McKesson Architect Blames Siloed AI Pilots for Surging Enterprise Costs

The AI Rush Is Making Contract Intelligence Cost More - Not Less: Why AI Fails Without Governed — Ai4

Ai4

Uncoordinated departmental AI deployments multiply software renewal expenses and drive failure rates to 95% by neglecting upfront data governance across business units.

AI-assisted summary of The AI Rush Is Making Contract Intelligence Cost More - Not Less: Why AI Fails Without Governed — original source. Check the original for full context.

Source published: 2026-08-19

Detailed notes

## Background Published on YouTube by the Ai4 channel on August 19, 2026, this session features Naresh Kommula, Lead IT Architect at McKesson. The presentation examines why the rush to deploy artificial intelligence often inflates enterprise costs through departmental data fragmentation and outlines how adopting a Governed Contract Data Layer (GCDL) and the Contract Obligation Intelligence Framework (COIF) can transform contracts into unified enterprise data assets.

## TL;DR Despite the ease of deploying modern AI models, enterprise contract intelligence initiatives frequently fail to scale and inflate technology costs due to siloed departmental pilots and uncoordinated data pipelines. Organizations routinely lose approximately 9% of their contract value because of poor post-execution operationalization, exacerbated by tool-first mindsets, fragmented clause ownership, and reactive compliance. To resolve this paradox, Naresh Kommula introduces the Contract Obligation Intelligence Framework (COIF) and Governed Contract Data Layer (GCDL), which establish upfront cross-functional alignment and a single source of truth for contract data. Shifting from disconnected point tools to a governed, continuous feedback loop eliminates duplicate extraction pipelines, reduces overall total cost of ownership, and builds a dependable foundation for agentic AI.

## Atmosphere & style The presentation carries a pragmatic, authoritative, and diagnostic tone characteristic of an enterprise architecture conference keynote. Kommula cuts through the prevailing generative AI hype by grounding his analysis in concrete financial stakes, sobering failure rates, and practical organizational friction. The progression moves deliberately from highlighting the hidden operational traps of rapid pilot deployments toward delivering a structured, systems-oriented blueprint for long-term data governance.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=XGSxWviHdc8&t=4s) The rapid rush to deploy enterprise AI is paradoxically driving up total costs rather than lowering them due to compounding complexity and fragmented silos. - [00:54](https://www.youtube.com/watch?v=XGSxWviHdc8&t=54s) Organizations lose an estimated 9% of total contract value annually due to poor post-execution operationalization, equating to $90 million in leakage for a $1 billion enterprise. - [03:00](https://www.youtube.com/watch?v=XGSxWviHdc8&t=180s) High AI pilot failure rates (such as 95% failing to scale in 2025 MIT data) stem from organizational and data bottlenecks rather than limitations in AI model capability. - [05:19](https://www.youtube.com/watch?v=XGSxWviHdc8&t=319s) The primary underlying driver of enterprise AI cost is fragmented, siloed implementations rather than token consumption or inference infrastructure overhead. - [07:48](https://www.youtube.com/watch?v=XGSxWviHdc8&t=468s) Intense market pressure drives departments to launch isolated AI pilots, resulting in overlapping tools, duplicate data pipelines, and compounding multi-million-dollar software renewals. - [09:54](https://www.youtube.com/watch?v=XGSxWviHdc8&t=594s) Legal, procurement, sales, and finance teams routinely run disconnected AI extraction and OCR pipelines on the exact same historical contract archives. - [13:21](https://www.youtube.com/watch?v=XGSxWviHdc8&t=801s) Contract AI initiatives fail due to four structural root causes: tool-first thinking, fragmented ownership, non-standardized contract terms, and reactive post-execution governance. - [17:10](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1030s) The Contract Obligation Intelligence Framework (COIF) provides an operating model—not a vendor tool—to establish continuous enterprise governance and cross-functional alignment. - [18:32](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1112s) A Governed Contract Data Layer (GCDL) establishes a single enterprise source of truth, eliminating redundant extraction pipelines for analytics, operational systems, and AI agents. - [20:23](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1223s) Upfront cross-departmental alignment creates a virtuous feedback loop, generating clean obligation data that enables predictable AI models, advanced predictive analytics, and lower TCO. - [21:40](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1300s) Organizations must resolve internal fragmentation and align adjacent departments (such as legal, finance, and procurement) before writing code or purchasing AI software.

## Detailed walkthrough Here is the detailed, faithful walkthrough of part 1 of the transcript.

---

[00:04](https://www.youtube.com/watch?v=XGSxWviHdc8&t=4s) The Paradox of the AI Rush: Higher Enterprise Costs

Naresh Kommula (Lead IT Architect at McKesson) opens his presentation by highlighting a growing paradox in modern enterprise technology: while the "AI rush" has made it dramatically easier to build and deploy contract intelligence solutions, it is simultaneously driving up overall costs rather than lowering them. Organizations everywhere are actively running AI pilots—some of which are successfully scaling, and others that are already in production and performing well. Yet beneath these surface-level successes, enterprise costs are escalating rapidly.

```mermaid accTitle: The Enterprise AI Paradox Flow flowchart LR subgraph Expansion["AI Pilot Proliferation"] A["Rapid AI Deployment"] --> B["Successful Departmental Pilots"] B --> C["Scale to Production"] subgraph Consequence["Enterprise Impact"] C --> D["Compounding Complexity & Data Silos"] D --> E["Rising Enterprise Costs"] end end ```

The core objective of this session is to uncover the underlying structural reasons behind this paradox and explore how organizations can shift toward an enterprise-wide mindset to resolve it.

- Organizations are scaling AI pilots successfully, yet overall enterprise operational and technology costs continue to rise. - The shift requires moving away from isolated departmental initiatives toward a unified, governed enterprise approach.

---

[00:54](https://www.youtube.com/watch?v=XGSxWviHdc8&t=54s) Unlocking the Lost 9% Value in Enterprise Contracts

Contracts represent the single most critical asset in any business because "that's where the money lives." Citing a World Commerce & Contracting (World CC) survey, Kommula notes that organizations routinely lose approximately **9% of their total contract value** due to poor post-execution operationalization. To illustrate the real-world financial stakes, he points out that a company with $1 billion in revenue is effectively losing around **$90 million** every year.

This revenue leakage does not happen because companies lack agreed-upon contractual terms. Rather, it occurs because operational teams struggle to operationalize and enforce those terms after the ink is dry. Whether in manufacturing, supply chain, software, or retail, contracts serve as the "bread and butter" of the business. Enterprise leaders recognize this challenge and continually attempt to eliminate revenue leakage, save money, and untap the full potential of their agreement portfolio.

- **9% Value Leakage:** Average lost contract value identified by World Commerce & Contracting research. - **$90M Impact:** Estimated financial loss for a $1 billion revenue enterprise due to poor operationalization. - **Universal Applicability:** Contract execution directly impacts every sector, including retail, software, manufacturing, and supply chain.

---

[03:00](https://www.youtube.com/watch?v=XGSxWviHdc8&t=180s) The AI Pilot Trap: Why 95% of Initiatives Fail to Scale

In response to these operational challenges, enterprise teams are launching continuous improvement initiatives and embedding AI everywhere to improve automation efficiency and generate advanced analytics. However, the reality of AI adoption reflects steep failure rates across the industry:

```mermaid accTitle: Industry AI Pilot Failure Rates Over Time flowchart LR subgraph Historical["2025 Realities"] A["MIT Survey 2025"] --> B["95% Pilots Fail to Scale or Reach Production"] end subgraph Present["2026 Projections"] C["Gartner Survey 2026"] --> D["60% Pilots Fail to Go Live or Deliver Target ROI"] end subgraph Future["2027 Agentic Era"] E["Gartner Forecast 2027"] --> F["40% Agentic Deployments Fail to Produce Adequate ROI"] end ```

Kommula stresses that these failures are not caused by bad teams, poor solutions, or underperforming AI models. Model capability is no longer the bottleneck—technological progress is moving so fast that solutions implemented just a year ago are already obsolete, with those same features being natively subsumed into newer frontier models and provider services.

- **2025 MIT Survey:** Approximately 95% of AI pilots failed to scale or move into production. - **2026 Gartner Benchmark:** Around 60% of pilots are unable to go live or deliver expected ROI. - **2027 Agentic Era:** Gartner projects that 40% of upcoming agentic AI deployments will still fail to generate sufficient ROI. - **Rapid Model Obsolescence:** Bespoke solutions built 12 months ago are frequently rendered obsolete by standard features in new frontier models.

---

[05:19](https://www.youtube.com/watch?v=XGSxWviHdc8&t=319s) The Ungoverned Data Trap and Common Cost Misconceptions

When exploring why AI solutions fail to scale, the most widely cited cause is **ungoverned data**. Enterprise leaders see this issue highlighted across industry talks and social media. Without governed data and a structured data layer, organizations cannot successfully deploy or scale advanced agentic solutions. However, while ungoverned data is a recognized issue, achieving alignment across large organizations remains a massive hurdle.

When asked what factors drive up the total cost of enterprise AI, industry discussions usually focus on immediate technical expenses: - **Token consumption costs** from high-volume LLM API calls. - **Inference infrastructure overhead** required to run complex models. - **Observability tools** marketed by exhibition hall vendors to monitor AI pipelines. - **Underlying model trust and safety mechanisms**.

While vendor solutions targeting observability and token reduction offer valid short-term optimizations, they fail to solve the primary underlying driver of enterprise cost: **fragmented AI solutions**.

---

[07:48](https://www.youtube.com/watch?v=XGSxWviHdc8&t=468s) How Fragmented AI Solutions Inflate Enterprise Overhead

Driven by the "AI rush," business and IT leaders feel intense pressure to deploy solutions quickly to demonstrate market responsiveness. "If you're not implementing a solution with AI," Kommula notes, "they see you as... not up to the market." This pressure causes departments to build AI capabilities in siloed, isolated pockets.

```mermaid accTitle: Departmental Silos vs Enterprise Cost Multiplication flowchart TD subgraph SiloedInputs["Fragmented Departmental Pilots"] L["Legal: OCR & Risk Extraction"] P["Procurement: Commitments & Vendor Terms"] S["Sales: Pricing & Rebates"] F["Finance & Compliance: Audit Models"] end subgraph CompoundingCost["Enterprise Oversight"] L --> C["Compounding Multi-Million Dollar Renewals"] P --> C S --> C F --> C C --> E["Elevated Total Cost of Ownership TCO"] end ```

While an isolated pilot might look cost-effective and deliver strong ROI within a single department, this fragmentation multiplies costs when viewed from an enterprise level. When CTOs, CIOs, or enterprise architects review annual budgets, they discover compounding $1 million to $10 million renewal fees across multiple departments for redundant, overlapping tools.

- **Market Pressure:** Fear of falling behind drives rapid, uncoordinated AI adoption across business units. - **Local Success vs. Global Cost:** Individual departmental wins obscure larger macro-level cost compounding. - **Compounding Renewals:** Redundant AI pilots create massive subscription renewal burdens for C-suite leaders.

---

[09:54](https://www.youtube.com/watch?v=XGSxWviHdc8&t=594s) The Siloed Reality of Contract Intelligence Across Departments

Contracts intersect virtually every organizational function, making contract intelligence a primary target for siloed AI projects. However, because each department views contracts through its own operational lens, teams routinely process the same historical documents independently:

Because vendor offerings and CLM (Contract Lifecycle Management) tools pitch standalone point solutions, departments execute their own "cheap pilots." While each pilot receives local funding approval, the resulting fragmentation leads to duplicate data pipelines, conflicting definition models, and ballooning enterprise software expenses.

---

[13:21](https://www.youtube.com/watch?v=XGSxWviHdc8&t=801s) Four Structural Root Causes Driving AI Cost Escalation

Kommula identifies four core structural failures that cause enterprise contract AI initiatives to break down:

```mermaid accTitle: Four Structural Failure Modes of AI Contract Initiatives flowchart TD subgraph RootCauses["Structural Failure Modes"] A["1. Tool-First Thinking"] B["2. Fragmented Ownership"] C["3. Weak Sand (Lack of Standardization)"] D["4. Post-Execution Governance"] end subgraph Impact["Enterprise Failure"] A --> E["Siloed Tools Without Alignment"] B --> E["Nobody Owns the Contract End-to-End"] C --> E["Creative Unstandardized Terms Break Ops"] D --> E["Reactive Governance Overhead"] end ```

1. **Tool-First Thinking:** In the rush to adopt AI, teams purchase software to solve immediate problems before building operational alignment with downstream teams. 2. **Fragmented Ownership:** A single 50-page contract contains quality, procurement, rebate, and regulatory terms. Because departments only care about their own specific clauses, "everyone cares about the contract, but no one cares as a whole." 3. **"Weak Sand" (Lack of Standardization):** Sales teams frequently draft highly creative, non-standard terms to close deals quickly. This lack of standardization creates immense complexity when downstream operational teams try to execute the contract. 4. **Governance After Execution:** Enterprise teams often delay governance until after contracts are executed. Unscrambling complex terms post-execution requires establishing specialized compliance teams to manually monitor operations, creating ongoing overhead.

- **Non-Standard Terms:** High-margin sectors like pharmaceuticals and banking suffer heavily from custom deal structures that break standard operational workflows. - **Governance Bottlenecks:** Managing compliance reactively post-execution requires dedicated monitoring teams, driving up enterprise headcount and operational costs.

---

[17:10](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1030s) The Solution: Contract Obligation Intelligence Framework (COIF)

To address these structural issues, Kommula introduces the **Contract Obligation Intelligence Framework (COIF)**. COIF is not another software product or vendor tool; it is an enterprise operating model designed to align stakeholders, streamline data flows, and reduce the overall cost of contract AI deployments.

```mermaid accTitle: Contract Obligation Intelligence Framework Flywheel flowchart TD subgraph Core["Continuous Governance Center"] CG["Continuous Enterprise Governance"] end subgraph Flywheel["COIF Flywheel Operating Model"] A["1. Contract Obligation Alignment"] --> B["2. Governed Contract Data Layer GCDL"] B --> C["3. Connected Contract Intelligence"] C --> A end CG --- Flywheel ```

The framework operates as a continuous flywheel structured around four core iterative principles: **Align**, **Govern**, **Intelligence**, and **Improve & Repeat**.

- **Not a New Tool:** Operates on top of existing software architectures to bring structural alignment across business units. - **Flywheel Architecture:** Replaces isolated, point-to-point integrations with a continuous operational feedback loop. - **Centralized Governance:** Anchors all operational decisions in a unified enterprise standard rather than siloed departmental logic.

---

[18:32](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1112s) The Core Components of the COIF Operating Model

The Contract Obligation Intelligence Framework consists of three interconnected operational components anchored by continuous governance:

1. **Contract Obligation Alignment:** Establishes cross-departmental agreement upfront. Before writing or executing a contract, teams evaluate how specific terms will impact downstream operational units such as administration, finance, invoicing, and order fulfillment. 2. **Governed Contract Data Layer (GCDL):** A single, unified data abstraction layer that standardizes contract definitions across the organization. Instead of sales, procurement, and finance building separate data extraction pipelines, all AI agents, analytical dashboards, and operational reports consume normalized data from this single source of truth. 3. **Connected Contract Intelligence:** Continuously cross-references formal contractual obligations against real-world operational actuals to generate actionable insights and flag risk exposure in real time.

```mermaid accTitle: Governed Contract Data Layer Ecosystem flowchart LR subgraph Consumers["Enterprise Data Consumers"] A1["Procurement Teams"] A2["Finance & Invoicing"] A3["Sales Operations"] A4["AI Agents & Analytics"] end subgraph CoreLayer["Unified Infrastructure"] GCDL["Governed Contract Data Layer GCDL"] end subgraph Source["Raw Data Inputs"] PDFs["Historical Contract Archives & PDFs"] end PDFs --> GCDL GCDL --> A1 GCDL --> A2 GCDL --> A3 GCDL --> A4 ```

By embedding **Continuous Governance** at the heart of this flywheel, organizations eliminate duplicate data processing, align departmental definitions, and establish a trusted, scalable foundation for future agentic AI capabilities.

- **Upfront Alignment:** Prevents downstream execution issues by reviewing contract structures before final execution. - **Single Source of Truth:** The GCDL eliminates redundant OCR scanning and disparate data models across enterprise tools. - **Real-Time Auditing:** Connected intelligence compares agreed contract terms against actual operational metrics to automatically detect revenue leakage and non-compliance.

[20:23](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1223s) The Virtuous Feedback Loop of the COIF Model

Establishing structural alignment early in the contract lifecycle fundamentally transforms both operational performance and the total cost of solution ownership. Under the Contract Obligation Intelligence Framework (COIF), upfront cross-functional alignment generates clean, high-quality obligation data around core contractual terms. That reliable data layer serves as the catalyst for deeper automation and advanced intelligence: machine learning models produce predictable, trustworthy results rather than erratic outputs, giving business units robust predictive analytics across their entire portfolio of contracts.

```mermaid flowchart TD accTitle: The Virtuous Feedback Loop of Governed Contract Intelligence subgraph Foundation["Governed Foundation"] A["Upfront Cross-Team Alignment"] --> B["High-Quality Obligation Data"] end subgraph Intelligence["Intelligence & Automation"] B --> C["Predictable AI Models & Automation"] C --> D["Advanced Predictive Analytics"] end subgraph Outcomes["Continuous Optimization"] D --> E["Standardized Contracts & Cost Reductions"] E -. "Continuous Feedback Loop".-> A end ```

The resulting intelligence initiates an ongoing compounding cycle across the organization. The insights extracted from governed contract data directly inform process improvements, accelerate contract standardization, and drive down operational waste. Because each phase operates as an interconnected feedback loop, the enterprise steadily improves its institutional knowledge over time, systematically shrinking operational expenses while elevating execution speed and accuracy.

- **Foundational Alignment:** High-quality data capture at the inception stage ensures that downstream models execute with precision and avoid governance drift. - **Predictive Capabilities:** Reliable obligation intelligence enables organizations to forecast contractual risks and performance milestones proactively rather than reacting post-signature. - **Continuous Feedback Mechanism:** Refined business processes and standardized clause libraries feed directly back into earlier lifecycle stages, lowering long-term total cost of ownership (TCO).

---

[21:40](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1300s) Fixing Fragmentation Before Rushing into AI

Before bringing the presentation to a close, Naresh Kommula imparts a crucial architectural principle for any team planning an artificial intelligence rollout, whether focused specifically on contract intelligence or general enterprise AI: address internal fragmentation before writing code or purchasing software. The prevailing instinct among business units is to rush ahead, acquiring or building siloed point solutions tailored solely to their immediate departmental pain points. That siloed mentality inevitably guarantees fragmented data pipelines and skyrocketing governance overhead across the wider enterprise.

Leaders must step back and prioritize alignment across departments before deploying tools. If achieving sweeping, enterprise-wide consensus immediately feels too daunting, organizations should begin by aligning departments that operate side-by-side—such as legal, procurement, finance, and operations. Kommula underscores that making this singular architectural commitment up front will dramatically reduce the total cost of enterprise AI ownership over the solution's lifetime. Concluding his address, he invites questions on the COIF model and thanks the audience, met with warm applause.

- **The Rush to Implement:** Departmental silos and hasty AI deployments duplicate data pipelines and obligation models, driving up costs rather than lowering them. - **Adjacent Alignment:** Bridging the gap between immediately adjacent business units provides an achievable, pragmatic starting point for unified contract governance. - **Cost Impact:** Upfront organizational alignment stands as the single most effective lever for curbing downstream AI maintenance, integration, and operational expenses.

## Notable quotes > [00:54](https://www.youtube.com/watch?v=XGSxWviHdc8&t=54s) that's where the money lives. — Naresh Kommula

> [07:48](https://www.youtube.com/watch?v=XGSxWviHdc8&t=468s) If you're not implementing a solution with AI, they see you as... not up to the market. — Naresh Kommula

> [13:21](https://www.youtube.com/watch?v=XGSxWviHdc8&t=801s) everyone cares about the contract, but no one cares as a whole. — Naresh Kommula

## Mentioned Naresh Kommula, McKesson, Ai4, World Commerce & Contracting, MIT, Gartner, Contract Obligation Intelligence Framework, Governed Contract Data Layer, Contract Lifecycle Management

## Video description Naresh Kommula - Lead IT Architect, McKesson

AI is making it easier than ever to build & deploy contract intelligence solutions, yet many organizations are experiencing rising complexity, governance overhead, and inconsistent AI outcomes. As AI expands across legal, procurement, finance, compliance, and operations, organizations often create multiple versions of contract intelligence, each with its own obligation models, governance processes, and data pipelines.

This session explores why contracts must be treated as governed enterprise data assets rather than unstructured documents. Attendees will learn how the Governed Contract Data Layer (GCDL) and Contract Obligation Intelligence Framework (COIF) can create a shared operational language for contractual obligations, improve compliance, reduce duplication, and provide a trusted foundation for scalable AI and emerging agentic capabilities.

[Source](https://www.youtube.com/watch?v=XGSxWviHdc8)

中文简报

McKesson架构师指孤岛式AI试点推高企业成本各部门各自为战的AI部署因忽视跨业务单元的前期数据治理，导致软件续费开支倍增，并将失败率推高至95%。尽管部署现代AI模型变得愈发简便，但由于孤立的部门试点和缺乏协调的数据管道，企业合同智能项目往往难以实现规模化扩展并推高了技术成本。由于签署后运营落实不力，再加上“工具优先”的心态、碎片化的条款归属以及被动应对合规等问题的加剧，企业通常会损失约9%的合同价值。为化解这一悖论，Naresh Kommula提出了合同义务智能框架（COIF）和受管合同数据层（GCDL），旨在建立前期跨职能协同机制以及合同数据的单一事实来源。从孤立的点状工具转向受治理的持续反馈回路，不仅消除了重复的提取管道，降低了总体拥有成本，还为智能体AI奠定了可靠的基础。由于复杂性不断累积以及碎片化孤岛的存在，仓促推进企业AI部署反而推高了总成本，并未如预期般实现降本。 (0:04) 由于签署后运营落实不力，企业每年预计损失合同总价值的9%，对于规模为10亿美元的企业而言，这相当于9000万美元的资金流失。 (0:54) AI试点的高失败率（例如MIT在2025年的数据显示有95%未能实现规模化）源于组织和数据瓶颈，而非AI模型本身的性能限制。 (3:00) 企业AI成本增加的核心深层动因是碎片化、孤岛式的实施路径，而非Token消耗或推理基础设施的开销。 (5:19) 巨大的市场压力促使各部门启动彼此孤立的AI试点，造成工具重叠、数据管道重复，并不断累加高达数百万美元的软件续费支出。 (7:48)

Reading summary

Naresh Kommula, Lead IT Architect at McKesson, argues that uncoordinated enterprise AI adoption inflates total operational costs because siloed departments deploy redundant tools over ungoverned data, rather than addressing post-execution contract leakage. Post-Execution Value Leakage Versus Pilot Failure Rates (0:54) Enterprise agreements represent a major point of financial exposure. Citing research from World Commerce & Contracting, Kommula notes that organizations lose an estimated 9% of total contract value due to poor post-execution operationalization—equating to an annual $90 million loss for an enterprise with $1 billion in revenue. This loss stems from operational teams struggling to enforce agreed-upon terms once contracts are executed across sectors such as supply chain, manufacturing, retail, and software. While companies deploy AI to curb these losses, initiatives frequently stall. Kommula cites an MIT 2025 survey indicating that 95% of AI pilots fail to scale or enter production, alongside a Gartner 2026 benchmark finding that 60% fail to go live or deliver expected ROI. Furthermore, Gartner projects that 40% of agentic AI deployments in 2027 will fail to achieve adequate ROI. Kommula asserts that underlying model capability is no longer the bottleneck, as rapid progress frequently renders bespoke solutions built just a year earlier obsolete. Departmental Fragmentation Drives Compounding AI Overhead (5:19) Industry discussions routinely attribute elevated enterprise AI costs to technical drivers like token consumption, inference infrastructure, and observability tools. Kommula contends that the true driver is organizational fragmentation fueled by market pressure to quickly implement AI point solutions. Individual business units launch isolated pilots that appear cost-effective locally but multiply software and governance costs at the macro level. Because contracts cross multiple business functions, teams independently process identical documents. Legal runs optical character recognition (OCR) scans on historical PDF archives to extract risk terms; procurement extracts purchasing commitments from the same files; sales tracks custom pricing and rebates; and finance monitors regulatory exposure. Enterprise technology leaders subsequently discover overlapping multi-million dollar annual renewals across disparate point solutions, creating duplicate data pipelines and inconsistent definition models. Structural Failures in Conventional Contract Management (13:21) Kommula identifies four structural failure modes behind escalating contract AI costs. First, organizations exhibit tool-first thinking by purchasing vendor software before establishing operational alignment. Second, contract ownership is fragmented: a single document contains terms relevant to multiple departments, meaning units prioritize narrow clauses while no single group manages the contract end-to-end. Third, deal structures suffer from weak standardization when sales teams introduce creative, non-standard terms to close deals quickly, which disrupts downstream workflows in complex sectors like pharmaceuticals and banking. Fourth, enterprises tend to govern reactively post-execution, requiring dedicated monitoring and compliance teams to manually reconcile operations after agreements are finalized. The Contract Obligation Intelligence Framework (17:10) To counter structural breakdowns, Kommula proposes the Contract Obligation Intelligence Framework (COIF), an operating model based on an iterative flywheel: Align, Govern, Intelligence, and Improve & Repeat. Rather than serving as another commercial software tool, COIF standardizes processes on top of existing architectures through three core components anchored by continuous governance. The model relies on upfront Contract Obligation Alignment to review downstream impacts before deal signing, a Governed Contract Data Layer (GCDL) to act as a unified data abstraction layer feeding normalized data to all models and teams, and Connected Contract Intelligence to audit formal obligations against operational actuals in real time. Establishing clean data upstream creates a feedback loop where standardized clause libraries and predictable model outputs systematically lower long-term total cost of ownership. Kommula concludes that enterprises must resolve internal operational fragmentation before investing in code or software point solutions. When broader enterprise-wide consensus cannot be achieved immediately, he advises starting by aligning immediately adjacent business units—such as legal, procurement, operations, and finance—to prevent duplicative pipelines and reduce long-term maintenance overhead.

精炼阅读摘要

McKesson 首席 IT 架构师 Naresh Kommula 指出，企业竞相部署人工智能虽降低了构建合同智能工具的技术门槛，却因部门割裂和数据缺乏治理导致总体拥有成本不降反升；破解这一困境的关键在于建立跨部门协同机制与统一的治理型数据层。降本悖论与合同履约中的规模化失败 (0:54) Kommula 强调，合同是承载企业资金流的核心资产，但履约执行力的欠缺导致了严重的收益泄漏。他引用国际商业与合同管理协会（World CC）的研究数据指出，全球企业因合同签署后的履约运营不善，平均损失高达合同总价值的 9%；以年营收 10 亿美元的企业为例，每年因履约不力造成的经济损失约为 9000 万美元。这种损失在零售、软件、制造及供应链等行业普遍存在，其根源并非缺乏明确的法定义务条款，而在于业务团队在签约后难以贯彻并追踪这些条款。面对运营挑战，众多企业试图借助 AI 实现流程自动化与高级分析，但行业实际落地却面临极高的失败率。Kommula 引用的调研显示：2025 年 MIT 调查指出约 95% 的 AI 试点未能成功规模化或投入生产；2026 年 Gartner 基准数据表明约 60% 的试点未能上线或实现预期 ROI；Gartner 进一步预测，到 2027 年仍将有 40% 的智能体（Agentic）部署无法达到目标回报率。他澄清，这并非源于模型性能不足，当前底层前沿模型的快速更迭已使算法本身不再是瓶颈，企业一年前定制开发的功能如今往往已被通用模型原生吸纳。成本误区与部门孤岛引发的开销激增 (5:19) 在分析 AI 成本激增的成因时，业内讨论往往聚焦于即时的技术开销，例如高频调用大模型 API 产生的 Token 消耗、复杂推理基础设施的算力投入、AI 可观测性监控软件以及底层安全机制的费用。Kommula 认为，尽管针对 Token 和监控工具的优化具有短期价值，但推动企业总成本飙升的真正核心因素是碎片化的点状解决方案。在“唯恐落后”的市场压力下，各业务部门纷纷建立互不连通的独立 AI 试点。这种孤岛式推进引发了严重的成本倍增效应。以合同处理为例，法务部门建立 OCR 试点扫描历史 PDF 以识别法律风险；采购部门使用独立流水线处理相同的文档以追踪采购承诺；销售部门单独提取定价与返利条款；财务及合规团队又各自构建审计模型。单个部门的试点看似成本低廉且具备局部收益，但在企业架构层面，重复的数据抽取流水线和不一致的模型定义，使得 C 级别管理层在年度预算复盘时面临多个部门数百万至上千万美元的重叠软件续约开销，大幅推高了企业的总体拥有成本（TCO）。制约合同智能化落地的四大结构性根因 (13:21) Kommula 归纳了阻碍企业合同 AI 取得成功的四大根本机制缺陷：首先是“工具优先思维”（Tool-First Thinking），业务团队在尚未与下游执行团队建立协同之前，便急于采购商用软件解决眼前痛点，导致系统上线后与实际业务脱节；其次是“权责碎片化”（Fragmented Ownership），一份数十页的商业合同通常混合了质量、采购、返利和监管等多维度条款，各部门仅关注自身相关的条款，形成了“人人都在意合同，但无人对其全生命周期负责”的真空状态。再次是“地基薄弱”（Lack of Standardization），为了快速促成交易，业务前端（如销售团队）频繁拟定高度个性化、非标准化的协议条款，在制药、银行业等高利润行业尤为严重，导致下游履约团队在执行时面临极大的复杂度和运营摩擦；最后是“事后被动治理”（Post-Execution Governance），企业通常将治理拖延至合同签署之后，往往需要临时建立专项合规团队进行人工监控与修补，从而持续消耗组织的人力与财务资源。合同义务智能框架与治理型数据层 (17:10) 为根治上述结构性缺陷，Kommula 提出了“合同义务智能框架”（COIF）。他明确强调，COIF 并非另一款软件工具，而是一套跨部门运作模型，围绕“对齐、治理、智能、改进与复用”四大迭代原则构成持续运转的治理飞轮。该框架由三大相互支撑的核心组件构成：一是“合同义务前置对齐”，要求在合同签署前评估条款对开票、财务、履约等下游部门的实际影响；二是“治理型合同数据层”（GCDL），建立统一的数据抽象层和标准化的合同定义规范，取代各部门重复建设的数据流水线，使所有 AI 智能体、分析看板和报表均消费归一化数据；三是“互联合同智能”，持续将法定义务与实际运营数据进行交叉核验，动态捕捉履约风险与收益漏洞。通过前期的高质量数据输入与跨部门对齐，下游 AI 模型的产出得以保持稳定可信，避免治理漂移。同时，由治理数据提炼出的标准化条款知识库与优化流程能够反哺拟约前端，形成持续降低 TCO 的良性循环。 Kommula 总结指出，企业在规划合同智能化或任何企业级 AI 项目时，最关键的架构原则是在编写代码或采购软件之前解决内部割裂。如果自上而下推动全企业级的统一治理难度过大，组织应优先推动法务、采购、财务等紧密相邻业务部门之间的横向对齐；唯有通过前期管理协同消除数据流水线的重复建设，才能真正控制 AI 部署的长期运维与管理成本。

Detailed notes — zh

## Background 本视频由频道 Ai4 于 2026 年 8 月 19 日发布在 YouTube 平台上。内容由 McKesson 首席 IT 架构师 Naresh Kommula 主讲，深入探讨了企业在部署合同智能 AI 方案时面临的复杂度攀升与治理开销激增困境，并介绍了如何通过受治理的合同数据层（GCDL）与合同义务智能框架（COIF）将合同转化为受治理的企业级数据资产。 ## TL;DR 演讲揭示了当前企业 AI 热潮背后的隐性成本悖论：碎片化与烟囱式的局部 AI 试点不仅难以规模化，反而推高了企业的总体拥有成本（TCO），导致高达 9% 的合同潜在价值因无法有效执行而流失。讲者深入剖析了工具优先思维、权责分散及缺乏统一标准等结构性根源，指出未治理的数据是 AI 失败的真正死穴。为此，他提出了合同义务智能框架（COIF）与受治理合同数据层（GCDL），倡导在早期建立跨部门语义对齐与持续治理机制，通过正向飞轮效应实现低成本、可扩展的企业级 Agentic AI 落地。 ## Atmosphere & style 整场演讲风格严谨务实、直击痛点，兼具技术架构师的全局视野与企业高管的成本敏锐度。讲者引用详实的前沿调研数据和生动的跨部门割裂案例，以沉稳笃定且具警示性的语调层层拆解当下 AI 盲目热潮中的隐性陷阱。叙事节奏明快有力、逻辑环环相扣，不仅给听众带来强烈的行业共鸣与认知颠覆，也为应对 AI 碎片化乱象提供了清晰明了的架构演进蓝图。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=XGSxWviHdc8&t=4s) 全球企业平均有 9% 的合同价值因签署后无法有效落地运营而白白流失，年营收 100 亿美元的企业每年损失可高达近 9 亿美元。 - [00:04](https://www.youtube.com/watch?v=XGSxWviHdc8&t=4s) 企业内部 AI 试点虽然看似成功，但由于缺乏统一的企业级架构视角，总体 IT 与运营成本不降反升。 - [04:00](https://www.youtube.com/watch?v=XGSxWviHdc8&t=240s) AI 规模化落地的最大瓶颈并非模型能力，而是数据未被治理；MIT 调查显示约 95% 的 AI 试点项目未能成功扩展进入生产环境。 - [06:42](https://www.youtube.com/watch?v=XGSxWviHdc8&t=402s) 各部门争相自建 AI 工具形成的烟囱式碎片化（Fragmented AI），导致严重的重复建设与复合成本暴涨。 - [09:25](https://www.youtube.com/watch?v=XGSxWviHdc8&t=565s) 合同数据在法务、采购、销售、财务等部门间严重割裂，针对同一份合同重复扫描提取与建模，推高了整体开销。 - [13:21](https://www.youtube.com/watch?v=XGSxWviHdc8&t=801s) 导致 AI 越用越贵的四大结构性根源包括：工具优先思维、分散的责任主体、缺乏标准化的流沙条款，以及高昂的事后治理成本。 - [17:48](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1068s) 合同义务智能框架（COIF）是一套运营模型而非单一工具，通过前端对齐、受治理数据层、连通智能与持续治理解决碎片化陷阱。 - [17:48](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1068s) 构建受治理的合同数据层（GCDL）作为单一事实来源，能够消除各部门重复购买与建模成本，为 Agentic AI 落地提供底座。 - [20:23](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1223s) COIF 框架依托高质量合同数据赋能端到端自动化与预测分析，形成流程优化与持续降低总体拥有成本（TCO）的良性闭环。 - [21:40](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1300s) 在仓促上线任何 AI 方案之前，必须优先解决数据与架构的碎片化问题，至少在业务紧密关联的相邻部门间建立统一标准与对齐机制。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=XGSxWviHdc8&t=4s) AI 浪潮下的隐性成本陷阱与合同资产困境演讲者 Naresh Kommula（McKesson 首席 IT 架构师）揭开了当前企业 AI 热潮背后的一个悖论：各大企业都在争相部署 AI 试点项目，且许多项目看起来相当成功甚至具备扩展性，但企业的总体技术与运营成本却不降反升。合同作为企业“资金流转的阵地”，其潜藏的经济价值极其惊人。根据 World Commerce 的调查数据，全球企业平均有高达 **9% 的合同价值** 未被有效开发或白白流失。对于一家年营收 100 亿美元的大型企业而言，这意味着每年会直接损失近 **9 亿美元** 的潜在收入。造成这种巨额流失的原因并非合同条款没谈好，而在于合同签署完成后，后端的业务运营团队无法有效地将其“行动化”与“ operationalize（运营落地）”，导致大量约定好的商业权益无法转化为实际收益。 - **核心矛盾**：AI 试点虽然成功上线，但由于缺乏统一的架构与企业级视角，反而推高了整体 IT 与运营成本。 - **合同价值流失率**：根据 World Commerce 调查，合同总价值的 **9%** 因后期无法有效执行而流失。 - **金额示例**：100 亿美元营收规模的企业，每年因合同执行力不足导致的流失可高达 **9 亿美元**。 --- ### [00:240] AI 试点成潮但难以规模化：从 MIT 到 Gartner 的预警数据虽然各大业务部门都在积极开展持续改进（CI）项目并引入 AI 工具，但从全行业数据来看，AI 试点的落地与投入产出比（ROI）前景并不乐观。演讲者引用了权威机构对 AI 项目成功率的统计与预测： ```mermaid flowchart LR accTitle: AI Pilot Success Rate Predictions (2025-2027) subgraph Y2025["2025 年 (MIT)"] A["95% 试点"] --> B["无法规模化/投入生产"] end subgraph Y2026["2026 年 (Gartner)"] C["60% 试点"] --> D["无法上线或达到预期 ROI"] end subgraph Y2027["2027 年 Agentic 时代"] E["40% Agent 解决方案"] --> F["难以产生足够的 ROI"] end end ``` 演讲者指出，AI 模型本身的技术能力早已不是瓶颈，模型演进速度极快，“一年前搭建的定制化解决方案，如今已被最新的前沿模型（Frontier Models）或原生服务直接覆盖”。真正的死穴在于 **数据未被治理（Ungoverned Data）** 以及缺乏统一的企业级规划。 - **MIT 2025 年调查**：约 **95%** 的 AI 试点项目未能成功扩展或进入生产环境。 - **Gartner 2026 年预测**：约 **60%** 的 AI 试点项目无法顺利上线或达到预期的 ROI。 - **Gartner 2027 年预测**：进入 AI Agent（智能体）时代后，仍有 **40%** 的 Agentic 解决方案无法创造足够的商业价值。 - **技术现状**：模型能力过剩且快速迭代，技术本身已不再是限制规模化的主因。 --- ### [06:42](https://www.youtube.com/watch?v=XGSxWviHdc8&t=402s) 烟囱式 AI（Fragmented AI）：企业成本激增的罪魁祸首除了 Token 消耗费用、推理算力成本以及模型信任度等常见的表面问题外，演讲者指出了一个遍布各类企业且极具隐蔽性的根源问题——**碎片化/烟囱式的 AI 解决方案（Fragmented AI Solutions）**。在“不搞 AI 就落后于市场”的集体焦虑下，各个部门的领导者争相引入各自的 AI 工具，由此形成了极其严重的部门壁垒。这种碎片化在单个部门内部看起来效率极高、ROI 也很亮眼，但站在 CTO、CIO 或企业架构师（Enterprise Architects）的高层视角来看，各个部门每年数百万甚至上千 shared 预算的叠加，造成了严重的重复建设与成本复合暴涨（Compounding Cost Impact）。 - **部门盲区**：单看法务、采购或销售部门，其自建的 AI 试点均表现良好，但放眼全公司则形成了严重的资源浪费。 - **高管困境**：CTO/CIO 在面对各部门每年数百万元的续约申请时，会发现碎片化工具带来的“隐性复合成本”已难以承受。 --- ### [09:25](https://www.youtube.com/watch?v=XGSxWviHdc8&t=565s) 多部门交织下的合同数据乱象合同是企业中风险最集中、价值最高，但也是 **治理最薄弱、碎片化最严重** 的领域。由于一份合同往往涉及法务、采购、销售、财务、合规及质量管理等多个部门，各部门在处理历史合同（Historical Contracts）时，均采取了“各自为政”的扫扫描与提取策略。 ```mermaid flowchart TD accTitle: Fragmented Contract Intelligence Ingest Pipelines DOC["历史合同 PDF 文档"] subgraph SiloedPipelines["各部门独立的 AI/OCR 管道"] Legal["法务团队"] -->|自建 OCR / 扫描| LegalModel["合同风险识别 Agent"] Procurement["采购团队"] -->|自建 OCR / 扫描| ProcModel["承诺与履行跟踪 Agent"] Sales["销售团队"] -->|自建 OCR / 扫描| SalesModel["返利与业绩分析 Agent"] Finance["财务/合规团队"] -->|自建 OCR / 扫描| FinModel["账单与条款审查 Agent"] end DOC --> Legal DOC --> Procurement DOC --> Sales DOC --> Finance ``` 法务关心历史合规风险，采购关注供应商承诺，销售看重返利与条款，财务关注扣款与发票。结果是，同一个 PDF 文件被四个部门重复调用不同的 OCR 工具扫描了四次，各自建立了不相通的数据模型。这种“便宜试点”积少成多，最终大幅推高了企业的总体运营开支。 - **多头扫描**：法务、采购、销售、财务等部门各自购买软件或使用 LLM 对同一份历史合同文本进行重复提取。 - **标准冲突**：由于缺乏统一的数据标准，各部门对同一个合同条款的理解与解析规则互不兼容。 --- ### [13:21](https://www.youtube.com/watch?v=XGSxWviHdc8&t=801s) 四大核心结构性病灶：为何 AI 越用越贵演讲者剖析了导致企业 AI 陷入“越搞成本越高”困境的四个底层结构性根源： | 根源问题 | 表现特征与后果 | |:--- |:--- | | **工具优先思维 (Tool-First Thinking)** | 盲目跟风“为了用 AI 而用 AI”，在未达成部门间业务对齐前就急于采购或搭建工具，忽略了跨部门协作与长远规划。 | | **分散的责任主体 (Fragmented Ownership)** | 一份 50 页的合同包含数十个维度的条款，各部门只关心属于自己的条款（如质量、采购、返利），“人人关注合同，却无人对合同整体负责”。 | | **缺乏标化与“流沙”条款 (Weak Sand / Lack of Standard)** | 销售或业务团队为了抢下订单而使用大量极其不标准、极其复杂的创意性条款，导致后端的运营与履约系统根本无法自动执行。 | | **事后治理成本过高 (Governance After Execution)** | 合同签署时缺乏管控，等履约出问题后才临时组建合规团队进行“事后追查”，不仅无法根治乱象，反而额外增加了治理团队的薪酬成本。 | --- ### [17:48](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1068s) 解决方案：合同义务智能框架（COIF）与治理数据层为彻底解决碎片化带来的成本陷阱，Naresh Kommula 提出了一套非工具性的企业级架构运营模型——**合同义务智能框架（Contract Obligation Intelligence Framework, COIF）**。该框架基于“飞轮架构”（Flywheel Architecture），强调通过治理与对齐来替代盲目的工具堆叠。 ```mermaid flowchart TD accTitle: Contract Obligation Intelligence Framework (COIF) subgraph COIF["合同义务智能框架 (COIF)"] direction TB A["1. 合同义务对齐 (Contract Obligation Alignment)"] --> B["2. 受治理的合同数据层 (Governed Contract Data Layer)"] B --> C["3. 连通的合同智能分析 (Connected Contract Intelligence)"] C --> D["改进与循环 (Improve & Repeat)"] D --> A CG(("核心控制：持续性治理 (Continuous Governance)")) CG -.- A CG -.- B CG -.- C end ``` COIF 框架包含三大核心支撑与一个中央控制引擎： 1. **合同义务对齐（Contract Obligation Alignment）**：在合同签署前及搭建系统前，先行明确前端条款对下游行政、财务、开票、履约团队的具体影响，奠定统一语义。 2. **受治理的合同数据层（Governed Contract Data Layer, GCDL）**：构建单一事实来源（Single Source of Truth）。无论上层是采购 Agent、财务 AI 还是日常报表，统一从该数据层提取合同数据，避免重复建模与定义。 3. **连通的合同智能分析（Connected Contract Intelligence）**：将静态合同条款与后端的实际业务执行数据（Actuals）进行动态比对，自动捕捉漏损与合规风险。 4. **持续性治理（Continuous Governance）**：作为飞轮的正中央引擎，持续监控并贯穿“对齐-治理-智能”全流程，从源头降低企业的 AI 部署与维护成本。 - **核心定位**：COIF 并非又一款软件工具，而是一套指导企业如何组织数据与业务协同的 **运营模型（Operating Model）**。 - **业务收益**：通过一次性定义与统一数据层，避免各部门重复购买 AI 算力与扫描工具，实现真正可扩展、低成本的 Agentic AI 应用。 <ElicitationsGroup message="如需进一步探讨此话题，您可以了解以下方向："> <Elicitation label="深入了解受治理的合同数据层 (GCDL) 的具体技术架构" query="请详细解释受治理的合同数据层 (Governed Contract Data Layer, GCDL) 的具体技术架构与数据模型设计。"/> <Elicitation label="对比解析传统的 CLM 系统与 COIF 运营模型的本质区别" query="请对比传统的合同生命周期管理系统 (CLM) 与讲者提出的合同义务智能框架 (COIF) 在运营理念与架构上的区别。"/> </ElicitationsGroup> ### [20:23](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1223s) COIF 框架的核心价值与正向闭环飞轮 Naresh Kommula 详细阐述了合同履约智能框架（COIF）与受治理合同数据层（GCDL）所带来的核心业务收益。他指出，该模型不仅能显著优化合同的履约表现和日常运营成本，还能从根本上降低企业拥有 AI 解决方案的总体拥有成本（TCO）。其底层逻辑在于：在流程初期建立跨部门的紧密对齐，能够直接催生出高质量的合同义务与条款数据；而坚实高质量的数据底座，进一步赋能了更深层次的合同智能与端到端自动化，确保底层 AI 模型能够稳定输出符合预期的精确结果，并为企业提供强大的合同预测性分析能力。这种智能分析能力反过来会驱动企业未来业务成果的持续改进。通过深度洞察合同履约情况，企业能够切实节省成本、重塑优化业务流程，并加速合同条款与模板的标准化。"每个环节都会形成闭环并反哺其他环节"，随着时间推移，企业将在持续演进的良性循环中不断压降成本、提升整体运营效率，这正是整个 COIF 模型的底层设计哲学。 ```mermaid flowchart TD accTitle: COIF 模型闭环演进机制 subgraph Alignment_Data["治理与数据基础"] A["跨部门早期协同对齐"] --> B["高质量合同义务数据"] end subgraph Intelligence_Automation["智能与模型赋能"] B --> C["深度合同智能与自动化"] C --> D["精准模型输出与预测分析"] end subgraph Optimization_Feedback["价值产出与持续迭代"] D --> E["业务流程优化与合同标准化"] E --> F["降低 TCO 与运营成本"] F -.->|"正向反馈反哺"| A end ``` - **数据质量提升**：前期对齐消除了语义偏差，直接沉淀结构化、高保真度的合同条款与履约义务数据。 - **预测与自动化**：依托高质量数据，预测分析能力得以精准落地，模型输出高度契合业务预期。 - **持续闭环优化**：智能化洞察持续推动合同标准化与流程改造，形成降低运营与 AI 拥有成本的良性循环。 --- ### [21:40](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1300s) 避免 AI 仓促上马与跨部门协同结语在演讲尾声，Naresh Kommula 向与会者分享了至关重要的最后建议。他告诫大家，在落地任何 AI 解决方案之前——不论是合同智能还是更广泛的生成式/通用 AI 项目——切忌急于求成、盲目在各自孤立的部门内仓促上线系统，而必须优先解决"数据与架构的碎片化问题"（Fragmentation Issue）。他强调，各部门应当主动推动跨组织的横向对齐；即使短期内无法做到全企业范围的统一，也至少应当协同相邻、业务紧密关联的部门。"仅仅是这一个决策，就能从企业整体层面大幅降低 AI 解决方案的拥有成本。"在全场热烈的掌声中，Naresh 结束了演讲，并开放现场问答交流。 - **核心治理建议**：在部署 AI 之前，先着手治理孤岛与碎片化，而非各自为政快速堆砌局部方案。 - **务实落地策略**：优先在具备直接业务往来的相邻部门之间建立统一的数据标准与对齐机制。 - **终极商业价值**：通过打破壁垒与共享框架，最大化摊薄企业级 AI 系统的研发与运维开销。 ## Notable quotes > [04:00](https://www.youtube.com/watch?v=XGSxWviHdc8&t=240s) 一年前搭建的定制化解决方案，如今已被最新的前沿模型（Frontier Models）或原生服务直接覆盖 — Naresh Kommula > [06:42](https://www.youtube.com/watch?v=XGSxWviHdc8&t=402s) 碎片化在单个部门内部看起来效率极高、ROI 也很亮眼，但站在 CTO、CIO 或企业架构师的高层视角来看，各个部门每年数百万甚至上千 shared 预算的叠加，造成了严重的重复建设与成本复合暴涨 — Naresh Kommula > [09:25](https://www.youtube.com/watch?v=XGSxWviHdc8&t=565s) 同一个 PDF 文件被四个部门重复调用不同的 OCR 工具扫描了四次，各自建立了不相通的数据模型 — Naresh Kommula > [13:21](https://www.youtube.com/watch?v=XGSxWviHdc8&t=801s) 人人关注合同，却无人对合同整体负责 — Naresh Kommula > [17:48](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1068s) COIF 并非又一款软件工具，而是一套指导企业如何组织数据与业务协同的运营模型 — Naresh Kommula > [20:23](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1223s) 每个环节都会形成闭环并反哺其他环节 — Naresh Kommula > [21:40](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1300s) 在落地任何 AI 解决方案之前——不论是合同智能还是更广泛的生成式/通用 AI 项目——切忌急于求成、盲目在各自孤立的部门内仓促上线系统，而必须优先解决数据与架构的碎片化问题 — Naresh Kommula > [21:40](https://www.youtube.com/watch?v=XGSxWviHdc8&t=1300s) 仅仅是这一个决策，就能从企业整体层面大幅降低 AI 解决方案的拥有成本 — Naresh Kommula ## Mentioned Naresh Kommula, McKesson, World Commerce, MIT, Gartner, Governed Contract Data Layer (GCDL), Contract Obligation Intelligence Framework (COIF), Agentic AI, CLM ## Video description Naresh Kommula - Lead IT Architect, McKesson AI is making it easier than ever to build & deploy contract intelligence solutions, yet many organizations are experiencing rising complexity, governance overhead, and inconsistent AI outcomes. As AI expands across legal, procurement, finance, compliance, and operations, organizations often create multiple versions of contract intelligence, each with its own obligation models, governance processes, and data pipelines. This session explores why contracts must be treated as governed enterprise data assets rather than unstructured documents. Attendees will learn how the Governed Contract Data Layer (GCDL) and Contract Obligation Intelligence Framework (COIF) can create a shared operational language for contractual obligations, improve compliance, reduce duplication, and provide a trusted foundation for scalable AI and emerging agentic capabilities. [Source](https://www.youtube.com/watch?v=XGSxWviHdc8)

Source quotes

[4:00] Naresh Kommula: 一年前搭建的定制化解决方案，如今已被最新的前沿模型（Frontier Models）或原生服务直接覆盖 [6:42] Naresh Kommula: 碎片化在单个部门内部看起来效率极高、ROI 也很亮眼，但站在 CTO、CIO 或企业架构师的高层视角来看，各个部门每年数百万甚至上千 shared 预算的叠加，造成了严重的重复建设与成本复合暴涨 [9:25] Naresh Kommula: 同一个 PDF 文件被四个部门重复调用不同的 OCR 工具扫描了四次，各自建立了不相通的数据模型 [13:21] Naresh Kommula: 人人关注合同，却无人对合同整体负责 [17:48] Naresh Kommula: COIF 并非又一款软件工具，而是一套指导企业如何组织数据与业务协同的运营模型 [20:23] Naresh Kommula: 每个环节都会形成闭环并反哺其他环节 [21:40] Naresh Kommula: 在落地任何 AI 解决方案之前——不论是合同智能还是更广泛的生成式/通用 AI 项目——切忌急于求成、盲目在各自孤立的部门内仓促上线系统，而必须优先解决数据与架构的碎片化问题 [21:40] Naresh Kommula: 仅仅是这一个决策，就能从企业整体层面大幅降低 AI 解决方案的拥有成本

Source quotes

[0:54] Naresh Kommula: that's where the money lives. [7:48] Naresh Kommula: If you're not implementing a solution with AI, they see you as... not up to the market. [13:21] Naresh Kommula: everyone cares about the contract, but no one cares as a whole.

Original transcript

[0:04] Uh today I'm going to speak about the AI rush. How the [snorts] AI rush is making the contact [0:09] intelligence costing more not less. I'm sure many of you in your organization implementing different [0:17] AI solutions, AI pilots. Some of the pilots may be scaling. Some of the pilots are already moving [0:23] into the productive and running very well. But all in all there is a underlying reasons [0:30] the pilots are getting successfully implemented solutions are even scaling but the cost is raising [0:36] more. So today I'm going to walk you through the what are the underlying reasons behind this [0:42] and how we could resolve and how we could make the enterprise mindset. [0:54] Uh it's not moving. Yeah. So before I dive into that uh first let me u [1:02] set on the what are we talking about the stakes contracts is one of the [1:09] important element in our any organization that's where the money lives so as per the world commerce [1:16] survey there is around 9% of the contract value is not being tapped by the organizations so 9% of [1:25] the contract value means if an organization with 9 billion or $1 billion revenue they're losing the [1:33] about $90 million of 90 million revenue they're losing it. So what it means it it is not because [1:42] they're not in the contract they're not in agreed terms but it's because their operations be after [1:49] the contracts are executed the operations actually not able to operationalize those contracts [1:54] and because of that it's costing them or their the revenue they're not able to capitalize. [2:00] So those are the stakes we're talking about when it comes to contract. every company whether it is [2:05] a manufacturer whether it is a supplier whether it is a software company or any retail company anyone [2:11] in the organization they always talk about this contract right so that is a bread and butter [2:16] so that those are the stakes we're talking about and um I'm sure it's not like it's not recognized [2:20] problem every leader trying to operationalize the contract so that they could untap the [2:27] potential of the contracts and they could they could identify the leakage save the money for [2:32] the organization Right. So with that there are programs are there to operationalize the contracts [2:40] and some of them are like I said delivering and many of the solutions are really working well. [2:46] But what is interestingly happening is with this AI rush, it is causing like lot of this [2:55] operational teams are started to implement more and more improvement projects right [3:00] like a continuous improvement projects and um with that we see AI is everywhere. [3:08] So you see every I mean that's where the all we are here right. So every team every organization [3:15] department are implementing the AI solutions to improve the automation efficiency and [3:21] generate advanced insights. But in reality what is happening based on the survey from MIT in 2025 [3:33] with this rush in the AI solutions there are about 95% of the [3:39] sol pilots couldn't scale or couldn't be productionize that is a last year survey from MIT [3:49] and around in 2026 there are around 60% of the [3:55] pilots are not able to go live or they are not able to scale to the expected ROI [4:03] and that is from the Gartner that's what they are predicting about it [4:08] in 2027 we are moving into the agentic era right and everyone is now gone are the days [4:14] with the chart apps gone are the days with the pilots now we are moving into the agentic world [4:20] but even in the the interesting thing is the Gartner survey says that [4:23] there are about 40% of the pilots or the solutions we are going to do in 2027 might not be able to [4:31] produce the enough ROI. So those are all the real numbers and the reason behas [4:38] there are a lot of reasons behind it right and it's not just because of the the teams are not [4:42] good or the solutions are not good it's not because of the models are not going well [4:47] everyone is already seeing the on the floor that there are models are performing very well [4:52] and you you see that uh techn is no more a problem right I mean uh there are every day there is a [4:57] new model is coming in the solution which we implemented a year ago is absolute now the same [5:02] solution is already coming part of the frontier model are the the services provided. Right? [5:09] So the there are core reasons behind why this the such no improvement in terms of the pilot failure. [5:19] So one of the reason as you know ungoverned data is one of the core reason I am sure you [5:27] heard about this everywhere you go every LinkedIn post every talk you see you might be hearing that [5:34] there is a there is a problem we are not able to scale our solutions because of the ungoverned data [5:39] and how many of you are resonate with this like yes ungoverned data is one of the problem no [5:49] Okay. So you might be hearing this a lot but I it and because of the data everywhere [5:55] we need to have a governed data and we need to have the a right data layer so that way we [6:00] need to have a right agentic solutions we could implement right but because of this [6:08] the ungoverned data is recognized problem. I see that many of the organizations are [6:12] some of them are in advanced phase some of them are in not in advanced phase of recognizing this [6:16] and implementing it. But the harder challenge or what we always see is in many organizations [6:22] we don't really we always recognize the problem but when it comes to scale large organizations [6:26] what happens is bringing in that alignment sometimes is becoming a harder challenge. So [6:32] definitely lot of the times what we see is the ungoverned data is [6:35] one of the core problem where the agentic era solutions will not be able to scale. [6:42] But apart from that there are other problems which could cost you more [6:47] which makes your AI solutions will be increasing the overall cost of the enterprise. Any guesses? [6:57] You might be hearing about the token cost. You might be hearing about the inference [7:03] or you might be hearing about the underlying trust. Right? Those are all the reasons [7:08] we always hear. Even in the when you go to the shop floor on the exhibition hall, we see that [7:14] there are lot of number of vendors talking about [7:18] we have the lot of solutions to observability. A lot of solutions for the token cost reduction. [7:24] So there are all good good observations, right? They're all like try to reduce your overall cost [7:28] of the AI solutions. But there is one problem I see consistently based on my conversations with [7:33] lot of leaders across the industry and based on my experience in this contact intelligence [7:38] there's one biggest problem it is I can resonate with a lot and also I see this is [7:43] not only in the contract intelligence but it's also across the different solutions [7:48] that is a fragmented AI solutions what I mean by that if you really look at our organization [7:58] lot of the rush at every leader, every business leader or IT leader in a [8:06] wanted to implement AI solutions, right? That is a rush and what is causing that rush? [8:13] They're running behind implementing a solution get some value and all of lot of the times what the [8:21] the interestingly if you're not implementing a solution with AI means they see you as oh [8:26] you are not up to the market right so that what is causing this because of that rush [8:33] the solutions are being implemented in a fragmented siloed manner [8:38] that is a fundamental one of the reason which I'm going to address today what how that is actually [8:44] creating a problem for an enterprise and that fragmentation looks simil simpler or smaller for [8:54] our own department we don't really see that when I am implementing a solution in my own department [8:58] I might feel that I'm really doing a good job and in fact in fact it is true so we do really [9:05] good job as our own our team as ourself but when it comes to when you look at an enterprise scale [9:11] this fragmentation creates cost. That's what I'm going to talk about how this fragmentation is [9:18] creating a issue for us as an organization and what is a framework I created to overcome that [9:25] here I wanted to come back to this onto the contracting topic because I know you are all [9:30] here AI but contracts is where the whole the risk is collide right we have an all the money is there [9:38] if you really understand your contracts well you can reduce your leakage you can reduce your risk [9:43] you can mitigate all your exposure to the complaints all of that is there that's where [9:48] everyone is implementing different solutions and but this is a where the most valuable [9:54] least governed and fragmented solution as I was telling you so if I go into the details generally [10:04] in contract intelligence contract touches about every department in an organization right so [10:11] we have a legal team impacts on uh contract they they are the ones who are drafting those contracts [10:17] they are the ones actually really starting the putting the first hand on the legal document right [10:21] so what we really see when you go and talk to any AI companies now they said oh you get a PDF copy [10:27] of your historical contracts we here are here to help you we scan it we give you the OCR copy [10:33] we give you the intelligence around it you go ahead and do do whatever you want so that you can [10:37] understand your risk in the historical contracts that is the biggest problem we see there are a lot [10:41] of Startups also evolved. Even the our regular CLM tools also implemented their own solutions [10:50] and the legal team is one of the starting point where they implemented the OCR. I can give you all [10:55] my thousands of documents from my historical last 10 years. Help me realize the value [11:03] and those pilots are successful too legal team. Right? Then when it comes to procurement team [11:07] they wanted to capture about their own commitments and how are they doing this uh commitment right [11:13] procurement also team m take the same PDF copies they again run this pilots on their own department [11:20] because legal team priorities are a little bit different than the procurement team department [11:24] similarly sales similarly finance and complaints and even many other departments [11:32] even the quality management team could be because everyone recognize there is a value [11:38] in the all our historical contracts sometimes like I mentioned in our beginning of the conversation [11:43] there is a so much value you could unpack if you really understand your contracts well so [11:47] that's where the rush was creating each department is creating their own pilots [11:53] so with this rush what is happening is there are many pilots but the fragmentation continue to [12:05] So what is happening in an enterprise or scale these pilots are individually successful [12:13] because as I said every department know what they wanted to do [12:17] and they know what they wanted to achieve with this pilot. they have a better ROI with that pilot [12:22] and they get an approval from funding and everyone is happy in their own department right but because [12:27] of these cheap pilots [clears throat] what's happening is the fragmentation is creating because [12:33] of this fragmentation is raising your enterprise cost that's what the the underlying results so [12:42] as a leader of the organization or a CTO or CIO or a enterprise SP roles you are sitting [12:50] And when you are in that level, when you are with the visibility, the lower level department teams [12:55] could could justify the cost. But when it comes to your approval, every every year you are coming [13:01] to a million dollar budget renewal or $10 million of renewal from each department. When you really [13:04] see what happens, the leaders at that level see okay, I have this pilot here, this pilot here, [13:10] that that cost is comp like compounding. So this fragmentation creating the compounding [13:16] impact and it's creating the cost. that is what the the problem. [13:21] So with that I'm trying to transition into a little bit [13:26] different on why are what are the underlying reasons right like I just uh speak about [13:30] some of the core structural problems which is driving us to that problem as an enterprise [13:37] uh teams. So one of the thing is the tool first thinking like what I say what I mean by that [13:46] as we discussed in the beginning because of this rush of the AI the teams are coming up with the [13:53] okay let's implement a solution AI is there let's solve the all the world problems right [13:57] and this is where the teams are implementing the the thinking about okay I'll go ahead [14:01] and implement the tool but they forget about that alignment with the different departments [14:05] maybe we should align with the different departments so those are the the tool first [14:08] thinking actually creates one of the thing because of this rush and the second thing is [14:14] the fragmented ownership is another underlying reason why this fragmentation uh creates [14:20] as I mentioned contracting is an interesting topic right if you really look at in a supply chain in [14:25] another department every that is a little bit cleaner but when it comes to contract [14:30] the biggest thing here is contract has hundreds of terms and every department have their own terms [14:38] geneneralally What happens in our organizations the one department does not necessarily care [14:43] about the other department terms which are written in the contracts. So if I have a 50page document [14:48] there are terms related to quality there are terms related to procurement and sales performance [14:53] rebates or regulatory compliance. What happened? Every department owns their own um [15:00] that that obligations. When it comes to can we have a consolidated solution, they don't [15:06] really necessarily see it because everyone cares about the contract but no one cares as an as a [15:14] whole like as a whole right like the one solution. So that fragmentation is one of the reason [15:20] and the second thing is the weak sand is one of the reason the third one. [15:25] So this what what I mean by that weeks and days is generally what happens whenever you have the lot [15:29] of contracts are writing in an organization the terms are really creative the sales team will go [15:35] go and talk to a customers and get a deal and they wanted to be creative so that they could win a [15:41] the customer right so with that what happens the obligations which are written in the contracts [15:46] are so complex many times they don't follow the standardization so like for example there is a in [15:53] a pharma industry or in a banking industry or in any industries where their commissions are there. [15:59] So they they create a they come up with a creative terms in a contract the they [16:03] so that it's very it becomes very problem for the organizations to operationalize in the post [16:08] operation of the contract. So the standardization is missing because of that rush okay I wanted [16:13] to execute the deal quickly so that I'll implement that's another fun underlying reason [16:18] and the fourth reason is governance after execution. So what I mean by that [16:26] when they're writing the contracts they always think about okay we will we are good with it's [16:30] all following the business is getting some money but when it comes to operationalizing the contract [16:36] they realize that okay this is becoming complex then they said okay let's implement some [16:40] governance process across the departments but by what happens because of that it didn't it creates [16:46] another governance team to be established to you continuously monitor that process unless and until [16:52] you stop in the thing that you if you don't do the standardization if you don't do the the cycle [16:58] it creates another cost. So these are all there are like there are more than what these four [17:03] structural problems but I I try to make it more to the our AI rush how this could be evolved. [17:10] So the idea is in order for us to as a enterprise leaders to address this. [17:17] So there are I come up with an operating model and that operating model actually helps you [17:23] in thinking a little bit differently than in a regular way. [17:28] What I mean by that it's not another tool I'm going to propose but it's an operating model [17:34] which is helps the organizations to think about this holistically and bringing that alignment [17:40] that reduces your overall cost of the solutions in the rush of the AI. [17:48] So this operate this model I call it as a contract obligation intelligent framework. [17:55] So it it it works on a core principles of the align, govern intelligence, improve and repeat. [18:03] So this is a flywheel architecture. It's a a methodological operating model [18:08] where as a as an organization bring that alignment across the board. So what it has [18:16] it is not a solution or technology. It is a operating model as I said. So you don't want [18:22] to bring another AI tool or another AI solution to solve this problem. There are enough technologies [18:27] available enough improvement is there in the solution. So the operating model is what [18:32] actually plays a important role. So the first one is talk about the contract application alignment. [18:39] The first component it talks about how as an organization we need to align [18:44] how if I'm writing a contract certain way how it is going to impact my downstream teams like [18:50] administrative teams, finance teams, invoicing teams, ordering teams how it impacts them. [18:55] So there are certain principles we have to follow to get that alignment first [19:01] and before you implement anything get the alignment first as an organization so that you [19:04] could foresee what are the challenges you're going to face. The second component is talks [19:10] about the governed contract data layer. So this is a layer where you talk about [19:15] you implement a governed data layer like you all know that the governance is one of the problem [19:20] right. So you need to have a a govern contact data layer which actually understand the contact ways [19:25] how the contacts are reading in different ways and you manage the data in a one govern layer so that [19:30] any AI agents you're building any AI solutions you're building even the regular operation [19:35] reports you're building it consumes the data from that govern layer rather than multiple places. [19:40] So the same procurement team come here, finance team comes here, sales team comes here, even all [19:45] the department teams comes here and get the data from there so that it's governed one definition [19:49] across the enterprise so that you don't have to repeat the definition again and again and again. [19:55] And the third component of this fly is talks about the connected contact intelligence. That's what I [19:59] was just touched upon briefly. So where you have this continuously you take your contracts and you [20:06] take your actuals and generate intelligence around this. So when you look at it on the center of that [20:12] is a continuous governance as I said right the governance is actually a critical player here. [20:17] It aligns all these three components of the flywheel and it helps you continuously monitor [20:23] your contract performance operational cost and also your overall cost of the solution ownership. [20:33] So the benefits of this model is you have a better alignment in the beginning which creates a high [20:41] quality obligation data when I say contract terms and you have the the high quality data will help [20:48] you enable the deeper intelligence and automation and so that you your models are perfect you get a [20:55] expected results and you have the better predictive analytics of around your contract [21:02] and that intelligence drive you the continuous improvement in future outcomes too. Right? So [21:08] what is happening here that intelligence helps you saving the cost or improving your business process [21:14] or standardizing your contracts etc. And each of these loop will [21:21] feed back to the other loop. So every time over a period of time you're as your organization [21:26] you're improving and you're reducing the cost and improving your operational efficiency. [21:30] So that is the underlying principle actually I wanted to explain about this the co model. [21:40] So one last thought before I close this conversation [21:44] before you implement any AI solution even when it comes to contact intelligence or general AI [21:52] try to fix that fragmentation issue rather than rushing and implementing a solution [21:56] in our own department try to bring that alignment across the organization so that maybe not across [22:02] organization at least the departments which are next to each other and I'm sure [22:07] that one decision will reduce the cost of your AI solution ownership as an enterprise. [22:14] So with that I'm actually completing the session. I'll open up any questions on the co model or [22:21] anything. Thank you so much for your time. I really appreciate you having him here. [applause]