Reltio Urges Shift to Real-Time Data Graphs to Scale Autonomous AI

The New Rules of Data Intelligence in the Age of AI With Reltio — Ai4

Ai4

Enterprises cannot scale autonomous agentic systems until they replace fragmented batch pipelines with unified real-time data graphs that eliminate human-in-the-loop bottlenecks.

AI-assisted summary of The New Rules of Data Intelligence in the Age of AI With Reltio — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by Ai4 on August 24, 2025, this presentation features Sriraj Rajaram, Director of Product Management at Reltio. According to the description, the session explores the new data strategies and enterprise foundations that technology leaders must adopt to support real-time, agentic AI decision-making at scale.

## TL;DR Sriraj Rajaram outlines the critical data management shifts required to transition from the Industrial Age to the Age of Intelligence, where autonomous AI agents replace manual human workflows. He presents four core rules for modern enterprises: establishing foundational data trust through unification rather than aggregation, architecting context-rich data graphs instead of static relational tables, prioritizing sub-millisecond edge velocity over raw storage volume, and packaging assets into reusable enterprise data products. In the concluding Q&A, he explains how modern zero-copy architectures and graph models overcome the historical failures of legacy 360-degree data initiatives without requiring burdensome ETL pipelines.

## Atmosphere & style Delivered during a morning 'cappuccino session,' the presentation carries an engaging, pragmatic tone that balances high-level enterprise vision with relatable personal anecdotes about travel delays and online shopping carts. Rajaram speaks with refreshing candor, openly referencing the 'dead bodies' of past failed 360-degree data initiatives to ground his technical arguments. The transition into audience Q&A feels collaborative and precise, offering a deep, authoritative dive into the architectural mechanics of graph modeling and zero-copy data unification.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=4s) The transition from the Industrial Age to the Age of Intelligence shifts the primary economic currency from physical asset accumulation to the organization and operationalization of data for autonomous machine decision-making. - [01:57](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=117s) Siloed, low-quality data acts as the 'kryptonite' of agentic AI, creating data debt that forces enterprises to keep humans in the loop and limits AI scalability. - [03:32](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=212s) Rule 1: Build foundational data trust through true data unification and automated entity resolution, rather than simple data aggregation into data lakes. - [05:02](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=302s) Rule 2: Architect data like a neural network using data graphs to provide AI agents with semantic relationship context rather than static, disconnected records. - [06:51](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=411s) Rule 3: Compete on data velocity rather than sheer volume by serving trusted data assets to the operational edge with sub-millisecond latency. - [09:02](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=542s) Rule 4: Package data assets into reusable, enterprise-wide data products to avoid rebuilding bespoke pipelines for every new AI and digital transformation cycle. - [10:46](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=646s) Historical 360-degree customer initiatives largely failed due to high ETL integration costs, monolithic architectures, and an inability to operationalize insights downstream in real time. - [14:46](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=886s) Graph-native architectures model entities, attributes, and interaction events together, establishing the holistic contextual web necessary for autonomous agent reasoning. - [16:27](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=987s) Zero-copy architecture eliminates the friction of physical data movement by accessing data in residence and layering business logic in minutes rather than spending years on data migration.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=4s) Introduction: From the Industrial Age to the Age of Intelligence

Opening the "cappuccino session" on the final morning of the event, Sriraj Rajaram introduces himself as a Director of Product Management at Reltio, an industry leader known for data unification and delivering 360-degree data views. He sets the stage by contrasting two fundamental economic paradigms: the Industrial Age versus the modern Age of Intelligence. While the Industrial Age relied on the accumulation of physical assets and control through proprietary methods, the Age of Intelligence shifts decision-making from humans to autonomous AI agents. In this new era, the primary currency is not physical accumulation, but how organizations organize, distribute, and operationalize their data and knowledge assets.

AI systems are moving beyond merely analyzing complex information or augmenting human analysis; in many operational domains, they are beginning to replace human decision-making entirely. This balance between augmenting human judgment and enabling true machine autonomy drives unprecedented speed and operational efficiency, but it also introduces profound enterprise challenges that require a rethinking of traditional data management.

[01:57](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=117s) Agentic AI in Action and the Threat of Data Debt

Autonomous agents are already visible in day-to-day operations. When Rajaram experienced a flight rescheduling on his way to the conference, an AI agent autonomously rebooked his hotel and car reservations without manual intervention. Similar automated workflows are transforming patient experience in healthcare, from managing prescription refill reminders to scheduling appointments with care providers. However, fully handing over mission-critical decisions to autonomous systems introduces operational risk, which explains why many organizations still insist on keeping a human in the loop.

Rajaram explains that humans remain in the loop largely because underlying enterprise data is siloed and untrusted: "Data is the kryptonite of agentic AI and AI automation." The rapid ease of software generation is accelerating an enterprise application explosion—organizations operating 50 applications today will soon manage 500, and those with 500 will see 5,000. This fragmentation widens the "data debt," defined as the widening chasm between what enterprise data is currently capable of doing and what it must deliver for autonomous operations.

```mermaid flowchart TD accTitle: Data Debt and the Path to Autonomous Systems subgraph Fragmentation["Enterprise Data Debt"] Apps["Application Explosion<br/>(50 to 5,000 Apps)"] Silos["Siloed & Poor-Quality Data"] Apps --> Silos end

subgraph Bottleneck["Current State"] HITL["Human in the Loop<br/>(Manual validation & risk mitigation)"] Silos --> HITL end

subgraph Solution["Reltio Data Intelligence Layer"] Unify["Governed Data Unification"] Graph["Context-Rich Data Graph"] Edge["Sub-Millisecond Edge Delivery"] Unify --> Graph --> Edge end

subgraph Autonomy["Scale & Competitive Advantage"] Agent["Autonomous Agentic AI<br/>(Zero Human Bottleneck)"] Edge --> Agent end

HITL -.->|Eliminated by Data Trust| Unify ```

[03:32](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=212s) Rule 1: Build Data Trust as the Foundation for Autonomous Systems

Boardrooms and executive teams across industries are focused on one core objective: how to scale AI for measurable business impact. As long as poor data quality forces a human to manually review every action, decision, or output, scaling remains constrained. Rajaram emphasizes that the first organization capable of scaling agentic workflows without requiring human validation will fundamentally reshape competitive dynamics in its industry. Achieving this requires establishing foundational data trust.

Reltio addresses this bottleneck by constructing a unified, trusted data layer that compiles and standardizes core profile entities—such as customers, products, and suppliers—from across disparate enterprise sources, making them available upstream and downstream. Rajaram makes a clear distinction: "Data unification is not the same as data aggregation." Aggregation simply gathers information into one place, whereas unification curates data through governed algorithms and automated business rules that establish continuous entity resolution and data integrity.

[05:02](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=302s) Rule 2: Architect Data Like a Neural Network

The legacy mindset of the Industrial Age treated data management as mere volume accumulation, leading companies to build massive data warehouses and data lakes without context. Rajaram shares an example of an industrial equipment manufacturer that invested heavily in a platform to accumulate vast streams of IoT sensor metrics from the field. Because the company failed to contextualize or act upon that raw telemetry, the accumulated data proved useless, and the manufacturer struggled competitively.

To support autonomous decision-making, data must be structured like a neural network—rich with relationship context rather than stored like static records in a filing cabinet. Reltio implements this using data graph technology. A data graph not only maintains core entity profiles (e.g., customer, product, supplier), but also models every interaction, transaction, and relationship connecting them. This graph functions as an enterprise-wide semantic layer, providing AI agents with the holistic contextual framework required to make independent, reliable business decisions.

- **Data Accumulation vs. Data Context**: Storing petabytes of raw metrics delivers no business value unless relationships and interactions are modeled semantically. - **Data Graph Architecture**: Connects profile entities, multi-dimensional attributes, and interaction histories into an interconnected semantic network for autonomous agents.

[06:51](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=411s) Rule 3: Compete on Velocity, Not Volume

Competitive advantage in the Age of Intelligence does not stem from possessing the largest repository of data; it comes from moving data intelligently in real time to the operational edge where actions occur. To illustrate the value of data velocity, Rajaram shares a personal e-commerce experience: while browsing an online store and hesitating over a purchase in his cart, the site's real-time engine detected the delay and instantly offered a 10% to 20% discount on the spot, successfully nudging him to complete the transaction. Conversely, another retailer sent an abandoned-cart discount email a full day after he left their website—long after his purchase intent had evaporated.

To enable instantaneous decision-making, Reltio provides a low-latency data delivery architecture called the "light-speed delivery network." This infrastructure distributes trusted enterprise data assets directly to operational touchpoints—including mobile applications, enterprise platforms, customer service portals, and e-commerce engines—with sub-millisecond performance.

```mermaid flowchart LR accTitle: Real-Time Data Velocity vs Latent Batch Processing subgraph RealTime["High Velocity Edge Delivery"] Browse1["User Hesitates at Cart"] --> Nudge["Instant Real-Time Context & Offer"] Nudge --> Conv["Immediate Purchase Conversion"] end

subgraph Delayed["Latent Batch Processing"] Browse2["User Abandons Cart"] --> Wait["24-Hour Latency / Batch Pipeline"] Wait --> Email["Next-Day Promo Email"] Email --> Lost["Lost Purchase Intent"] end ```

[09:02](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=542s) Rule 4: Adopt an Enterprisewide Data Products Approach

Enterprises are grappling with AI integration while still actively navigating broader digital transformation initiatives that started years ago with social, mobile, analytics, and cloud (SMAC). With technology moving rapidly from generative AI to agentic AI, organizations cannot afford to rebuild bespoke data pipelines for every emerging technology wave. Instead, leaders must establish a unified, enterprise-wide strategy that treats data assets as reusable data products.

By packaging core assets—such as customer data or supplier data—into modular, governed data products, enterprises ensure that a single validated asset can be shared seamlessly across marketing, sales, finance, operations, and back-office functions. Even when initiating an implementation around a single targeted use case, designing data assets as products ensures that downstream initiatives across diverse business units can immediately leverage the same trusted data foundation.

- **Digital Transformation Continuity**: Reusable data foundations prevent organizations from having to rebuild data infrastructure as new AI paradigms emerge. - **Enterprise Data Products**: Modular assets (e.g., Customer Product, Supplier Product) deliver simultaneous value across marketing, finance, sales, and supply chain domains.

[10:46](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=646s) Summary and Q&A: Overcoming the Failure of Historical 360 Initiatives

Rajaram summarizes the four core rules of data intelligence: 1. **Build Data Trust**: Eliminate human bottlenecks by ensuring underlying data accuracy. 2. **Architect Data Like a Neural Network**: Structure context-rich data graphs rather than static filing cabinets. 3. **Compete on Velocity, Not Volume**: Deliver trusted assets to the edge with sub-millisecond latency. 4. **Treat Data as Enterprise Data Products**: Build reusable assets across business units.

He notes that Reltio outlines ten complete rules in their dedicated white paper and opens the floor for audience questions.

The first question addresses how organizations can overcome the historical failures associated with building context-rich 360-degree data views. Rajaram acknowledges that "360 initiatives are like an enigma—everybody wants it, nobody can have it... history is filled with failures and dead bodies of 360 initiatives." Historically, these projects failed due to three factors: exorbitant ETL integration costs, monolithic architectures burdened by legacy tech debt, and the inability to deliver curated insights downstream to operational systems in real time. Modern technology overcomes these barriers through zero-copy data sharing (eliminating physical consolidation), composable architectures (bringing computing capabilities directly to data residence), and real-time agentic execution.

[14:46](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=886s) Q&A: Graph Modeling vs. Relational Databases

An audience member asks for clarification on what it means to architect data as an interconnected network, inquiring whether this refers to relational database tables with embedded business logic or an object-oriented model.

Rajaram clarifies that the architecture relies on graph theory rather than standard relational tables linked by foreign keys. In a graph-native data structure, the architecture manages three core components: - **Entities**: Core business objects such as customers, products, and suppliers. - **Attributes**: Granular metadata tied to entities, such as physical locations and contact details. - **Interactions**: Discrete business events, such as a customer placing a support call or a supplier completing a delivery at a specific site.

Connecting entities, attributes, and temporal interactions within a unified graph provides the deep contextual relationship web that AI agents need to reason autonomously. An attendee begins asking a follow-up question regarding how this differs from legacy 360 approaches in enterprise environments carrying 15 years of accumulated data silos across multiple lakehouses and data warehouses, setting up the deeper technical discussion.

[16:27](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=987s) Eliminating Data Movement with Zero-Copy Architecture

Addressing how this new approach fundamentally differs from legacy paradigms, Sriraj Rajaram reflects on the longstanding obstacle enterprise data initiatives have faced: the sheer friction of data consolidation. Traditionally, organizations struggle with massive volumes of data scattered across siloed systems, compounded by domain-specific business logic embedded directly within the datasets. Rajaram shares an illustrative real-world example from a customer engagement where a value of "zero didn't mean zero, it meant something else" due to underlying business logic that required interpretation and contextual understanding rather than raw ingestion.

To overcome these historical bottlenecks, modern data management leverages zero-copy architecture, eliminating the need to physically migrate or duplicate datasets across environments. Instead of moving data from one physical repository to another, enterprises can access data in residence, interpret its nuances, and layer critical business rules and semantic logic directly on top of it. This architectural shift dramatically accelerates the timeline for delivering unified, comprehensive 360-degree customer and entity views—transforming what used to be a year-long data migration slog into an operation accomplished in minutes.

```mermaid flowchart LR accTitle: Zero-Copy Data Unification Flow subgraph S1["Data in Residence"] A["Source Data"] B["Embedded Business Logic"] end subgraph S2["Reltio Zero-Copy Layer"] C["In-Place Data Access"] D["Context & Logic Layering"] end subgraph S3["Output"] E["Real-Time 360 View"] end

A --> C B --> D C --> E D --> E ```

- **In-Place Access Over Physical Relocation**: Zero-copy technology queries and analyzes datasets where they natively reside, removing the overhead, latency, and synchronization issues tied to ETL pipelines. - **Embedded Logic Resolution**: Enables enterprises to accurately parse and layer intricate operational logic—such as contextual numeric values—without corrupting or misrepresenting source records. - **Drastic Reduction in Time-to-Value**: Organizations can bypass up to a year of pipeline engineering and data transport to assemble trusted 360-degree views in a matter of minutes directly within Reltio.

## Notable quotes > [00:00](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=0s) The old rules of data no longer apply. To lead in the Age of Intelligence, enterprises must adopt new data strategies that move at the speed of AI. — Sriraj Rajaram

> [01:57](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=117s) Data is the kryptonite of agentic AI and AI automation. — Sriraj Rajaram

> [01:57](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=117s) Organizations operating 50 applications today will soon manage 500, and those with 500 will see 5,000. — Sriraj Rajaram

> [03:32](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=212s) The first organization capable of scaling agentic workflows without requiring human validation will fundamentally reshape competitive dynamics in its industry. — Sriraj Rajaram

> [03:32](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=212s) Data unification is not the same as data aggregation. — Sriraj Rajaram

> [05:02](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=302s) Storing petabytes of raw metrics delivers no business value unless relationships and interactions are modeled semantically. — Sriraj Rajaram

> [06:51](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=411s) Competitive advantage in the Age of Intelligence does not stem from possessing the largest repository of data; it comes from moving data intelligently in real time to the operational edge where actions occur. — Sriraj Rajaram

> [10:46](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=646s) 360 initiatives are like an enigma—everybody wants it, nobody can have it... history is filled with failures and dead bodies of 360 initiatives. — Sriraj Rajaram

> [16:27](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=987s) zero didn't mean zero, it meant something else — Sriraj Rajaram

## Mentioned Sriraj Rajaram, Reltio, Ai4, SMAC (Social, Mobile, Analytics, and Cloud)

## Video description Sriraj Rajaram, Director, Product Management, Reltio

Abstracts

The old rules of data no longer apply. To lead in the Age of Intelligence, enterprises must adopt new data strategies that move at the speed of AI— in real-time - and that are connected and built for continuous decision-making. In this session, Principal, Product Management Sriraj Rajaram shares the new rules of data that enterprises need to adopt in the Age of Intelligence, and the key capabilities and enterprise data foundation that CAIOs, CIOs and IT leaders need build across their organizations to be ready for the operational realities of AI at scale. Attend this session to learn how an agentic data fabric of unified, trusted data assets accelerates AI transformation & learn more about real-world strategies for success.

[Source](https://www.youtube.com/watch?v=L6iVGTxu9Lg)

中文简报

Reltio呼吁转向实时数据图谱以规模化扩展自主AI 企业唯有用消除人工介入瓶颈的统一实时数据图谱取代碎片化批处理管道，才能实现自主智能体系统的规模化扩展。 Sriraj Rajaram概述了从工业时代迈向智能时代所必需的关键数据管理变革；在智能时代，自主AI智能体将取代人工工作流程。他为现代企业提出了四大核心准则：通过统一而非聚合来建立底层数据信任、构建富上下文的数据图谱而非静态关系表、优先保障亚毫秒级边缘速度而非单纯追求存储容量，以及将资产打包为可复用的企业数据产品。在最后的问答环节中，他阐释了现代零复制架构与图模型如何在无需繁重ETL管道的情况下，克服传统360度数据举措的历史性失败。 经济模式正从物理资产积累转向数据的组织与自主运营，而企业应用爆发引发的数据孤岛与数据债务，已成为阻碍 Agentic AI 规模化落地的核心瓶颈。 (1:57) 企业需通过自动化实体解析建立数据信任，并利用类神经网络的数据图谱（Data Graph）连接实体、属性与交互，为 AI Agent 自主推理提供语义上下文支持。 (5:02) 智能时代的数据竞争取决于流转速度而非体量，通过低延迟交付网络在亚毫秒级内将可信数据资产分发至边缘触点，才能捕捉实时业务转化机会。 (6:51) 企业应将核心数据资产封装为标准化的企业级数据产品，确保跨部门复用，避免在 Agentic AI 等技术迭代中重复构建定制化管道。 (9:02) 针对历史上 360 度画像项目因高昂 ETL 成本而频繁失败的困境，现代方案采用零复制（Zero-Copy）架构就地处理数据与业务逻辑，将长期迁移工程缩短至数分钟。 (16:27)

Reading summary

Sriraj Rajaram, Director of Product Management at Reltio, argues that scaling autonomous agentic AI requires moving beyond legacy data accumulation to build trusted, contextualized, and low-latency data foundations. Resolving Data Debt to Remove Human Bottlenecks (1:57) In the shift from the Industrial Age to the Age of Intelligence, enterprise decision-making is increasingly transferred to autonomous software agents. While agentic workflows already handle consumer tasks like travel rebooking and healthcare reminders, organizations continue to keep humans in the loop to mitigate operational risks stemming from fragmented, untrusted data. Rajaram warns that rapid software generation is driving an enterprise application explosion—projecting environments will expand from 50 applications to 500, or from 500 to 5,000—which widens the enterprise "data debt." Trust and Semantic Context via Graph Architecture (3:32) Rajaram contends that achieving competitive scale with AI requires eliminating manual validation by establishing foundational data trust. Rather than basic data aggregation, this demands true unification through governed algorithms, continuous entity resolution, and automated business rules. Accumulating raw volume without context yields little value, as illustrated by an industrial manufacturer whose uncontextualized IoT telemetry failed to support operations. Relational Limitations and Graph-Based Entity Context (14:46) To provide the reasoning framework autonomous agents require, data must be structured as an interconnected semantic network using graph theory rather than standard relational tables linked by foreign keys. In a graph-native architecture, systems connect core business entities (such as customers and suppliers), granular attributes (such as locations), and discrete temporal interactions (such as customer calls or deliveries). This connected graph operates as an enterprise semantic layer, providing the contextual relationship web necessary for independent automated decision-making. Velocity at the Edge and Reusable Data Products (6:51) Competitive advantage in autonomous operations relies on data velocity rather than data volume. Moving trusted information to operational touchpoints with sub-millisecond latency enables instantaneous actions, such as presenting immediate purchase incentives to hesitating digital shoppers rather than sending delayed abandoned-cart emails after purchase intent has evaporated. Furthermore, organizations navigating waves from generative to agentic AI must treat core assets as modular, governed data products to avoid building bespoke pipelines for every new initiative. Overcoming Historical 360 Failures with Zero-Copy (16:27) Historically, comprehensive 360-degree data initiatives frequently failed due to prohibitive ETL pipeline costs, monolithic legacy debt, and an inability to operationalize insights in real time. Modern architectures address these historical points of failure by adopting zero-copy data sharing, which eliminates physical data consolidation. By accessing source data directly in residence and layering semantic logic and business rules over existing datasets without duplication, enterprises can compress 360-degree data modeling timelines from year-long pipeline projects into minutes. While zero-copy architecture and graph structures resolve long-standing integration bottlenecks, fully realizing agentic autonomy depends on an enterprise's ability to consistently operationalize data products across disparate legacy environments without reintroducing manual validation.

精炼阅读摘要

Reltio产品管理总监Sriraj Rajaram认为，企业进入智能时代的核心挑战在于底层数据孤岛与数据债务制约了智能体AI（Agentic AI）的自主决策。企业必须通过建立数据信任、构建神经网络式图谱、提升实时分发速率以及推行企业级数据产品，摆脱对人工验证环节的依赖并实现规模化扩张。 化解数据债务以实现智能体自主化 (1:57) 在智能时代，决策权正逐步从人类转移至自主AI代理，但多数机构因底层数据割裂与不可信，仍然被迫保留“人在回路”（Human in the Loop）机制。Rajaram指出，企业应用软件生成门槛降低导致系统数量激增，企业维护的应用程序可能从50个跃升至500甚至5,000个，这加剧了数据割裂并形成了所谓的“数据债务”。只要低质量数据依然要求人工对每个输出进行核查，AI的规模化业务价值就会受到实质性阻碍。 区分数据统一与简单聚合以建立信任 (3:32) 规模化运行智能体工作流的前提是建立基础数据信任。Rajaram明确指出，数据统一（Data Unification）不等同于简单的数据聚合（Data Aggregation）。单纯的聚合仅是将零散信息搬运到一个位置，而统一则是利用治理算法与自动化业务规则，持续执行实体解析与数据完整性校验，将分散在各部门的客户、产品与供应商等核心实体档案标准化，进而为上下游系统提供可靠的数据底座。 采用图模型赋予数据上下文关系 (5:02) 传统工业时代思维倾向于无节制地累积数据量，如某工业设备制造商重金搭建平台收集海量IoT传感器遥测数据，却因缺乏上下文关联而无法产生业务价值。Rajaram主张企业应像神经网络一样组织数据，利用数据图谱（Data Graph）替代静态表格。该图谱将核心实体（客户、产品、供应商）、细粒度属性以及离散交互事件（如客服通话、现场交付）联结为语义网络，使AI代理具备自主推理所需的全局业务上下文。 以毫秒级时效取代体量竞争 (6:51) 智能时代的竞争优势不再取决于数据仓库的绝对容量，而取决于将可信数据实时推送到业务边缘的流转速度。例如在电商结账犹豫时即时提供折扣可促成交易，而延迟一天的批量邮件往往错失购买意图。Reltio通过光速分发网络架构，使受治理的企业数据资产能够以亚毫秒级的低延迟直达移动应用、运营系统与客户服务门户，支持前端业务引擎的瞬时决策。 以零复制架构打破360度视图历史困境 (16:27) 过去全方位360度视图项目常因高昂的ETL成本、遗留技术债务以及无法实时传递洞察而屡遭失败。传统整合往往还面临底层嵌入业务逻辑的解读困难，例如某些数据集中“0”因特殊规则并不代表数值零。借助零复制（Zero-Copy）架构，企业无需物理迁移或复制数据，可直接在数据驻留处进行就地访问与语义逻辑叠加，从而将长达一年的管道工程压缩至几分钟内交付。 通过将核心数据打包为跨部门复用的标准化数据产品，企业能够避免在技术演进中反复重建底层管道。然而，如何在保有遗留数据孤岛的复杂IT架构中平稳落地零复制和图谱语义层，仍是企业从传统分析向全自主智能体决策转型时需要持续面对的实施课题。

Detailed notes — zh

## Background 该视频由 Ai4 频道于 2025 年 8 月 24 日发布在 YouTube 平台，记录了 Reltio 产品管理总监 Sriraj Rajaram 的主题演讲。页面简介指出，本次演讲旨在分享企业在智能时代必须采纳的数据智能新法则，探讨企业高管应如何构建统一、可信的代理式数据编织（Agentic Data Fabric）以支撑规模化 AI 的落地与持续决策。 ## TL;DR 演讲指出，当前企业正从注重资产积累的工业时代迈入强调数据分发与自主行动的智能时代，而数据孤岛与信任缺失引发的“数据债务”已成为 Agentic AI 的最大阻碍。为了实现无需人工介入的规模化自主决策，讲者提炼了数据智能的四大法则：建立作为自主基石的数据信任、采用图模型构建神经网络式富上下文语义层、依托极速网络竞逐交付速度而非单纯体量，以及在全企业推行标准化数据产品思维。针对传统 360 度全景视图因繁重搬迁而屡屡失败的困境，企业应结合零复制（Zero-Copy）技术与可组合架构，在数据驻留地就地赋能 AI 智能体执行业务闭环。 ## Atmosphere & style 演讲以轻松幽默的“卡布奇诺早会”调侃开场，整体氛围专业干练、务实且富有前瞻性。讲者巧妙结合在线购物实时挽留与滞后邮件等生活化案例，将枯燥的数据架构理论讲得通俗生动。随后的问答环节互动热烈，讲者直面传统数据仓库与 360 项目的沉疴顽疾，以条理分明的技术洞察给出了清晰犀利的破局方案。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=4s) 人类正在经历从工业时代向智能时代的范式转变，竞争核心由囤积实体资产转变为高效分发数据资产并驱动 AI 代理进行自主决策。 - [02:30](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=150s) 数据质量低下与数据孤岛引发的数据债务是 Agentic AI 的“氪石”，导致企业因缺乏信任而不得不长期依赖人工审核介入。 - [03:32](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=212s) 数据智能第一法则：建立数据信任作为自主系统的底层基石，率先在无人工介入下规模化运行业务流程的企业将重塑行业格局。 - [03:32](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=212s) 数据统一绝非简单的数据聚合，必须通过严谨的数据治理与算法规则对跨系统数据进行清洗、匹配并维护一致的核心档案。 - [05:02](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=302s) 数据智能第二法则：像神经网络一样构建数据架构，摒弃文件柜式的静态堆积，通过数据图谱建立富含上下文的企业语义层。 - [06:51](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=411s) 数据智能第三法则：企业的竞争优势取决于数据转化为行动的速度而非数据体量，需要将受信任数据以亚毫秒级延迟交付至业务边缘。 - [06:51](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=411s) 企业需要制定统一且具有凝聚力的数据资产投资策略，确保数据具备通用性与复用性，从容应对生成式 AI 与 Agentic AI 等多轮技术浪潮。 - [09:55](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=595s) 推行“数据产品（Data Products）”思维，将客户、产品和供应商等核心数据封装为标准资产，供跨部门业务流复用以释放最大价值。 - [12:14](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=734s) 传统 360 度视图屡屡失败的原因在于物理数据集成成本极高、深受遗留单体系统拖累且难以实时将数据赋能至一线业务终端。 - [12:14](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=734s) 现代零复制技术（Zero-Copy）、可组合架构与 Agentic AI 的结合，使算力能够下沉至数据驻留地，彻底破局传统 360 视图的落地困境。 - [14:46](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=886s) 通过整合实体（Entities）、属性（Attributes）与动态交互（Interactions）的图数据模型，超越传统主外键范式构建高关联度的企业知识网络。 - [14:46](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=886s) 借助零复制技术，企业无需花费数月搬运散落在数仓和数据湖中的历史数据，几分钟内即可在原始驻留地解析隐式业务逻辑并构建业务层。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=4s) 工业时代到智能时代：从资产积累到自主行动 Reltio 产品管理总监 Sriraj Rajaram 在演讲开场时幽默地称这是一场“卡布奇诺早会”，并简要介绍了 Reltio 作为数据管理和数据统一（Data Unification）领域领导者的背景。他指出，当前人类社会正在经历从“工业时代（Industrial Age）”向“智能时代（Age of Intelligence）”的深刻范式转移。工业时代的核心驱动力是利用专有方法积累和控制实体资产；而在如今的智能时代，核心资产转变为知识资产与数据资产，关键不在于盲目囤积，而在于如何高效分发并将其转化为行动。 AI 系统正在处理庞大且复杂的企业级信息，其角色已经从辅助人类决策（Human Decision-making）逐步升级为在某些场景下直接替代人类决策。这种“增强人类判断”与“赋予机器自主性”的双重能力，极大提升了业务运转的速度与效率。例如，在差旅场景中，航班延误后 AI 代理（AI Agent）能够自主触发并重订酒店与租车；在医疗患者体验中，代理可以自动发送处方补药提醒或自主重新预约护理人员。 ```mermaid flowchart LR accTitle: 工业时代与智能时代的范式对比 subgraph 工业时代 A["实体资产积累"] --> B["专有方法控制"] --> C["人类主导决策"] end subgraph 智能时代 D["数据/知识资产"] --> E["分发与行动化"] --> F["AI 代理自主决策"] end ``` --- ### [02:30](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=150s) 数据债务：Agentic AI 的“氪石” 尽管 AI 决策展现出巨大潜力，但企业面临的核心痛点是对代理决策缺乏信任，因此仍然不得不保留“人机协同（Human-in-the-loop）”。保留人类监督的根本原因在于数据孤岛（Data Silos）与低劣的数据质量。Sriraj 形象地指出，“数据是 Agentic AI 和 AI 自动化的氪石（Kryptonite）”。 随着 Agentic AI 的普及，企业内部的应用正在呈现爆炸式增长——原本只有 50 个应用的企业会激增到 500 个，拥有 500 个应用的则会暴增到 5,000 个，因为创建 AI 应用的门槛已经大幅降低。应用泛滥导致“数据债务（Data Debt）”不断加剧，即“企业能够用数据做到的事情”与“数据本应交付的业务价值”之间的鸿沟被持续拉大。 - **核心矛盾**：缺乏数据信任迫使企业保留人工审核以核对动作与决策的准确性。 - **数据债务定义**：企业当前的数据能力与业务实际需求之间的差距。 - **应用爆炸**：创建 AI Agent 门槛降低导致企业内孤岛化应用数量呈几何级数增长（50 $\rightarrow$ 500 $\rightarrow$ 5,000）。 --- ### [03:32](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=212s) 数据智能第一准则：建立作为自主系统基石的数据信任 企业董事会与高管层当前最关心的议题是“如何规模化扩展 AI 的影响力（Scale AI for impact）”。Sriraj 提出了 AI 时代数据智能的**第一条准则**：必须将“数据信任（Data Trust）”构建为企业自主系统的底层基石。谁能率先在无需人工介入（No Human-in-the-loop）的情况下规模化运行代理流程，谁就将重塑所在行业的竞争格局。 这也是 Reltio 的核心专业所在。Reltio 专注于构建统一且值得信赖的数据层，从全企业范围内编译客户、产品、供应商等统一档案（Unified Profiles），并无缝分发给上下游系统。Sriraj 特别强调，“数据统一（Data Unification）”绝不等于简单的“数据聚合（Data Aggregation）”。Reltio 不是盲目堆砌信息，而是通过严格的数据治理流程和算法规则，智能决定如何清洗、匹配并统一维护数据资产。 --- ### [05:02](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=302s) 数据智能第二准则：像神经网络一样构建数据架构 数据智能的**第二条准则**是：像构建神经网络一样架构企业数据。过去许多企业沉迷于数据积累，建造了庞大的数据仓库（Data Warehouses）和数据湖（Data Lakes）。Sriraj 举了一个工业制造企业的反面案例：该厂商投入巨资建立平台收集市场上部署的海量 IoT 设备指标，但由于缺乏将指标转化为行动的能力，这些海量数据变成了沉睡的废料，导致该企业如今经营陷入困境。 智能时代的关键是赋予数据“丰富的上下文（Context Rich）”。Reltio 平台采用了“数据图谱技术（Data Graph Technology）”，它不仅沉淀客户、供应商和产品等静态档案实体，还将它们之间的所有交互与关系进行关联映射。数据图谱本质上构成了横跨整个企业的**语义层（Semantic Layer）**，为 AI 代理提供进行自主商业决策所需的完整上下文信息，这是 Agentic 流程在企业内规模化落地的唯一路径。 ```mermaid flowchart TD accTitle: Reltio 数据图谱语义层架构 subgraph 实体层 P1["客户 Profile"] P2["产品 Profile"] P3["供应商 Profile"] end subgraph 关系与上下文 R1["动态交互数据"] R2["业务关联关系"] end subgraph 决策消费层 Semantic["数据图谱（语义层）"] Agent["AI 代理自主决策"] end P1 --- R1 P2 --- R1 P3 --- R2 R1 --> Semantic R2 --> Semantic Semantic --> Agent ``` --- ### [06:51](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=411s) 速度压倒容量：毫秒级交付网络与全企业投资策略 企业的竞争优势来源于将数据转化为行动的速度，而非单纯的数据量大小。“关键不在于谁拥有的数据更多，而在于谁能更聪明、更实时地调度数据，并将其交付至业务处理的最边缘（Edge of processing）”。Sriraj 分享了自己的两次在线购物经历： 1. **实时拦截挽留**：在某个电商网站犹豫不决、反复进出购物车时，系统实时感知到了迟疑，当即弹出 10% 到 20% 的折扣券，瞬间打破犹豫促成下单。 2. **滞后的冷启动**：在另一家网站同样犹豫未买，直到离开网站一天后才收到“购物车遗留物品折扣”的邮件，但此时购买冲动早已消退，商机已经流失。 为了解决数据时效性问题，Reltio 推出了“光速数据交付网络（Light-speeded Delivery Network）”，能够在**亚毫秒级（Sub-millisecond）**的极低延迟下，将企业受信任的核心数据资产推送至移动端、企业应用、独立站点或电商引擎等边缘计算节点。 最后，Sriraj 提出了**第三条准则**：制定具有凝聚力的全企业范围数据资产投资策略。许多企业在数字化转型（社交、移动、分析、云）尚未彻底完成之际，就迎来了生成式 AI（Generative AI）与代理式 AI（Agentic AI）的连环冲击。只有建立统一的企业级数据资产管理方法，才能确保数据资产具有通用性与复用性，避免在未来每次面临全新技术浪潮时都要被迫推倒重来。 ### [09:55](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=595s) 数据产品方法与四大数据智能法则总结 在企业级数据战略中，推行“数据产品”（Data Products）思维至关重要。企业需要将客户数据、供应商数据等核心资产封装为标准化的数据产品，供组织内不同部门的业务流程和系统消费。即使最初仅从单一用例切入，也必须考量同一份数据资产如何在市场营销、销售、财务、运营及后台等更广泛的业务板块解锁潜在价值。Reltio 正是通过这种数据产品方法，赋能了众多各行业头部客户的日常业务运转。 ```mermaid flowchart TD accTitle: 数据智能的四大核心法则 subgraph CoreRules ["数据智能四大法则"] R1["1. 建立数据信任<br/>(减少人工介入，改变竞争格局)"] R2["2. 神经元网络化架构<br/>(富上下文图谱，摒弃文件柜式存储)"] R3["3. 竞逐速度而非体量<br/>(将数据推向计算边缘，实时触发行动)"] R4["4. 企业级协同投资<br/>(全面推行数据资产即产品)"] end R1 --> R2 --> R3 --> R4 ``` Sriraj 对数据智能的四大核心法则进行了系统性提炼： - **建立数据信任（Build Data Trust）**：若缺乏信任，系统便永远需要人类介入（Human in the loop）；谁能率先最大限度减少人工干预，谁就能彻底重塑行业竞争格局。 - **像神经网络一样构建数据架构（Architect Data Like a Neural Network）**：让数据富含上下文关联（Context Rich），而非像“文件柜”一样被动堆积，从而明确掌握实体之间的动态关联以支持行动。 - **竞逐速度而非单纯追求体量（Compete on Velocity, Not Volume）**：将数据直接输送到业务处理的最前端边缘，以最高速度驱动决策与行动。 - **建立企业级协同投资的数据产品体系（Enterprise-wide Data Products Approach）**：打破部门孤岛，系统化地将数据视作可复用的数据产品进行统一治理与输出。 --- ### [12:14](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=734s) 传统 360 度视图为何屡屡失败与现代技术的破局 在问答环节，有观众提问企业在收集富上下文信息时面临诸多挑战，业界有何最佳实践。Sriraj 指出，行业长期以来推崇的“360 度全景视图”（Customer 360 / Product 360 / Supplier 360）宛如一个“谜团”（Enigma）——每个人都梦寐以求，却几乎无人能够真正达成，历史舞台上堆满了 360 项目失败的惨痛案例。其根本原因在于过去的底层技术无法支撑其实现路径：传统的 360 项目高度依赖数据集成，必须将散落在各处的数据物理搬运至中央存储库，这不仅技术复杂度极高、成本昂贵，还深受单体架构系统和巨额技术债务的拖累；更棘手的是，即便构建出了 360 视图，企业也难以将其快速下发至一线业务终端（如客服工作台），导致业务价值无法闭环。 ```mermaid flowchart LR accTitle: 传统 360 痛点与现代技术应对 subgraph Traditional ["传统 360 困境"] T1["高成本/高复杂度的物理数据搬运"] T2["单体遗留系统与沉重技术债"] T3["数据难以实时触达下游业务端"] end subgraph Modern ["现代技术破局"] M1["零复制技术 (Zero-Copy)<br/>数据就地保留与共享"] M2["可组合架构 (Composable Architecture)<br/>算力下沉至数据驻留地"] M3["Agentic AI (智能体)<br/>实时打通上下文并执行业务价值"] end T1 -.->|技术演进| M1 T2 -.->|技术演进| M2 T3 -.->|技术演进| M3 ``` 如今技术终于迎头赶上，彻底打破了上述瓶颈： - **集成难题**：通过**零复制技术（Zero-Copy Data Sharing）**，数据无需搬迁，在原始驻留地即可实现跨系统共享与访问。 - **业务价值传递**：快速演进的 **Agentic AI（智能体 AI）** 能够实时获取数据上下文并驱动下游动作，无需再耗时费力地在各系统间做大规模数据同步。 - **单体架构改造**：依托**可组合架构（Composable Architecture）**，企业可以将特定的应用能力直接部署到数据驻留的地方，实现“算力就地就近运行”。 --- ### [14:46](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=886s) 图模型架构与零复制重构 360 视图 观众进一步追问所谓“网络化互联数据设计”究竟是指传统关系型数据库的外键关联还是内嵌业务逻辑。Sriraj 澄清道，以 Reltio 为例，底层采用的是**图数据模型（Graph Model）**而非传统的关系型数据库。数据被拆解为三个核心维度： 1. **实体（Entities）**：如客户、产品、供应商； 2. **属性（Attributes）**：如地理位置、联系方式； 3. **交互（Interactions）**：如某客户发起的客服呼叫、某供应商在特定地点的货物交付。 将实体、属性与交互行为全部通过图谱进行交织关联，便构成了具备丰富上下文的知识网络。必须超越单纯依靠主外键（Foreign Keys）定义从属关系的传统关系型范式，进入以连接为核心的图论世界。 针对观众提出“面对积累了 15 年、散落在各数仓/数据湖中的海量孤岛数据，这种方式与传统 360 有何本质区别”的疑问，Sriraj 深入剖析道：传统 360 项目往往耗费长达一整年的时间仅仅用于“搬运数据”，并且还要艰难理清隐藏在数据里的隐性业务逻辑（例如在某些遗留系统中“0”并不代表数字零，而是带有特殊的业务含义）。而借助现代**零复制技术（Zero-Copy）**，企业无需搬迁数据，几分钟内即可在数据驻留地直接建立访问连接、解析隐式逻辑并在其上构建业务层，以极高的效率完成 360 上下文的构建。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=4s) 工业时代的核心驱动力是利用专有方法积累和控制实体资产；而在如今的智能时代，核心资产转变为知识资产与数据资产，关键不在于盲目囤积，而在于如何高效分发并将其转化为行动。 — Sriraj Rajaram > [02:30](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=150s) 数据是 Agentic AI 和 AI 自动化的氪石（Kryptonite）。 — Sriraj Rajaram > [03:32](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=212s) 谁能率先在无需人工介入（No Human-in-the-loop）的情况下规模化运行代理流程，谁就将重塑所在行业的竞争格局。 — Sriraj Rajaram > [03:32](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=212s) 数据统一（Data Unification）绝不等于简单的数据聚合（Data Aggregation）。 — Sriraj Rajaram > [05:02](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=302s) 像构建神经网络一样架构企业数据。 — Sriraj Rajaram > [06:51](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=411s) 关键不在于谁拥有的数据更多，而在于谁能更聪明、更实时地调度数据，并将其交付至业务处理的最边缘（Edge of processing）。 — Sriraj Rajaram > [09:55](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=595s) 让数据富含上下文关联（Context Rich），而非像‘文件柜’一样被动堆积，从而明确掌握实体之间的动态关联以支持行动。 — Sriraj Rajaram > [12:14](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=734s) 360 度全景视图宛如一个‘谜团’——每个人都梦寐以求，却几乎无人能够真正达成，历史舞台上堆满了 360 项目失败的惨痛案例。 — Sriraj Rajaram > [14:46](https://www.youtube.com/watch?v=L6iVGTxu9Lg&t=886s) 必须超越单纯依靠主外键定义从属关系的传统关系型范式，进入以连接为核心的图论世界。 — Sriraj Rajaram ## Mentioned Sriraj Rajaram, Reltio, Ai4, Agentic AI, Data Graph Technology（数据图谱技术）, Zero-Copy（零复制技术）, Composable Architecture（可组合架构）, 360 度全景视图（Customer 360）, 语义层（Semantic Layer）, 光速数据交付网络（Light-speeded Delivery Network） ## Video description Sriraj Rajaram, Director, Product Management, Reltio Abstracts The old rules of data no longer apply. To lead in the Age of Intelligence, enterprises must adopt new data strategies that move at the speed of AI— in real-time - and that are connected and built for continuous decision-making. In this session, Principal, Product Management Sriraj Rajaram shares the new rules of data that enterprises need to adopt in the Age of Intelligence, and the key capabilities and enterprise data foundation that CAIOs, CIOs and IT leaders need build across their organizations to be ready for the operational realities of AI at scale. Attend this session to learn how an agentic data fabric of unified, trusted data assets accelerates AI transformation & learn more about real-world strategies for success. [Source](https://www.youtube.com/watch?v=L6iVGTxu9Lg)

Source quotes

[0:04] Sriraj Rajaram: 工业时代的核心驱动力是利用专有方法积累和控制实体资产；而在如今的智能时代，核心资产转变为知识资产与数据资产，关键不在于盲目囤积，而在于如何高效分发并将其转化为行动。 [2:30] Sriraj Rajaram: 数据是 Agentic AI 和 AI 自动化的氪石（Kryptonite）。 [3:32] Sriraj Rajaram: 谁能率先在无需人工介入（No Human-in-the-loop）的情况下规模化运行代理流程，谁就将重塑所在行业的竞争格局。 [3:32] Sriraj Rajaram: 数据统一（Data Unification）绝不等于简单的数据聚合（Data Aggregation）。 [5:02] Sriraj Rajaram: 像构建神经网络一样架构企业数据。 [6:51] Sriraj Rajaram: 关键不在于谁拥有的数据更多，而在于谁能更聪明、更实时地调度数据，并将其交付至业务处理的最边缘（Edge of processing）。 [9:55] Sriraj Rajaram: 让数据富含上下文关联（Context Rich），而非像‘文件柜’一样被动堆积，从而明确掌握实体之间的动态关联以支持行动。 [12:14] Sriraj Rajaram: 360 度全景视图宛如一个‘谜团’——每个人都梦寐以求，却几乎无人能够真正达成，历史舞台上堆满了 360 项目失败的惨痛案例。 [14:46] Sriraj Rajaram: 必须超越单纯依靠主外键定义从属关系的传统关系型范式，进入以连接为核心的图论世界。

Source quotes

[0:00] Sriraj Rajaram: The old rules of data no longer apply. To lead in the Age of Intelligence, enterprises must adopt new data strategies that move at the speed of AI. [1:57] Sriraj Rajaram: Data is the kryptonite of agentic AI and AI automation. [1:57] Sriraj Rajaram: Organizations operating 50 applications today will soon manage 500, and those with 500 will see 5,000. [3:32] Sriraj Rajaram: The first organization capable of scaling agentic workflows without requiring human validation will fundamentally reshape competitive dynamics in its industry. [3:32] Sriraj Rajaram: Data unification is not the same as data aggregation. [5:02] Sriraj Rajaram: Storing petabytes of raw metrics delivers no business value unless relationships and interactions are modeled semantically. [6:51] Sriraj Rajaram: Competitive advantage in the Age of Intelligence does not stem from possessing the largest repository of data; it comes from moving data intelligently in real time to the operational edge where actions occur. [10:46] Sriraj Rajaram: 360 initiatives are like an enigma—everybody wants it, nobody can have it... history is filled with failures and dead bodies of 360 initiatives. [16:27] Sriraj Rajaram: zero didn't mean zero, it meant something else

Original transcript

[0:04] Good morning everyone. Uh welcome to [0:06] this cappuccino session uh you know on [0:08] the first day uh first session of the [0:10] last day but uh my name is Siraj as I [0:12] was introduced director of product here [0:14] at Railto. For those who may not know [0:16] Railto is a household name in data [0:18] management we are the industry leaders [0:20] in terms of data unification providing [0:22] 360deree views of data and such. More on [0:25] that later. uh but today we're going to [0:27] talk about a couple of insights we've [0:29] gleaned from the uh implementations [0:31] we've done around AI and agentic AI [0:33] implementations and how data plays into [0:36] that. So hopefully we can uh share some [0:38] of those insights with you today and [0:40] then uh continue the conversation later [0:42] on. So uh before we begin I just to set [0:45] the context we're going to talk about [0:47] two paradigms that we know right the [0:49] world we knew it which is also known as [0:52] the industrial age uh and the world as [0:54] we know it now which is the age of [0:56] intelligence. The industrial age was [0:59] built on accumulation of physical assets [1:01] and control using proprietary methods. [1:04] Uh and that was the key driving force in [1:06] the industrial age. Today however where [1:08] we are today is in the age of [1:09] intelligence which is completely [1:11] different. Uh it is shifting decisions [1:14] from humans to AI agents and unlike the [1:17] industrial age where we were trying to [1:19] accumulate physical assets here it is [1:21] about knowledge assets and data assets [1:23] and not just about accumulating them but [1:25] how we distribute and actionize them. Um [1:28] AI systems today as you have heard [1:30] throughout this conference are analyzing [1:32] vast amounts of information complex [1:34] amounts of information. They're not just [1:36] augmenting human decision-m but they're [1:39] actually replacing human decision-m in [1:41] some cases. Right? So this dual [1:43] capability of augmenting human judgment [1:46] while also enabling machine autonomy uh [1:50] leads to greater speed and efficiency [1:51] but it pre presents a certain amount of [1:54] challenges and we're going to talk about [1:55] that. you're already starting to see [1:57] some of these uh I'm sure in a plethora [2:00] of examples today or later on I mean [2:02] earlier on in this at this event you've [2:04] seen examples where for example agents [2:06] have uh you know augmented customer [2:09] service for example my flight got [2:11] rescheduled on my way in here so [2:14] autonomously an AI agent rebookked my [2:16] hotel rebook my car reservation and so [2:18] on and so forth you're also seeing that [2:19] in patient experience you know uh from [2:21] sending refill reminders to rescheduling [2:24] appointments or booking appointments [2:25] with your care provider. All of these [2:27] are being handled by agents. But with [2:30] all of this agentic decision making, [2:33] there is a level of risk involved and [2:36] organizations such as yourselves may not [2:38] be fully willing to trust the agent with [2:41] that decision-m process. This is why we [2:43] still have humans in the process. But [2:47] what we're seeing is that humans are in [2:48] the process because there is siloed data [2:52] or issues with data. Data is the [2:54] kryptonite of agentic AI and AI [2:57] automation. Uh why is that? Because you [3:00] know you have poor quality data across [3:02] your enterprise in a plethora of apps [3:04] you know and with agentic AI the app [3:06] explosion is only going to get worse. If [3:08] you had 50 apps today you're going to [3:09] have 500. If you have 500 today you're [3:11] going to get 5,000 because it's so easy [3:13] to create these apps. Uh and this is [3:15] what we call as the data debt. The gap [3:18] between what you're able to do with your [3:19] data and what it needs to deliver. So [3:22] that is the data debt and that gap is [3:24] only going to get wider as you see fast. [3:27] This is the exact time where you need to [3:29] be operating faster. This brings us to [3:32] the first rule of data intelligence that [3:35] you need to enable this AI uh [3:37] capabilities. The first rule is that you [3:40] need to build data trust as the [3:42] foundational layer for autonomous [3:44] systems in your enterprise. The [3:46] conversation in every boardroom with [3:48] every executive is how to scale AI for [3:51] impact. Right now there is a human in [3:53] the loop as I said earlier because you [3:55] don't trust your data. The human in the [3:56] loop is making sure that that data is [3:58] right, the action is right, the decision [4:00] is right. The first organization that [4:03] can scale agentic processes without the [4:06] need for a human in the loop is the one [4:08] that's going to reshape the competitive [4:10] dynamics in that industry. And the key [4:12] to that is, as we said, building data [4:14] trust. This is where my organization [4:17] Railio comes in. We specialize in this. [4:19] We are the leaders in this. We we [4:22] specialize in building a unified trusted [4:24] data layer that com that compiles [4:27] profiles such as customers, products, [4:30] suppliers and so on and so forth from [4:32] across your organization and makes it [4:34] available to downstream and upstream [4:36] systems. I want to be very clear data [4:39] unification is not the same as data [4:42] aggregation. Right? That's a very [4:44] different concept. We're not just [4:45] compiling information. We're curating [4:47] information within Realio through [4:49] governed processes and through [4:50] algorithms that actually determine how [4:53] to unify and keep data together. So now [4:56] that you've unified your data, if you [4:58] follow the first rule, what's next? [5:02] The second rule of of uh of data [5:05] intelligence in the in the age of AI is [5:07] to architect your data like a neural [5:08] network. U in the past you know a lot of [5:13] organizations in the industrial age as I [5:14] referred to focused on data accumulation [5:17] building large warehouses building data [5:20] links uh so it was all about hey how [5:23] much information do I have do I have the [5:25] the lowest level of information uh to [5:27] take there was a an industrial [5:29] manufacturer uh were built um some an a [5:33] platform to accumulate various metrics [5:36] from IoT devices that they deployed in [5:38] the in the marketplace [5:40] But they didn't action on that insight. [5:42] So that data accumulation that they had [5:44] in these vast warehouses was not of much [5:47] use and and that manufacturer isn't [5:49] doing so well today. The difference lies [5:51] between not accumulating data but making [5:54] the data context rich like do you know [5:57] what exact what is the level of [5:58] information the quality of information [6:00] that you have and that's why you need [6:02] the data to be context rich. This is [6:04] where Relio within Relio's platform we [6:07] follow what we call as a data graph [6:09] technology. The data graph not only [6:11] accumulates various profile information [6:14] like who's your customer, who's your [6:15] supplier and what products do you have, [6:17] but it also links interactions between [6:20] all of them. It links the relationships [6:22] between all of them. So you're able to [6:24] get a very rich contextual view of your [6:27] data and the attributes that are [6:28] associated with it. In effect, the data [6:31] graph becomes the semantic layer for the [6:34] data across your enterprise. So AI [6:36] agents that you enable have a complete [6:39] context that they require to make [6:41] autonomous decisions for your business. [6:43] This is the only way agentic processes [6:46] can scale across your enterprise. [6:51] Now competitive advantage comes from [6:54] converting data into action faster than [6:57] your competitors. It's not about who has [6:59] more data. It's about who moves data [7:02] smarter and in real time and delivers it [7:04] to the edge of your processing where [7:06] it's needed. For example, the other day [7:09] I was shopping online and I was on the [7:11] fence about buying a particular product. [7:13] Uh I was dillydallying, you know, in the [7:15] shopping cart. I was I was in the [7:17] shopping cart. I was out of the shopping [7:18] cart. I was looking at other products [7:20] that was recommended to me. So on and so [7:22] forth. At that instant, the e-commerce [7:25] site that I was browsing on presented an [7:27] option to me saying, "Hey, here I'm [7:29] discounting it by 10% or 20%." It made [7:32] the decision for me, right? Like, while [7:34] it it it recognized that I'm interested [7:36] in this product, but I'm delaying about [7:38] it. Here is something to move me past [7:40] that that inertia into actually [7:42] purchasing the product. a similar [7:44] experience I had with on another [7:45] e-commerce site where I was similarly on [7:48] the edge about buying it. But then I [7:49] left the website because I didn't get [7:51] any there was nothing that that [7:52] interacted with me to say, "Hey, you [7:54] know, here's something to help you make [7:55] that decision." Now, a day later, I get [7:57] an email saying, "Here's hey, we noticed [7:59] that you left something in your shopping [8:00] cart. Here is something to help you, you [8:03] know, if here's a discount coupon." But [8:04] it's too late. I'm already past that [8:06] moment. I'm already past the purchasing [8:08] moment. So this is where data velocity [8:11] not volume is critical. Can you get the [8:14] data that is actionable to the edge of [8:17] that processing? Then you can make [8:19] decisions happen faster. [8:21] This is where with RTO we have something [8:23] what is called as light speeded delivery [8:25] network. A lot of customers lot of uh [8:29] you know products that you use on a [8:30] day-to-day basis. A lot of companies [8:32] that you interact with are using this [8:34] particular technology to deliver their [8:37] organizational trusted data assets at [8:40] very low latency to the edge of [8:42] processing. So whether it is your mobile [8:44] app or whether it is your enterprise [8:46] application or your website or your [8:48] e-commerce engines those those [8:50] applications are getting trusted data [8:53] from railio in submillisecond [8:56] performance. This is what we call as [8:58] light speeded data delivery network. [9:02] And the last rule for today that I'm [9:03] going to leave you with is creating a [9:06] comp cohesive enterprisewide [9:09] investment approach for managing your [9:11] data assets. A lot of you are navigating [9:14] the AI challenge that is being presented [9:17] with you. But we aren't even done with [9:18] the digital transformation challenge. I [9:20] mean it was just a few years ago that we [9:22] started with social, mobile, analytics [9:24] and cloud. And while we are still going [9:26] on that journey, here is the next set of [9:28] challenges, AI. And more challenges are [9:30] going to come our way. Today it's [9:32] yesterday it was generative AI, today it [9:34] is agentic AI, God knows what it is [9:35] tomorrow, but all of these are coming [9:38] your way sooner than later. So having an [9:40] enterprisewide approach to managing your [9:43] data assets is going to ensure that [9:46] those data assets don't need to be [9:48] repurposed. I mean, don't need to be [9:50] remade every time a new initiative comes [9:52] along. investing in this in this logic [9:55] of having enterprisewide initiatives and [9:57] taking a data products approach. So you [9:59] know ensuring that your customer data is [10:02] a data product or your supplier data is [10:03] a data product that can then be consumed [10:06] by different processes and systems [10:08] across your organization is going to [10:10] actually help yield the best results for [10:12] you. uh even if you start with one use [10:14] case uh it's important to consider how [10:17] that same data asset can unlock outcomes [10:20] and benefits impact elsewhere in the [10:22] organization right so this is for [10:25] example how Relto is unlocking the data [10:28] products whether it is marketing sales [10:31] finance operations and back office we've [10:34] unlocked various use cases using the [10:36] data products approach for a number of [10:38] logos that you see there and many of [10:40] these are are familiar logos common [10:42] logos that you you interact with on a [10:44] day-to-day basis. [10:46] So, um you know, just to summarize the [10:50] four rules of data intelligence, first [10:53] is to build data trust. Without trust, [10:56] you know, you're going to have a human [10:57] in the loop all the time. So, the first [10:59] company or the first organization that [11:02] can minimize the human intervention is [11:04] the one that's going to change the [11:05] competitive dynamics of that industry. [11:08] The second rule is to architect data [11:10] like a neural network. Make it context [11:13] rich, not accumulate not not like a [11:15] filing cabinet, you know, where you're [11:17] just storing data, but actually make the [11:19] data context rich. So you can action on [11:21] it. You know what's happening between [11:23] various entities of data that you have. [11:26] The third rule is to compete on velocity [11:29] not volume. Right? So if you can deliver [11:33] your data to the edge of processing, [11:35] more actions are going to happen at that [11:37] edge with that data. The last rule is to [11:40] create a cohesive enterprisewide [11:43] investment approach where you're [11:44] treating data as data products. There [11:46] are more rules than this. You know, we [11:48] we we've got more rules, but time only [11:50] allows me to discuss so much with you. [11:51] If you would like to learn more about [11:53] these rules and any other approaches [11:54] that we have towards data intelligence, [11:57] please visit us at this booth or feel [11:58] free to scan the QR code and download [12:00] our white paper on the the 10 rules [12:04] actually we have around data [12:05] intelligence. So I'm happy to pause here [12:08] and we've got about eight or nine [12:10] minutes left. Happy to take any [12:11] questions that the audience may have. Um [12:14] I'll just repeat the question uh for [12:15] others. Uh the question is that uh in [12:18] the organization there's the challenge [12:20] about collecting this data and creating [12:22] this context rich information and what [12:23] are the best practices or approaches to [12:25] do that. Um so a common approach to do [12:28] this is what you may have heard of as [12:30] 360 initiatives right like 360deree view [12:32] of data whether it is customer 360 [12:34] product 360 supplier 360 but 360 [12:37] initiatives are like an enigma everybody [12:39] wants it nobody can have it right and [12:41] everybody's tried it and it's just not [12:43] succeeded it's there's the history is is [12:46] filled with failures and dead bodies of [12:48] 360 initiatives and the reason for that [12:50] is that uh technology hasn't caught up [12:53] to that so far the reason 360 [12:55] initiatives first of all require a lot [12:56] of integration. You need to bring data [12:58] from various parts of your organization [13:00] to a central repository and then you [13:02] know create that 360deree view. [13:04] Integration by itself is very expensive [13:06] and it it's it's very technologically [13:08] complex. Right? Second is um um there's [13:12] a lot of uh monolithic applications and [13:15] you know there's they have huge amounts [13:16] of technology debt. So that's again a [13:19] big challenge that you have. And then [13:21] the the last problem that I see is that [13:24] deriving business value once you built [13:26] all this 360deree of view of data [13:29] deriving business value is even more [13:30] complex because once you create the 360 [13:33] view or this context view how do you [13:35] deliver it downstream you know where the [13:37] actual application like you know your [13:39] customer and you know how your customer [13:40] is behaving with you but that data is no [13:43] use until you deliver it to the service [13:44] application where where that customer is [13:46] interacting technology has finally [13:48] caught up to it. How how is that? Uh [13:50] let's talk about the first problem of [13:52] integration. Now you have zero copy [13:54] approaches. You can leave data where it [13:55] is and you have zero copy data sharing. [13:58] So technology finally allowing you to [13:59] overcome the integration challenge. The [14:01] second challenge about delivering [14:03] business value that agentic AI is now [14:06] you know scaling so fast that uh you [14:09] know you can you don't have to [14:12] necessarily move data from the 360 [14:14] application to a downstream application. [14:16] Agent AI can do that real time for you. [14:18] And the last thing about monolithic [14:20] structures of existing applications now [14:22] you have what is called as composable [14:23] architecture where you take certain [14:25] components like within relio you can [14:27] take a certain component and put it [14:28] where your data is. So you have the [14:30] power of the application at the data [14:32] residence not necessarily having to move [14:34] data to the application. So these are [14:36] you know some approaches obviously every [14:38] every challenge is a little bit nuanced [14:40] but generally now technology has caught [14:42] up to solve this challenge. [14:46] >> Sure. So when you say design your data [14:49] as an interconnected network, what does [14:51] that entail? Is that like are we talking [14:53] relational databases in a bunch of [14:56] tables like is that business logic [14:58] inside the data or [15:01] >> um yeah I mean so within within Railto [15:04] we store data as a graph right so it's [15:07] uh you have if you look at the elements [15:09] of data you have entities customer [15:12] product supplier so on and so forth and [15:14] then you have attributes associated with [15:16] them for example their location their [15:18] contact information and then you have [15:20] interactions associated with them this [15:22] customer made a service call or this [15:25] supplier you know made a delivery at so [15:27] and so location. Now when you interlink [15:29] all of this using a graph that is what [15:31] is the context rich graph that we're [15:33] talking about. [15:35] Okay. So, so think about think about we [15:38] we have to move beyond the relational [15:40] world where we have foreign keys and [15:42] identifiers into a graph model where [15:46] we're talking about connections between [15:47] data assets and not just you know this [15:49] one is is the parent of that one [15:54] more you could you could yeah in in a in [15:56] a in a in a programming parlance yes [15:58] object-oriented but it's more graph [16:00] theory [16:03] >> so in an environment that has you know [16:05] 15 years of data everywhere. How is that [16:08] different than a 360 initiative? [16:10] >> Um, so you have data everywhere. [16:12] >> Correct. So you have you have data [16:14] assets that are in silos or in in [16:17] multiple warehouses or lakehouse [16:20] 360s would be let's bring it all [16:22] together. But correct [16:23] >> I just heard you said earlier about [16:24] those are work and they kind of dive. [16:27] Yeah. So how is this different? [16:29] >> Sure. Sure. I mean so the what I was [16:32] alluding to earlier is that the [16:33] challenge has always been bringing all [16:35] that data together right you have so [16:37] much data the volume of data and then [16:39] there's rules hidden within the data for [16:40] example I worked with a customer where [16:42] zero didn't mean zero it meant something [16:44] else there was a business logic behind [16:46] that so now what you don't what you what [16:48] we're doing is there is a the thing of [16:51] zero copy where you don't have to [16:53] necessarily move the data from one [16:55] location to another we can access data [16:57] in residence through uh through a [17:00] technique and then we can understand [17:01] that and layer business logic on that. [17:03] So we can construct 360 much faster. I [17:07] mean you would spend a year just moving [17:08] data. Now you can spend minutes [17:11] accessing data where it is and you can [17:13] get that consolidated wave with zero [17:15] copy technologies. And within relative [17:17] we have zero copy technologies there.