TCS Rejects 'Fail Fast' AI as 3,000 Enterprise POCs Miss ROI Targets

From Insight to Autonomy Across the Enterprise — Ai4

Ai4

Tata Consultancy Services argues that corporate AI investments fail to generate returns because organizations deploy technical agents in isolation without first redesigning the end-to-end business value chain.

AI-assisted summary of From Insight to Autonomy Across the Enterprise — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by the Ai4 channel on August 29, 2025, this presentation features Kathirvelan Ganesan, Solutions Head for the AI Practice at Tata Consultancy Services (TCS). The session details how organizations can move beyond isolated generative AI experiments to transform end-to-end enterprise value chains through autonomous systems and human-machine collaboration.

## TL;DR Kathirvelan Ganesan discusses why thousands of enterprise generative AI proofs of concept have stalled in delivering measurable business ROI, urging organizations to transition from point solutions to full value-chain transformation. He highlights the necessity of completing upfront business design and persona mapping before drafting technical specifications, warning that the startup mantra of 'failing fast' is hazardous for enterprise-scale rewrites. Using real-world examples across contact centers and healthcare, he demonstrates how varied agent archetypes and human-in-the-loop guardrails operate in practice. Finally, he outlines TCS's partnership and co-funding programs with AWS to support enterprise Process-to-Agent (P2A) initiatives.

## Atmosphere & style The presentation maintains a pragmatic, seasoned, and consulting-driven tone that cuts through generative AI hype with enterprise realities. Ganesan presents with warm, approachable authority, pairing cautionary architectural anecdotes with structured operational frameworks. The session moves at a dense, rapid pace before arriving at an abrupt, polite cutoff by the moderator when the session clock expires.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=4s) Tata Consultancy Services (TCS) operates at massive global enterprise scale, with approximately 600,000 associates across 55+ countries delivering AI transformation across major industry sectors. - [01:59](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=119s) While thousands of GenAI POCs have transitioned into production via soft launches, enterprises still struggle to realize expected business metrics and ROI due to siloed implementations. - [01:59](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=119s) Rapidly shifting foundation models and frameworks prevent enterprise leaders from stepping back to align technological investments with core business outcomes. - [04:20](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=260s) TCS organizes enterprise transformation around three strategic pillars: Reimagining Business Value Chains, Autonomous Systems, and Human-AI Collaboration. - [04:20](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=260s) Value chain transformation is structured across three lenses: Value (experience and growth), Path (knowledge synthesis and assurance), and Realize (breaking down IT, operations, and business silos). - [04:20](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=260s) Autonomous systems require a combined stack of institutional data, models (LLMs/SLMs), and multi-agent frameworks to power the orchestrated 'New IT'. - [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) Business design must precede technical design; jumping directly into technical documents and isolated agent code leads to integration and orchestration failure across complex workflows. - [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) Startup mantras like 'fail fast' are ill-suited for enterprise-scale value chain transformation; organizations must 'think big, start small, and course-correct'. - [13:32](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=812s) Transforming the contact center requires moving beyond isolated chatbots to end-to-end orchestration using platforms like 'Contact Synergy' with embedded quality and compliance human-in-the-loop checkpoints. - [13:32](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=812s) Agent architectures must be matched to operational complexity, ranging from simple and model-based reflex agents to goal-based and utility-based agents. - [16:26](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=986s) In highly regulated domains like healthcare, end-to-end value chain automation demands auditable human oversight to ensure patient safety and HIPAA/GDPR compliance. - [17:50](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1070s) TCS and AWS partner deeply to de-risk enterprise AI adoption through programs like the GenAI Rapid Program, POC co-funding, and the Process-to-Agent (P2A) initiative.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=4s) Introduction and the Scale of Tata Consultancy Services

Kathirvelan Ganesan opens his presentation by acknowledging that his full name can be tricky to pronounce, warmly inviting the audience to simply refer to him as "GK or Kathir, whatever you want." Serving as the Solutions Head for the AI Practice at Tata Consultancy Services (TCS), he begins by polling the room on their familiarity with TCS. As part of India's largest multinational conglomerate, the Tata Group, TCS operates at massive enterprise scale, delivering consulting, IT services, and comprehensive business solutions across more than 55 countries with a global workforce of around 600,000 associates and over 200 delivery centers worldwide.

The practice spans virtually every major enterprise sector, including banking, financial services, life sciences, healthcare, manufacturing, and communications, media, and information services (CMI). Kathirvelan frames the session's core agenda: examining how TCS is actively helping global organizations transform their end-to-end business value chains using generative AI and agentic systems, while openly addressing the day-to-day practical hurdles and real customer experiences encountered along the way.

- **Global Presence:** Operating across 55+ countries with 200+ global delivery centers. - **Workforce:** Approximately 600,000 associates worldwide. - **Industry Footprint:** Banking, Financial Services, Healthcare, Life Sciences, Manufacturing, and CMI.

---

[01:59](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=119s) From GenAI Hype to Reality: Moving Beyond Isolated POCs

Generative AI has decisively moved past its initial hype cycle and transitioned into reality; enterprises universally acknowledge that AI is fundamentally reshaping the business landscape. In the early wave of adoption, organizations rushed to build proofs of concept (POCs) and prototypes. However, moving those prototypes into production proved exceedingly difficult. To bridge the gap, enterprise teams adopted soft-launch strategies—deploying GenAI within isolated departments, evaluating preliminary results, and gradually expanding outward. Through this staged approach, more than 3,000 POCs have now successfully transitioned into production across the enterprise ecosystem.

Yet, a deeper systemic frustration has surfaced. Despite successfully pushing thousands of POCs into production, businesses are discovering that they are still not achieving their expected business metrics and return on investment. Compounding this challenge, the underlying foundational models and development frameworks are shifting on a daily basis, leaving enterprise leaders feeling as though "they don't have real breathing time to sit back and think how they can get the business metrics around what they have invested." Organizations are now recognizing that siloed, standalone point applications cannot deliver structural impact. True value only emerges when generative and agentic AI are deeply embedded across the entire enterprise value chain.

```mermaid flowchart LR accTitle: Evolution of Enterprise GenAI Adoption subgraph Wave1["Phase 1: Experimentation"] A["Standalone Prototypes & POCs"] --> B["Departmental Soft Launches"] end subgraph Wave2["Phase 2: Production Stagnation"] B --> C["3,000+ POCs in Production"] C --> D["Limited Business Outcome & Metric Deficits"] end subgraph Wave3["Phase 3: Value Chain Transformation"] D --> E["End-to-End Enterprise Value Chain Integration"] end ```

---

[04:20](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=260s) The Three Strategic Pillars: Reimagine, Autonomous, and Human-AI Synergy

To steer enterprises away from superficial automation and toward structural transformation, TCS organizes its methodology around three strategic pillars: **Reimagining Business Value Chains**, building **Autonomous Systems**, and fostering **Human + AI Collaboration**. The objective is not merely automating mechanical, routine tasks, but fundamentally re-architecting how value flows across an organization through three distinct lenses: *Value*, *Path*, and *Realize*.

The *Value* lens concentrates on elevating customer and employee experience, unlocking net-new revenue growth through usage-based models and digital twins, and optimizing structural operational costs. The *Path* lens focuses on knowledge synthesis—extracting structured, unstructured, and deeply embedded tacit human knowledge across the enterprise—alongside design intelligence and end-to-end business assurance to establish model and data trust. Finally, the *Realize* lens drives the structural convergence of previously segregated units; AI dissolves traditional departmental silos between data teams, operations teams, and business units, uniting them into integrated, autonomous execution loops.

Autonomous systems are built upon three core operational components—data (structured, unstructured, and institutional tacit knowledge), models (both Large Language Models and Small Language Models), and autonomous agents (leveraging Amazon Bedrock agents, CrewAI, or Strands depending on enterprise context). This architecture powers what Kathirvelan terms the "New IT"—not just writing new software code, but ensuring that disparate systems, models, and workflows are completely orchestrated, chained, and assured to operate reliably. Within this ecosystem, human capabilities and agentic abilities complement one another: humans provide contextual judgment, empathy, and high-stakes decision-making under uncertainty, while agents handle complex orchestration under rigorous compliance guardrails.

- **Value Lens:** Customer and employee experience, growth via new business models, and lean cost efficiency. - **Path Lens:** Synthesis of tacit and explicit institutional knowledge, design intelligence, and business assurance. - **Realize Lens:** Breaking down organizational walls between IT, operations, and business functions through continuous convergence. - **Autonomous Core Stack:** Enterprise data assets, LLMs/SLMs, and multi-agent frameworks (e.g., AWS Bedrock, CrewAI).

---

[09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) Why Business Design Must Precede Technical Design

Kathirvelan emphasizes that before writing a single line of technical specification, teams must establish a robust business design. To illustrate why, he shares an instructive case study of an enterprise client that sought to revolutionize its entire business value chain. Eager to see results, the client bypassed business design, performed only a high-level review of their workflow stages, jumped straight into drafting a Technical Design Document (TDD), wrote hasty user stories, and developed individual agents. While every single agent performed well in isolation, the entire initiative collapsed the moment they were integrated: the agents failed to orchestrate, communication broke down, and the project failed to yield its promised business outcomes.

To prevent such failures, TCS enforces a disciplined business design phase that aligns agents along four critical parameters: clarity of purpose (defining the precise goal and task composition of each agent), enablement tools (selecting appropriate prompt strategies and foundation models), trust and accountability (embedding strict safety guardrails and regulatory compliance), and scalability with feedback (integrating human-in-the-loop checkpoints, confidence scoring, and continuous evaluation). Rather than isolated development, business leaders, operators, and AI architects co-create the target state using purpose-built accelerators, including the Business Value Canvas, Agent Persona Canvas, workflow maps, use-case prioritization matrices, and guardrail frameworks.

Reflecting on modern software methodologies, Kathirvelan warns against misapplying startup mantras to enterprise-scale initiatives: "People used to say, 'fail fast and move to the next one,' but that is not applicable here when we come to the business value chain transformation." Rushing into complex end-to-end integrations without architectural clarity risks expensive failure. Instead, organizations should adopt the philosophy to "think big, start small, keep progressing, do course correction, and keep progressing." Only once this business blueprint is crystallized does the technical team construct the Technical Design Document, Agent Instruction Documents, Agent Manifest JSON schemas, human-in-the-loop (HITL) specifications, and test suites.

```mermaid flowchart TD accTitle: Business Design to Technical Architecture Transition subgraph BusinessDesign["1. Upstream Business Design (Foundational)"] A["Business Value Canvas & Workflow Mapping"] --> B["Agent Persona Canvas & Purpose Clarity"] B --> C["Guardrails, Compliance & HITL Checkpoints"] end subgraph TechnicalDesign["2. Downstream Technical Architecture"] C --> D["Technical Design Document (TDD)"] D --> E["Agent Instruction Documents & Agent Manifest JSON"] E --> F["HITL Specs, Unit Testing & Orchestration"] end ```

---

[13:32](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=812s) Real-World Implementation: Transforming the Contact Center Value Chain

To ground these architectural principles in practice, Kathirvelan highlights TCS's work in revamping end-to-end contact center value chains. Initially, the client had deployed isolated GenAI point solutions—such as standalone self-service customer chatbots to enhance customer experience, alongside smart assist tools designed to help human agents retrieve knowledge during calls. While helpful, these disconnected tools left critical workflow gaps unaddressed. The client soon realized that real business transformation required interconnecting every stage of the customer interaction lifecycle.

In response, TCS deployed its proprietary "Contact Synergy" platform to orchestrate the entire contact center journey from intake to post-call resolution. By mapping the entire value stream, the team pinpointed exactly where autonomous agents could operate independently and where mandatory human-in-the-loop interventions were required—such as quality engineering reviews to validate critical failures, compliance checks, and sensitive data masking audits.

Kathirvelan points out that enterprise architectures cannot rely on a "one-size-fits-all" model. System designers must carefully match the exact agent architecture to each functional task across the value chain, categorizing them appropriately based on operational complexity:

- **Simple Reflex Agents:** For direct, deterministic rule-based triggers and condition-action execution. - **Model-Based Reflex Agents:** For tracking state and context across multi-turn interactions. - **Goal-Based Agents:** For executing multi-step user requests oriented toward explicit operational targets. - **Utility-Based Agents:** For balancing trade-offs, preferences, and optimal paths under competing constraints.

---

[16:26](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=986s) Transforming Healthcare: The Patient Care Value Chain

Kathirvelan next details TCS's ongoing work modernizing the healthcare continuum via its specialized Patient Care Experience platform. Much like in customer support, early healthcare initiatives began with narrow, standalone GenAI tools for answering patient inquiries. However, meaningful impact demanded unifying the entire patient care value chain—from initial patient intake and triage through clinical documentation, treatment coordination, and follow-up care.

Because healthcare carries immense regulatory and patient safety obligations, human-in-the-loop integration is paramount. The platform systematically establishes auditable checkpoints to ensure strict adherence to HIPAA and GDPR requirements, guaranteeing that autonomous agents never execute sensitive medical determinations without physician or clinician oversight. The underlying architecture applies varied orchestration patterns—such as sequential chaining and parallel agent execution—depending on clinical requirements, allowing hospitals to achieve incremental, risk-managed transformation across their existing care workflows.

---

[17:50](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1070s) TCS and AWS Strategic Partnership and Ecosystem Advantages

Turning to execution, Kathirvelan outlines how TCS pairs its proprietary delivery frameworks with cloud hyperscalers, placing particular emphasis on its deep strategic alliance with Amazon Web Services (AWS). Supporting enterprises from initial experimental sandboxes to enterprise-grade AI production platforms, TCS blends AWS-native capabilities like Amazon Bedrock with leading open-source model frameworks, applying TCS-engineered governance wrappers to enforce security, compliance, and responsible AI guardrails.

Kathirvelan outlines several key figures and exclusive collaborative programs that underpin this alliance, designed to help clients de-risk and fund enterprise AI adoption:

- **Alliance Scale:** More than 500 joint AWS enterprise clients worldwide and over 25,000 AWS-certified TCS associates. - **GenAI Competency:** Recognized as an accredited AWS Generative AI Competency Partner. - **GenAI Rapid Program:** A structured AWS assessment framework utilized to rapidly evaluate enterprise readiness, technical architecture, and value potential for GenAI initiatives. - **Dedicated POC Co-Funding:** Mutual investment programs where TCS and AWS provide co-funding for enterprise clients to build and validate novel, production-oriented proofs of concept. - **Process to Agent Initiative:** A newly launched collaborative framework focused on systematically decomposing complex enterprise business processes and mapping them into robust, autonomous agentic workflows.

[19:53](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1193s) Funding P2A Initiatives and Session Wrap-Up

Addressing opportunities for enterprise transformation, Kathirvelan Ganesan touches on how organizations can advance their existing workflows toward greater automation. For clients operating diverse legacy or established processes, teams looking to transition those operations through a P2A initiative can apply for and secure dedicated funding to support the journey toward autonomous systems.

Before the topic can be expanded further, the session moderator steps in to signal that the allotted presentation time has officially expired, politely cutting in with, "that's all that's all the time we have. I'm sorry." Kathirvelan smoothly wraps up his presentation, thanking the audience and inviting attendees to continue the conversation in person or digitally. He directs participants to visit the Tata Consultancy Services exhibit on the conference floor or reach out via email to explore enterprise AI implementations further.

- **Exhibition Booth:** Attendees are invited to connect directly with the TCS team at TCS booth 546. - **Follow-Up Inquiries:** Open outreach through Kathirvelan Ganesan's email address for deeper discussions on AI-driven process transformation.

## Notable quotes > Human Insight Meets AI Brilliance: Transforming the Future. — Kathirvelan Ganesan

> Discover how the collaboration between human creativity and machine precision is fundamentally evolving how organizations operate. — Kathirvelan Ganesan

> Don’t miss this chance to find out how to lead with intelligence! — Kathirvelan Ganesan

> [00:04](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=4s) GK or Kathir, whatever you want. — Kathirvelan Ganesan

> [01:59](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=119s) they don't have real breathing time to sit back and think how they can get the business metrics around what they have invested. — Kathirvelan Ganesan

> [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) People used to say, 'fail fast and move to the next one,' but that is not applicable here when we come to the business value chain transformation. — Kathirvelan Ganesan

> [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) think big, start small, keep progressing, do course correction, and keep progressing. — Kathirvelan Ganesan

> [19:53](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1193s) that's all that's all the time we have. I'm sorry. — Session Moderator

## Mentioned Kathirvelan Ganesan, Tata Consultancy Services (TCS), Tata Group, Ai4, Amazon Web Services (AWS), Amazon Bedrock, CrewAI, Strands, Contact Synergy, Patient Care Experience platform, HIPAA, GDPR, Business Value Canvas, Agent Persona Canvas, GenAI Rapid Program, Process to Agent (P2A)

## Video description Kathirvelan Ganesan – Solutions Head, AI Practice, Tata Consultancy Services

Abstract: Human Insight Meets AI Brilliance: Transforming the Future. Join us for an engaging discussion where we explore the powerful fusion of human insight and AI-driven intelligence. Discover how the collaboration between human creativity and machine precision is fundamentally evolving how organizations operate. Learn from real-world examples of transformation enabled by redesigned business models, autonomous systems, empowered workforces, and trusted partnerships. Experts will share actionable strategies to harness the combined power of human and AI to unlock new levels of productivity and innovation. Don’t miss this chance to find out how to lead with intelligence!

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

中文简报

3000个企业POC未达ROI目标，TCS摒弃“快速失败”AI模式 Tata Consultancy Services 认为，企业 AI 投资之所以未能产生回报，是因为各机构在未预先重新设计端到端业务价值链的情况下孤立地部署了技术智能体。 Kathirvelan Ganesan 探讨了为何数千个企业生成式 AI 概念验证未能带来可衡量的业务 ROI，并呼吁机构从单点解决方案转向全价值链转型。他强调在拟定技术规范前必须完成前置业务设计与画像映射，并警告称初创公司奉行的“快速失败”信条对企业级重构而言极具风险。结合联络中心和医疗领域的实际案例，他展示了多样化智能体原型以及人机协同护栏在实践中的运作方式。最后，他概述了 TCS 与 AWS 为支持企业“流程到智能体”（P2A）计划而开展的合作与联合资助项目。 尽管已有超过 3,000 个生成式 AI POC 通过软启动进入生产环境，但由于孤立应用及基础模型的频繁更迭，企业仍难以获得预期的业务指标与投资回报率。 (1:59) TCS 围绕重塑业务价值链、自主系统和人机协作三大支柱推进企业级转型，通过价值、路径和实现三层视角消除 IT、运营与业务之间的部门孤岛。 (4:20) 业务设计必须先于技术架构设计，初创公司“快速失败”的模式并不适用于复杂的企业价值链改造，企业应遵循“放眼大局、小步起步、持续修正”的原则。 (9:10) TCS 与 AWS 达成深度战略合作，拥有超过 500 家联合企业客户和逾 25,000 名认证员工，并通过联合出资 POC 及“Process to Agent”计划推动业务流程向自主智能体转型。 (17:50)

Reading summary

Kathirvelan Ganesan, Solutions Head for the AI Practice at Tata Consultancy Services (TCS), argues that enterprises fail to capture expected returns from generative AI when they rely on isolated point solutions rather than embedding orchestrated, agentic systems across their entire business value chains. The Production Paradox of Isolated AI (1:59) Drawing on TCS's perspective across more than 55 countries and approximately 600,000 associates, Ganesan observes that enterprise generative AI has moved past initial experimentation. By using departmental soft launches to bridge the gap between prototyping and deployment, organizations have successfully transitioned more than 3,000 proofs of concept (POCs) into production. Despite this milestone, Ganesan notes that enterprises are still not achieving their expected business metrics or return on investment. This stagnation is compounded by daily shifts in foundational models and development frameworks, which leave leaders feeling they lack "real breathing time" to evaluate how their investments translate into business outcomes. According to Ganesan, standalone departmental applications cannot deliver structural impact; achieving meaningful value requires embedding generative and agentic AI across end-to-end enterprise workflows. Architecting Autonomous Systems and Human-AI Synergy (4:20) To move beyond basic task automation, TCS structures enterprise transformation around reimagining business value chains, building autonomous systems, and designing human-AI collaboration. Ganesan analyzes this shift through three lenses: generating *Value* via improved experiences, cost optimization, and usage-based or digital-twin revenue models; establishing a *Path* by synthesizing structured, unstructured, and tacit human knowledge alongside business assurance; and *Realizing* transformation by dissolving traditional silos between data, operations, and business units. These autonomous systems combine enterprise data assets, large and small language models, and multi-agent frameworks such as Amazon Bedrock agents, CrewAI, or Strands. Ganesan defines this layer as the "New IT," where the primary challenge is not merely writing code, but orchestrating, chaining, and assuring disparate models and workflows. Within this architecture, autonomous agents manage complex execution under strict compliance guardrails, while humans contribute empathy, contextual judgment, and high-stakes decision-making under uncertainty. Why Business Design Must Precede Technical Specifications (9:10) Ganesan warns that building multi-agent architectures without an upstream business blueprint leads to integration failure. He cites an enterprise client that bypassed business design, briefly reviewed its workflows, and immediately drafted a Technical Design Document (TDD) and user stories to build individual agents. Although each agent functioned properly in isolation, the initiative collapsed upon integration because the agents could not communicate or orchestrate across the value chain. To prevent such breakdowns, TCS mandates a co-created business design phase before any technical specifications—such as TDDs, Agent Instruction Documents, or Agent Manifest JSON schemas—are written. Using tools like the Business Value Canvas and Agent Persona Canvas, teams align agents across four parameters: clarity of purpose, model and prompt enablement, regulatory trust and guardrails, and scalability through human-in-the-loop (HITL) feedback and confidence scoring. In Ganesan's view, the startup mantra to "fail fast" does not apply to enterprise value chain transformation; instead, organizations must "think big, start small, keep progressing, do course correction." End-to-End Orchestration in Customer Care and Healthcare (13:32) Ganesan illustrates these principles in customer service and healthcare, where clients initially deployed disconnected chatbots and agent-assist tools that left major workflow gaps. In contact centers, TCS replaced isolated tools with its "Contact Synergy" platform to orchestrate the full interaction lifecycle. Rather than relying on a uniform agent design, the architecture matches specific operational tasks to simple reflex, model-based reflex, goal-based, or utility-based agents, while inserting mandatory human reviews for quality engineering, compliance, and sensitive data masking. Similarly, in healthcare, TCS's Patient Care Experience platform connects patient intake, triage, clinical documentation, treatment coordination, and follow-up care. Because healthcare requires strict adherence to HIPAA, GDPR, and patient safety standards, the system uses sequential chaining and parallel execution alongside auditable human-in-the-loop checkpoints, ensuring clinicians oversee all sensitive medical determinations. To execute these transformations at scale, TCS pairs its governance wrappers and open-source frameworks with Amazon Web Services (AWS), leveraging an alliance that encompasses over 500 joint enterprise clients and 25,000 AWS-certified TCS associates. Through joint offerings—including the GenAI Rapid readiness assessment, co-funded POCs, and the newly launched Process to Agent (P2A) initiative—the partners help enterprises decompose legacy processes into autonomous workflows. Ganesan noted that clients can apply for dedicated funding to transition established operations through the P2A program, though the session's allotted time expired before he could elaborate on how those funding mechanisms operate.

精炼阅读摘要

塔塔咨询服务公司（TCS）AI实践解决方案主管Kathirvelan Ganesan指出，企业生成式AI已跨越初步实验阶段，但大量孤立的概念验证并未转化为预期的商业回报；企业唯有坚持业务设计先于技术实现，通过多智能体系统与人机协同重塑端到端业务价值链，才能实现从数据洞察迈向系统自主运营。 概念验证困局与端到端价值链重构 (1:59) 企业在早期生成式AI采纳浪潮中多采用部门内软启动策略，目前已有超过3000个概念验证（POC）推进至生产环境。然而，Ganesan指出，企业管理者普遍遭遇系统性挫折：这些投入并未带来预期的商业指标改善与投资回报。与此同进，底层基础大模型和开发框架每天都在更新迭代，使管理层缺乏充分的时间进行战略反思与回报评估。孤立的单点应用无法对企业产生结构性影响，生成式AI与智能体系统必须深度嵌入至完整的企业业务价值链中，才能释放实际价值。 自主系统架构：三大支柱与三重视角 (4:20) 为推动企业超越常规任务自动化，TCS提出了以重构业务价值链、构建自主系统以及促进人机协同为核心的三大战略支柱，并通过三大视角进行落地。其中，“价值视角”聚焦于提升客户与员工体验、利用基于使用量模式和数字孪生开拓新增长点，并优化结构性成本；“路径视角”着眼于知识综合，提炼企业内部的结构化、非结构化及隐性知识，配合设计智能与业务保障以建立模型信任；“实现视角”则致力于打破数据、运营与业务部门间的传统壁垒，促进多部门融合为闭环执行流。 企业自主系统的核心技术栈建立在企业数据资产、大语言模型（LLM）与小型语言模型（SLM），以及多智能体框架（如Amazon Bedrock Agents、CrewAI或Strands）之上。Ganesan将这一架构定义为“新IT”，其核心不再是单纯编写代码，而是对系统、模型与工作流进行编排、链接与可靠性保障。在此体系中，人类负责在高度不确定性场景下提供同理心、经验判断与高风险决策，智能体则在严格合规护栏内执行复杂流程调度，形成互补协同。 业务设计前置：多智能体编排的防错机制 (9:10) 在系统架构设计中，Ganesan强调业务设计必须先于技术规范。他引用了一起企业失败案例：某客户试图重构其业务价值链，因急于求成而跳过业务设计，仅粗略梳理流程便直接编写技术设计文档与单体智能体。尽管每个智能体在孤立测试中表现优异，但在系统集成时却因缺乏编排机制和通信断裂导致整体崩溃。Ganesan据此提出警告，初创企业常用的“快速失败（fail fast）”口号不适用于复杂企业价值链重构，大型企业应践行“宏观思考、微观起步、持续推进并纠偏”的渐进路径。 为防范集成风险，TCS制定了严格的前置业务设计规范，要求在编码前依据四大维度界定智能体：明确目标定位、配备赋能工具、确立信任与问责合规机制，以及构建包含人类在环（HITL）检查点与置信度评分的可扩展反馈环。业务团队与AI架构师需借助业务价值画布、智能体角色画布与用例优先级矩阵共同绘制目标蓝图，随后技术团队方可展开技术设计文档（TDD）、智能体指令文档、Manifest JSON架构及单元测试集的编写。 垂直场景工程化：智能体分级与人机协同 (13:32) 在客户联络中心场景中，TCS通过其Contact Synergy平台重构了从用户接入到通话后处理的全流程。早期的单点客服机器人与坐席辅助工具因缺乏协同而存在流程断点，而全链路重塑不仅实现了流程串联，还在质量工程审计、合规核查及敏感数据脱敏等关键节点设立了强制性的人类在环审核。Ganesan强调企业不可依赖一刀切模型，而应根据操作复杂度匹配不同类型的智能体，包括处理确定性规则的简单反射智能体、追踪多轮交互上下文的模型反射智能体、执行多步目标的基于目标智能体，以及在多重约束下进行权衡决策的基于效用智能体。 在医疗健康领域，TCS的患者关怀体验平台覆盖了从初诊分诊、临床记录到治疗协调和随访的全流程。由于医疗行业受到HIPAA及GDPR等严格法律与患者安全法规约束，系统采用串行或并行链式编排，并在关键医疗决策环节设置具备可追溯性的强制人类在环节点，严禁智能体在未经医生或临床专业人员审核的情况下独立执行敏感医疗判定，从而在风险可控的前提下实现增量式转型。 云服务生态赋能与P2A项目迁移机制 (17:50) 依托分布在全球55个以上国家、拥有约60万名员工及200多个交付中心的业务网络，TCS将其工程框架与主流云厂商结合。在与Amazon Web Services（AWS）的战略合作中，双方拥有超过500家联合企业客户，TCS配备了逾2.5万名获得AWS认证的技术人员，并获得了AWS生成式AI能力认证。双方将Amazon Bedrock原生功能及开源模型与TCS的企业治理封装相结合，以确保安全、合规与负责任AI护栏的落地。 为了降低企业落地的技术与资金门槛，TCS与AWS推出了生成式AI快速评估计划（GenAI Rapid Program）以及针对新型生产级POC的联合注资机制。此外，双方最新推出了“流程至智能体”（Process to Agent, P2A）专项倡议，该计划旨在协助企业系统化拆解复杂的遗留业务流程，并将其重构为稳健的自主智能体工作流，企业可通过申请专属资助来支持这一迁移过程。 Ganesan总结认为，企业向自主系统转型的核心在于摆脱技术驱动的盲目尝试，依托严密的业务设计和生态支持实现端到端闭环；但由于会议现场时间限制，针对传统业务向P2A迁移的具体资助申请门槛及更深层的执行阻力未能详尽展开，企业在实际推进流程解构时仍需面对复杂的异构系统兼容性考验。

Detailed notes — zh

## Background 本视频由行业频道 Ai4 于 2025 年 8 月 29 日发布在 YouTube 平台。内容为塔塔咨询服务（TCS）人工智能业务部解决方案负责人 Kathirvelan Ganesan 的主题演讲，主要探讨如何将人类洞察与 AI 智能深度结合，通过重构业务模式、自主系统及人机协同推动企业实现从洞察到自主运营的全面转型。 ## TL;DR 针对企业在生成式 AI 探索中陷入单点 PoC 泛滥且缺乏实质业务回报的困境，讲者指出必须将单点应用跃升为端到端业务价值链的系统性重塑。TCS 提出了涵盖价值链重构、自主系统与人机协同的三大战略支柱，并强调“业务设计必须优先于技术设计”，避免仓促编码带来的系统集成风险。结合联络中心（Contact Synergy）与医疗患者关怀两大标杆案例，演讲展示了如何深度结合人在回路（HITL）机制与智能体网络实现合规高效的自主流转。最后，讲者介绍了 TCS 与 AWS 联合推进的快速评估计划与“流程转智能体”创新行动，为企业落地提供全链路支持。 ## Atmosphere & style 演讲风格务实干练且架构清晰，讲者以亲和幽默的自我介绍开场，随后迅速切入大型企业在 AI 转型中的深层阵痛。整场叙事兼备战略咨询的全局透视与工程交付的实战沉淀，讲者不避讳早期试错教训，诚恳分享业务逻辑重于技术堆砌的深刻洞见。随着联络中心与医疗案例的层层展开，内容节奏紧凑严密，传递出一种从技术炒作回归商业本质的理性与笃定。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=4s) 塔塔咨询服务（TCS）拥有约 60 万名员工，业务遍及全球 55 个国家，具备深厚的企业级 IT 服务与生成式 AI 全球落地能力。 - [01:29](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=89s) 生成式 AI 已步入产业现实阶段，但大量企业在原型投产后遭遇“价值困局”，孤立的单点 PoC 应用难以带来预期的量化业务回报。 - [01:29](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=89s) 面对日新月异的技术迭代，释放 GenAI 真实成效的突破口在于将其深度嵌入到端到端的整条企业价值链中，而非继续开发分散的原型。 - [04:20](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=260s) TCS 提出重塑企业价值链的三大战略支柱：价值链重构（Reimagine）、自主系统（Autonomous）以及人机协同（Human + AI Collaboration）。 - [04:20](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=260s) 从价值、路径与实现三大透视维度切入，AI 正在打破数据、运营与业务之间的壁垒，推动技术与运营在统一智能底座上深度融合。 - [07:11](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=431s) 自主系统技术内核依托于数据、模型（LLM/SLM）与智能体三大组件，“新 IT”扮演中枢神经角色，负责无缝串联并提供确定性保障。 - [07:11](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=431s) 人类与 AI 是共生互补关系：人类提供复杂环境下的判断力与同理心，智能体提供高速规模化执行能力，需在明确安全护栏下协同作业。 - [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) 智能化变革必须坚持“业务设计优先于技术设计”，传统单点敏捷的快速试错在端到端系统集成中容易引发数据断层与业务逻辑冲突。 - [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) 价值链重构需遵循“大处着眼、小处着手、持续纠偏、稳步推进”原则，在编码前明确智能体目标、工具赋能、合规护栏与人在回路（HITL）机制。 - [13:32](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=812s) 联络中心实践（TCS Contact Synergy）表明，必须将全链路串联并针对具体业务场景匹配智能体架构，在质检与合规节点严格嵌入 HITL。 - [16:26](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=986s) 医疗患者关怀重塑需将 HIPAA/GDPR 合规与 HITL 置于最高优先级，通过智能体串行链式或并行协同拓扑兼顾临床严谨性与服务效率。 - [17:50](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1070s) TCS 联合 AWS 提供 GenAI 快速评估计划、PoC 专项资金及“流程转智能体”（Process to Agent）行动，全面降低企业智能化转型门槛。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=4s) 讲者与塔塔咨询服务（TCS）简介 演讲者 Kathirvelan Ganesan 走上讲台，以轻松幽默的方式开场。他笑着向听众介绍，自己的全名可能有点拗口难读，大家叫他“GK”或者“Kadir”都可以。作为塔塔咨询服务（Tata Consultancy Services，简称 TCS）人工智能业务部（AI Practice）的解决方案负责人（Solutions Head），他首先向在场观众提问，有多少人听说过 TCS 这家企业。 看到台下积极的响应，GK 简要介绍了 TCS 的全球规模与业务版图。TCS 隶属于印度最大的跨国集团之一——塔塔集团（Tata Group）。目前，TCS 拥有约 60 万名员工（associates），为全球约 55 个国家和地区的客户提供涵盖战略咨询、IT 服务以及端到端业务解决方案的支持。其业务覆盖了从银行与金融服务、生命科学、医疗保健、制造业到通信传媒（CMI）等几乎所有的垂直行业领域，并在全球各地建立了超过 200 处交付中心。 - **组织规模**：隶属于塔塔集团，拥有约 60 万名专业员工。 - **全球布局**：业务遍及全球 55 个国家，运营超过 200 处全球交付中心。 - **行业覆盖**：涵盖金融银行、医疗与生命科学、制造业、电信媒体等几乎全部关键领域。 ### [01:29](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=89s) 从 PoC 狂热到业务现实：生成式 AI 的价值困局 GK 迅速切入主题，指出本次分享的核心在于探讨 TCS 如何赋能全球大型企业，借助生成式 AI（GenAI）与智能体（Agentic）技术重塑其业务价值链，并分享 TCS 团队在实战中面对的真实挑战与客户经验。回顾 AI 的发展轨迹，生成式 AI 已经彻底走出了早期的炒作周期（hype cycle），成为切实改变世界的产业现实，这一点在业界已达成广泛共识。 然而，企业在拥抱 GenAI 的过程中普遍经历了一段曲折的探索期。初期，客户热衷于构建各类概念验证（PoC）与独立原型，却在尝试将这些成果推向生产环境时频频受阻。为了打破僵局，很多企业采取了部门内部“软发布”（soft launch）的策略，先在一个业务部门小范围试点、验证产出，再逐步向其他部门横向推广。虽然业界至今已有超过 3000 个 PoC 成功迈入生产部署阶段，但新的严峻挑战随之浮现：即便系统已经成功上线，企业管理层却发现它们未能带来预期的核心业务回报与量化价值。 更为棘手的是，底层的模型架构与技术框架几乎每天都在迭代演化。客户在目不暇接的技术更新中疲于奔命，甚至“连坐下来喘口气、静心评估既有投资之业务指标的时间都没有”。面对这种困境，领先的企业开始幡然醒悟：与其继续打造零散、孤立的单点 GenAI 原型应用，不如探索如何将生成式 AI 深度嵌入到端到端的整条企业价值链中，而这正是释放实质业务成效的真正突破口。 ### [04:20](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=260s) 驱动价值链重塑的三大战略支柱与透视维度 为了指引企业系统性重塑业务价值链，TCS 提炼出三大战略支柱：价值链重构（Reimagine）、自主系统（Autonomous）以及人机协同（Human + AI Collaboration）。GK 强调，AI 绝不仅仅是完成某些简单的自动化任务，它的根本使命在于全面重新设计业务流程。TCS 从“价值”（Value）、“路径”（Path）和“实现”（Realize）这三个透视维度切入，对价值链展开全局重构。 ```mermaid flowchart TD accTitle: 三大战略支柱与价值链重构架构 subgraph S1["战略支柱 Strategic Pillars"] P1["重构价值链 (Reimagine)"] P2["自主系统 (Autonomous)"] P3["人机协同 (Human + AI)"] end subgraph S2["三大透视透镜 Three Lenses"] L1["价值维度 (Value Lens)<br/>客户/员工体验 · 业务增长 · 精益成本"] L2["路径维度 (Path Lens)<br/>知识综合 · 设计智能 · 业务保障"] L3["实现维度 (Realize Lens)<br/>技术与运营收敛 · 自主智能体 · 人类微调"] end P1 --> S2 P2 --> L3 P3 --> L3 ``` 在“价值”维度上，企业关注的是体验提升（涵盖客户体验与员工体验）、业务增长（构建基于用量计费的新商业模式、数字孪生与新兴服务）以及成本精益化。“路径”维度则包含三大抓手：首先是知识综合（Knowledge Synthesis），即如何从组织内部提取并融合结构化、非结构化数据以及深层次的隐性知识（tacit knowledge）；其次是设计智能（Design Intelligence），利用 AI 实现流程设计的自动化与智能化；最后是业务保障（Business Assurance），确保对模型输出与业务资产建立充分的信赖与合规机制。而在“实现”维度上，最深刻的变革莫过于技术与运营的融合（convergence），AI 正在彻底打破过去数据团队、运营团队与业务部门之间的深层壁垒，推动各方在统一的智能底座上协同作业。 ### [07:11](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=431s) 自主系统构成要素与“新 IT”人机协作范式 深入到自主系统的技术内核，GK 指出该体系依托于三大核心组件的有机结合：数据、模型与智能体。数据层面涵盖了企业环境内部庞杂的结构化数据、非结构化内容以及专家头脑中的隐性知识；模型层面不仅包括前沿的大语言模型（LLM），也兼顾灵活高效的小语言模型（SLM）；而在智能体（Agents）层面，TCS 除了依托 AWS Bedrock Agents 等云厂商原生方案构建系统外，还会根据客户的具体业务场景扩展接入 CrewAI、Strands 等先进智能体框架。 这些要素共同喂养并驱动着自主系统与重构后的价值链。GK 明确界定，此处所说的“新 IT”（New IT）并非指传统的软件编码流程，而是指 IT 系统如何扮演中枢神经的角色，将所有模型、数据与智能体无缝集成串联，并为全流程的稳定运行提供确定性保障。 与此同时，全新的工作方式亟需在人类与 AI 之间建立深层次的协同效应。人类的独特优势在于判断力、同理心决策以及在高度不确定与模糊环境下的决断能力；而智能体则长于高速处理、规模化执行与多任务串联。两者不仅不是相互替代的关系，反而需要在明确的安全护栏（guardrails）与合规策略约束下，实现真正的“人机共生”。 ### [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) 业务设计先于技术设计：为何单点敏捷在价值链重塑中失灵 结合一线的实战教训，GK 详细剖析了为什么企业在推进智能化变革时必须坚持“业务设计优先于技术设计”（Business Design over Technical Design）。他分享了一个真实案例：某大型客户希望全面重构整条业务价值链，但技术团队一开始就直接跳进了详细的技术设计文档（TDD）编写中。他们粗略地对整条价值链的各个阶段做了高阶梳理，草草拆解出若干用户故事（user stories），便迅速着手编码并部署了对应的自主智能体。 起初，这些智能体在各自孤立的环节中表现十分优异，然而灾难在系统集成阶段爆发了。当所有智能体被串接在一起时，由于前端缺乏对业务逻辑与交互上下游的全局考量，各种冲突与数据断层频频出现，客户根本无法获得预期的端到端业务成果。这一惨痛教训表明，必须在编码前投入足够的时间做好业务设计，厘清各个智能体的目标构成、工具赋能策略、可信合规护栏以及人在回路（HITL）的标定与反馈机制。 对于端到端价值链重构而言，传统的互联网敏捷教条“快速失败并转向下一个”（fail fast and move to the next one）完全不再适用。企业绝不能在底座不稳的情况下仓促试错，而是必须在前期沉下心来，全面剖析完整价值链，拉通各方利益相关者进行共创。针对实施策略，GK 提炼了一句核心原则：“大处着眼，小处着手，持续纠偏，稳步推进”（Think big, start small, keep progressing, do the course correction and keep progressing）。只有在业务设计完全明晰之后，再推进技术设计文档（TDD）、智能体指令文档（Agent Instruction Document）、智能体配置清单（Agent Manifest JSON）、人在回路规范（HITL Specification）以及对应的单元测试方案编写，整个系统工程方能行稳致远。 - **业务设计核心要素**： - **明确目标（Clarity of Purpose）**：定义每个智能体在价值链中的确切职责与任务边界。 - **工具赋能（Enablement Tools）**：制定清晰的 LLM 选型与提示词工程（Prompt Strategy）路线。 - **可信与问责（Trust & Accountability）**：嵌入企业级安全护栏与合规审计机制。 - **扩展与反馈（Scalability & Feedback）**：建立人在回路（HITL）机制与置信度评估标准。 - **加速器与模板工具**：运用业务价值画布（Business Value Canvas）、智能体角色画布（Agent Persona Canvas）、工作流映射图与护栏矩阵等标准化资产。 ### [13:32](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=812s) 实战案例：重塑联络中心全价值链（TCS Contact Synergy） 为了具象化展示端到端重塑的威力，GK 分享了 TCS 为客户重塑联络中心（Contact Center）价值链的标杆案例。起初，团队也是在局部场景摸索 GenAI 应用——例如仅针对改善客户体验开发了自助问答聊天机器人，或者针对客服坐席开发了辅助智能助手（Smart Assist）。然而，当把这些点状方案推向客户现场时，大家意识到孤立的功能无法解决系统性问题，必须将联络中心从呼入、交互到后续处理的全链条串联起来。 为此，TCS 进行了全链路梳理，并落地了自主研发的“TCS Contact Synergy”解决方案。该方案清晰划分出哪些环节完全由智能体自主驱动，哪些环节必须引入人机协同。例如，在质检工程（QE）环节验证关键操作失误，或是在合规审计中对敏感数据进行脱敏核查时，系统在关键节点都严格嵌入了人在回路（HITL）机制。 GK 提醒道，价值链智能化转型绝无“放之四海皆准”（one size fits all）的现成模板，必须依据企业的特有流程量身定制。在设计智能体时，必须根据业务需求精准匹配其行为架构——无论是简单的反射型智能体（simple reflex agent）、基于模型的智能体（model-based agent），还是基于目标的智能体（goal-based agent）乃至基于效用的智能体（utility-based agent），唯有架构与场景严丝合缝，才能将价值链的潜能彻底释放。 ### [16:26](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=986s) 实战案例：重塑端到端医疗患者关怀价值链 紧接着，GK 介绍了 TCS 正在推进的另一个重量级案例——重构患者关怀（Patient Care）价值链，并推出了专属的“TCS 患者关怀体验”（TCS Patient Care Experience）解决方案。与联络中心类似，团队最初同样是从简单的点状 GenAI 问答切入，但很快发现医疗关怀是一项高度连续且敏感的严肃服务，碎片化的功能无法满足临床与患者的真实诉求。 在医疗健康领域，人在回路（HITL）的重要性被提升到了最高优先级。系统必须严密遵循包括 HIPAA 以及 GDPR 在内的各项医疗隐私与数据合规法则。因此，TCS 深度审视了医疗机构的全部端到端工作流，识别出每一个高风险检查点并无缝介入 HITL 流程，确保每一份诊断建议或关怀方案都在人类专家的把关下生成。根据临床业务的逻辑拓扑，智能体之间灵活采用串行链式调用（chaining）或并行处理（parallelization）的设计模式，在严守合规底线的同时，大幅提升医疗服务流转的效率与温度。 ### [17:50](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1070s) TCS 与 AWS 深度协同：从加速评估到智能体创新 在演讲尾声，GK 重点阐述了 TCS 与亚马逊云科技（AWS）构建的战略生态优势。作为 AWS 的战略合作伙伴与生成式 AI 能力认证伙伴（GenAI Competency Partner），TCS 仅在 AWS 领域就拥有超过 500 家标杆客户，并培养了超过 2.5 万名获得 AWS 专业认证的技术专家。双方不仅能够为客户从零构建 AI 实验沙箱（Playground），更具备搭建企业级生产化 AI 平台的全栈能力，并深度结合了 TCS 自研的安全负责任 AI 框架与智能体加速器。 GK 透露，依托双方紧密的联合伙伴机制，企业客户可以获得极为稀缺的实战支持与资金扶持。其中重点包括： - **GenAI 快速评估计划（GenAI Rapid Program）**：借助 AWS 专项项目，快速为企业的生成式 AI 构想进行业务成熟度与技术可行性诊断。 - **GenAI PoC 联合专项资金**：在客户探索高价值的新型创新 PoC 时，TCS 可联合 AWS 为其申请专项 PoC 资金支持，降低企业的早期试错成本。 - **“流程转智能体”创新行动（Process to Agent Initiative）**：这是 TCS 近期启动的前沿计划，旨在系统性地帮助企业梳理既有业务流程，将其直接解构并跃升为具备自主执行能力的智能体集群网络。 ### [19:53](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1193s) 推动流程自主化倡议与展会交流收尾 在演讲的最后阶段，Kathirvelan Ganesan 探讨了推动企业落地转型的实际支持方案。他指出，针对客户现有的各类不同业务流程，如果希望将其升级转化为“流程走向自主化”（P2A）倡议，团队可以协助进行申请并争取获取所需的项目资金与资源支持，从而帮助企业加速跨越从业务洞察到自主系统的演进过程。 随后现场主持人提醒演讲时间已到（"That's all the time we have. I'm sorry."），Ganesan 随即向全场致谢并作结语。他热情欢迎有兴趣进一步探讨的与会者随时到展区沟通，大家可以前往塔塔咨询服务公司（TCS）的现场展台，或者直接通过电子邮件与他取得联系，共同交流企业智能化转型与自主系统的实践经验。 - **展台与联系信息**：TCS 在大会现场设有 546 号展位（Booth 546），参会人员可前往展台或通过邮件与讲者团队进一步接洽合作与方案咨询。 ## Notable quotes > [01:29](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=89s) 连坐下来喘口气、静心评估既有投资之业务指标的时间都没有 — Kathirvelan Ganesan > [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) 业务设计优先于技术设计（Business Design over Technical Design） — Kathirvelan Ganesan > [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) 快速失败并转向下一个（fail fast and move to the next one） — Kathirvelan Ganesan > [09:10](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=550s) 大处着眼，小处着手，持续纠偏，稳步推进（Think big, start small, keep progressing, do the course correction and keep progressing） — Kathirvelan Ganesan > [13:32](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=812s) 放之四海皆准（one size fits all） — Kathirvelan Ganesan > [17:50](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1070s) 流程转智能体创新行动（Process to Agent Initiative） — Kathirvelan Ganesan > [19:53](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1193s) 流程走向自主化（P2A） — Kathirvelan Ganesan > [19:53](https://www.youtube.com/watch?v=wSKq-CgYjFw&t=1193s) That's all the time we have. I'm sorry. — 现场主持人 ## Mentioned Kathirvelan Ganesan (GK / Kadir), 塔塔咨询服务 (Tata Consultancy Services / TCS), 塔塔集团 (Tata Group), 亚马逊云科技 (AWS), AWS Bedrock Agents, CrewAI, Strands, TCS Contact Synergy, TCS 患者关怀体验 (TCS Patient Care Experience), HIPAA, GDPR, Ai4 ## Video description Kathirvelan Ganesan – Solutions Head, AI Practice, Tata Consultancy Services Abstract: Human Insight Meets AI Brilliance: Transforming the Future. Join us for an engaging discussion where we explore the powerful fusion of human insight and AI-driven intelligence. Discover how the collaboration between human creativity and machine precision is fundamentally evolving how organizations operate. Learn from real-world examples of transformation enabled by redesigned business models, autonomous systems, empowered workforces, and trusted partnerships. Experts will share actionable strategies to harness the combined power of human and AI to unlock new levels of productivity and innovation. Don’t miss this chance to find out how to lead with intelligence! [Source](https://www.youtube.com/watch?v=wSKq-CgYjFw)

Source quotes

[1:29] Kathirvelan Ganesan: 连坐下来喘口气、静心评估既有投资之业务指标的时间都没有 [9:10] Kathirvelan Ganesan: 业务设计优先于技术设计（Business Design over Technical Design） [9:10] Kathirvelan Ganesan: 快速失败并转向下一个（fail fast and move to the next one） [9:10] Kathirvelan Ganesan: 大处着眼，小处着手，持续纠偏，稳步推进（Think big, start small, keep progressing, do the course correction and keep progressing） [13:32] Kathirvelan Ganesan: 放之四海皆准（one size fits all） [17:50] Kathirvelan Ganesan: 流程转智能体创新行动（Process to Agent Initiative） [19:53] Kathirvelan Ganesan: 流程走向自主化（P2A） [19:53] 现场主持人: That's all the time we have. I'm sorry.

Source quotes

Kathirvelan Ganesan: Human Insight Meets AI Brilliance: Transforming the Future. Kathirvelan Ganesan: Discover how the collaboration between human creativity and machine precision is fundamentally evolving how organizations operate. Kathirvelan Ganesan: Don’t miss this chance to find out how to lead with intelligence! [0:04] Kathirvelan Ganesan: GK or Kathir, whatever you want. [1:59] Kathirvelan Ganesan: they don't have real breathing time to sit back and think how they can get the business metrics around what they have invested. [9:10] Kathirvelan Ganesan: People used to say, 'fail fast and move to the next one,' but that is not applicable here when we come to the business value chain transformation. [9:10] Kathirvelan Ganesan: think big, start small, keep progressing, do course correction, and keep progressing. [19:53] Session Moderator: that's all that's all the time we have. I'm sorry.

Original transcript

[0:04] My name is Kadir Vang since it's very [0:06] tough to pronounce maybe you can call me [0:08] as GK or Kadir whatever you want and so [0:13] I'm I'm from TCS [0:16] solution head from AI practice. So [0:22] how many of you heard about TCS data [0:24] consultancy services? [0:27] >> That's great. So just give you a quick [0:29] intro. CTC is part of the Tata group. [0:32] Okay. So one of the largest [0:33] multinational group in India and the [0:37] thing is like we we have 100k [0:41] around 600k associates and we have been [0:44] doing um consulting IT services and then [0:49] business solutions almost we have been [0:54] um supporting around 55 countries across [0:57] the globe and then we have been involved [1:01] in developing solution solutions, [1:03] services across all business groups [1:07] starting from banking, finance, life [1:10] science, healthcare, manufacturing, CMI [1:13] all okay almost all we have been [1:16] supporting and uh the other important [1:19] thing is we have uh 200 delivery 200 [1:23] plus delivery centers across the uh [1:26] globe. Okay. [1:29] So today I'm going to talk about how TCS [1:33] is helping [1:34] global uh organization [1:38] in transforming the business value chain [1:41] powered by jai and agentic and and also [1:46] I'm going to talk about some of our [1:48] practical challenges and customer [1:51] experiences what we have what we are [1:53] facing in the day-to-day life right so [1:55] that that's a whole agenda [1:59] So we know about uh this thing okay um [2:02] AI where it all started it started from [2:05] the maybe uh everybody knows about that [2:08] that AI is not no more in the hype state [2:12] it has moved to the reality that is AI [2:15] is changing the world that everybody [2:17] have started accepting that okay so when [2:20] when jenna wave started like what what [2:22] happened [2:24] the thing is people the customers have [2:27] started doing the PC's prototypes and [2:31] all but really they struggled to move to [2:34] production. So then what the strategy [2:36] they took is okay so [2:40] after after doing the PC's and [2:42] prototypes they have done the soft [2:44] launch they have um soft launch within [2:47] the departments okay they have launched [2:50] and then made sure that they are getting [2:51] the right outcome and then slowly then [2:55] moved to the different departments okay [2:58] so right now we are in the stage right [3:00] around 3,000 plus PC's have been moved [3:03] to production [3:05] Okay. And then the thing is they have [3:08] moved to the production but now after [3:11] moving to the production see the [3:13] businesses like the customers have [3:14] started realizing they are not getting [3:16] the right business outcome even after [3:19] moving to production successfully the [3:21] PC's and then again at the same time you [3:25] might be knowing that there have been [3:28] the changes in the the frameworks models [3:32] every day it keeps changing and then the [3:34] customers are really feeling uh like uh [3:37] um how they don't have real breathing [3:40] time to sit back and think how they can [3:42] get the business metrics around what [3:45] they have invested. So now the [3:48] organizations have started realizing [3:50] like instead of just move just going for [3:54] small standalone prototypes okay say uh [3:59] gen applications they want to move from [4:01] that to how they can embed J gen jenna [4:05] across the enterprise value chain that's [4:07] what they have started thinking and then [4:09] they have already started working on [4:11] that so that is where the real uh like [4:14] the outcome they are whatever they are [4:17] expecting is going to come. So with this [4:20] set intro like what the thing is um for [4:25] transforming the business value chain. [4:28] Okay. So what from TCS we are [4:30] considering these three strategic [4:32] pillars reimagine autonomous and then [4:36] human plus AI collaboration. So this is [4:40] about a is not just an auto just doing [4:43] some automation task fundamentally it is [4:46] going to how redesign the whole business [4:48] value chain okay along three lenses AI [4:52] path value and then realize that we [4:54] going to see the autonomous is something [4:56] like where the data models and agents [4:58] are going to play a role and making the [5:01] work more autonomous at the same time [5:04] with the help of human beings and again [5:06] the third one is about the human plus AI [5:08] Okay, where they augment human [5:11] capabilities with AIdriven insights and [5:13] capabilities. See the thing is it is not [5:15] only about the the faster resolution but [5:18] how the synergy should happen between [5:20] the human and A. So the with these three [5:23] pillars like maybe the first one is [5:26] about the reimagining value chains like [5:30] we we see that in three lenses value [5:34] path and then the realize. So the value [5:37] path and realize is something like as [5:40] you know the experience growth cost [5:42] experience is nothing about the customer [5:43] experience and then the employee [5:45] experience the growth is about the [5:48] building the new business models usage [5:50] based services digital teams and all and [5:52] again the cost is always how to make it [5:55] lean and then more efficient. Okay. And [5:58] the a path is about the knowledge [6:00] synthesis is about like see we we have [6:03] we know like the knowledge is structured [6:06] unstructured and the tacet knowledge how [6:08] we can extract that from the [6:09] organization. Design intelligence is [6:11] about how we can make it [6:15] how we can make it um by applying a AI [6:20] how we can automate that that is the [6:21] other thing. The third one is about the [6:24] business assurance where the business [6:26] assurance is something where right now [6:29] everything like from models uh from um [6:33] whatever the models data everything we [6:36] we we need to have the trust and [6:38] assurance. Okay. So that is the other [6:40] key thing we are looking for. A realizes [6:43] something like we as we know like the [6:46] convergence integrating tech and ops see [6:49] right now AI has broken the barriers [6:52] right now as you know like it is not [6:54] like where earlier it's a data team uh [6:57] operations team business team now it is [6:59] breaking the barriers and then everybody [7:02] is converging through AI that is the [7:04] biggest thing is happening again [7:05] autonomous human finetuning [7:08] as we we go we see as we go so [7:11] autonomous system is the three key [7:12] components like data, models, agents [7:15] like play a role and again when we [7:18] develop solutions we see what is the [7:20] partner ecosystem we have okay from AWS [7:23] to we are the strategic partners of AWS [7:26] and then apart from that we have a [7:28] partnership with other uh companies as [7:31] well and then the data if you see like [7:34] the data could be a structured [7:35] unstructured and then the tacitu [7:37] knowledge what we have in the [7:39] enterprises and then models is about [7:43] let it be LLMs or SLMs as well and then [7:46] agents could be like where [7:48] we develop solutions with bedrock agents [7:50] but at the same time we we have the uh [7:54] we we will util based on the customer [7:56] needs we we have crew or strands that we [8:01] can extend. So if you see the whole [8:03] holistic picture here, this will feed to [8:05] the reimagined value chains of [8:07] autonomous system. The new IT, new it is [8:09] not about the new software development [8:11] I'm talking about here. It is about how [8:14] this IT is being helping okay to make [8:17] all all the things integrated and [8:20] chained and then giving the assurance of [8:22] working in the way what it wants. That [8:24] is a new it. And then again new ways of [8:26] working people have to really find out [8:28] like how the human and a can collaborate [8:30] and work together [8:33] and this pretty much you you you might [8:36] know about how humans and agents here. [8:38] So key things okay the humans are very [8:40] good in the judgment and then they are [8:42] good in making uh decisions or empathy [8:45] decisions and all and also some of the [8:49] uncertain conditions like uh where human [8:53] is good in taking the decision then [8:55] agents these these are the areas where [8:58] they can collaborate and work together [9:00] and again the ensuring guard rails and [9:04] policy compliance and all we have to [9:05] apply [9:08] So [9:10] with this maybe in another three slides [9:12] is going to be a couple of three slides [9:13] where it's going to be really [9:15] interesting where I'm going to share the [9:17] real experience what we faced with the [9:18] customers. Okay. So the thing is [9:23] why business design? I'm not talking [9:25] about technical design here because see [9:27] one of the customers okay they want to [9:30] transform the complete value chain of [9:31] the uh business value chain. So they [9:34] jumped into developing the technical [9:37] design document and then they have what [9:41] they did is they have done the highle [9:43] analysis of the complete value chain and [9:46] then each stages and that come up they [9:49] came up with the quick uh user stories [9:51] and then based on that they have [9:53] developed the agents and then they have [9:55] started deploying. So everything was [9:57] working well but the thing is the the [10:00] agents were working individually well [10:03] but when the real challenge happened [10:05] when they start integrating it really [10:08] start the challenges they face and they [10:10] are they they were not able to get the [10:11] right outcome what they expected okay so [10:14] that is the reason like before jumping [10:17] in directly okay we we need to focus on [10:20] the business design okay from from the [10:23] CC for for instance the clarity of [10:26] purpose is each agent what is the goal [10:28] and composition of task enable [10:30] enablement tools is something like the [10:32] LLM or PROM strategy we have to be clear [10:34] with that and then trust accountability [10:37] is like guard rails and then uh and also [10:40] about the other compliance task and then [10:43] scalability feedback is about the HITL [10:46] and then the other confidence factors [10:48] evaluation all we have to take into [10:50] consideration and again we will be [10:52] facing okay a lot of challenges about So [10:55] what are the ambiguous goals and then [10:58] the role mapping and all see these the [11:01] challenges what we face but what we have [11:03] done is we spent enough time in [11:06] analyzing the complete value chain and [11:08] then we start started involving the [11:10] stake uh holders from the beginning [11:13] co-created with co-created with the [11:14] stakeholders and then we we we followed [11:17] the complete steps and then the inter [11:20] enterprise impact obviously you might be [11:22] knowing the main thing is about improved [11:25] integration and then the humans AI [11:27] synergy and what this for this we have [11:30] used our accelerators and templates like [11:33] business value canvas agent persona [11:35] canvas workflow maps use case [11:36] prioritization guard matrix and again as [11:39] as I told you we have we have been [11:43] involved in developing applications [11:44] across horizontal and uh vertical. So [11:49] the the the key thing key point here is [11:52] when we develop gen applications the [11:54] prototypes and all yes fine we can build [11:57] quickly and then move to production [11:58] finally but that it be a siloed [12:01] application but here we are looking for [12:03] the complete end to end value chain [12:04] application. So there people used to say [12:08] okay fail fast and move to the next one [12:10] but that is not applicable here when we [12:12] come come to the business value chain [12:13] transformation. So we need to spend [12:15] time. We need to spend time. We need to [12:18] spend time and then right end to end we [12:21] have to be clear with the business [12:22] design and then we have to get in. Okay. [12:25] But at the same time it is not like end [12:28] to end we have to do big bang and all. I [12:30] could say uh think big start small keep [12:34] progressing do [12:37] keep uh do the course correction and [12:39] keep progressing. That's what I could [12:41] say from the business value uh [12:42] transformation. So the with this uh [12:47] business uh because I'm not going to [12:49] cover the technical design because once [12:51] the business design is clear then all [12:54] comes about the technical design is [12:55] something like where we we used to uh [12:58] create starting from technical design [13:01] document then agent instruction document [13:04] then um J [13:07] man agent manifest JSON that we will be [13:11] creating and after that the hi L spec we [13:14] would be creating and then the [13:15] corresponding unit test plans and [13:18] scripts you'd be creating. So that is [13:19] for the technical design perspective. [13:21] Okay. But before that this is the base [13:24] where we have to be clear when we go for [13:27] any end to-end business value chain [13:29] transformation. [13:32] So this is another experience like what [13:36] I want to share like one of the [13:38] customers we have been building the [13:40] contact center value chain where end to [13:42] end we have been initially we we have [13:46] built we have been building the j [13:48] application in pockets like for [13:50] improving the customer experience alone [13:53] we have built some of the self-help chat [13:55] bots and all then again we have built [13:58] some of the applications in uh for [14:00] improving the agent experience. Okay. [14:03] For for instance giving smart assist for [14:06] agents that sort of thing. But then then [14:10] when we started applying to the [14:12] customers then they have started [14:13] realizing that many more points needs to [14:16] be connected to get the complete outcome [14:18] of this uh transformation transforming [14:20] the contact center value chain. So then [14:23] what then we realized okay then we then [14:27] we realized and then we we thought okay [14:29] end to end we need to consider and then [14:31] what we did is complete analysis we did [14:35] and this is the steps we followed for [14:37] the customer and then we in fact we have [14:40] our own solution or application so it's [14:44] called a contact synergy TCS own [14:45] solution we have got okay so we have [14:48] built and and if you see [14:52] You know clearly like where are the [14:54] areas where agents can play a role and [14:57] where are the areas where agents and [14:59] human needs to collaborate. For [15:02] instance, QE means to validate the [15:04] critical fails or masking audits. Each [15:06] and every stage of the value chain we [15:08] identify and then we have put HITL [15:11] process in place. Again the steps [15:14] whatever we have given is not the [15:15] journey flow is not the complete still [15:17] lot more steps are there but I have just [15:19] give given you a snapshot. So with this [15:23] we have designed for the customer and [15:25] then right now we have then the customer [15:28] have started realizing [15:30] the [15:32] right uh right outcomes from this whole [15:35] value chain. Okay. So the the other [15:38] thing is about the [15:42] the one size fit all approach won't help [15:44] it defi based on the organization [15:47] slightly it would differ based on the uh [15:51] based on the uh process flow what they [15:55] are adopting. Okay. And again we have to [15:58] define the we have to design the agents [16:01] based on their customer needs. Okay. Uh [16:04] the thing is for instance uh the agents [16:07] could be a simple reflex agent or model [16:10] based agent or goal based agent or [16:12] utility based agent and we have to [16:14] clearly identified okay so then only [16:17] we'll get the real benefits of what what [16:19] exactly we are aiming for. Okay, [16:26] this is another one the transforming [16:28] patient care uh value chain where we for [16:30] one of the customer we have been [16:32] currently working on again we have [16:35] developed initially gen applications [16:37] like patient care and then simple sort [16:40] of gen applications but then the [16:43] customers have started realizing we have [16:45] to apply end to end so we have been [16:47] involved in building the end to end [16:50] application and it's called patient care [16:53] experience application for of TCS. Again [16:57] here uh the thing is lot of HITL is [17:02] involved because as you know in [17:03] healthcare like like let it be a GDPR [17:06] compliance, HIPPA compliance we have to [17:08] meet that and wherever we we have to [17:10] identify the checkpoints and then [17:12] include HITL so that human and agent can [17:16] work together and then come up with the [17:18] right output right outcome. Again as I [17:21] said this is the incremental progress [17:23] and then we have to study the [17:25] organization the entire workflow and [17:27] then keep on keep on uh embedding the AS [17:32] wherever required again the the value it [17:36] could be a the design type also it could [17:39] value it could be a chaining or whatever [17:41] the parallelization [17:44] based on the needs we will apply and [17:46] then make sure that it is being done [17:47] Okay. [17:50] So how TCS we can help like building the [17:53] use cases and then info workflow scale [17:56] connected or intelligence and we see we [18:01] we work uh closely with AWS uh okay we [18:04] are the strategic partners for AWS and [18:08] right from setting up the playground to [18:11] setting up the enterprise AI platform. [18:13] So we we work on leveraging all the AWS [18:17] services and also all all the other [18:20] opensource services as well models based [18:22] on the customer need and we have our own [18:24] frameworks and accelerators in place [18:27] starting from for the responsibility [18:29] secure application build agent framework [18:30] and all. [18:33] The other thing is just before winding [18:35] up I want to touch upon what like how [18:38] the TCS differentiation and AWS [18:40] advantage is going to really add uh [18:44] value to the customers here if you see [18:48] uh we are the strategic partners and [18:50] then we have 500 plus customers AWS [18:53] customers alone and then we have 25,000 [18:55] associates uh like certified in AWS and [18:59] then we are geni competency partner and [19:01] while working with WS we have got a lot [19:04] of [19:06] so we we see there are a lot of [19:07] advantage we have for exclusively [19:10] partner initiatives we have for instance [19:12] jenna rapid program one program is there [19:16] in AWS where we can utilize that for [19:19] rapid assessments of the gen initiatives [19:21] that is one thing the second one is for [19:24] any of the P initiatives we will get the [19:26] funding okay when we when we work along [19:29] with AWS with the customers we will get [19:31] the funding from AWS The second one is [19:33] about the Geni P funding where for any [19:36] of the new innovative PC's we always [19:40] work with the customers and then see [19:42] like how we can leverage AWS and get the [19:44] P funding that the second one and the [19:46] third one the recent one what we have [19:48] started is the process to agent [19:49] initiative where in this initiative if [19:53] if if any of the customers like where we [19:56] have they have different processes in [19:57] place if we want to convert that P2A [20:00] initiative we can we can apply and get [20:02] the required funding. So I think [20:05] >> that's all that's all the time we have. [20:06] I'm sorry. [20:08] >> So that's it from my end. Always you can [20:11] uh connect us in the booth TCS booth 546 [20:14] we have the booth and then reach me at [20:17] my email id. Thank you.