation）……这正是目前聪明资本和头部顶尖银行重金布局的深水区。 [10:00] Pranjal Daga: 第三阶段则是预计在 2027 年实现的‘自主化’（Autonomy），它将带来 100 倍的效率质变。 [13:30] Pranjal Daga: 要么选择拥抱 AI 技术成为银行业生态中的赢家……要么固步自封……沦为商学院案例库里的警示教材。

Source quotes

[0:00] Pranjal Daga: By 2027, AI will fundamentally transform how every lending decision is made, creating what I call 'The Underwriting Singularity' – the point where AI becomes faster, more accurate, and more insightful than any human underwriter. [0:00] Pranjal Daga: Processing 300,000 financial statement line items manually would take a team of human credit analysts roughly two years. Accend executes this same workload in under 30 days using a human-in-the-loop AI model. [0:00] Pranjal Daga: The legacy system misclassified a negative $1 million expense item as a positive $75 million asset. This single sign-inversion error transformed a distressed financial profile into a phantom credit surplus. [0:00] Pranjal Daga: Subsequent audits across other financial institutions revealed that similar foundational errors affected more than 10% of audited loan portfolios, forcing risk committees to base major capital allocations on fiction rather than deep-rooted facts. [5:00] Pranjal Daga: Central to Dimon's strategy was the concept of the 'fortress balance sheet'—an operating posture focused on preparation rather than attempting to predict the exact timing of a macro downturn. [5:00] Pranjal Daga: Instead of five manual risk check-ins per day, institutions equipped with human-in-the-loop AI can conduct 500 continuous risk reviews daily. [10:00] Pranjal Daga: The first stage, Augmentation, delivers a 3x speed improvement... The second stage, Automation, unlocks a 10x improvement... The final stage, Autonomy, slated to arrive by 2027, will deliver a 100x improvement. [10:00] Pranjal Daga: Modern AI-driven lenders can evaluate a complex $50 million commercial credit application between 2:00 PM and 4:00 PM on the same day, whereas traditional institutions still take weeks to complete initial spreading. [15:00] Pranjal Daga: By 2027, approximately 80% of all commercial lending decisions will be executed through AI-driven underwriting engines. [15:00] Pranjal Daga: Lenders face a stark binary choice over the next two years: embrace autonomous underwriting to become the JPMorgan Chase of the modern era, or resist adoption and face the fate of Bear Stearns, Lehman Brothers, and Washington Mutual.

Original transcript

[0:00] - Hello everyone. Thank you so much firstly for being here. I really appreciate you giving time. It's a very busy conference. So today, like Jillian mentioned, we are going to talk about the underwriting singularity. Thank you. Now what exactly does that mean? We'll go into the details about that. Thank you. AI is the word of the conference. but AI isn't optional anymore in financial services. Let's take an example. Yeah. Our company, it's called Accent, And we help fintechs and banks accelerate their business customers' underwriting. Last month, we processed over 300,000 line items from financial statements. Now, if your analysts are doing exactly the same thing, It will take them two years to do that. and we are able to do that with the human in the loop and AI approach. in less than 30 days. But what's shocking here is that the legacy providers, which a lot of you banking and lending execs would be using right now in your day-to-day operations. they had a big big big issue. You see, they calculated the net income incorrectly, in one of the cases that led to a catastrophic lending decision. So it's not just about efficiency that you see going from two years to 30 days, but it's about making sure that your decisions are high quality, and accurate. Thank you. of Let's go a little bit deeper into this $76 million nightmare. You see here, that we are taking an example with a different customer of ours. It's a bank, does a lot of lending, we took five different cases of their legacy provider as a backtest. And what we found was something that actually made my blood go very cold. In this you can see the mistake of counting expenses as positive as against negative. So you see the mistake that you see here, is the correct value is negative one million dollars however that service identified as a positive 75 million dollars Now, this is a mistake. This is just one particular mistake. You can see how this can just blow up in your entire portfolio and make you question things from the get-go. So this mistake was basically by a customer of ours that had a three-year contract with this legacy provider. Some of you might have already guessed which one that is. And then they're able to now just move away even after paying for those three years. accuracy perspective. Thank you. Now, this is not just one financial institution's problem. We had another financial institution that we work with, and with them, we found that there are these mistakes that are happening for more than 10% of their portfolio. Yes. you're basically creating those decisions based on fiction. rather than deep-rooted facts, which should be the case for every lending and underwriting decision. What it does It's not only the three digit million dollar potential risk that you see here. but it also leads to portfolio killing mistakes. why do I call them portfolio killing is it's not just the capital or the financial losses [5:00] but it leads to reputational losses. And these reputation losses will lead to mistrust in the financial ecosystem. Thank you. Thank you. So what will we talk about today? We talk about three things. By the way, in this talk, my focus or my goal is to talk about facts and reality only. So the first thing that we'll talk about is this $76 million error. And why? This is not something that is happening to some financial institutions. it's a very high chance that it might be happening in your financial institution. and you might not be aware of that. The second thing that we'll talk about is the Jamie Dimon playbook. Jamie Dimon is the CEO of JPO and Chase. And in 2008, he created a strategy, which if we look back at, we can say that it's not just making us think about AI as the future of risk management It makes us believe that AI is something that's like in the present you need to survive. And the third thing that we'll talk about is to talk about singularity. And why 2027 is the year according to me, where we'll have all of the lending decisions made with some sort of AI. It'll be human in the loop, but AI will be very responsible for it. And there's going to be like a few different type of financial institutions. One who will win, And the others, who will be written down in business school, cases. So, let's get started. We talk about Jamie Dimon. A few weeks ago, I was at the acquired podcast live taping in New York where Jamie Dimon was there. And a couple of things that he talked about were focused on a lot on how he has learned so much from a few days of the 08 crisis. And he's applying all of those things now in his day-to-day while running Chase. So let's go back to 2008. financial crisis is happening a JP Morgan shares is coming across as a winner While the other financial institutions, Lehman Brothers, Bear Stearns, Washington Mutual, and many others, are just failing. So what are the things that Jamie Dimon and the team did at that point? to make sure that they are able to survive. The one important thing he mentioned in that live taping was that the credit and risk committees happened five times a day instead of five times or instead of like every five weeks it happened every single day five times. And the other thing that he mentioned was something that the industry is very well aware of now and everybody's talking about. which is the fortress balance sheet. Now what is that fortress balance sheet? It's basically talking about preparation is so much more important than actually prediction of the O8 crisis. So he was able to prepare and Chase was able to prepare on those things as again just predicting that when will that happen? so a few things that have happened in this Imagine that the OIT crisis is happening right now. Thank you. What Jamie Dimon at that stage did not talk about is those five risk check-ins that he was doing daily with the executives with the team was basically based on manual work by a few individuals in the company. So it took a lot of time. People are still going to spreadsheets, still making their way into things, and then working on everything with that. Now let's say AI is doing like all of those things in AI. like a day total. So. [10:00] Instead of five risk check-ins a day, you can do 500 risk check-ins in one single day. the level of scalability that you can achieve with that is Next or nothing. Thank you. So now, what do we focus on? We've talked about this, the case study of running 300,000 line items and processing that, analyzing that. in a single month. If you see here, what we did was our human-in-the-loop AI process these, like, all these things within 30 days. If your analysts were doing that, it'll take them 600 days without any sickness, no vacations, and AI's accuracy is actually higher. By the way, this is not just AI. This is human in the loop AI. So it's a marriage between the two that we're talking about. Thank you. Now the thing is it's not just focused on what's more efficient. It's actually focused on how you can be faster than the players that are beating you or your competition. Thank you. Now there are three stages of transformation. The first stage is augmentation. So you already have like a few things. You already, for example, you are analyzing financial statements. You're able to do that with like 3x the processing time. This is somewhere where people think it's like the finish line, but actually things are getting just started, right? So it's like the first gate, It's the beginning. The next stage is automation. And that automation, you get like 10x improvement, but that 10x improvement is happening because you're doing like the entire spreading or the analysis process, doing a lot of things with that. But here is where the smart money from the smart banks and smarter AI institutions and financial institutions is going in. And the third one is autonomy. This is what we are thinking about seeing in the next couple of years by 2027. So 100x improvement, things will get much faster and us humans will be focused on building relationships and have edge cases. So committees and making decisions will be much, much, much faster. for 50 million dollars within two hours there's like 2 p.m. to 4 p.m. within two hours they were able to get to an assessment now of course it's a 50 million dollar deal so they will not make the judgment and the decision that at least at that point in time in 2025 might see it in 2027 or something around that time frame But if you take a look at the traditional lender, it's basically having like all of these things within a specific time frame. Thank you. I'm just going to move on to the next one, which is how you can have the compound advantage. The compound advantage that you have is basically going from more data, more data to more transactions, better decisions, and then you have more AI. So it's like basically what you see here is like a 4D chess move where you can have more and more things associated with it. And let's say if you're doing like 10 deals or 10 loans per month, as against that, if you have a human-in-the-loop AI in your portfolio, you'll be doing 10x or 100x of that portfolio. So it's the compounding that's going to help you out. Thank you. Thank you. Now, what I want all of you to take away from this conversation is basically your Monday morning action plan. So, changing everything is not the way to go. You've got to go like step by step. So, first things first, just analyzing what might be the error rate. Do we have this $76 million error? Checking those things out, step number one. Step number two, how can we figure out what's the highest volume use case that I have? And then starting out with that. Don't change everything. Start with one particular portfolio segment. And then measure speed. if you're getting things done in hours instead of days or weeks, if that's not the case you're still like behind the last thing and very important thing to notice here is everything is [15:00] about efficiency. Thank you. a lot of things here that we are talking about is about acquisition. How can you get more market share? as again just thinking about having quick turnaround times. What we have focused on in 2027 prediction: credit decisions. The industrial average will become six hours or less. The market leaders will make decisions within an hour. Second, financial analysis. Right now it takes like more than eight hours. For a standard one, for complex ones, it takes like longer. we should be able to reduce that down to one hour. Highest accuracy, more than what we see with expert humans. And then mark dominance. We see like 80% of all commercial lending decisions will be made by this. So if you are using AI, this will become a table stakes for you for all of your lending decisions. Now, as I mentioned before, the choice is yours. One choice could be that you're on the side, which is the winner. in the banking ecosystem. You're using AI, you're making decisions, winning more deals, making things more faster efficient, and having more market share. versus on the other side where you might think about yourself as like the best turns, Washington Mitchell or Lehman Brothers. you're going to get in some ways disrupted because of that. So the one thing to notice is not to shy away from utilizing these new tech, and then focusing on the other one. Yeah, this is about me. Thank you so much for coming to the conversation. You can reach out to me on my email. or if at all you're interested in having a conversation. We could not get enough time for questions, but I'll be around, and I'd love to talk to everybody, so please come towards me, and we can have a quick chat. Thank you.