NFL Overhauls Operating Model to Scale Generative AI Across 32 Clubs

Fireside Chat with Chi Ogbuehi, National Football League (NFL) and Celeny Da Silva, MAKERS AI — Ai4

Ai4

Layering artificial intelligence directly onto legacy marketing workflows fails without standardized metadata governance and disciplined operating models to prevent system drift.

AI-assisted summary of Fireside Chat with Chi Ogbuehi, National Football League (NFL) and Celeny Da Silva, MAKERS AI — original source. Check the original for full context.

Source published: 2026-08-22

Detailed notes

## Background Published on YouTube by the Ai4 conference channel on August 22, 2026, this video documents an enterprise fireside chat between Celeny Da Silva, Founder and Chief Executive of MAKERS AI, and Chi Ogbuehi, Vice President of Marketing Technology and Consumer Products at the National Football League (NFL). The session explores how the NFL operationalizes artificial intelligence to scale fan engagement across global markets.

## TL;DR Chi Ogbuehi outlines the NFL's overarching martech strategy, emphasizing that artificial intelligence cannot scale enterprise-wide without first modernizing organizational operating models and underlying workflows. Rather than automating broken processes, the league established a unified intake 'front door' and a universal metadata taxonomy across 32 clubs and 22 countries. Ogbuehi details the essential role of human-in-the-loop governance—highlighting how human reviewers caught quality degradation during the live 2026 NFL Draft—and explains how the league drives 'mass intimacy' for fans. Rather than fixating on token or compute costs, the NFL measures AI value through production cycle times, throughput scalability, and first-time-right asset quality.

## Atmosphere & style The conversation carries a polished, pragmatic, and candid tone, striking a balance between high-level executive strategy and unvarnished operational realities. Da Silva guides the discussion with an inquisitive, collegiate energy, while Ogbuehi provides grounded, experience-tested insights with refreshing candor about the limitations of raw AI models and past deployment hitches. It feels like an authentic insider briefing on enterprise transformation, replacing speculative generative AI hype with actionable architectural discipline.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=DnzndoCPAes&t=4s) The NFL's massive global footprint of 32 clubs and over 200 million fans across 22 countries led to the foundational enterprise principle: 'AI doesn't scale, operating models do.' - [01:50](https://www.youtube.com/watch?v=DnzndoCPAes&t=110s) An 'always-on' year-round calendar and surging audience diversity pushed legacy creative production pipelines past capacity, driving the pragmatic need for AI automation. - [06:10](https://www.youtube.com/watch?v=DnzndoCPAes&t=370s) A disciplined, phased deployment pipeline (Concept to Pilot MVP to Real-World Execution) ensured the NFL only scaled workflows that demonstrated measurable viability. - [06:10](https://www.youtube.com/watch?v=DnzndoCPAes&t=370s) Scaling AI required a tailored talent strategy that blended creative designers with technologist practitioners to engineer modular, reusable visual assets. - [08:08](https://www.youtube.com/watch?v=DnzndoCPAes&t=488s) To avoid accelerating broken workflows, the NFL instituted a centralized 'front door' intake mechanism to govern all incoming campaign and content requests. - [08:08](https://www.youtube.com/watch?v=DnzndoCPAes&t=488s) A standardized enterprise metadata taxonomy established central consistency across the league while maintaining local autonomy for individual clubs and international regions. - [10:30](https://www.youtube.com/watch?v=DnzndoCPAes&t=630s) Human-in-the-loop review remains non-negotiable for enterprise AI governance, serving as an essential quality safety net while systems learn and mature. - [10:30](https://www.youtube.com/watch?v=DnzndoCPAes&t=630s) During the 2026 NFL Draft live test, AI output quality degraded on Day 2 after a successful Day 1, demonstrating why real-time generative pipelines require active human oversight before public distribution. - [12:34](https://www.youtube.com/watch?v=DnzndoCPAes&t=754s) The NFL leverages AI to deliver 'mass intimacy'—providing timely, deeply personalized fan interactions at scale based on the premise that 'everyone footballs differently.' - [13:53](https://www.youtube.com/watch?v=DnzndoCPAes&t=833s) AI ROI is evaluated through outcome-oriented metrics such as production cycle time, asset throughput, and 'first time right' quality rather than compute or token costs. - [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) Driving enterprise AI adoption requires embedding capabilities seamlessly into existing workflows rather than forcing teams to learn isolated standalone tools.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=DnzndoCPAes&t=4s) Welcoming Chi Ogbuehi: The NFL's Massive Scale and the Operating Model Thesis

Opening the Ai4 session, Celeny Da Silva, Founder and CEO of MAKERS AI, frames football as far more than just athletic competition—it is an identity, a community, and an intense emotional bond. For the National Football League (NFL), supporting that relationship requires managing extraordinary operational breadth: a single league governing 32 individual clubs, engaging over 200 million fans across 22 countries. Every single week of the season, the organization faces the staggering task of transforming raw, on-field moments into hyper-relevant, localized content tailored for diverse international markets. Da Silva poses the central dilemma confronting enterprise technology leaders: what does it take to move artificial intelligence past mere experimentation into a trusted, enterprise-grade capability? She introduces the guiding philosophy articulated by the league itself: "AI doesn't scale, operating models do."

Da Silva welcomes Chi Ogbuehi, Vice President of Marketing Technology and Consumer Products at the NFL, greeting him warmly and noting with a laugh that he traveled all the way from Greece to join the discussion. Ogbuehi chuckles, clarifying that while he was indeed abroad, he returned over the weekend just in time for the conference. Da Silva quickly pivots into the day-to-day realities of league operations, asking Ogbuehi to pull back the curtain on what actually occurs across marketing and content teams between game days—specifically on a typical Tuesday during the active season—and where AI currently fits into that machinery.

- **League footprint:** 1 overarching league, 32 franchise clubs, over 200 million fans, spanning 22 countries. - **Core philosophy:** "AI doesn't scale, operating models do."

---

[01:50](https://www.youtube.com/watch?v=DnzndoCPAes&t=110s) The "Always-On" Fan Engine: From Global Diversity to Broken Workflows

Ogbuehi begins by clarifying his team's purview: leading marketing technology across emerging tech, operating architectures, and the digital platforms that drive fan engagement both domestically and globally. While casual viewers only see the three-hour spectacle on the field, an intricate, continuous operation hums behind the scenes to deliver varied content across disparate channels, languages, and time zones. As the NFL’s international presence has expanded into 22 countries, its audience profile has diversified dramatically, seeing explosive growth among female audiences and grassroots participants through initiatives like flag football. However, serving a deeply global audience brings immense operational friction, demanding varied formats, multi-language translations, and nuanced cultural localization, all while fans demand immediate, real-time personalization.

```mermaid flowchart TD accTitle: NFL Content Supply Chain and Operational Model subgraph Intake["Intake & Governance"] Req["Content Requests"] --> FD["Unified 'Front Door'"] FD --> Meta["Standardized Metadata Framework"] end

subgraph Generation["Production & Processing"] Meta --> AI["AI Content Generation & Automation"] AI --> RT["Real-Time Feeds (e.g. Draft / Live Games)"] end

subgraph Governance["Human-in-the-Loop Review"] RT --> HITL{"Quality & Brand Review"} HITL -->|"Approved"| Dist["Multi-Channel Distribution (Clubs, International, Social)"] HITL -->|"Degraded / Flagged"| Fix["Manual Remediation"] Fix --> Dist end ```

The sheer velocity and volume of fan expectations eventually pushed legacy production setups past their breaking point. Ogbuehi emphasizes that their pivot toward AI was never about chasing a flashy technical novelty or inserting an algorithm for its own sake; it was a pragmatic response to business demand overwhelming traditional creative capacity. He explains that fan engagement is fundamentally "always on." The narrative does not kick off on kickoff Sunday; it starts months earlier with the off-season draft, schedule releases, and training camp storylines. Once the regular season begins, production shifts into an unforgiving, real-time pace where timely reactions are mandatory. Facing this unrelenting velocity, the NFL recognized that traditional creative pipelines could no longer keep pace without rethinking the operational backbone.

- **Global reach:** Content localized across 22 countries with cultural adaptations. - **Audience demographics:** Surging engagement across women and international flag football communities. - **Year-round calendar:** Off-season milestones (NFL Draft, schedule drop, training camps) transition seamlessly into high-frequency, in-season real-time content.

---

[06:10](https://www.youtube.com/watch?v=DnzndoCPAes&t=370s) Why AI Doesn't Scale Without an Operating Model: MVP Lessons and People Strategy

Addressing his mantra that "AI doesn't scale, operating models do," Ogbuehi details the hard-earned lessons behind the phrase. Rather than leaping recklessly from a high-level concept directly to full enterprise-wide rollout, the NFL executed a measured, phased methodology: concept validation, pilot MVP (Minimum Viable Product), and controlled real-world deployment. Crucially, the organization did not attempt to scale every experiment; they deliberately scaled only what proved viable. The MVP phase served as an unvarnished audit, highlighting not just technical friction, but exposing gaps across people, processes, and technology.

On the organizational side, the technology team realized that tooling alone achieves nothing without internal alignment and cultural buy-in. They actively recruited champions across departments, empowered power users, and introduced a hybrid profile of talent—pairing creative designers with technologist practitioners capable of decomposing visual assets into flexible, modular content that could be dynamically adapted, reused, and scaled across campaigns.

> "AI doesn't scale, operating models do."

- **Deployment framework:** Structured pipeline progressing from Concept $\rightarrow$ Pilot MVP $\rightarrow$ Real-World Execution. - **Selective scaling:** Only operationalized workflows that demonstrated measurable efficiency and accuracy during the MVP phase. - **Talent strategy:** Embedded cross-functional champions and blended creative-technologist roles to build modular, re-usable design assets.

---

[08:08](https://www.youtube.com/watch?v=DnzndoCPAes&t=488s) Rebuilding the Foundation: The Intake Front Door and Unified Metadata Strategy

When Da Silva asks how AI reshaped the creative supply chain from initial concept to execution, Ogbuehi underscores their most critical architectural realization: teams cannot simply bolt modern AI on top of fragmented, broken workflows. Attempting to automate undisciplined processes only accelerates failure. The NFL had to take a deliberate step back to establish a clean operational foundation, defining exact standards for how marketing teams interact with the tech stack day to day.

To establish governance and eliminate ad-hoc fragmentation, the team instituted an operational "front door"—a centralized intake mechanism through which all content, campaign, and asset requests must flow. This structural gateway enforced repeatable patterns, consistency, and compliance before any automation occurred. Simultaneously, the NFL was forced to completely overhaul its metadata strategy. Scaling content across the central league office, 32 distinct club franchises, and diverse international territories proved impossible without a shared, universal taxonomy. The league had to carefully balance centralization and autonomy, establishing standardized metadata rules to maintain system consistency while allowing clubs the federation and localized flexibility needed to preserve their authentic local voice.

- **Intake mechanism:** Implemented a single "front door" request entry point to govern workflows, standardize inputs, and ensure repeatable output patterns. - **Foundational prerequisite:** Refused to automate flawed processes; established an enterprise-wide metadata taxonomy first. - **Federated balance:** Standardized data architecture at the league level while granting local autonomy for clubs and international regions to safeguard authentic tone.

---

[10:30](https://www.youtube.com/watch?v=DnzndoCPAes&t=630s) Trust, Governance, and Human-in-the-Loop: The 2026 NFL Draft Test Case

Turning to the question of institutional trust, Da Silva asks what benchmarks must be reached before an enterprise can confidently trust an automated capability at scale. Ogbuehi explains that trust is earned incrementally as models learn, making human-in-the-loop oversight non-negotiable across the NFL's production chain. Every AI-generated asset undergoes rigorous human validation before being distributed to the public. However, Ogbuehi cautions that oversight must be engineered strategically so that reviewers act as quality safeguards rather than operational bottlenecks.

To illustrate this dynamic, Ogbuehi shares a real-world example from the 2026 NFL Draft, where the league hooked generative AI pipelines into live, real-time data feeds for the very first time. As draft picks were announced from the podium, the AI engine instantly generated tailored social graphics and copy for immediate review and publication by the social team. During round one, the system performed flawlessly with high-fidelity output. But by day two of the draft, the content quality began noticeably degrading. Because human reviewers were actively embedded in the loop, the team caught the anomalies and remediated the assets prior to broadcast, averting public-facing errors. Ogbuehi concludes that while models cannot be blindly trusted out of the box, disciplined human-in-the-loop governance provides the safety net required while systems learn and mature.

- **Live trial:** First end-to-end integration of generative AI with real-time sports data feeds executed during the 2026 NFL Draft. - **Failure point caught:** Day 1 output produced high-quality content; by Day 2, automated output experienced noticeable quality degradation. - **Governance principle:** Maintained human reviewers to catch and fix pipeline drift before public distribution, avoiding operational bottlenecks while safeguarding brand trust.

[12:34](https://www.youtube.com/watch?v=DnzndoCPAes&t=754s) Achieving "Mass Intimacy" for Unique Fan Bases at Scale

When asked how fans perceive the difference in how the NFL communicates with them today, Chi Ogbuehi frames the effort around the organization's overarching mission: modern sports fans expect interactions that are timely, relevant, and deeply personalized. The core challenge for the league is delivering that level of personalization at immense scale. Ogbuehi articulates this ambition through the concept of "mass intimacy"—the capability to interact with fans in ways that feel custom-tailored to their personal affinity, acknowledging that "everyone footballs differently." Because individual fans connect with the sport, teams, and players across varied touchpoints, AI enables the league to simultaneously serve millions while preserving that distinct, one-on-one personal feel.

Celeny Da Silva reflects on this idea, emphasizing that the greatest power of sports and the NFL is enabling human connection and shared love for the game. However, building out this degree of bespoke global personalization requires significant technical and organizational investment, prompting the question of how the NFL measures tangible business returns.

```mermaid flowchart LR accTitle: NFL Mass Intimacy Engine subgraph FanUnderstanding["Fan Understanding"] A["Diverse Fan Base 'Everyone footballs differently'"] end subgraph AIPlatform["AI & MarTech Platform"] B["Mass Intimacy Engine Personalization at Scale"] end subgraph Outcomes["Fan Experience"] C["Timely Delivery"] D["Relevant Content"] E["Deeply Personal Interactions"] end A --> B B --> C B --> D B --> E ```

[13:53](https://www.youtube.com/watch?v=DnzndoCPAes&t=833s) Measuring Business Value Beyond Cost and Token Credits

When evaluating the return on investment for AI, Ogbuehi explains that the NFL focuses squarely on outcomes and direct problem-solving rather than getting bogged down in micro-level consumption costs like compute credits and token pricing. The primary business metrics the NFL tracks include:

- **Production Cycle Time:** Tracking the total latency from the moment a content or campaign request is submitted until the finished asset reaches the fan. - **Throughput & Scalability:** Assessing whether the martech stack can generate sufficient variations and asset volumes to satisfy diverse global fan segments. - **Asset Quality & "First Time Right":** Measuring how frequently AI-generated or assisted deliverables meet production standards without requiring repetitive manual corrections.

While financial considerations naturally shape broader capital allocation decisions, Ogbuehi emphasizes that real business value drivers—speed, scale, quality, and fan base expansion across diverse languages and international markets—must guide whether and where the NFL continues to invest.

[15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) Core Principles for Operationalizing AI Across MarTech

Closing the discussion, Da Silva asks for the top principles required to successfully operationalize AI across a large-scale marketing technology ecosystem. Ogbuehi highlights two foundational requirements:

1. **Operating Model Alignment:** The organizational operating model must harmonize people, processes, and technology. As Ogbuehi highlights, AI does not act as a standalone solution; rather, "once you do that, AI becomes an accelerator." 2. **Designing for Adoption Up Front:** Organizations must actively minimize barriers to entry. Instead of forcing teams to adopt isolated standalone applications, the NFL redesigned end-to-end workflows so AI capabilities integrate seamlessly into existing day-to-day operations, ensuring employees do not feel like they have to "learn yet another tool."

Da Silva wraps up the fireside chat by underscoring Ogbuehi's central thesis on deploying enterprise AI at the NFL: technology alone is insufficient—"AI doesn't scale. Operating models do."

## Notable quotes > [00:04](https://www.youtube.com/watch?v=DnzndoCPAes&t=4s) AI doesn't scale, operating models do. — Celeny Da Silva

> [06:10](https://www.youtube.com/watch?v=DnzndoCPAes&t=370s) AI doesn't scale, operating models do. — Chi Ogbuehi

> [12:34](https://www.youtube.com/watch?v=DnzndoCPAes&t=754s) everyone footballs differently — Chi Ogbuehi

> [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) once you do that, AI becomes an accelerator. — Chi Ogbuehi

> [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) learn yet another tool. — Chi Ogbuehi

> [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) AI doesn't scale. Operating models do. — Celeny Da Silva

## Mentioned Chi Ogbuehi, Celeny Da Silva, National Football League (NFL), MAKERS AI, Ai4, NFL Draft

## Video description Celeny Da Silva - Founder & Chief Executive, MAKERS AI Chi Ogbuehi - Vice President, Marketing Technology and Consumer Products, National Football League (NFL)

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

中文简报

NFL重构运营模式，在32支球队规模化推广生成式AI 若缺乏标准化的元数据治理和规范的运营模式来防止系统漂移，直接将人工智能套用在传统营销工作流上将难以奏效。 Chi Ogbuehi概述了NFL的整体营销技术（martech）战略，强调若不首先推进组织运营模式和底层工作流的现代化，人工智能就无法在整个企业范围内实现规模化应用。该联盟并未盲目自动化存在缺陷的流程，而是在32支球队和22个国家之间建立了统一的需求接入“前门”（front door）以及通用的元数据分类体系。Ogbuehi详细阐述了人机协同治理的关键作用——重点提到人工审核员如何在2026年NFL选秀直播期间及时发现画质下降问题——并解释了联盟如何为球迷构建“大规模亲近感”。NFL并未单纯局限于Token或算力成本，而是通过制作周期时间、吞吐量扩展能力以及资产一次性合格率来评估AI的价值。 NFL覆盖22个国家、32支球队和超2亿名球迷的庞大全球版图，催生了核心企业原则：“扩展的不是AI，而是运营模式。” (0:04) 全年无休的“全时在线”日程安排以及受众群体日益增长的多样性，使传统的创意制作流程超出负荷，从而推动了对AI自动化的务实需求。 (1:50) 规范且分阶段的部署流程（从概念构思到试点MVP，再到实际落地执行）确保NFL仅推广已被证实具备可衡量可行性的工作流。 (6:10) 实现AI规模化需要量身定制的人才战略，将创意设计师与技术实践人员紧密结合，从而打造模块化、可复用的视觉资产。 (6:10) 为避免加速运转存在缺陷的工作流，NFL设立了集中的“前门”准入机制，用于统筹管理所有传入的营销活动与内容需求。 (8:08)

Reading summary

Chi Ogbuehi, Vice President of Marketing Technology and Consumer Products at the National Football League (NFL), argues that artificial intelligence cannot scale enterprise operations on its own; rather, successful deployment requires redesigning underlying operating models, workflows, and governance. Operational Bottlenecks in the Always-On Fan Calendar (1:50) The NFL oversees 32 franchise clubs and interacts with over 200 million fans across 22 countries. The league's audience has diversified significantly with growth among female fans and grassroots participants in flag football, compounding the operational challenge of delivering localized, multi-language, and culturally adapted content. Because fan engagement operates as an always-on cycle—spanning off-season events like the draft, schedule releases, and training camps directly into live regular season game days—the velocity and volume of required content overwhelmed the league's legacy creative pipelines. Centralized Governance and Foundational Metadata Taxonomy (8:08) To prevent modern AI from merely automating broken, fragmented processes, the NFL established a structured operational baseline. The league instituted an intake "front door" mechanism through which all content and campaign requests must flow, establishing repeatable patterns and compliance standards before introducing automation. Human-in-the-Loop Safeguards: The 2026 Draft Case (10:30) Because model trust is built incrementally, the NFL requires human-in-the-loop validation for all AI-generated assets prior to public distribution. During the 2026 NFL Draft, the league linked generative AI pipelines to real-time sports data to produce instant social media graphics as draft picks were announced from the podium. While the system operated with high fidelity during round one, asset quality deteriorated noticeably on day two; embedded human reviewers detected the degradation and corrected the materials before broadcast. Personalization at Scale and Business Outcome Metrics (12:34) The NFL aims for "mass intimacy," defined by Ogbuehi as delivering personalized, timely interactions at scale under the premise that "everyone footballs differently." Rather than evaluating AI through micro-level compute costs or token credits, the league tracks operational performance indicators: production cycle time from intake to delivery, asset throughput and scalability across global territories, and "first-time-right" asset quality that avoids repetitive manual intervention. Designing Workflows for Organic Employee Adoption (6:10) To achieve long-term adoption, the NFL pursued a phased progression from concept validation to pilot minimum viable products (MVPs), scaling only those workflows that demonstrated measurable efficiency and accuracy. Cross-functional departmental champions and hybrid creative-technologist roles were introduced to assemble modular design assets. Ogbuehi emphasizes embedding AI capabilities directly into existing daily work patterns so that staff are not forced to manage isolated standalone applications. The NFL's experience highlights that enterprise-level AI functions primarily as an operational accelerator rather than an independent solution, remaining reliant on human oversight, clean input architecture, and disciplined organizational processes.

精炼阅读摘要

在Ai4大会上，美国国家橄榄球联盟（NFL）营销技术与消费产品副总裁Chi Ogbuehi与MAKERS AI创始人Celeny Da Silva提出核心论点：人工智能本身无法实现规模化，唯有运营模型才能实现规模化；企业若要将AI由概念验证转化为可靠的企业级能力，必须实现人、流程与技术的系统性协调，而非在破碎的工作流上强行套用算法工具。规模化受众诉求与“大众亲密度”引擎 (1:50) NFL必须管理涵盖1个中心联盟、32家俱乐部特许经营权以及遍布22个国家的逾2亿名球迷的庞大运营版图。随着国际版图扩张、女性球迷激增以及腰旗橄榄球等基层运动的发展，球迷受众群呈现出高度多样化的特征。Ogbuehi指出，由于“每个球迷体验橄榄球的方式各不相同”，联盟的运营并非局限于比赛日的三小时赛程，而是涵盖选秀、赛程发布、训练营等节点的“全天候”叙事流程。一旦进入正赛周期，内容生产更是进入高频、实时的节奏，需要应对跨语种翻译、文化本土化及个性化交互的多重压力。面对激增的内容需求，传统的内容创作流水线逼近极限。Ogbuehi强调，联盟采用AI并非出于技术猎奇，而是解决业务需求超出传统产能瓶颈的务实手段。联盟致力于通过AI构建“大众亲密度”（Mass Intimacy）引擎，使内容在面向数以亿计受众进行大规模分发的同时，仍能保持如同量身定制般的个人连接感与高相关性。流程治理底座：统一入口与联邦元数据 (8:08) 针对AI技术在创意供应链中的落地，Ogbuehi指出核心痛点：企业绝不能将现代AI直接套用在碎片化或存在缺陷的工作流程之上，自动化紊乱的流程只会加速系统崩溃。为此，NFL首先对日常营销团队与技术栈的交互方式进行了基础重构，设立了中央运营“前门”（Front Door）作为集中的需求摄入端口，强制要求所有内容、营销活动及素材申请通过统一通道接入，在自动化介入前确保输入模式与合规标准的统一。与此同时，联盟彻底重构了全组织的元数据体系（Metadata Strategy）。为了在中央联盟办公室、32支俱乐部及全球多个国际市场间实现高效内容分发，NFL制定了通用的分类法标准，同时赋予俱乐部与各地区必要的联邦化自主权，在维持全局数据一致性的同时，保护各支球队在本地化运营中独特的真实语调。渐进式验证与复合型人才协同模式 (6:10) 在实施路径上，NFL没有采取激进的全局铺开策略，而是严格推行“概念验证 $\rightarrow$ 最小可行产品（MVP）试点 $ ightarrow$ 受控真实部署”的分阶段落地框架。Ogbuehi解释道，MVP阶段的作用是对人员、流程与技术进行客观审计，联盟坚持“选择性扩展”原则，仅将证明具备衡量效益与准确性的工作流进行工程化推广。在组织人才层面，NFL认识到单靠技术工具无法驱动采纳，因此在各业务部门培育跨职能推动者与“超级用户”。此外，联盟组建了创意设计师与技术专家协同的复合型人才团队，将原本一体化的静态设计资产拆解为模块化、可复用且具备高适应性的设计组件，使得AI系统能够在不同营销活动中高效重组与动态调用这些素材。人机协同治理与现场实时数据压测 (10:30) 在建立对自动化系统的信任方面，Ogbuehi认为算法的可靠性是在模型演进过程中逐步累积的，因此在NFL的生产链中，人机协同（Human-in-the-loop）是不可妥协的底线，但这一机制必须被合理设计为质量防护网，而非造成效率延误的瓶颈。所有AI生成的素材在正式对外发布前，均须经过严格的人工审核。在2026年NFL选秀大会的实测中，联盟首次将生成式AI流水线接入现场实时体育数据流：当选秀名单宣布时，AI引擎即时生成定制社交图文素材供审核。在首日第一轮选秀中，系统生成表现完美；但到选秀第二天，自动化产出内容出现了明显的质量劣化。得益于流程中嵌入的人工审核机制，团队在公开分发前及时拦截并修复了异常资产，避免了公开面向受众的品牌错误，证实了在模型演进期间严密人工治理的重要性。业务价值指标：突破代币与算力成本思维 (13:53) 在评估AI投资回报率（ROI）时，Ogbuehi说明NFL将目光投向核心业务成果与直接问题解决，而非纠缠于算力点数或代币（tokens）定价等微观技术消耗指标。联盟关注的核心指标包括：衡量从需求提交到素材触达球迷整体耗时的“生产周期时间”（Production Cycle Time）；评估能否生成满足全球细分群体海量素材变体的“吞吐量与扩展性”（Throughput & Scalability）；以及衡量内容无需反复人工修正即可达标的“一次性成功率”（First Time Right）。此外，在推动全系统采纳时，NFL遵循“从一开始就为采纳而设计”的原则，重构端到端流程，将AI功能内嵌至现有日常工具中，消除技术门槛，避免强推孤立的新软件而增加员工学习负担。该对话总结了NFL以运营模型为核心的AI落地经验，强调在不打破既有工作流的前提下通过流程重构与人机治理实现业务增效。但需要指出的是，该模式高度依赖中心化审核体系与强大的品牌协同治理能力；面对未来呈指数级增长的多语种与实时交互需求，如何确保人工审核机制不演变为新的效率瓶颈，以及在更广泛的国际文化语境中持续保障模型输出精度，仍是该体系长期运作面临的现实考验。

Detailed notes — zh

## Background 该内容由 Ai4 频道于 2026 年 8 月 22 日发布在 YouTube 平台上。视频记录了 MAKERS AI 创始人兼首席执行官 Celeny Da Silva 与国家美式橄榄球联盟（NFL）营销技术与消费产品副总裁 Chi Ogbuehi 展开的一场炉边对话，深入探讨了 NFL 如何跨越全球多元受众的超大规模挑战，构建可信、高效的企业级 AI 营销技术与运营模式。 ## TL;DR 对话围绕“AI 本身无法规模化，运营模式才能”这一核心论断，系统复盘了 NFL 应对全球 22 国、超 2 亿球迷即时个性化内容需求的转型历程。Chi Ogbuehi 阐明了从 MVP 试点到全联盟落地的递进路径，重点剖析了如何通过引入技术型创意人才、重构统一接入前端与底层元数据标准，以及在关键链路嵌入人机协同机制来确保生成质量。NFL 借助 AI 推动“大规模亲密度”的内容生产，在价值衡量上摆脱了微观 Token 成本的局限，全面锚定交付周期压缩与资产吞吐量等业务结果，并强调必须在系统设计之初就面向团队的原生采纳进行规划。 ## Atmosphere & style 整场炉边对话洋溢着专业、务实且洞见深刻的行业交流氛围。两位嘉宾之间的互动自然流畅，既有探讨体育情感共鸣时的由衷赞叹，又有复盘真实落地挑战（如 2026 年选秀大会实时生成质量劣化）时的坦诚与克制。听众不仅能感受到顶级体育商业联盟在驾驭前沿技术时的严谨架构思维，更能体会到一种以业务结果为导向、拒绝技术盲从的理性力量。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=DnzndoCPAes&t=4s) NFL 拥有庞大的全球球迷生态，覆盖 1 个联盟、32 支球队、22 个国家和地区及超 2 亿球迷，传统线性工作流已无法承载爆发式的跨语言、跨文化即时内容需求。 - [01:36](https://www.youtube.com/watch?v=DnzndoCPAes&t=96s) NFL 探索 AI 的出发点不是追逐新技术，而是以具体业务瓶颈为导向，核心是一场由 AI 驱动的“运营模式转型之旅”。 - [04:32](https://www.youtube.com/watch?v=DnzndoCPAes&t=272s) NFL 的球迷互动处于“全天候开启（always-on）”状态，从休赛期的长期叙事铺垫无缝切换到赛季中的高强度实时响应，AI 成为解锁规模化与交付速度的关键。 - [06:10](https://www.youtube.com/watch?v=DnzndoCPAes&t=370s) 践行“AI 本身无法规模化，运营模式才能”，NFL 遵循概念验证到 MVP 试点再到业务执行的渐进路线，坚持只规模化被验证有效的部分（scale what worked）。 - [06:10](https://www.youtube.com/watch?v=DnzndoCPAes&t=370s) 在人员层面，企业需寻找内部业务拥护者，并重点引入兼具创意设计与工程落地能力的“技术型创意人才（creative technologists）”。 - [08:11](https://www.youtube.com/watch?v=DnzndoCPAes&t=491s) 重构流程与数据底座：前端设立统一接入“前门（front door）”，底层统一元数据标准，并兼顾联盟全局一致性与本地俱乐部特色授权。 - [10:38](https://www.youtube.com/watch?v=DnzndoCPAes&t=638s) 建立人机协同（Human-in-the-loop）机制与治理规则，在关键风控节点把关；在 2026 年选秀大会实测中成功拦截了 AI 出现的质量衰退，避免品牌公关风险。 - [12:34](https://www.youtube.com/watch?v=DnzndoCPAes&t=754s) 提出“大规模亲密度（mass intimacy）”理念，认识到“每个人体验橄榄球的方式截然不同”，借助 AI 打破传统广播模式，实现亿级球迷规模下的专属个性化互动。 - [13:44](https://www.youtube.com/watch?v=DnzndoCPAes&t=824s) 衡量 AI 商业价值超越微观 Token 与算力成本，聚焦生产周期缩短、资产吞吐量提升及“一次做对（first-time right）”的交付质量等宏观业务成效。 - [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) MarTech 生态落地的核心法则是以运营模式为骨架释放加速倍增效应，并在系统架构之初“全面面向实际采纳而设计（design for adoption upfront）”，避免给团队增加额外学习负担。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=DnzndoCPAes&t=4s) 炉边对话开场：NFL 的全球规模与核心命题活动由 MAKERS AI 创始人兼首席执行官 Celeny Da Silva 开场主持。对于广大球迷而言，美式橄榄球早已超越了单纯的体育比赛范畴，它代表着身份认同、社区联结与深刻的情感共鸣。而对国家美式橄榄球联盟（NFL）来说，提供这种极致的球迷体验意味着必须在惊人的超大规模下持续运转：一个联盟、32 支球队俱乐部、横跨全球 22 个国家和地区、覆盖超过 2 亿名球迷。在赛季的每一个周，NFL 都必须将赛场上瞬息万变的战况转化为高度相关、本地化的数字内容与体验，精准触达不同市场的数百万受众。面对如此庞大的受众与跨区域需求，如何将人工智能从早期的局部实验，转变为整个组织能够信赖并大规模落地的核心能力？Celeny 引用了 NFL 的一句核心理念：“AI 本身无法规模化，运营模式才能（AI doesn't scale, operating models do）”，并邀请本场特别嘉宾——NFL 营销技术与消费产品副总裁 Chi Ogbuehi 登台。Chi 刚结束在希腊的假期回到现场，随后就 NFL 的幕后内容机制与技术探索展开了深度分享。 - **NFL 运营规模指标**：1 个联盟、32 支球队、全球 22 个国家/地区、超 2 亿球迷。 - **核心讨论主题**：打破 AI 实验局限，通过运营模式创新支撑企业级规模化与可信应用。 --- ### [01:36](https://www.youtube.com/watch?v=DnzndoCPAes&t=96s) 庞大的全球球迷生态与打破传统工作流壁垒 Chi Ogbuehi 介绍了其所负责的营销技术团队职能。该团队主要统筹新兴技术的引入、运营模式的搭建以及底层技术平台的演进，全面支撑 NFL 在美国本土及国际市场的球迷互动。Chi 谈到，公众在谈及 NFL 时往往只关注赛场上的激烈比拼，很少有人能看到幕后为了支撑这些球迷体验所付出的系统性努力。每天、每周，团队都在通过不同的渠道、不同的语言版本向全球各个市场分发多元化的内容与体验。随着 NFL 的全球化扩张，目前的球迷画像呈现出前所未有的多元性：女性球迷群体大幅增长，腰旗橄榄球（Flag football）等衍生运动也迅速普及。然而，全球化与多元化也带来了极高的复杂度。团队必须面对多语言翻译、多渠道格式适配以及不同地区的文化细微差异（cultural nuances）。与此同时，球迷对内容的即时性（immediacy）、时效性与高度个性化提出了极高的期望。Chi 直言，这种对内容爆发式增长的需求已经彻底超越了传统工作流的承载极限，“传统工作流已经完全无法支撑这种需求”。这促使 NFL 开始探索 AI 的应用价值。他特别强调，NFL 并非为了追逐新技术而生硬地将 AI 塞进组织，而是为了解决具体的业务瓶颈；这根本上不是一场单纯的“AI 旅程”，而是一场由 AI 驱动的“运营转型之旅（operational transformation journey）”。 ```mermaid flowchart TD accTitle: NFL 球迷内容需求与传统工作流矛盾示意图 subgraph FanDemand["球迷端需求爆发"] D1["全球 22 国多元文化与多语言"] D2["极高即时性与个性化期待"] D3["多格式跨渠道分发"] end subgraph Bottleneck["传统生产瓶颈"] B1["传统线性工作流"] B2["产能与响应速度无法跟上需求"] end subgraph Solution["转型路径"] S1["以业务问题为导向引入 AI"] S2["业务运营模式根本性重塑"] end FanDemand --> B1 B1 --> B2 B2 --> Solution ``` --- ### [04:32](https://www.youtube.com/watch?v=DnzndoCPAes&t=272s) 全天候球迷互动与内容供应链的实时演进在回应日常运营节奏时，Chi 指出 NFL 的球迷互动处于一种“全天候开启（always-on）”的状态。与球迷的每一次接触都在创造新的互动机会，而赛场内外每一个新事件的发生，都会引发球迷对即时、有意义回应的新期待。这种互动绝不仅限于周日的正赛阶段，早在赛季开始前的休赛期便已全面拉开序幕：球迷们紧密追踪选秀大会（Draft）、赛程发布（Schedule release）以及季前训练营（Training camp），这些节点都在持续为常规赛蓄势。随着新赛季临近，球迷的注意力逐步转向深入研究对阵赛程、探索个性化观赛途径、追踪球队叙事主线与喜爱的球星动态。而一旦正式进入赛季，内容生产模式便立刻切换为高强度的近乎“实时（real time）”状态。正是因为需要全年不间断地应对如此庞大、复杂的实时内容吞吐量，NFL 才将 AI 视作解锁规模化交付与生产速度（scale and velocity）的关键契机。 - **休赛期蓄势关键节点**：NFL 选秀大会、赛程正式发布、夏季训练营。 - **赛季周期运作节奏**：从休赛期的长期叙事铺垫，快速切换为赛季中的全天候实时内容响应。 --- ### [06:10](https://www.youtube.com/watch?v=DnzndoCPAes&t=370s) 验证与落地：从 MVP 试点到人、流程与技术的重塑针对“AI 本身无法规模化，运营模式才能”这一核心论断，Chi 解释道，这是团队在实战探索中获得的最深刻教训。NFL 采取了非常克制且结构化的落地路径，绝不从一个初步概念直接盲目推广到全企业级规模，而是遵循“概念验证 -> 试点最小可行产品（MVP） -> 真实世界业务执行”的递进路线，并坚持“只规模化被验证有效的部分，而非盲目规模化一切（scale what worked）”。在 MVP 阶段，团队不仅明确了哪些技术可行、哪些不可行，更重要的是摸清了在人员（People）、流程（Process）和技术（Technology）三个维度必须发生的深刻转变： ```mermaid flowchart LR accTitle: MVP 验证及三维重构体系 subgraph Phase["落地演进路径"] P1["概念验证 Concept"] --> P2["试点 MVP"] P2 --> P3["真实业务执行 Execution"] P3 --> P4["有效模式规模化 Scale What Worked"] end subgraph Pillars["三维重构要素"] direction TB E1["人员 People 寻找内部拥护者 引入技术型创意人才"] E2["流程 Process 摒弃旧有流程叠加 构建标准化统一入口"] E3["技术/数据 Tech 统一底层元数据标准 兼顾联盟一致与本土特色"] end P2 -.-> Pillars ``` - **人员维度的变革**：必须在组织内部快速建立认同（buy-in），物色并赋能各部门的业务拥护者（champions）；同时重点吸纳“技术型创意人才（creative technologists）”，即既懂创意设计，又掌握工程能力、能将视觉设计转化为可复用、可自适应、可规模化扩展数字资产的复合型人才。 --- ### [08:11](https://www.youtube.com/watch?v=DnzndoCPAes&t=491s) 工作流重构：建立统一入口与底层元数据标准化在流程层面，Chi 强调过去走过最大的弯路就是试图直接把 AI 技术生硬地“套”在现有的老旧工作流之上。这种做法很快在 MVP 阶段碰壁并证明完全无法规模化。团队不得不停下脚步，重新构建干净坚实的操作底座（clean foundation）。为了理顺日常协作，团队在前端设立了统一的需求接入“前门（front door）”，确保全联盟和各业务线的所有内容制作请求均通过标准化通道流转，从而实现生产模式的可复用性、交付一致性以及合规治理。与此同时，团队对底层元数据战略（Metadata Strategy）进行了彻底重构。要在 32 支球队俱乐部及跨国海外市场之间协同放大 AI 效能，所有系统必须基于同一套标准元数据语言进行对话。团队清晰界定了哪些元素需要全联盟集中标准化以确保品牌资产一致性，哪些部分需要向下授权（federate）给各俱乐部与本地团队，以保留其原汁原味的本土文化特色。 --- ### [10:38](https://www.youtube.com/watch?v=DnzndoCPAes&t=638s) 信任建立与人机协同：2026 年选秀大会的质量衰减实录谈及企业级 AI 规模化落地前必须具备的要素，Chi 强调核心依然是如何围绕 AI 去架构和运行业务，以及建立对 AI 能力的信任机制。由于 AI 模型的输出需要时间持续迭代与调优，NFL 始终在整个生产链路中保持“人机协同（Human-in-the-loop）”机制：所有由 AI 生成的内容资产在面向公众正式发布前，必须经过人工审核把关。但为了防止人工审核沦为新的生产瓶颈，团队仅在关键风控节点精细化嵌入治理规则与人工审查。 Chi 分享了 2026 年 NFL 选秀大会（2026 Draft）的真实案例。这是联盟首次将 AI 系统直接接入实时数据流：随着现场选秀结果逐一揭晓，AI 能够根据实时数据源瞬间生成配套的多媒体内容，供社交媒体团队快速分发。在选秀前几轮，AI 生成的内容质量非常精准优秀；然而进行到选秀第二天时，团队敏锐地发现 AI 输出的内容质量开始发生明显的劣化与衰退（degrade）。得益于事先部署的人机协同审核机制，团队在这些瑕疵内容流向公域前及时进行了拦截与修正，避免了潜在的公关与品牌风险。Chi 总结道，企业绝不能从一开始就对 AI 抱有盲目信任；只有随着模型不断学习、持续稳定后，人工干预的程度才能逐步审慎后撤，这对于维护球迷体验与品牌信任而言至关重要。随后，对话切入到 AI 如何进一步赋能与重塑端到端的球迷体验。 ### [12:34](https://www.youtube.com/watch?v=DnzndoCPAes&t=754s) 大规模亲密度：以 AI 重塑球迷的个性化连接当 Celeny Da Silva 问及当下的球迷究竟如何切身感受到这种技术变革，以及 NFL 在球迷群体中观察到了哪些具体反馈时，Chi Ogbuehi 选择从联盟的最高战略目标切入。他指出，现代球迷对 NFL 的互动体验抱有极高期待——每一次触达都必须做到足够即时（timely）、紧密相关（relevant），并且展现出深度的个性化（deeply personalized）。然而，真正的技术挑战在于如何在数以亿计的庞大球迷基数上“大规模”（at scale）实现这一点。为了具象化这一理念，Chi 提出了一个核心概念——“大规模亲密度”（mass intimacy）。他生动地解释道：“每个人体验和热爱橄榄球的方式都是截然不同的（everyone footballs differently）。”每个球迷都有着独一无二的观赛动机、偏好和交互习惯。NFL 正在借助 AI 技术打破过去粗放的大众广播模式，让系统有能力在同一时间并行运作，向每一位球迷提供仿佛一对一专属定制的内容与互动，让他们在浩瀚的粉丝海洋中依然能感受到强烈的个人归属感。Celeny 对此深表赞同，感叹体育与 NFL 最迷人的特质正是赋予全人类建立情感纽带、共同分享热爱与激情的无形力量。 ```mermaid flowchart LR accTitle: NFL 大规模亲密度实现路径 subgraph CoreGoal ["核心诉求"] A["球迷多元化诉求 'Everyone footballs differently'"] end subgraph AIEngine ["AI 赋能支撑"] B["高相关性与即时交互"] C["大规模内容个性化"] end subgraph Outcome ["最终体验"] D["大规模亲密度 (Mass Intimacy)"] end A --> B A --> C B --> D C --> D ``` ### [13:44](https://www.youtube.com/watch?v=DnzndoCPAes&t=824s) 衡量商业价值：超越 Token 成本，聚焦业务结果与产出效率针对高规格个性化体验背后必然伴随的技术成本，Celeny 追问 NFL 迄今为止究竟如何衡量 AI 项目的实际商业价值。Chi 直言，NFL 的评估重心完全锚定在具体问题的解决成效（outcome-driven）上，而非单纯的技术指标。在实际衡量中，他们确立了几个关键的业务成效维度：首先是**生产周期的缩短**（reduction in production cycle time），即从业务端提出一个内容创意或营销需求，到该资产最终制作完成并推送到球迷眼前，端到端的流转耗时究竟被压缩了多少；其次是**生产吞吐量**（throughput），重点衡量系统是否具备在极短窗口期内，为多元渠道海量生成各类创意变体与衍生资产（variations and assets at scale）的能力；第三是**交付质量**，其中“一次做对”（first-time right）的合格率是评估生成能力的重要标尺。在谈及成本控制时，Chi 坦承成本确实是任何企业做出投资决策的关键考量，但 NFL 并没有让自己过早地陷入对底层细枝末节的纠结之中。他特别强调：“我们目前并没有过度聚焦在微观的成本模型、算力点数（credits）或 Token 消耗上，当下的战略重心是其为整个业务带来的真正价值。”正是这些核心业务驱动力在指引联盟判断未来是否值得继续加码投入。Celeny 随之补充指出，这种能够跨越不同国家文化、不同语言环境进行超个性化触达的能力，本质上已经在实质性地推动 NFL 全球球迷版图的扩张与高粘性互动。 - **生产周期指标**：严密监测从接收制作需求到资产最终触达球迷的全链路交付时长。 - **资产吞吐能力**：评估 AI 能否低成本、高并发地生成跨渠道的海量创意衍生版本。 - **质量保证体系**：关注内容生成的首次准确率（first-time right），降低人工返工损耗。 - **投资决策导向**：现阶段以宏观业务增益引导资本配置，不被微观 Token 计费模式束缚脚步。 ### [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) AI 规模化落地的核心法则：组织运营模式与原生式采纳设计在对话接近尾声时，Celeny 邀请 Chi 总结在庞大的营销技术生态（MarTech ecosystem）中成功落地、规模化运营 AI 时最重要的两到三项核心原则。Chi 几乎没有犹豫，脱口而出反复强调的词汇：“运营模式（Operating model），这绝对是最重中之重的一点。” Chi 阐述道，技术本身并不是万灵药，企业必须首先深刻理解如何将组织内的人员（people）、业务流程（processes）和技术工具与恰当的运营模式相互对齐；唯有底层的运营骨架理顺之后，AI 才能真正作为“超级加速器”（accelerator）释放倍增效能。紧接着，他抛出了第二条核心原则——**在系统设计之初就必须全面面向‘实际采纳’而设计**（design for adoption upfront）。为了从根本上降低团队对新技术的抵触和准入门槛，NFL 在搭建其营销技术生态时，并未生硬地叠加外挂式工具，而是彻底重塑了端到端的工作流，使整个交互过程顺畅得如呼吸般自然。Chi 特别强调，必须极具前瞻性与针对性，“绝不能让业务团队觉得他们又得被迫去学习掌握另一款繁琐的工具软件”。整场对话在两人欢快的致意中迎来尾声。Celeny 用一句精辟凝练的金句为全场研讨画上句号：“AI 本身无法自发规模化，能够实现规模化的是成熟的运营模式（AI doesn't scale. Operating models do）。”这也是 NFL 能够跨越复杂的企业组织架构、在全球范围内稳步铺开 AI 技术的最关键启示。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=DnzndoCPAes&t=4s) AI doesn't scale, operating models do — Celeny Da Silva > [01:36](https://www.youtube.com/watch?v=DnzndoCPAes&t=96s) 传统工作流已经完全无法支撑这种需求 — Chi Ogbuehi > [01:36](https://www.youtube.com/watch?v=DnzndoCPAes&t=96s) operational transformation journey — Chi Ogbuehi > [06:10](https://www.youtube.com/watch?v=DnzndoCPAes&t=370s) scale what worked — Chi Ogbuehi > [08:11](https://www.youtube.com/watch?v=DnzndoCPAes&t=491s) clean foundation — Chi Ogbuehi > [08:11](https://www.youtube.com/watch?v=DnzndoCPAes&t=491s) front door — Chi Ogbuehi > [12:34](https://www.youtube.com/watch?v=DnzndoCPAes&t=754s) mass intimacy — Chi Ogbuehi > [12:34](https://www.youtube.com/watch?v=DnzndoCPAes&t=754s) everyone footballs differently — Chi Ogbuehi > [13:44](https://www.youtube.com/watch?v=DnzndoCPAes&t=824s) first-time right — Chi Ogbuehi > [13:44](https://www.youtube.com/watch?v=DnzndoCPAes&t=824s) 我们目前并没有过度聚焦在微观的成本模型、算力点数（credits）或 Token 消耗上，当下的战略重心是其为整个业务带来的真正价值。 — Chi Ogbuehi > [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) Operating model — Chi Ogbuehi > [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) design for adoption upfront — Chi Ogbuehi > [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) 绝不能让业务团队觉得他们又得被迫去学习掌握另一款繁琐的工具软件 — Chi Ogbuehi > [15:26](https://www.youtube.com/watch?v=DnzndoCPAes&t=926s) AI doesn't scale. Operating models do — Celeny Da Silva ## Mentioned Celeny Da Silva, Chi Ogbuehi, MAKERS AI, National Football League (NFL), Ai4, Flag football, NFL Draft ## Video description Celeny Da Silva - Founder & Chief Executive, MAKERS AI Chi Ogbuehi - Vice President, Marketing Technology and Consumer Products, National Football League (NFL) [Source](https://www.youtube.com/watch?v=DnzndoCPAes)

Source quotes

[0:04] Celeny Da Silva: AI doesn't scale, operating models do [1:36] Chi Ogbuehi: 传统工作流已经完全无法支撑这种需求 [1:36] Chi Ogbuehi: operational transformation journey [6:10] Chi Ogbuehi: scale what worked [8:11] Chi Ogbuehi: clean foundation [8:11] Chi Ogbuehi: front door [12:34] Chi Ogbuehi: mass intimacy [12:34] Chi Ogbuehi: everyone footballs differently [13:44] Chi Ogbuehi: first-time right [13:44] Chi Ogbuehi: 我们目前并没有过度聚焦在微观的成本模型、算力点数（credits）或 Token 消耗上，当下的战略重心是其为整个业务带来的真正价值。 [15:26] Chi Ogbuehi: Operating model [15:26] Chi Ogbuehi: design for adoption upfront [15:26] Chi Ogbuehi: 绝不能让业务团队觉得他们又得被迫去学习掌握另一款繁琐的工具软件 [15:26] Celeny Da Silva: AI doesn't scale. Operating models do

Source quotes

[0:04] Celeny Da Silva: AI doesn't scale, operating models do. [6:10] Chi Ogbuehi: AI doesn't scale, operating models do. [12:34] Chi Ogbuehi: everyone footballs differently [15:26] Chi Ogbuehi: once you do that, AI becomes an accelerator. [15:26] Chi Ogbuehi: learn yet another tool. [15:26] Celeny Da Silva: AI doesn't scale. Operating models do.

Original transcript

[0:04] Good afternoon everyone. Uh, welcome to AI4. I am Salin Das Silva, founder and [0:10] CEO of Makers AI. For fans, football is more than a game. It's an identity, [0:18] a community connection, emotional connection. For the NFL, delivering that experience means [0:26] operating at an extraordinary scale. One league, 32 clubs, more than 200 million [0:33] fans across 22 countries. Every week of the season, the NFL must turn what happens on [0:40] the field into relevant localized experience that engage millions of fans across markets. [0:47] So that does so what does it take to turn AI from experimentation into capability that the [0:56] organization can trust and scale in the NFL's own words AI doesn't scale operating model zoo [1:09] please welcome our very very special guest she [1:16] uh vice president of marketing technology and consumer products at NFL. Thank you. Welcome. [1:23] We're very happy to have you. Especially came all the way from Greece to see us. Well, [1:29] I came back holiday over the weekend, but yes. So, um G, um take us inside the work. It is Tuesday [1:36] during the season. What has to happen across the NFL's marketing and content operation to keep [1:42] fans engaged between one Sunday and the next? And where does AI sit in that process today? Sure. Um, [1:50] I'm [clears throat] going to take a few steps back and just talk about one quick thing. So, I head up [1:54] a lead marketing technology for the NFL where my team is responsible for the emerging technologies, [2:02] the operating models, the the platforms that power how we engage our fans domestically and [2:10] internationally as well. And I'll also add that one thing that most people when people think about [2:16] the NFL, most people think about what happens on the field. Um, they don't see what happens [2:21] behind the scenes and how it all comes together to actually make those fan experiences possible. Um, [2:30] every week and every day, we're always delivering different types of experiences and content to [2:35] our fans across different channels, across different markets, across different languages. [2:41] Um, and as we've continued to expand, we're now in 22 countries. As we've continued to expand, [2:50] our our fan base has become one of the most diverse. Um, we have lots of women who are [2:56] fans. We have flag football. We have one of the most diverse fan bases ever. Um, but I [3:02] would add that that with that global fan base and that global diversity brings its own share [3:08] of complexity. right now. We're thinking about, you know, different languages, different formats, [3:16] all these different cultural nuances. Oh, and by the way, um we're in an environment where fans are [3:22] expecting um us to engage with them in ways that are timely, relevant, and highly personalized. [3:30] So when you think about the scale and you think about the growing demand and the the the demand [3:35] and expectations for immediiacy, for timeliness, and for you know relevant content, it's gotten [3:42] to a point where that demand has outpaced um our ability to like it's completely the our [3:48] traditional workflows can can no longer support that demand. And that's where we started to look [3:54] at AI. Um, I would add that AI, well, we weren't looking at AI as this new technology and how do [4:00] we place this new technology in organization. We were actually looking at AI and trying to figure [4:05] out how can AI help us solve a practical business problem. And so, so just wanted to provide that [4:13] that context and you know in a lot of ways I wouldn't even say this is an AI journey that [4:18] we're on. I would say it's more of an operational transformation journey of which AI is playing a [4:23] big role. Um yeah and then and then going back to your question [laughter] um for us fan engagement [4:32] is always on right so every day that we interact with our fans or every day kind of creates new [4:42] opportunities um for how we engage our fans and as new moments unfold that also leads to new [4:50] expectations around how we can respond to fans. uh meaningfully and quickly and so it's always on and [4:57] this interaction with our fans doesn't start on Sunday right it starts you know way before [5:05] the season even even kicks off right so during the off season fans are following the draft and [5:11] schedule release and and training camp and so many other moments that help build anticipation for the [5:19] season and then as the season is approaching uh a lot of fans are exploring um you know schedules [5:26] and um diving into like more personalized ways to watch following storylines and their favorite [5:32] players and so forth. And then once we get into the season, speaking of content, we actually start [5:38] to shift into more real time. And so we're always always continuously engaging our fans whether it's [5:44] in real time or not. So just wanted to provide more context and that also provides um sheds [5:50] more context on the the complexity of just the volume of content that we always put out there. [5:57] Um and where again AI we started to look at AI as a very interesting opportunity to help us really [6:04] deliver that scale and velocity. Yeah. Yeah, that takes me to uh next question is when you said AI [6:10] um doesn't scale or uh operate operating models do. What does that mean in practice and what [6:15] proved the most challenging um to operationalize so far? Sure. Um so that statement or that phrase [6:22] AI doesn't scale operating operating models do actually came from one of the biggest lessons [6:29] we learned in this journey. Um so we didn't we were very intentional about how we approached [6:37] our journey. We didn't go from concept to enterprise scale. We took a more structured [6:42] approach from concept uh we did a pilot MVP um and then moved into real world execution. But the only [6:51] things we scaled we scaled what we we didn't scale everything. We scaled what worked. Um, [6:56] and so AI, when I the MVP phase taught us a lot, right? There's some things that worked and there's [7:04] some things that didn't work. But one of the key things we learned through the MVP phase was of [7:09] course what worked, but also what needed to change from a people perspective, a process perspective, [7:14] and a technology perspective. And so from a people perspective, we quickly realized that there was a [7:20] need to buy gain buyin quickly. And so we started to look for champions within the organization, [7:27] empower users. And then we started to onboard creative people as well as technologists, [7:33] people who had the capability to take a creative and convert that or take a creative or a design [7:40] and convert that into content that can be reused, adapted, and scaled. And that's very fascinating [7:47] because that takes um takes us to my next question is is like just to walk us through the supply [7:52] chain but like from the idea to the moment of the experience and the content what you just mentioned [7:56] but where has AI reshaped that workflow and where must human judgment still guide the decision which [8:03] is sure talent and yeah yeah sure no that that's a great question so for us um the first step wasn't [8:11] again this wasn't an AI journey this was really redefining ing how we work. And so the first step [8:17] from a people, process and technology perspective was again that was how we redefined how we worked [8:22] from a process perspective. We realized we couldn't layer AI on top of existing workflows, [8:27] right? We needed to set establish a clean foundation and if we didn't work off of a clean [8:33] foundation, the AI was not going to work in our favor. So again, it started first with and that's [8:38] where the AI doesn't scale operating models do. We had to figure out the correct operating model [8:43] first. And then we started to build upon that. So once we defined the operating model, the next [8:49] step was to look at um the next question we asked ourselves is what does what should it look like? [8:55] How should our teams from a day-to-day perspective operate within this ecosystem. And so that's where [9:03] you know things like you know creating a front door where all requests come through to help [9:08] us ensure we achieve the repeatable patterns, the consistency, the governance. started to [9:14] started to play a role um in addition to um yeah go ahead but going through the process like what [9:19] have you learned um that did not translate into enterprise capability is there anything specific [9:26] you can share yeah so that's it's going back to the MVP there when we first moved uh launched [9:33] the MVP we really thought we could layer on AI on top of existing processes and we quickly learned [9:41] you were not going to scale Right. So it required that we take a step back and revisit our approach. [9:47] Um it also required the other thing that we did is that we you know again we we layered on AI [9:54] on top of existing uh capabilities realized that was not going to help us scale. We had to take a [9:59] step back and even revisit our metadata strategy. Right? So if we were going to scale this across [10:05] the league and to clubs and international markets, we had to make sure we're working off of the same [10:11] standard metadata language. And so we have to take a step a step back, look at the operating model, [10:17] look at our metadata strategy, look at how what made sense to standardize to gain consistencies [10:24] and where we wanted to federate um in terms of what was authentic. So yeah. Yeah. What that's [10:30] great actually um takes to my next um question is what what must be in place before you trust [10:38] um capability at scale. What were the elements that you would say one or two two things that [10:44] must be in place before uh moving forward with that approach? Um it comes back down to how you [10:52] want to run AI or how you want to design or run the business around AI. So what must be in place [10:58] comes back down to the operating model really being able to figure it out. [snorts] Um from a [11:03] trust perspective I AI gets better and better with time. So for us we still have a human in the loop [11:10] in the entire process. We have humans review um our assets and our content before it's publicly [11:18] published. But we don't want the human in the loop to be a bottleneck. So we've plugged in governance [11:23] and human reviews where it makes sense. And I'll give you a quick example. For the 2026 draft, [11:29] it was the first time we plugged our AI into real-time sources. And what that meant was as [11:35] the draft picks were being announced, the AI was immediately creating the content that our social [11:40] team was publishing. But it's a good thing we had a human in the loop because what was happening [11:45] is that the first few rounds of the draft pick, the AI was doing it. It was on point. It was the [11:51] quality, everything was great. But we started to notice by like day two of the draft, the quality [11:56] started to degrade. Oh wow. And we were able to catch that on time and fix it before we sent those [12:03] um the content. We publish the content out uh for for public consumption. So a human the loop [12:09] is still important. Um I don't know that you can trust AI from the get-go, but it gets smarter and [12:15] when it as it gets smarter and as it learns, you can start to take a step back. It's definitely [12:20] very critical for um experience, fan experience and trust as well. So that takes us to a shift [12:28] to talk a little bit about the fan experience and um what is an AI making the fan experience better [12:34] today and how do the fans feel that difference if you can uh give us examples of what have been [12:40] done and what have you observed from the fan base. Yeah, I like to talk about that through the lens [12:46] of our main goal. You know, I mentioned earlier that fans expect us to interact with them in ways [12:52] that are timely, that are relevant, um, and are that are deeply personalized, right? Um, and what [12:59] AI is helping us do, and we have to do that and do that at scale. And I like to talk about this [13:04] through the lens of mass intimacy, right? Being able to deliver personalized content, being able [13:11] to interact with our fans in a way that feels deeply personal for each fan because everyone [13:16] footballs differently. Our fans are very unique and they interact with us in very different ways. [13:20] So, how can we drive mass intimacy where every fan is feeling, you know, that we're interacting with [13:27] them in a way that is relevant and personal, but be able to do that sim simultaneously at scale, [13:33] you know, I love that. I love that. And one thing I love about sports in the NFL is the connection [13:38] that it allows empowers humanity to basically connect and share the experience and love. So, [13:44] this is fascinating. Um but um that comes with a price. Sure does. So my next question is how [13:53] does the NFL measure business value? Um or has so far obviously it's um AI is not that um yeah uh so [14:04] we're we're looking at the outcome. So we have a problem we're trying to solve and we're measuring [14:09] business value based on that. So reduction in production cycle time. How long is it taking from [14:15] the time a request comes in till it gets to the fan? What does that cycle time look like? Um in [14:21] terms of the throughput, are we able to create the variations and the assets that scale? Um quality, [14:28] are we getting is like first time first time right plays a role there. I'll also add that cost is [14:33] definitely a big consideration for for companies because you need to it it it plays a role in how [14:39] you make your investment decisions. But for us that is a part of a broader uh set of things we're [14:47] analyzing from an AI perspective. We're looking at some of these key drivers I talked about because [14:53] that also will inform next steps. So we're not too focused right now on cost and credits and c the [15:00] cost models and tokens. Our f focus right now is on the value to the business. some of the things I [15:07] highlighted because we know that that is actually going to inform next steps and inform how we [15:12] choose to or not to invest and also expanding your fan base. Yeah. Um obviously because you are able [15:19] to personalize to so many different countries and and different languages but much more engaging. Um [15:26] so and finally uh from your experience at NFL um what are the two three principles that matter most [15:33] when oper operational malize AI across a marketing technology ecosystem um there are a few things I [15:42] keep saying operating model operating model yes that's that's a big one um you need to really [15:47] know how to align people processes technology with the right operating model and once you do [15:52] that AI becomes an accelerator Um the other one I will add is really you need to design for adoption [15:59] upfront right so reducing those barriers to entry and what we did when we designed our ecosystem we [16:05] did it in a way that um made it feel more natural we really redesigned the entire workflow that made [16:11] it feel more natural so people aren't feeling like they have to learn yet another tool so just [16:17] being able to bear in mind like how you design the ecosystem and design for adoption up front [16:24] I That will be another one that I will call out to be more intentional. Yes. And that uh brings [16:32] us to an end. Oh, okay. Okay. I think we're about um ch Thank you for taking taking us inside what [16:43] um it takes to operationalize AI at scale across a complex enterprise like the NFL. You heard it [16:50] here. AI doesn't scale. operating models do. Chi, thank you very much. Yeah, thank you.