Google Cuts Chrome Ad Time 70% as AI Bottlenecks Shift to Compliance

Applying AI in Performance Marketing: Practical Lessons from Operating at Scale — Ai4

Ai4

While generative AI accelerates performance ad creation cycles from 30 days to 10, overall time-to-market remains stalled because enterprise legal and approval workflows cannot match automated output volume.

AI-assisted summary of Applying AI in Performance Marketing: Practical Lessons from Operating at Scale — original source. Check the original for full context.

Source published: 2026-09-04

Detailed notes

## Background Published on YouTube by Ai4 on September 4, 2026, this fireside chat features James Croom (Senior Director of Marketing for Google Chrome), Wesley ter Haar (Co-founder & Chief AI Officer at Monks), and moderator Kristin Stoller (Editorial Director at Fortune). The session focuses on practical operational lessons and organizational shifts required when integrating generative AI into global performance marketing workflows at scale.

## TL;DR While generative AI has dramatically collapsed asset production cycles—slashing Google Chrome's localization turnaround from thirty days to ten—enterprise go-to-market velocity remains bottlenecked by traditional legal and compliance review workflows. Navigating this shift requires organizations to evolve from chaotic, fragmented pilots toward standardized operational capabilities while maintaining human oversight for upstream strategic vision and emotional intuition. Additionally, the rise of AI-mediated discovery is forcing marketers to rethink digital strategy, treating brand websites as clean knowledge bases for autonomous agents and driving conversational relevance across open social platforms.

## Atmosphere & style The discussion is candid, grounded, and refreshingly sober, trading speculative industry hype for the unvarnished realities and organizational 'scar tissue' of enterprise implementation. The rapport between Croom and ter Haar is collegial and pragmatic, marked by sharp tactical anecdotes and lighthearted banter over competing AI tools. Listening in feels like sitting inside a frank, high-level debrief between seasoned marketing operators who are actively managing the friction between rapid machine generation and human-speed corporate governance.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=4s) While AI has drastically accelerated creative asset generation, overall go-to-market speed remains constrained by traditional, slow-moving legal, risk, and enterprise compliance reviews. - [01:28](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=88s) Beyond tech giants, early enterprise AI adoption is led by margin-compressed sectors like FMCG, CPG, and automotive, which need AI to manage the exponential explosion of channel-specific digital content demands. - [02:36](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=156s) Google Chrome's localized performance marketing workflow reduced campaign turnaround time by nearly 70% (from 30 to 10 days) and cut direct production costs by roughly 27% across 40+ global locales. - [02:36](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=156s) High-velocity AI production enables marketing teams to continuously 'daisy chain' the two-week novelty performance bump of fresh creative before audience ad fatigue sets in. - [06:06](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=366s) Scaling AI in creative agencies requires fundamentally overhauling company culture from subjective personal preferences to structured, process-driven operations. - [06:06](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=366s) Unlike startups that can risk brand damage with unvetted creative, large enterprises must implement strict governance and guardrails to prevent low-quality AI 'slop' from reaching markets. - [09:35](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=575s) AI agents excel at absorbing vast briefing contexts and generating downstream creative variations, allowing human marketers to focus upstream on strategic ideation, emotional nuance, and creative intuition. - [09:35](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=575s) AI tools can be deployed across large brand portfolios to audit and verify live creative against concurrent campaigns, preventing contradictory messaging or redundant visuals. - [11:52](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=712s) To prevent 'death by a thousand pilots' and redundant internal tooling, organizations should consolidate around a standardized core of mission-critical AI capabilities while maintaining controlled room for specialized experimentation. - [14:13](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=853s) Agency value is shifting from billing for raw production hours toward navigating complex enterprise approval workflows and providing seasoned domain judgment. - [14:13](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=853s) The most effective AI marketing talent model pairs junior AI-native operators who master new tools with seasoned domain experts who supply strategic direction. - [16:55](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1015s) Marketing is expanding beyond human audiences to optimizing directly for large language models and autonomous agents acting as search and recommendation intermediaries. - [17:38](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1058s) Because LLMs rely on fresh discussions from open platforms like Reddit, X, and YouTube, brands in low-conversation categories face new hurdles in establishing algorithmic visibility. - [18:38](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1118s) Brand websites are transforming into structured knowledge bases for AI agents, making content pruning and knowledge base hygiene critical to prevent models from retrieving obsolete information. - [18:38](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1118s) As autonomous AI agents make purchasing decisions based on utility and economic logic rather than emotional persuasion, marketers must fundamentally rethink how brand equity functions.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=4s) The Paradox of Speed: Creation vs. Enterprise Governance

The session opens with moderator Kristin Stoller setting the stage: after years of promises that artificial intelligence would completely revolutionize marketing, what is actually measurably faster in mid-2026, and what has stubbornly remained the same? James Croom points directly to content production as the clearest win—assets can now be produced at a vastly accelerated clip. However, overall go-to-market speed has not accelerated at anywhere near the same pace because traditional review cycles, such as legal and compliance sign-offs, have completely failed to keep up with the sheer volume of generated assets.

Wesley ter Haar reinforces this operational reality, describing a "quite paradoxical situation." While AI has successfully collapsed the manual labor required to transition from an initial insight to in-channel creative assets, the operational bottleneck has simply shifted down the line. Because sign-offs, risk assessments, and enterprise review workflows have not sped up proportionally, work can theoretically be done instantly, yet getting it out the door still stalls.

```mermaid flowchart LR accTitle: AI Content Generation Bottleneck Flow subgraph Accelerated ["AI-Accelerated Phase"] A["Raw Insights"] --> B["Asset Generation"] end subgraph Bottleneck ["Enterprise Bottleneck"] B --> C["Legal & Compliance Review"] C --> D["Stakeholder Sign-offs"] D --> E["In-Market Launch"] end ```

[01:28](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=88s) Early Adopters and Post-Pilot Momentum Across Industries

When surveying the broader industry landscape, ter Haar highlights two distinct categories of businesses leading enterprise AI adoption. First are the technology giants who are practically obligated to adopt early—engaging in what he terms "a little bit of dog fooding or drinking your own champagne." Second are industries operating under intense margin compression and growth mandates, particularly Fast-Moving Consumer Goods (FMCG), Consumer Packaged Goods (CPG), and automotive brands.

These consumer goods and automotive categories have been heavily weighed down by the complexity of modern digital advertising. The exponential explosion of channel-specific content demands previously forced marketing organizations to spend their days "project managing instead of marketing." As a result, these sectors have rapidly moved past exploratory experiments into what ter Haar calls "post-pilot activities," deploying AI workflows at genuine enterprise scale.

[02:36](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=156s) Case Study: Scaling Global Performance Ads for Google Chrome

Croom walks through a concrete implementation from his team at Google, working alongside Monks to market the Chrome browser and its new Gemini features across 50 to 60 global locales. Operating with a lean team primarily based in San Francisco, the challenge was to create hyper-targeted performance ads across dozens of languages—where certain individual countries require over 10 distinct local language variations.

```mermaid flowchart TD accTitle: Chrome Performance Ad Production Pipeline A["Gemini Feature Launch"] --> B["AI-Assisted Concepting\n(Personas, Use Cases, Prompts)"] B --> C["Human-Crafted Master Template"] C --> D["AI Variant Generation\n(Micro-segments e.g. US Racket Sports)"] D --> E["Automated Localization\n(40+ Locales with Cultural Nuance)"] E --> F["Performance Feedback Loop\n(Continuous Iteration)"] ```

The workflow utilizes AI during the early concepting phase to brainstorm personas, use cases, and prompt structures. The team then crafts an "ideal ad" master template with heavy human oversight. From this master template, AI generates vast variant matrices tailored to specific micro-segments (such as US-based racket sports enthusiasts). The system then localizes the creative across 40 locales, adjusting cultural references (such as recognizing where tennis lacks popularity) before pushing assets live and feeding performance data back into the loop.

- **Turnaround time reduction**: The production cycle shrank from approximately 30 business days (which involved tedious back-and-forth emails with overseas linguists, localized research, and manual proofreading) down to just 10 business days during the pilot—a nearly 70% decrease in turnaround time. - **Direct cost savings**: Production costs dropped by roughly 27%, though Croom notes financial savings are secondary to the massive advantage in deployment speed. - **Combatting ad fatigue**: Rather than relying purely on aggregate performance lift, the team uses production velocity to "daisy chain" the initial two-week novelty performance bump that new ads experience before audience ad blindness sets in.

[06:06](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=366s) Transforming Agency Culture from Preferences to Processes

Reflecting on Monks' journey since early 2023, ter Haar acknowledges that adopting AI across a creative agency was far harder than simply distributing software licenses and conducting training sessions. The agency had to fundamentally overhaul its operating philosophy, pivoting from a culture driven by subjective creative preferences to one governed by rigorous, structured processes—a transition that produced significant "scar tissue" within a legacy creative business.

The strategic goal is delivering "work that people love and that machines prefer." While generative tools enable micro-targeting and hyper-personalization for consumers, the sheer volume and refresh rate satisfy algorithmic distribution engines. On the client side, Croom emphasizes that while zero-marketer startups can afford to move recklessly, enterprise giants face severe brand, compliance, and commercial risks. If a startup fails, "you just change a company's name and start again," whereas a flagship product like Chrome demands stringent enterprise safeguards to prevent low-quality "slop" from reaching audiences.

[09:35](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=575s) Upstream Human Intuition Meets Downstream Agentic Context

When deciding where to inject AI into the creative lifecycle, ter Haar advocates for a non-nostalgic stance, noting that technology can assist across the entire workflow as long as skilled operators remain in the lead. A primary operational advantage is an AI agent's ability to maintain vast contextual awareness. While human creative teams notoriously overlook lengthy briefs and "never" read the accompanying 60-page background documents, AI systems digest full documentation instantly, allowing human creatives to apply their intuition, taste, and emotional instincts ("fingerspitzengefühl") on top of complete context.

Croom adds that Google maintains a clear division of labor:

- **Upstream strategic ideation**: Big brand platforms, core messaging, and creative concepts remain anchored in human intuition. - **Downstream production**: Generation, variant rendering, and asset resizing are heavily automated. - **Cross-campaign verification**: AI is deployed to audit live creative against roughly 20 other active Google brand campaigns, ensuring distinct products avoid conflicting messaging or redundant visual examples.

[11:52](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=712s) Navigating "Death by a Thousand Pilots" and Organizational Chaos

Addressing the common trap of pilot proliferation, ter Haar notes that unbounded experimentation eventually hits diminishing returns because underlying AI models collapse disparate tasks into nearly identical workflows. Monks shifted from an uncoordinated explosion of pilots into a focused framework centered around five core operational capabilities. This consolidation avoids internal corporate politics, where competing teams develop duplicate tools and become protective over their specific pilots.

```mermaid flowchart TD accTitle: Evolution of Pilot Strategy subgraph Phase1 ["Early Stage: Fragmentation"] P1["Pilot A"] P2["Pilot B"] P3["Pilot C"] P4["Pilot D"] end subgraph Phase2 ["Current Operating Balance"] C1["Standardized Core\n(Top 5 Mission-Critical Capabilities)"] C2["Controlled Innovation Chaos\n(Use-Case Specific Experiments)"] end Phase1 --> Phase2 ```

Croom notes that managing this transition requires careful cultural handling. Forcing premature standardization can demoralize teams who have poured effort into specialized tooling just because an alternative is marginally faster or cheaper. For now, Google balances standardized production engines with a degree of "exciting innovative chaos," permitting multiple overlapping tools when tailored to specific team needs.

[14:13](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=853s) The Evolution of Agency Value and the Rise of AI-Native Talent

Revisiting his past statement that 65% of agency tasks could be automated, ter Haar reflects on what clients actually pay for. Speed must not come at the expense of quality, but more importantly, agency fees often cover navigating enterprise complexity rather than raw production hours. He shares an anecdote where Monks compressed a 6-to-7-month global campaign into an astonishing 12-week sprint to inspire a client organization, only for the actual deployment to take the standard timeline because the client's internal approval infrastructure could not digest work at that velocity.

Croom concludes with an observation on marketing talent dynamics. Value is coalescing around two distinct profiles: younger, AI-native talent who intuitively manipulate modern tools, and seasoned domain experts who supply critical strategic judgment. While many enterprises attempt to in-house the raw technology and hire junior AI operators, pairing deep agency domain expertise with AI-accelerated execution creates an exponentially higher impact per person.

[16:55](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1015s) Marketing for Algorithms: From Human Consumers to LLMs

The panel touches on how domain expertise in brand management, creative vision, and messaging remains essential within marketing teams and agencies alike, before shifting focus to a fundamental industry change: marketing directly to algorithms rather than just human consumers. Kristin Stoller shares a personal anecdote from moving apartments over the weekend, recalling how she turned to Claude to ask for the fastest kitchen table delivery that could arrive in under three weeks. When she notes that the results were not particularly impressive, James Croom playfully interjects that she should have tried Gemini instead.

Kristin uses this experience to prompt a deeper question on how marketing strategies and creative workflows must evolve now that AI systems mediate consumer discovery. The shift forces marketers to reconsider how ads, content, and brand presence are structured when the immediate "audience" evaluating recommendations is a large language model.

```mermaid flowchart TD accTitle: Marketing Shift from Human Audiences to Agentic Discovery subgraph HumanInput["User Query & Discovery"] U["Consumer Prompt / Search"] end

subgraph LLMProcessing["AI / Agent Layer"] LLM["Large Language Model / AI Agent"] KB["Web Knowledge Bases & PR Archives"] SOC["Open Social Channels (Reddit, X, YouTube)"] end

subgraph BrandOutput["Brand Action & Delivery"] OUT["Synthesized Recommendation to User"] end

U --> LLM KB --> LLM SOC --> LLM LLM --> OUT ```

---

[17:38](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1058s) Social Freshness, LLM Retrieval, and Category Challenges

Addressing the mechanics of AI discovery, James Croom highlights that large language models rely heavily on fresh, real-time data, which is primarily harvested from open social channels and active public discussions. While clarifying that he is speaking from general industry observation rather than insider Google intelligence, he emphasizes that social presence across platforms like X, Reddit, and YouTube has become massively more important. Ensuring that consumers and creators are actively discussing a brand across these networks directly influences how models surface and cite products.

However, James points out a sharp disparity across product categories. While tech products like Google Chrome are naturally discussed, debated, and recommended across digital communities, non-conversational categories face a steeper hurdle. Consumers rarely go online to casually recommend everyday items or sensitive purchases, such as prescription medication, to their peers. For brands operating in industries with minimal organic social banter, the urgent mandate is determining creative ways to establish relevance and ignite genuine discussions in open forums where algorithms listen.

- **Primary Data Sources for AI Freshness:** Active discussions on open platforms like Reddit, X (formerly Twitter), and YouTube serve as key signals for LLMs needing up-to-date information. - **The Category Gap:** Tech products benefit from high organic discussion volume, whereas brands in low-chatter categories (e.g., healthcare, pharmaceuticals, standard utility goods) must engineer new strategies to remain relevant in AI-indexed spaces.

---

[18:38](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1118s) Websites as Agent Knowledge Bases and Brand Identity in an Agentic Landscape

Wesley ter Haar expands on how digital ecosystems must adapt, noting that brand websites are effectively transforming into "a knowledge base for agents." Historically, companies rarely purged legacy web pages, preserving old material indefinitely for posterity. Today, however, outdated PR announcements and deprecated messaging can be ingested by LLMs and reflected in generated answers. While many in the industry anticipated that content strategy and strategic copywriting would be among the first disciplines automated away, Wesley argues the exact opposite: the need to curate, prune, and structure accurate, agent-ready knowledge bases completely reframes and elevates the importance of content strategy.

Looking toward the broader future, Wesley reflects on how traditional brand real estate is collapsing directly into agentic interfaces. This raises profound questions about the nature of branding itself. As autonomous agents increasingly make decisions grounded in economic logic and utility rather than emotional persuasion, marketers must fundamentally rethink how a brand creates value when the consumer's emotional connection is mediated by a rational algorithm. Kristin Stoller closes the fireside session by thanking the panelists for their grounded, practical insights on operating AI at enterprise scale.

- **Knowledge Base Hygiene:** Brands must actively manage their web footprints to prevent LLMs from retrieving and hallucinating outdated PR messages or obsolete product information. - **Reframing Content Strategy:** Strategic copywriting and content architecture are shifting from one-off creative outputs to continuous knowledge base maintenance for autonomous agents. - **Rational Agents vs. Emotional Branding:** Traditional brand equity relies on emotional resonance, whereas AI agents prioritize "economically logical decisions," requiring a deep strategic pivot for brand marketers.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=4s) quite paradoxical situation — Wesley ter Haar

> [01:28](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=88s) a little bit of dog fooding or drinking your own champagne — Wesley ter Haar

> [01:28](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=88s) project managing instead of marketing — Wesley ter Haar

> [01:28](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=88s) post-pilot activities — Wesley ter Haar

> [06:06](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=366s) work that people love and that machines prefer — Wesley ter Haar

> [06:06](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=366s) you just change a company's name and start again — James Croom

> [09:35](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=575s) fingerspitzengefühl — Wesley ter Haar

> [11:52](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=712s) death by a thousand pilots — Wesley ter Haar

> [11:52](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=712s) exciting innovative chaos — James Croom

> [18:38](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1118s) a knowledge base for agents — Wesley ter Haar

> [18:38](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1118s) economically logical decisions — Wesley ter Haar

## Mentioned Wesley ter Haar, James Croom, Kristin Stoller, Monks, Google, Chrome, Gemini, Fortune, Ai4, Claude, Reddit, X, YouTube

## Video description Wesley ter Haar - Co-founder & Chief AI Officer, Monks James Croom - Senior Director Marketing - Chrome, Google Kristin Stoller - Editorial Director, Fortune

In this fireside chat, James Croom and Wesley ter Haar will share practical perspectives on applying AI in performance marketing within a large, globally distributed marketing organization. The discussion will focus on operational learnings from working at scale across multiple markets, teams, and approval environments, with examples anchored in Chrome-specific work.

Topics will include how AI can support faster creative development, more efficient production workflows, and increased agility in refreshing marketing assets. The session will also examine the organizational side of adoption, including the investment required to build new capabilities and adapt ways of working as tools and processes continue to evolve.

Designed as a candid but practical conversation, the session will highlight what teams are learning as they integrate AI into modern marketing operations, while staying focused on grounded use cases and implementation realities rather than broad industry speculation.

[Source](https://www.youtube.com/watch?v=b3PH5D4T3dI)

中文简报

谷歌将Chrome广告制作周期缩短70%，AI瓶颈转向合规审核尽管生成式AI将效果广告制作周期从30天缩短至10天，但因企业法务和审批流程跟不上自动化产出的体量，整体推向市场的进度依然受阻。尽管生成式AI大幅压缩了素材制作周期——将Google Chrome的本地化交付周期从30天压缩至10天——但企业的整体上市速度依然受制于传统的法务与合规审查流程。应对这一转变，企业需要从混乱、零散的试点阶段转向标准化的运营能力，同时在上游战略愿景和情感直觉层面保持人工把关。此外，AI主导的内容发现模式的兴起，正迫使营销人员重新审视数字化战略，将品牌网站打造为供自主智能体调用的规范知识库，并在开放的社交平台上提高对话的相关性。尽管 AI 显著加快了创意资产的生成速度，但大型企业的整体上市周期仍受制于法务和合规等传统审查流程，呈现出“制作提速但推进滞后”的矛盾现象。 (0:04) Google Chrome 与 Monks 合作的全球效果营销工作流将制作周期缩短了近 70%（从 30 个工作日减至 10 天），直接制作成本降低约 27%，并通过高频产出应对广告疲劳。 (2:36) 在大型代理商推行 AI 需将企业文化从主观创意偏好转向严密的结构化流程，企业级品牌更需建立严格的安全护栏以防范低质劣造内容。 (6:06) 团队应明确人机分工：AI 擅长全面吸收大型简报并批量生成下游素材变体，人类营销人员则在上游专注于战略洞察、情感直觉与跨战役审核。 (9:35) 随着 AI Agent 逐渐介入消费决策，品牌官网正演变为 Agent 的知识库，促使营销重心转向清理维护实时知识库并提升开放社交平台的讨论度。 (18:38)

Reading summary

While generative AI drastically compresses creative production timelines, enterprise adoption remains constrained by legacy legal reviews, organizational inertia, and the emerging need to optimize content for algorithms rather than human consumers alone, according to Google's James Croom and Monks' Wesley ter Haar. The Production Paradox and Global Scale (0:04) Artificial intelligence has sharply reduced the manual labor required to turn initial concepts into channel-ready assets, yet overall go-to-market speed remains stalled by enterprise compliance, legal reviews, and sign-off workflows. Wesley ter Haar describes this as an operational paradox where the bottleneck has merely shifted downstream. High-volume asset creation outpaces traditional corporate approval mechanisms, meaning assets generated almost instantly still wait weeks for commercial release. In a concrete deployment, Google and Monks marketed Chrome's Gemini features across 50 to 60 global locales using a centralized, lean team. AI assisted in early concepting and localized master templates across 40 target regions, adjusting for cultural nuances such as regional sports preferences. This workflow reduced turnaround times from 30 business days to 10—a nearly 70% decrease—while trimming production costs by roughly 27%. James Croom noted that the primary commercial benefit was not direct cost savings, but deployment velocity, enabling teams to continuously refresh creative to counter audience ad fatigue and exploit the initial two-week engagement lift of new ads. Balancing Human Intuition with Agentic Automation (9:35) Transitioning creative organizations into AI workflows requires moving from subjective personal preferences to structured processes, an evolution that ter Haar noted created substantial internal organizational friction. To prevent brand degradation or low-quality output, Google maintains a clear operational division: upstream strategic ideation, core messaging, and creative concepts remain anchored in human judgment, while downstream tasks like resizing, variant generation, and localization are automated. AI is also used for cross-campaign compliance, cross-referencing live assets against roughly 20 other Google campaigns to prevent conflicting messaging. Both speakers caution against uncoordinated experimentation, commonly termed 'death by a thousand pilots.' As foundational models improve, disparate specialized tools often collapse into similar capabilities. Monks addressed pilot fatigue by consolidating exploratory projects into five core operational capabilities. Croom explained that Google balances standardized production engines with controlled experimentation, warning that enforcing rigid standardization too early can demoralize teams that developed specialized custom tools. Agency Value and Talent Bifurcation (14:13) As execution becomes commoditized, agency commercial value is shifting away from raw billable hours toward navigating enterprise operational friction. Ter Haar noted an instance where an agency compressed a seven-month global campaign into a 12-week sprint, yet actual deployment still followed standard enterprise timelines because the client's internal approval infrastructure could not digest the volume at that pace. Internally, marketing talent is bifurcating into two high-value roles: junior, AI-native operators who intuitively manipulate tools, and experienced domain specialists who supply essential brand governance and strategic taste. Croom observed that while many enterprises attempt to simply bring raw tools in-house with junior personnel, combining experienced strategic guidance with AI-accelerated workflows delivers substantially greater output per person. Optimizing Content for Agentic Discovery (17:38) As consumers increasingly use AI models for discovery and product evaluation, marketing strategies must adapt to appeal directly to algorithms. Croom observed that large language models rely on fresh data drawn heavily from open social platforms like Reddit, X, and YouTube. While conversational sectors like technology naturally generate organic discussions, non-conversational categories—such as healthcare or utilities—face greater difficulty establishing relevance within AI retrieval indexes. Concurrently, corporate websites are transforming from static archives into knowledge bases for autonomous agents. Ter Haar argued that leaving outdated press releases or deprecated product details online risks algorithmic ingestion and distorted AI answers. Consequently, content strategy and copywriting are shifting from producing one-off creative copy to actively curating and maintaining accurate brand data for rational, utility-driven agent evaluation. A persistent tension remains between the speed of algorithmic generation and the speed of human governance. As decision-making shifts from emotionally driven consumer choice to utility-focused agentic discovery, brands must resolve internal compliance bottlenecks and restructure their digital footprints for machine evaluation.

精炼阅读摘要

James Croom与Wesley ter Haar指出，人工智能已显著压缩营销素材的制作周期，但企业法务与风控审批的滞后使整体交付陷入瓶颈；在规模化落地中，营销核心正转向“以人类直觉把握上游战略、以算法推进下游变体生成”，并重构面向大语言模型（LLM）检索的内容架构。生产提速与企业治理滞后的矛盾 (0:04) James Croom与Wesley ter Haar指出，截至2026年中期，AI在营销领域最明确的突破在于内容生产提速，将洞察转化为渠道创意素材的繁复手工大幅压缩。然而，整体产品上市（GTM）周期并未实现同等加速。主要瓶颈转移到了下游的企业级治理流程，包括法务合规审核、风险评估以及多方利益相关者确认，传统的人工审批速度完全无法跟上爆炸式增长的AI资产体量。 ter Haar将这一现象描述为“自相矛盾的困境”：技术理论上支持即时交付，但实际落地仍因审批机制而停摆。此外，Croom强调，零营销人员的初创公司可以不顾合规风险敏捷试错，但大型企业必须部署严格的安全防护机制，防止低劣生成物损害核心品牌声誉与商业利益，这使得大型机构在释放AI全速潜力时面临天然的治理阻力。 Chrome案例：素材生成与降本增效 (2:36) 在推广Google Chrome及其Gemini新功能的实践中，Croom团队与Monks在50至60个全球市场落地了AI驱动的效果广告流程。该工作流先由AI辅助进行画像构建、场景构思和提示词设计，再由人工监督打造“理想广告”主模板；随后系统基于该模板生成海量针对细分微客群（如美国拍类运动爱好者）的变体矩阵，并自动化适配逾40个地区的本土化语言与文化语境，最后将投放数据反馈至迭代回路。在这一试点项目中，端到端制作周期从原先约30个工作日（涉及与海外语言学家的繁复沟通和人工校对）骤降至10个工作日，缩短了近70%；直接制作成本削减了约27%。Croom指出，财务节省仅是次要收获，更深层的商业价值在于极速部署能力——通过高频轮换素材“串联”新广告前两周的受众新奇度红利，有效破解了数字营销中普遍存在的受众广告疲劳问题。组织文化重塑、人机分工与试点收敛 (6:06) ter Haar回顾了Monks自2023年以来的转型历程，指出在创意代理机构推行AI并非单纯采购工具，而是迫使机构从“主观创意偏好”转向“严密结构化流程”的深度哲学重构。在人机分工上，团队强调发挥AI对海量文档背景信息的即时解析能力，让资深创意人员在此基础上专注注入直觉、品味与情感判断。Google的分工同样清晰：上游宏观品牌平台与核心构思依托人类直觉，下游素材变体与尺寸缩放交由自动化，并引入AI对约20个并行品牌活动进行交叉审核，防范信息冲突。针对企业普遍陷入的“过度试点之死”与内部政治消耗，ter Haar指出，由于底层大模型会使不同任务流程趋同，无序试验会迅速遭遇边际效应递减，Monks据此将发散的试点收敛为5项核心能力标准。Croom则补充，过早强推标准化可能挫伤团队积极性，因此当前Google在标准生产引擎与受控的“创新混乱”之间维持平衡，同时将初级AI原生人才与资深行业专家深度绑定，以实现单兵产出的指数级放大。面向LLM检索与知识库的内容重构 (16:55) 随着AI代理介入消费者的探索与购买决策，营销对象正在从纯人类受众扩展至大语言模型。Croom基于行业观察指出，LLM极度依赖实时且新鲜的数据，而这些数据主要抓取自Reddit、X（前Twitter）和YouTube等开放社交平台上的活跃讨论。这导致品牌在开放社交网络上的声量，直接决定了模型是否及如何推荐该品牌。然而行业间存在明显分化：科技产品天然具备高讨论度，而处方药或常规消耗品等低社交属性类别，则面临更难在公开算法索引中激发真实讨论的严峻挑战。 ter Haar进一步提出，品牌官网正在演变为“面向AI代理的知识库”。过去企业往往永久保留陈旧网页，但如今过期的公关通稿与旧产品信息会被LLM抓取并生成错误推荐，这使得知识库的清理、剪枝与结构化维护变得至关重要，内容策略与战略文案撰写的战略地位不降反升。未来传统品牌阵地将进一步融入代理界面，当理性的大模型基于经济效益和实用逻辑做出中介决策时，传统的感性品牌塑造逻辑必须迎来底层变革。本次研讨揭示了AI在大型营销体系中从局部试验走向规模化应用的深层张力：虽然内容产出与变体定制已实现工业化提速，但企业法务合规审批链条的滞后、算法中介化的消费决策逻辑，以及对理性决策代理的品牌价值传达，仍构成尚未完全解决的长期挑战。

Detailed notes — zh

## Background 该视频由 YouTube 频道 Ai4 于 2026 年 9 月 4 日发布。视频内容为一场炉边对话，由《财富》杂志编辑总监 Kristin Stoller 主持，嘉宾为 Monks 联合创始人兼首席 AI 官 Wesley ter Haar 与 Google Chrome 浏览器营销高级总监 James Croom，重点围绕大型全球化营销组织在效果营销中规模化应用 AI 的实操经验、生产流程与组织变革展开深入探讨。 ## TL;DR 对话深入剖析了 AI 在现代营销中“生产极度提速”与“下游审核与落地滞后”的核心矛盾，指出企业正从概念验证迈向解决实际业务压力的后试点时代。James Croom 结合 Chrome 浏览器全球效果营销的落地实战，展示了借助 AI 模板化生成和微细分本地化实现制作周期缩短 70%、成本降低 27% 并有效缓解广告疲劳的经验。Wesley ter Haar 则从代理机构视角强调，企业必须从“偏好驱动”转向“流程驱动”，利用 AI 承载海量上下文，释放人类直觉与审美突破。此外，双方探讨了在标准化与局部创新混沌间的权衡、AI 原生人才与行业专家的协作，以及品牌未来如何面向大模型与理性 AI Agent 重构内容资产与知识库策略。 ## Atmosphere & style 整场对谈氛围务实、坦诚且充满行业一线实战的锐气与幽默，彻底摒弃了技术乌托邦式的空洞宣讲。James 的沉稳严谨展现出跨国科技巨头在风控与品牌声誉上的克制，而 Wesley 则言辞犀利风趣，毫无保留地分享了机构转型过程中的“伤疤组织”与试错教训。两位嘉宾在主持人行云流水的引导下抛接自然，在互相调侃（如推介 Gemini）中碰撞出深刻洞见，让整场讨论既有直击组织痛点的清醒审视，又洋溢着拥抱范式转移的探索兴奋感。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=4s) AI 虽极大提升了内容生成速度，但由于下游法务、合规等审核流程无法同步匹配，导致营销资产整体进入市场的时间（Go-to-Market Time）并未成比例缩短。 - [01:28](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=88s) 率先规模化落地 AI 的行业主要为两类：出于验证技术而自用（dogfooding）的科技公司，以及受数字广告复杂度与增长压力逼迫的快消品和汽车等传统行业。 - [02:30](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=150s) Chrome 浏览器的实战表明，通过 AI 辅助构思与微细分模板生成，极精简的团队即可覆盖 50 至 60 个全球市场，制作周期缩短近 70%，成本降低约 27%。 - [02:30](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=150s) 高频快速的 AI 资产生成能力有效解决了受众“广告疲劳”（Ad Fatigue）问题，让团队能够通过持续串联（daisy-chain）不间断捕获上线前两周的高效转化红利。 - [06:05](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=365s) 代理机构与营销团队推进 AI 转型不能仅靠发放工具，而必须从高度依赖个人主观审美的“偏好驱动”彻底转向标准化的“流程驱动”。 - [06:05](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=365s) 成熟大品牌与初创团队在 AI 营销上的核心区别在于风控体系，大企业必须在高速生成管线上叠加严密机制，杜绝粗劣的“AI 垃圾内容”（slop）以保护品牌资产。 - [09:47](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=587s) 人机协同实现新分工：AI 擅长承载和记忆全量庞杂的背景上下文，人类专家则专注于直觉判断、审美把关与突破性创意洞察。 - [09:47](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=587s) 在拥有众多子品牌的大型企业中，AI 能极其高效地进行跨品牌多活动交叉审核，防止素材重复或传播信息相互矛盾。 - [12:14](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=734s) 企业应警惕“千百试点之死”（death by a thousand pilots），避免各部门重复造轮子与领地内耗，应将资源收敛聚焦于少数核心能力，同时在局部保留适度的创新探索混沌。 - [14:13](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=853s) 随着执行工时被 AI 压缩，代理商的核心商业价值正转向协助客户管理复杂的企业内部流程与决策节奏。 - [16:06](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=966s) 企业最具竞争力的人才结构是将精通大模型的“AI 原生代”与具备深厚策略和品牌把控力的“资深行业专家”相结合，由 AI 放大行业专家的商业影响力。 - [17:09](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1029s) 营销对象正在由“人类”拓展至“算法与 Agent”，Reddit、X 和 YouTube 等开放社交平台上的真实讨论已成为大模型抓取信息并生成消费推荐的关键源头。 - [17:09](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1029s) 企业官网的角色正在重构为面向大模型的“Agent 知识库”，品牌需系统性清理与维护动态内容，以适应未来基于经济理性与逻辑对比的 AI Agent 决策场景。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=4s) AI 赋能营销的现状与瓶颈：生产提速与审核滞后的矛盾圆桌对话在 2026 年年中展开。主持人 Kristin Stoller（《财富》杂志编辑总监）开场指出，过去至少三四年间，整个行业都在宣扬“AI 将彻底颠覆营销”。她向两位嘉宾抛出第一个问题：在各自的组织中，有哪件事因 AI 而显著提速，又有哪件事依然原地踏步？ Google Chrome 浏览器营销高级总监 James Croom 指出，衡量提速最明显的环节无疑是内容制作（content production），团队现在能以极高的频率快速产出物料。然而，整体的“进入市场时间”（Go-to-Market Time）并没有如愿大幅缩短。核心原因在于下游的审核流程完全跟不上激增的内容量——法务合规（legal and compliance reviews）等审核环节虽然有所改进，但效率远未达到匹配生成速度的水平。 Monks 联合创始人兼首席 AI 官 Wesley ter Haar 对此深有同感。他表示，团队已经成功压缩了从“挖掘洞察”到“制作渠道营销资产”之间的绝大部分人工操作。然而瓶颈发生了转移：“所有事情都能做得快得多，但实际落地却没有快得多”，行业陷入了一种极为矛盾的困境，审批与法务流程的滞后拖慢了最终的上线节奏。 --- ### [01:28](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=88s) 行业落地现状：从被迫“尝鲜”到高压行业的规模化落地 Wesley 结合 Monks 跨越科技、零售、娱乐、汽车等多个行业的经验，梳理了目前 AI 在营销领域落地最成功的场景与整体行业格局。他认为，率先推进 AI 营销的行业主要分为两类：第一类是“不得不先走一步”的科技公司，这既是出于内部试用自身技术（dogfooding / drinking your own champagne）的需要，也是为了验证产品；第二类则是面临严峻财务压力与增长压力的传统行业，例如快消（CPG / FMCG）以及部分汽车制造业。 Wesley 分析，数字广告的极度复杂化正在拖垮这些传统企业——爆炸式增长的内容需求导致大量营销团队将精力消耗在繁琐的“项目管理”上，而非真正的“做营销”。因此，这些受压迫最深的行业正在迅速跨过概念验证阶段，全面进入“后试点时代”（post-pilot），开始在真实业务中规模化运用 AI。 --- ### [02:30](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=150s) Chrome 浏览器的实战案例：AI 驱动的全球化精准效果营销 James 带来了 Google 内部利用 AI 推进 Chrome 效果营销（Performance Marketing）的真实案例。该项目的核心目标是在全球范围内让用户理解 Chrome 的新功能（如最新集成的 Gemini 功能）并促成下载使用。项目面临的挑战是：常驻旧金山的一个精简小团队，需要覆盖全球 50 至 60 个市场，处理海量语种（部分单一国家甚至涉及 10 种以上语言）。 ```mermaid flowchart TD accTitle: Chrome 浏览器 AI 效果营销工作流 accDescription: 展示从功能洞察与模板构建到多变量生成、本地化及数据反馈迭代的全流程。 subgraph Ideation ["1. 概念与模板构建"] A["新功能输入 (Gemini 特性)"] --> B["AI 辅助构思 (用户画像 / 使用场景 / Prompt)"] B --> C["高标准主模板制作 (高度人工把关与审美控制)"] end subgraph Production ["2. 批量生成与本地化"] C --> D["微细分变体生成 (如美国球拍运动爱好者)"] D --> E["跨语种智能本地化 (40+ 区域 / 剔除文化不相关元素)"] end subgraph Delivery ["3. 投放与效果迭代"] E --> F["渠道分发与投放"] F --> G["捕获前2周新广告红利 (New Ad Bump)"] G --> H["性能数据回流 (Daisy-chain 持续轮换)"] H -.-> B end ``` 在与 Monks 的合作中，团队在概念构思阶段引入 AI 探索目标画像、应用场景与提示词（prompts），随后由人工严格把关并设计出理想的广告母版（Template）。依托该模板，团队能够批量生成面向特定微细分人群（micro-segments）的定制视频与图像变体（例如专门针对美国网球/匹克球等球拍运动爱好者的定制广告）。随后，系统将内容本地化至 40 多个区域市场，并结合当地文化习惯自动调整内容（例如在网球不普及的国家替换对应运动）。 - **制作周期大幅缩短**：单轮资产从立项到上线，由以往的约 30 个工作日压缩至试点期间的 10 个工作日，时间缩短近 70%。以往周期冗长是因为需要反复与印度等当地母语员工邮件校对沟通并重做调研。 - **制作成本显著下降**：整体成本降低了约 27%。 - **解决广告疲劳（Ad Fatigue）**：效果营销中通常存在上线前两周效果极佳、随后受众产生“广告盲区”（Ad Blindness）导致转化下滑的现象。高频快速的制作能力让团队能够将这“前两周的红利期”持续串联（daisy chain），实现持续高效的增长拉动。 --- ### [06:05](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=365s) 组织变革与规模化挑战：从“偏好驱动”到“流程驱动”与品牌风险控制 Wesley 分享了代理机构在推动 AI 转型过程中的挫折与经验。Monks 早在 2023 年 2 月就全面拥抱 AI，但最初误以为转型仅仅是“提供工具权限、培训与团队赋能”。实际操作中发现，要真正将生产力收益转化为客户价值极其困难，甚至给团队带来了不少“伤疤组织”（scar tissue）。作为创意机构，Monks 必须从过去高度依赖个人主观偏好的“偏好驱动”（preference-heavy）模式，彻底转向标准化的“流程驱动”（process-heavy）模式。Wesley 总结，现代营销的终极目标是打造“让人类喜爱、让机器算法偏好”（work that people love and that machines prefer）的内容。 ```mermaid flowchart LR accTitle: 初创团队与大型企业在 AI 营销应用上的对比 accDescription: 对比零营销人初创公司的高风险承受力与大型企业严格的风控机制。 subgraph Startup ["初创企业 (Zero-Marketer Startups)"] S1["单人/无专业背景主导"] --> S2["极高生成速度"] S2 --> S3["试错成本低 (失败可直接更名重来)"] end subgraph Enterprise ["大型成熟品牌 (如 Google Chrome)"] E1["跨国多层级协作"] --> E2["企业级风控体系 (商业 / 合规 / 品牌声誉)"] E2 --> E3["杜绝劣质内容 (Anti-Slop) (维护全球数十亿用户资产)"] end ``` James 随后指出了成熟大企业与当前行业热议的“零营销人初创团队”（zero-marketer startups）之间的本质差异。小公司如果尝试激进 AI 营销导致翻车，大不了“换个公司名字重新开始”；但对于 Chrome 这样拥有全球海量用户的成熟品牌，绝不容许出现任何偏差。大企业应用 AI 面临的最大阻力在于风险防范（risk mitigation），包括商业风险、合规风险与品牌形象风险。团队必须建立严密的企业级监管机制，确保推向市场的每一份内容都能真正取悦用户，而不是沦为泛滥粗劣的“AI 垃圾内容”（slop）。如何将企业现有的严格审核机制，平滑地叠加在高速运转的 AI 生成管线上，是当前最核心的落地挑战。 ### [09:47](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=587s) 摆脱怀旧情绪：让 AI 承载上下文，人类专注直觉突破面对技术带来的行业巨变，Wesley ter Haar 强调团队必须保持“非怀旧”（non-nostalgic）的心态。Monks 已经提供创意服务超过 25 年，从业者很容易陷入“过去我们一直这么做”的经验主义惯性中。尽管 AI 技术进步迅速且几乎无处不在，但最终产出的质量高度取决于工作流顶端负责主导的人（person in the loop / in the lead）。无论优秀的策略师还是创意人员，都在借助 AI 更快地产出更具深度的方案，而 AI 最核心的价值之一正是解决创意流程中的“上下文断层”。 Wesley 幽默地调侃了创意人员的工作习惯：如果给创意团队发一份简报（brief），他们能把简报正文读完就谢天谢地了，更别提附带的 60 页背景文档——“100% 没人会看”。而 AI Agent 最擅长的恰恰是完整记忆并消化所有上下文信息。当 AI 牢牢稳固上下文，人类专家便能释放精力，专注于发挥人类独有的经验、直觉与“指尖感觉”（德语词汇 *Fingerspitzengefühl*，意指敏锐的下意识洞察），去抓住那些让人起鸡皮疙瘩、真正带来突破性创意的灵感瞬间。 ```mermaid flowchart LR accTitle: 人机协同创意工作流 subgraph AI["AI Agent 核心职责"] A1["消化 60+ 页背景文档"] A2["全量承载上下文与细节"] A3["加速执行与原型生成"] end subgraph Human["人类专家核心价值"] H1["直觉与审美洞察"] H2["情绪共鸣 (Goosebumps)"] H3["突破性战略决策"] end AI --> B["人机协同产出"] Human --> B ``` - **人类与 AI 的分工**：AI 负责底层庞杂文档的上下文记忆与自动化执行；人类负责顶层大想法（big idea）、品牌核心主张与审美把关。 - **全生态交叉审核**：James Croom 指出，Google 内部常年在市场上有约 20 个品牌并行运作。AI 极其擅长交叉审核所有品牌的活动，防止出现相互矛盾的信息或重复的营销素材，免去了营销人员最繁琐低效的筛查工作。 --- ### [12:14](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=734s) 告别“千百试点之死”：在标准化与创新混沌之间权衡当主持人 Kristin Stoller 提到业内常有企业“同时跑着 76 个 AI 试点项目”时，Wesley 直言许多公司最初都会经历“死于千百试点”（death by a thousand pilots）的阶段。由于 AI 技术具备将多种碎片化任务整合进单一工作流的能力，如果缺乏统筹，不同团队很容易各自重复造轮子。Monks 最初经历了一段激进的创新大爆发，随后果断转向结构化收敛，将全公司的核心精力与资源聚焦在最关键的 5 项核心能力上。Wesley 提醒，试点过多的另一大隐患在于容易卷入企业内部的部门政治与领地争夺，六个团队都在做类似的东西，最终谁也不愿放弃自己的试点。 James 则从 Google 的实践出发，阐述了“何时该标准化推进”与“何时该保留局部混乱”之间的微妙平衡。如果团队耗费大量精力为某个试点搭建了自研工具，随后公司仅因为另一个方案“稍微更好、更快或更便宜”就强推全员标准化，往往会极大打击一线团队的积极性。因此在现阶段，Google 很多领域仍处于“快速试错、探索技术边界”的阶段，允许针对不同具体场景存在多个大致相似但各有侧重的工具版本，在保持一定“令人兴奋的创新混沌”（exciting innovative chaos）中持续收获业务突破。 --- ### [14:13](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=853s) 压缩人工耗时之后：代理商真正收费的核心是“管理企业节奏” 针对此前提出的“代理商 65% 的收费任务可被 AI 取代”这一言论，Wesley 进行了深入澄清。他指出，执行层面的手动工时确实被大幅压缩，速度本身就是一种竞争力，但绝不能以牺牲质量为代价。更重要的是，在实际大规模落地中，代理商真正获得报酬的核心价值往往在于“帮客户管理复杂的企业内部流程”。 Wesley 分享了一个真实的合作案例：团队曾为某家大型跨国企业（非 Google）打造一套全球营销战役，原本按传统方式需要耗时 6 到 7 个月的工作，团队借助 AI 仅用 12 周便完成了全流程端到端的交付，本意是想激发客户内部拥抱 AI 的热情。然而在实际落地推进时，整个项目的最终上线时间依然与传统流程一模一样——因为大型跨国企业的内部审批、协同与决策节奏根本无法在 12 周的超高频节奏下运转。这并不是对客户的批评，而是大型组织的客观现实。由此可见，单纯压缩人工制作耗时，并不能直接消除企业组织运转的结构性瓶颈。 --- ### [16:06](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=966s) AI 原生代与资深行业专家的结合优势在人才与团队结构方面，James 指出当前企业最有价值的两类人才正在形成互补合力： 1. **AI 原生代（AI-native）**：多为职场新人，对大模型与自动化工具具有天然的直觉，能够极致挖掘 AI 的生产力上限。 2. **资深行业专家（Domain Experts）**：可能尚未完全精通 AI 提示词与工具底层，但拥有深厚的品牌管理、核心文案设计、创意把控与营销策略积淀。 James 特别为传统代理商从业者注入信心：“如果你来自代理商，完全不需要急着转行。”许多大公司试图通过招聘精通 AI 的新人将技术能力全面内化（in-house），但这并不能替代真正的行业专业壁垒。当专业经验得到 AI 速度与生产力的放大时，单个人才释放的商业影响力反而会呈指数级增长。 --- ### [17:09](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1029s) 为算法而营销：面向大模型与 Agent 生态重塑品牌资产讨论转向了一个前沿议题：当消费者越来越频繁地向 AI（如 Claude 或 Gemini）寻求消费建议时，营销人员不仅是在面向“人类”营销，更是在面向“算法与大模型”营销。主持人分享了自己搬家时在 Claude 里询问“3 周内能送达的最快餐桌”的经历，James 顺势幽默插话：“你应该试试 Gemini！” ```mermaid flowchart TD accTitle: 面向大模型与 Agent 的营销生态 subgraph InputChannels["LLM 动态信息抓取源"] S1["Reddit 讨论社区"] S2["X (Twitter) 实时舆论"] S3["YouTube 视频与评论"] end subgraph AgentLogic["Agent 认知与决策层"] L1["LLM / Agent 知识库检索"] L2["逻辑与经济理性决策"] end subgraph BrandAsset["企业资产战略调整"] B1["官网转型为 Agent 知识库"] B2["清理/重构过时 PR 历史沉淀"] B3["低社交属性品类的主动声量构建"] end InputChannels --> L1 BrandAsset --> L1 L1 --> L2 ``` 关于如何应对“算法级营销”的转变，两位嘉宾给出了关键观察： - **社交阵地成为大模型获取新鲜信息的主渠道**：大语言模型高度依赖开放社交网络获取最新资讯。X、Reddit 以及 YouTube 等平台上的真实用户讨论，直接影响大模型给出的推荐结果。对于像 Chrome 这样的科技产品，用户天然会在社交媒体上讨论；但对于药品等缺乏日常自发讨论的品类，品牌必须重新思考如何在公开讨论场中建立相关性。 - **企业官网的角色重构为“Agent 知识库”**：过去企业官网的旧内容、历史 PR 稿件通常为了归档而永久留存，但在 AI 时代，旧信息随时可能被大模型检索并混入回答中。内容策略与战略文案撰写不仅不会消亡，反而需要升级为系统性构建、维护面向 Agent 的动态知识库。 - **理性 Agent 决策下的品牌定义重构**：传统品牌营销高度依赖触发人类消费者的非理性情绪共鸣；然而当未来的购买决策被更多讲求经济理性、逻辑对比的 AI Agent 接管时，品牌如何在冰冷的逻辑计算中保持吸引力与不可替代性，是全行业亟待探索的核心课题。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=4s) 所有事情都能做得快得多，但实际落地却没有快得多 — Wesley ter Haar > [06:05](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=365s) 让人类喜爱、让机器算法偏好 — Wesley ter Haar > [09:47](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=587s) 如果给创意团队发一份简报，他们能把简报正文读完就谢天谢地了，更别提附带的 60 页背景文档——100% 没人会看 — Wesley ter Haar > [12:14](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=734s) 死于千百试点 — Wesley ter Haar > [12:14](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=734s) 令人兴奋的创新混沌 — James Croom > [16:06](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=966s) 如果你来自代理商，完全不需要急着转行 — James Croom > [17:09](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=1029s) 你应该试试 Gemini！ — James Croom > [06:05](https://www.youtube.com/watch?v=b3PH5D4T3dI&t=365s) 换个公司名字重新开始 — James Croom ## Mentioned Wesley ter Haar, James Croom, Kristin Stoller, Monks, Google, Google Chrome, Gemini, Fortune, Claude, Reddit, X (Twitter), YouTube ## Video description Wesley ter Haar - Co-founder & Chief AI Officer, Monks James Croom - Senior Director Marketing - Chrome, Google Kristin Stoller - Editorial Director, Fortune In this fireside chat, James Croom and Wesley ter Haar will share practical perspectives on applying AI in performance marketing within a large, globally distributed marketing organization. The discussion will focus on operational learnings from working at scale across multiple markets, teams, and approval environments, with examples anchored in Chrome-specific work. Topics will include how AI can support faster creative development, more efficient production workflows, and increased agility in refreshing marketing assets. The session will also examine the organizational side of adoption, including the investment required to build new capabilities and adapt ways of working as tools and processes continue to evolve. Designed as a candid but practical conversation, the session will highlight what teams are learning as they integrate AI into modern marketing operations, while staying focused on grounded use cases and implementation realities rather than broad industry speculation. [Source](https://www.youtube.com/watch?v=b3PH5D4T3dI)

Source quotes

[0:04] Wesley ter Haar: 所有事情都能做得快得多，但实际落地却没有快得多 [6:05] Wesley ter Haar: 让人类喜爱、让机器算法偏好 [9:47] Wesley ter Haar: 如果给创意团队发一份简报，他们能把简报正文读完就谢天谢地了，更别提附带的 60 页背景文档——100% 没人会看 [12:14] Wesley ter Haar: 死于千百试点 [12:14] James Croom: 令人兴奋的创新混沌 [16:06] James Croom: 如果你来自代理商，完全不需要急着转行 [17:09] James Croom: 你应该试试 Gemini！ [6:05] James Croom: 换个公司名字重新开始

Source quotes

[0:04] Wesley ter Haar: quite paradoxical situation [1:28] Wesley ter Haar: a little bit of dog fooding or drinking your own champagne [1:28] Wesley ter Haar: project managing instead of marketing [1:28] Wesley ter Haar: post-pilot activities [6:06] Wesley ter Haar: work that people love and that machines prefer [6:06] James Croom: you just change a company's name and start again [9:35] Wesley ter Haar: fingerspitzengefühl [11:52] Wesley ter Haar: death by a thousand pilots [11:52] James Croom: exciting innovative chaos [18:38] Wesley ter Haar: a knowledge base for agents [18:38] Wesley ter Haar: economically logical decisions

Original transcript

[0:04] out there. Okay, here we go. All right. Well, thank you all. Uh, welcome and thank you to my [0:08] two panelists for being here with me. So, James and Wes, it's mid 2026. I feel like for at least [0:15] 3 years or more, people have been telling us like AI is going to transform marketing. It's going to [0:20] be totally different. You won't recognize it. So to start, can you walk me through name one thing [0:26] in your organization that is measurably faster than it was three years ago because of AI and [0:31] one thing that has stayed the same? James, we're going to start with you. Sure. I mean, [0:35] there's tons of stuff. I'd say um the one of the top things of course for is content production, [0:41] uh which we're able to now get out a much greater clip. Uh I'd say get some go to market time isn't [0:45] necessarily as much quicker as we want. Uh, and a lot of that's because of the review [0:49] processes haven't really kept up with the amount of content that we're making. So, legal reviews, [0:53] compliance reviews, etc. Not they've got faster, but they're not anywhere near quick enough. Okay. [0:58] Yeah, very similar. I think we've been able to collapse a lot of the manual effort that it takes [1:03] to go from an insight into in channel assets or marketing. Uh, the bottleneck moves to your point. [1:11] I don't think the process of signoffs, the process of legal checks, the process of reviews has sped [1:18] up at the same level. So you're in this uh quite paradoxical situation where everything can be done [1:23] a lot faster, but it doesn't necessarily happen a lot faster. Yeah, I know we are going to dive [1:28] deep into those bottlenecks today and how to solve them. But first, Wes, can you kind of give us a [1:32] lay of the land because you work across so many industries? You work across tech, retail, [1:37] entertainment, auto. where is AI and marketing actually working the best today and what is your [1:43] view on the landscape? So I think the industries that went first are either industries that have [1:51] to go first like tech companies because there is a little bit of dog fooding or drinking your own [1:56] champagne depending on how you how you think and talk about it. And then we saw it in industries [2:00] that I think uh are just under pressure. Pressure financially and pressure to grow. Uh CPG quite [2:08] clearly, FMCG, little bit of car industry. I think those industries are also being almost [2:15] dragged down through the complexity of digital advertising, right? There's been such an explosion [2:20] of content needs that a lot of those marketing organizations are now project managing instead [2:25] of marketing. So, we've seen a lot of uh what I would call post-pilot activities there where [2:30] people are doing it for real at scale. Excellent. James, I want to go to you because you know I love [2:36] a real life example. Um you brought one here with us today. Can you talk to me about the [2:41] project you're working on at Google right now that is most exciting you and why? Sure. So, [2:46] yeah, I run uh marketing for Chrome browser and I've got a modest playlist I wanted to run [2:50] you through. Hopefully, you'll be able to see some of the ads. So we've been um one of the [2:54] key things we do is performance marketing which is trying to get people at scale to understand [2:58] uh new features and then also download and use uh Chrome browser. So we we operate in a ton of [3:03] different local like 50 60 local around the world, tons of different languages. Some countries have [3:08] like 10 plus languages we're operating in and it's a pretty small team primarily based in uh [3:12] San Francisco who's doing the work. So here's like the actual ads that we we made using this process. [3:17] So working with Wes's crew at Monks and these are some performance ads that we put together. So [3:22] uh we used AI during the concepting phase. Um hey here are all the features that we've got coming [3:26] out. This is for um for uh some new uh Gemini features that we we had launching. Here's some new [3:32] features that we come out come out. What personas might we want to go after? Uh what use cases might [3:36] we want to talk about? What prompts might we do? And then we create a template uh which is like a [3:41] the ideal ad. There's a lot of human oversight in this part, but we create a template and then we [3:46] use that template to generate a ton of different variants. So we can see here there's a video with [3:52] a prompt in it. Um this is all uh you know AI AI generated with some human oversight with [3:56] um with images and then uh we can create that for different personas. So we're able to like make it [4:02] them very very targeted. So making it a ad for people who are into racket sports in the US for [4:08] instance. Practically that wasn't possible in the past but now because you can generate so much more [4:11] content so much more quickly um using uh using AI we're able to make this sort of micro segments. So [4:17] we produce all of this in English. We then um use it to localize in 40 different loces. You know, [4:22] it knows certain we've ensured that you know, tennis is not a popular sport in all countries. [4:26] So, you know, you make sure that it's relevant for the for the right audience. Um and then we [4:30] can get that to market and then get results back and then do it again with new um with with the [4:34] data that we get from um from the performance marketing. And and from the time you start [4:39] like inception of this from from the time you're ready to roll it out, how long is that process? [4:44] So it previously was like about 30 business days uh which is not awesome uh because uh a lot a lot [4:51] of that is like back and forth emailing people in India who speak the language to make sure that you [4:56] got it right doing local doing uh new research for insights um for what might resonate and then [5:00] actually making all the stuff and proof reading everything. Uh we now got this down during the [5:04] pilot down to 10 business days and I I'm confident we can get it much quicker in the end. So it was [5:08] a what is that like a nearly 70% decrease in time. It also cost us a lot less money. Um, [5:15] which cost us about 25 30% less money. I think it was 27% less. That's actually not the most [5:20] exciting part though. Uh, because it doesn't cost that much to make performance ads. What's [5:24] really exciting is the speed you can operate at. That's And what's the reception been like [5:28] in the market to this? So, they work very well. Um, the I we don't yet have data that it is [5:34] uh improved performance in aggregate. It has reduced cost and it's got us to market more [5:39] quickly. Uh what it what it has done though is we get that new you get this bump when you [5:44] a lot of you are probably performance and growth marketers. You get this like new ad bump when you [5:47] get something new into market where it works really well for the first few weeks and then [5:51] you people get ad blind and it performs less well. Uh but we're able to like daisy chain [5:55] those those those first two weeks of performance again and again. Um and so we're expecting it to [6:00] have a lot of um of lift in impact but right now it's primarily speed times marketing cost [6:05] benefits. Very very impressive. Um, and Wes, I want to come to you. You know, I know you and [6:10] your agency definitely had a hand in this as well and and worked really well, but I'm I'm [6:13] also curious for you to share with us maybe one big win that your agency has had recently in with [6:19] using AI and marketing and then conversely any challenges you've had or something that didn't [6:23] land like you thought it would. Everything has gone perfect. No. Uh, you ready? No, to be honest, [6:30] the because we we raised our hand quite early uh at least within our industry um February of 2023, [6:38] uh we said, "Hey, this is clearly going to change everything and we're going to go figure this [6:42] out." I think we initially um thought it was going to be easier. I think we were thinking [6:48] it was going to be a lot of sort of uh access to tooling, training, and and team enablement. [6:53] It's actually extremely difficult to uh capture productivity like productivity is quite a [7:01] uh a vague sort of concept. So for us to get that productivity gain which then translates back into [7:08] um upside for our client, we really had to change our ways of working. We had to move from a company [7:15] that was very um sort of preference heavy to something that was a lot more process heavy, which [7:19] isn't necessarily easy for a creative business. So that that was just a lot of scar tissue. Uh we're [7:26] in a good sport. I think there's still a lot more work to do. The big successes are these types of [7:31] things. I I think in the end when you think about marketing, you're trying to do work that people [7:37] love and that machines prefer. I think this is a great example of being able to target at the micro [7:44] level. Really going after people's uh um sort of interests and what they're enthusiastic about in [7:50] a much more personalized way, which is great, but then to your point, we can do it at such [7:55] speed and scale that the algorithms keep giving us that bump. Right. So for me, it's how to use the [8:00] technology on both sides. Uh and we're seeing lots of massive massive wins there. And Jay, [8:06] I want to talk about the challenges for you too. could you walk me through, you know, [8:10] what what is the biggest challenge you're facing? Is it, you know, maybe something that didn't land [8:14] the way you thought it would or is there any other type of challenge that you think are really hard [8:17] for you right now? So, yeah, we see um at this conference, I've, you know, heard a lot of people [8:22] talking about these like zero marketer startups, right, where it's somebody overseeing marketing [8:26] who doesn't have a necessary deep expertise in it or like a single CMO running, you know, [8:30] a bunch of people's um work. Uh, and I think that works well for for uh small startups. Um I haven't [8:37] seen it yet work for large companies. You know we're a big successful um brand with Chrome and [8:41] you know one of the world's most popular products and I haven't seen quite that yet. Uh the thing [8:46] um that has been the biggest challenge for us I mentioned earlier has been uh basically like [8:50] risk mitigation. It's you know commercial risk um compliance risk and then also brand risk. you [8:56] know, it's a we have big popular brands and we need to do a lot of like oversight to make sure [9:00] that the quality of the work we're putting into market is delighting people and isn't slop. Uh, [9:05] and you know, it's less risk when you're smaller if you like do really if it goes really badly [9:09] wrong, you just change a company's name and start again. Uh, but you can't do that with like a with [9:13] a big Google doesn't allow you to do that. What we had we periodically had a new brand sometimes. Uh, [9:18] but yeah, Chrome has been a long time. Don't want to don't want to stuff that up. So, um I'd say [9:21] that's been the um that's been the hardest bit is like applying enterprise processes on top of this [9:27] like new very fast um new very fast um uh content generation thing. This is for you both. How do you [9:35] decide or evaluate where in the creative process you're going to inject AI? Like at what what [9:41] point and what is the signal or the metric that tells you, okay, we need to do X here or Y here? [9:47] I mean, we we try and be um non-nostalgic. So, it's very easy because we're in year 26, right? [9:58] So, we've we've sold creative services for well on 25 years. So, it's very easy to be nostalgic about [10:06] this is how we do things. Um I personally believe the technology is progressing so quickly and is in [10:12] such a great spot that you can use it everywhere. But there is a huge spectrum in outputs dependent [10:18] on the the person on top of the process, the person in the loop or in the lead. So we have [10:24] great strategists that are using technology. We have great creatives that are using technology and [10:29] they're doing that to get to something more quickly but also get to something that has [10:33] more context. Uh I'm not sure how many creatives they're in the room. I love creatives and creative [10:39] teams. Uh if I send them a brief, I'm happy if they end up reading the whole brief. never [10:44] happens. Never happens. I'm 100% sure they didn't read the additional 60 pages of documentation. [10:50] So the working with agents that are amazing in holding that context, but then still having that [10:55] sort of human gut feeling and expertise and just the the finger speedul that you sometimes have [11:01] where you get a little bit of um goosebumps and you go this is breakthrough like that [11:07] combination has been super strong for us. James, anything to add from the Google perspective? [11:13] Yeah, I mean I'd say we're also finding that the more upstream work is still much more focused on [11:19] human uh intuition. So what is the big idea around the around the creative? What is the big brand [11:25] message? Uh the production is definitely being much more automated and AI is great for that. [11:30] Um and then I said the other like very useful thing is um uh we we're finding it's very good [11:36] at cross-checking against all the other stuff our company is doing. So, you know, we may we [11:40] have potentially like 20 brands in market at any one point and you want to make sure that they're [11:44] not sending conflicting messages or using the same examples and, you know, I think AI is great for [11:48] that. It's a very tedious job for marketers to do and it's something like people are very happy to [11:52] outsource. Yeah. I also want to ask, you know, on the last panel I was moderating on about an hour [11:57] ago, one of the panelists said to me, you know, I have 76 pilots running at any given time and we're [12:03] trying to figure out how which one do you which one do you decide to push forward. I'm sure it's [12:08] very similar for the two of you. I'd love to hear how many pilots you've got running, but how do [12:12] you decide which ones to to push forward? Which are worth your time? It is a little bit death [12:18] by a thousand pilots. I think like to be honest, that's where we started and then um realistically [12:26] because the technology sort of collapses a lot into single workflows, you end up with [12:32] most people building very similar things. Um so we went from this sort of explosion of innovation [12:39] u which was really meaningful for the company because it made everybody rethink how we work to [12:44] something that became a lot more structured which is these are the the key five things that we need [12:49] to do really well as a business and we're going to put very concentrated effort and energy behind [12:54] those five things instead of having it be very fragmented. When I look at a lot of our clients, [13:00] I think the risk sometimes with these pilots is that they play into organizational structures and [13:07] penels and politics and then people get very precious about their pilot. Um I think that's [13:14] one of the great challenges in this landscape is that you'll end up with six teams that have [13:18] the exact same thing, which one wins, etc., etc. And actually building on that, I'd say [13:25] it's been really difficult to figure out when is a moment when you need to standardize and when is it [13:31] actually fine to let everyone keep doing their own stuff. Uh it's like can be very demotivating for [13:35] teams to be working on on a pilot for a long time. They built their own tooling around it and then [13:40] um you know somebody else has one which is like very very slightly better, faster, [13:44] cheaper and they feel like all their work you know is thrown away. And sometimes it's actually we've [13:49] been finding it's fine to have sometimes multiple versions of roughly the same thing which are [13:53] optimized for like slightly different use cases where we're still in the innovation move fast [13:58] um invent new things see what the technology can do phase and we're much less in the like [14:03] standardiz roll out phase. There are a few areas where we're standardized but mainly it's still [14:08] you know somewhat exciting innovative chaos uh and getting some great results out of it. Yeah, [14:13] makes sense. was I I want to come back to you because, you know, I've read a lot of [14:17] the the things you've said over the years and and I think it was a year or two ago you said, [14:21] you know, 65% of the tasks that agencies are paid for could be done by AI. Do you still [14:28] believe this? And what does this mean for for you and your agency's future? I hope somebody has an [14:32] answer for me during these next few days. So I I think what you see uh which sort of goes back to [14:40] the bottleneck conversation we had at the start um a lot of the work can be massively sped up. Um [14:49] and then I think it's very important to understand uh speed can be its own quality in some cases but [14:56] uh speed should not be happening to the detriment of quality right so that that's always a bit of [15:01] a balancing act I think realistically uh what you sort of start figuring out as you try and [15:06] implement this at scale is a lot of what clients pay us for is to manage their process uh we we [15:14] had a funny example a few months ago I won't name any names wasn't Google uh where we did what is [15:22] normally 6 to 7 months of work and we did it in 12 weeks end to end global campaign everything [15:28] done and we really did it as um to make a um the organization that we were working uh for sort of [15:36] enthusiastic about adoption. Um, and now that we're actually doing it, it's still taking the [15:42] exact same time it would take in more traditional ways because the organization cannot operate at [15:48] that cadence. That's not a critique. That's just a reality of large enterprise organizations. So, [15:53] while a lot of our work is massively compressed at the manual labor effort, um, it's not necessarily [16:01] the bulk of what we're being paid for by most of our clients. That's interesting. Maybe building [16:06] on that uh a what some of the panelists touched on this earlier this morning. Um we're we're finding [16:13] of course that two types of people add a lot of value whether they're you know from excellent [16:17] agency partners like the like monks or whether they're internal uh is there's people who are AI [16:22] native and lots of the time they're newer to the workforce and then there's people who are experts [16:26] who maybe don't know AI as well. Uh and usually what you're bringing in with an agency is a ton of [16:31] expertise. Uh and so I do think there's a you know if there's bunch of folk from agencies I wouldn't [16:36] be immediately jumping career now because the expertise you're bringing when um it's amplified [16:40] with the speed uh and production of AI is I think probably going to be a lot more impactful per [16:45] person. Um a lot I I see a lot of the big uh companies in-ousing the tech like having new [16:52] people who really know how to use AI and trying to do that in-house because they believe that that's [16:55] the competitive advantage. I think it is, but also the actual expertise around brand management, [17:01] around messaging, around creative is also, you know, extremely key to have um in your team, [17:06] whether that's in an agency or an one conversation I had this morning and something I'm personally [17:10] interested in. You know, I I moved apartments this weekend. I went to Claude and typed in what is the [17:15] fastest kitchen table that I could get that'll get me there in 3 weeks. It wasn't that great [17:20] with the responses, but it made me think Gemini. Yeah, I know. I need I need to try Gemini. Um, [17:25] but it made me think of how you both are now marketing not just for for humans, but also [17:29] for algorithms. And how does this change how you approach a project or how does this change the [17:34] way you creatively think about how you're you're designing the the ads and the marketing you're [17:38] doing? I have a um so a known thing is that um uh LMMs take a lot of they need fresh information and [17:48] the way they tend to get their fresh information is on discussion and open social channels. that [17:52] seems to be a thing that people are talking about. I don't know any special knowledge on this working [17:55] at Google, but like you know, it seems to be known in the industry. And so, you know, social as a [18:00] channel is massively more important now. Making sure that people are talking about your products [18:04] uh on X on Reddit um and on YouTube specifically also is very key. Uh the challenging thing is like [18:10] for some categories, you know, it's sort of easy when you're when you're marketing a tech product [18:14] like you're marketing Chrome. It's a thing that people naturally talk about in their channels. [18:18] Um it's I can imagine it's quite hard though if you're marketing a product which isn't naturally [18:21] spoken about. People don't like recommend to their friends like medication necessarily or like you [18:26] know or or lots of other categories which are sort of which are less discussed in those channels. Um [18:31] and so you know if you're in those industries where there's less social discussion I' I'd [18:34] probably be trying to figure out how to make your your brand and product relevant in those places. [18:38] Yeah I think you're also um having to rethink what the role of your web environment is. it [18:46] becomes more of a knowledge base for agents. Um, we had an interesting example where historically [18:52] you never really delete things from the web like that's it's for posterity, but now a lot of old PR [18:59] uh messages can still be pulled into LM's output. So your um which is is funny because I think a lot [19:06] of people are making maybe expecting content strategy and uh the strategic copyrightiting [19:12] that's part of that to be one of the first parts of the industry that goes away. I actually think [19:16] that completely reframes the ongoing knowledge base building for an agentic landscape. I also [19:22] think the um the reality is that a lot of the the space brand currently turns up is [19:28] sort of collapsing into agentic interfaces. So a lot of rethinking on what is a brand in [19:33] a landscape where you have agents sort of taking uh more economically logical decisions and less [19:39] emotional like there's a lot of lot of thinking that still needs to happen there. Yeah. Well, [19:43] great advice. So, I wish we had more time, but thank you both so much for for joining me.