Tealium Warns AI Agents Cut Search Traffic, Bypass Brand Tracking

Disruption or Destruction? 5 Trends That Will Change Your Customer Data Strategy & Marketing Future — Ai4

Ai4

As autonomous AI engines shift consumer search and memory upstream, brands must abandon static demographic personas for real-time situational modeling and server-side data architectures to retain conversion visibility.

AI-assisted summary of Disruption or Destruction? 5 Trends That Will Change Your Customer Data Strategy & Marketing Future — original source. Check the original for full context.

Source published: 2025-08-27

Detailed notes

## Background Published on YouTube by Ai4 on August 27, 2025, this talk features Zack Wenthe, Director of Product Marketing at Tealium. According to the page description, the session explores five emerging trends that threaten to blindside enterprise marketing organizations and outlines how to rethink customer data strategies before traditional methods become obsolete.

## TL;DR Zack Wenthe explains how answer engines, agentic browsing, and upstream persistent AI memory are dismantling traditional search discovery, clickstream tracking, and linear journey mapping. He argues that marketers must move away from static demographic personas to focus on real-time situational context and immediate next-action micro-conversions. To maintain attribution and customer engagement in this fragmented environment, organizations must adopt server-side tracking, leverage data clean rooms, capture zero-party context, and deploy LLMs for behavioral personalization.

## Atmosphere & style Delivered as a brisk, high-energy morning conference session in Las Vegas, the presentation strikes a balanced, pragmatic tone anchored by a 'no panic, no snooze' mindset. Wenthe blends lighthearted pop-culture references with sharp industry warnings about plummeting click-through rates and stock hits, creating an engaging executive briefing. It feels focused, direct, and actionable, transforming overwhelming AI disruption into practical, engineering-minded next steps.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=27f16xgmmx4&t=4s) AI disruption is fundamentally challenging foundational customer data architectures, requiring data and marketing leaders to proactively adapt rather than panic or ignore the shifts. - [01:35](https://www.youtube.com/watch?v=27f16xgmmx4&t=95s) The shift toward answer engines and agentic search reduces traditional search engine click-through rates by 50% when AI summaries resolve queries directly. - [01:35](https://www.youtube.com/watch?v=27f16xgmmx4&t=95s) Brands must treat AI assistants as discovery channels by structuring content into concise, scannable facts and leveraging press releases for faster indexing. - [05:32](https://www.youtube.com/watch?v=27f16xgmmx4&t=332s) Personalization is moving upstream into persistent AI agent memory, bypassing brand-side recommendation engines and shifting user preference storage to consumer AI platforms. - [05:32](https://www.youtube.com/watch?v=27f16xgmmx4&t=332s) Marketers should adopt co-personalization strategies by enabling consumers to export brand preferences, curated lists, and product summaries directly into their personal AI assistants. - [09:03](https://www.youtube.com/watch?v=27f16xgmmx4&t=543s) Fragmented, high-intent visits from AI agents are replacing linear clickstreams, rendering traditional pixel-based retargeting and multi-touch attribution models ineffective. - [09:03](https://www.youtube.com/watch?v=27f16xgmmx4&t=543s) Organizations must transition away from brittle client-side tracking pixels to server-side Conversion APIs (CAPI) and collect direct zero-party context at checkout. - [11:22](https://www.youtube.com/watch?v=27f16xgmmx4&t=682s) Real-time situational context and immediate in-session behavior consistently override static demographic personas and historical propensity scoring. - [11:22](https://www.youtube.com/watch?v=27f16xgmmx4&t=682s) Marketing optimization should pivot from long-horizon 'Next Best Action' models to securing the immediate next action and answering micro-moment queries in the first 30 seconds of a visit. - [15:31](https://www.youtube.com/watch?v=27f16xgmmx4&t=931s) Linear customer journey maps are collapsing as brand-agnostic consumers prioritize speed and utility, requiring brands to eliminate checkout friction and deploy LLMs for real-time objection handling. - [18:33](https://www.youtube.com/watch?v=27f16xgmmx4&t=1113s) Modernizing data infrastructure requires immediate audits of agent-assisted traffic, robust server-side identity stitching, and data clean room partnerships to measure external conversions. - [19:09](https://www.youtube.com/watch?v=27f16xgmmx4&t=1149s) LLMs can enrich behavioral segmentation by turning raw transaction data into rich contextual narratives combined with external environmental factors like weather and seasonality. - [20:18](https://www.youtube.com/watch?v=27f16xgmmx4&t=1218s) Enterprise leaders should navigate AI disruption with proactive curiosity and active experimentation rather than fear or complacency.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=27f16xgmmx4&t=4s) Rethinking Customer Data Strategy in the Age of AI

Kicking off the second day of a conference in Las Vegas is always an uphill battle, as attendees balance early morning sessions against late dinners and nightlife. Opening the morning session, Zack Wenthe, Director of Product Marketing at Tealium—a customer data platform (CDP), data integration, and data orchestration provider—addresses enterprise data leaders on the realities facing modern marketing. While artificial intelligence is frequently pitched through the lens of boundless excitement and opportunity, the session focuses instead on the disruptive forces that threaten to derail existing data architectures, pipelines, and long-term marketing strategies.

Setting the tone with a nod to *Ted Lasso*, the presentation operates under a strict principle: "There's two buttons I never like to push. Panic or snooze." This is not an alarmist proclamation that the sky is falling, but a practical breakdown of five emerging trends observed across enterprise customer bases. These shifts challenge foundational assumptions about how customer data is gathered, processed, and activated, requiring marketing and data leaders to proactively adapt their architectures before their existing playbooks become obsolete.

---

[01:35](https://www.youtube.com/watch?v=27f16xgmmx4&t=95s) Trend 1: Answer Engines and Agentic Browsing

The primary discovery pathway for consumers is fundamentally shifting away from legacy search behavior. For decades, the discovery loop consisted of entering keywords into Google, scanning a list of ten blue links, clicking through to brand websites, and navigating content. The rise of answer engines, agentic browsing tools like Perplexity's Comet browser, and autonomous agent frameworks from OpenAI is dismantling this pattern. Search engines and AI assistants now summarize answers directly in the search interface, resolving user queries without requiring an external site visit.

```mermaid flowchart TD accTitle: Search Discovery Evolution accDescription: Comparison between traditional search engine discovery and AI agentic search discovery.

subgraph Legacy["Traditional Search Discovery"] L1["User Query on Google"] --> L2["Search Engine Results Page"] L2 --> L3["Click-Through to Brand Website"] L3 --> L4["Long-Path Content Investigation & Lead Capture"] end

subgraph Agentic["Agentic & Answer Engine Discovery"] A1["User Natural Language Prompt"] --> A2["AI Summary / Agentic Synthesis"] A2 -->|50% CTR Drop| A3["Direct Answer in Engine (Zero-Click)"] A2 -->|High Intent Referral| A4["Direct Downstream Action / Checkout"] end ```

The data reflects severe consequences for zero-click interactions: when an AI summary is shown on Google, click-through rates drop by 50%. While AI-driven referral traffic to brand sites has experienced explosive growth—up 357% year-over-year—it still accounts for roughly one billion data points compared to Google's daily volume of approximately 119 billion. However, this shift disproportionately harms publishing and content-heavy brands whose acquisition funnels rely on top-of-funnel blog posts and educational materials. A major marketing-focused enterprise recently suffered a 30% drop in its stock price following Q2 earnings, directly citing soft organic search visibility and a decline in organic traffic.

- **Treat AI Assistants as Channels:** Tag and isolate agent-referred visits in data pipelines and customer data platforms to systematically compare agentic traffic versus human traffic. - **Publish Assistant-Readable Facts:** Reformat content into scannable summaries, bulleted takeaways, and concise factual snippets that AI scrapers and indexers can easily ingest. - **Leverage Press Releases:** Structured press releases are currently being indexed significantly higher and faster by Google’s answer engines than standard SEO blog content.

---

[05:32](https://www.youtube.com/watch?v=27f16xgmmx4&t=332s) Trend 2: Agentic Memory and Upstream Personalization

As users interact regularly with platforms like ChatGPT, Perplexity, and Microsoft Copilot, these systems continuously accumulate persistent contextual memories about individual preferences, affinities, and constraints. This dynamic shifts the personalization layer upstream into the hands of the consumer and the AI platforms, circumventing brand-side personalization engines. AI providers are aggressively competing to control this persistent memory layer, fulfilling the long-promised vision of an omnipresent, truly personalized assistant that Siri and Alexa originally failed to deliver.

With ChatGPT processing roughly 2.5 billion prompts every day—representing an engagement volume equivalent to roughly 20% of the world's population prompting daily—the scale of upstream data capture is unprecedented. While Apple Intelligence is still maturing, Apple’s track record with iOS 14.5 (App Tracking Transparency) proves how quickly platform-level privacy controls can disrupt marketing attribution and tracking. Marketers must recognize that when consumer preferences live inside private LLMs, traditional onsite recommendation engines lose visibility.

```mermaid flowchart LR accTitle: Upstream Personalization Shift accDescription: Diagram showing consumer data moving into agent memory rather than brand-side CDPs.

subgraph ConsumerAI["Client-Side Agentic Layer"] U["Consumer"] -->|Prompts & Preferences| AM["Agent Memory (ChatGPT / Perplexity)"] AM -->|Filters & Evaluates| SR["Curated Answer / Product Selection"] end

subgraph BrandLayer["Brand Ecosystem"] SR -->|Direct High-Intent Visit| BP["Brand Transaction Portal"] BP -.->|Bypassed| BR["Legacy Brand Recommendation Engine"] end ```

- **Upstream Migration of Data:** Personalization occurs before the user ever lands on a brand property, rendering historical onsite profile building less relevant. - **Co-Personalization Strategy:** Offer exportable brand preferences, curated lists, and favorite product summaries that consumers can save directly into their personal AI agents. - **Real-World Impact:** Rather than browsing Netflix or Hulu recommendation carousels, consumers increasingly rely on automated Friday digests generated by tools like Perplexity to curate streaming shows, comedy specials, and podcasts tailored specifically to their profile.

---

[09:03](https://www.youtube.com/watch?v=27f16xgmmx4&t=543s) Trend 3: Fragmented Measurement and the Death of the Clickstream

As consumers spend less time conducting exploratory, top-of-funnel research on brand websites, the traditional concept of an uninterrupted clickstream is breaking down. The historic marketing model—tracking an unknown visitor across multiple sessions, building an engagement profile, offering an introductory discount, retargeting them via display ads, and nurturing them through conversion—is giving way to disjointed, "random acts of engagement."

When users arrive on a site via an AI agent, they often appear spontaneously with high intent, leaving attribution models unable to trace the preliminary touchpoints that influenced the sale. Marketers can no longer depend on lengthy client-side browsing trails to drive automated customer journeys or trigger retargeting pixels.

- **Implement Server-Side APIs:** Shift immediately away from client-side JavaScript pixels toward Conversion APIs (CAPI) to reliably pass transaction data back to ad networks and prevent wasted ad spend on already-converted users. - **Capture Zero-Party Context:** Add low-friction intake questions during checkout or onboarding to ask how users arrived (e.g., whether they utilized AI search engines). - **Run Holdout Experiments:** Establish incrementality testing and holdout groups to quantify how AI discovery tools are altering baseline customer acquisition.

---

[11:22](https://www.youtube.com/watch?v=27f16xgmmx4&t=682s) Trend 4: The Situation vs. the Person in Behavioral Modeling

For decades, enterprise marketing has prioritized static customer personas, demographic brackets, and historical propensity scoring under the assumption that past demographic segments reliably predict future actions. However, as data fragmentation obscures user identity and creates gaps in predictive models, contextual and situational factors dramatically outweigh static user traits.

This dynamic is demonstrated by the classic behavioral study detailed in the literature surrounding *The Person and the Situation*. In the experiment, seminary students were instructed to walk across campus to deliver an urgent presentation. Half the students were given 60 minutes to reach the building, while the other half were told the room had moved and they had only five minutes to get there. En route, every student encountered an individual in distress asking for help. Despite their shared theological background and moral alignment, the students under time pressure rarely stopped, whereas those with ample time stopped to assist. The situational context—time pressure—completely overrode the intrinsic traits of the individuals.

```mermaid flowchart TD accTitle: Situational Context vs Static Personas accDescription: Flowchart illustrating how real-time in-session context overrides historical demographic models.

subgraph StaticModel["Legacy Persona Modeling (Ineffective)"] D["Demographics & Persona"] --> P["Historical Propensity Score"] P -->|Fails in fragmented journeys| M["Generic Marketing Message"] end

subgraph ContextModel["In-Session Context Optimization (Tealium Model)"] S["Active Buyer Situation / Mode"] --> B["In-Session Behavior (First 30s)"] B -->|Overrides Historical Data| NA["Immediate Next-Action Micro-Conversion"] end ```

A Tealium enterprise customer demonstrated this principle in production: despite maintaining vast repositories of historical demographic data and complex propensity-to-buy algorithms, their system now discards those models within 30 seconds of a visitor landing on the website. Instead, the platform dynamically personalizes the experience entirely around the real-time, in-session behavior of that specific visit, achieving significantly higher conversion rates.

- **Shift from Next Best Action to Immediate Next Action:** Rather than predicting a hypothetical sale weeks away, optimize UI elements to secure the immediate next click, sustain session engagement, or answer a single question in the micro-moment. - **Account for Situational Modes:** Consumers alter their buying behavior entirely depending on context, such as back-to-school rushes, holiday shopping, or routine maintenance purchases. - **Deploy Zero-Party Micro-Surveys:** Replace lengthy click paths with lightweight, 1-to-3 question prompts (e.g., "What are you here for today?") to instantly route users to exact answers without forcing them to navigate junk content.

---

[15:31](https://www.youtube.com/watch?v=27f16xgmmx4&t=931s) Trend 5: The Collapse of Linear Customer Journey Maps

The standard practice of designing rigid, linear customer journey maps inside journey orchestration platforms is failing because consumers no longer follow prescribed paths. Accelerated by the COVID-19 pandemic—which forced historically offline demographics into ecommerce, grocery delivery, and digital brand switching—buyers have become largely brand agnostic and comfortable navigating their own nonlinear purchasing paths.

While elaborate journey diagrams remain visually impressive to enterprise marketers, actual consumers demand immediate speed and utility. If a buyer encounters friction, they switch alternatives instantly. Because incoming traffic is shifting toward lower overall volumes with substantially higher intent, digital properties must be optimized for execution velocity rather than forced brand storytelling.

- **Eliminate Purchase Friction:** When high-intent buyers arrive directly from agentic discovery engines, strip away unnecessary intermediate steps and streamline checkout workflows. - **Proactive LLM Objection Handling:** Deploy large language models to answer product and service questions contextually in real time based on active browsing clicks, resolving purchase hesitations on the fly without requiring new static landing pages.

---

[17:49](https://www.youtube.com/watch?v=27f16xgmmx4&t=1069s) The Monday Action Plan: Implementing Immediate Data Shifts

Rather than attempting an unmanageable overhaul of the entire martech stack, marketing and data leaders must focus on concrete, operational changes starting Monday morning.

- **Audit and Tag Agent-Assisted Traffic:** Establish tracking parameters across all data collection pipelines and CDPs to isolate visits driven by AI search engines, agents, and answer engines, establishing a monthly review process to monitor traffic directionality. - **Deploy Server-Side Tracking and Conversion APIs:** Prioritize server-to-server data pipelines to ensure accurate transaction attribution, prevent audience targeting leakage, and maintain clean measurement feeds independent of browser client-side limitations.

[18:33](https://www.youtube.com/watch?v=27f16xgmmx4&t=1113s) Modern Data Infrastructure: Server-Side Tracking, Data Clean Rooms, and Identity Stitching

Client-side pixel tracking has become increasingly unreliable, meaning that when a consumer completes a purchase on a device, there is a strong probability the tracking pixel fails to fire altogether. Relying on client-side mechanisms is "iffy at best," making server-side tracking essential for enterprise data strategies. Moving to server-side data collection allows organizations to reliably pass transactional profile information directly to ad networks and stitch unified customer profiles with consistent identity resolution.

```mermaid flowchart LR accTitle: Modern Data and Personalization Flow subgraph Capture["1. Server-Side Infrastructure"] A["Customer Transaction"] --> B["Server-Side Collection"] B --> C["Ad Network Profiles & Data Clean Rooms"] end

subgraph Intelligence["2. AI & Context Enrichment"] C --> D["LLM Behavioral Context"] E["External Attributes (Weather, Season)"] --> D end

subgraph Activation["3. Advanced Segmentation"] D --> F["Dynamic Segmentation & Personalization"] end ```

Marketers must also navigate the challenge of understanding downstream conversions in an environment where autonomous AI agents handle transactions end-to-end. To address this loss of visibility, brands should actively test and partner with data clean room solutions to analyze customer conversion behaviors securely without relying on legacy tracking.

- **Server-side tracking**: Replaces brittle client-side pixels to ensure complete transactional fidelity with advertising platforms. - **Identity consistency**: Stitches cross-channel user profiles into a unified customer record. - **Data clean rooms**: Unlocks conversion attribution and measurement even when AI agents execute transactions across external ecosystems.

---

[19:09](https://www.youtube.com/watch?v=27f16xgmmx4&t=1149s) AI-Powered Behavioral Personalization and Context Enrichment

Personalized content strategies must shift toward behavioral personalization that actively influences on-site consumer actions. Behavioral psychology insights—such as studies conducted in traditional shopping malls showing how subtle environmental and behavioral cues lengthen dwell times and increase spending—apply directly to the digital realm. Marketers can mirror these dynamics online by using Large Language Models (LLMs) to synthesize and interpret customer activity.

Instead of analyzing clicks or cart additions in isolation, teams can deploy LLMs to generate rich, contextual descriptions of basket combinations and hospitality interactions. These narratives feed into advanced segmentation engines alongside external contextual data points surfaced by AI, such as local weather patterns and seasonal trends at the exact moment of purchase. Capturing these situational dimensions reveals underlying consumer motivations that traditional data models completely miss.

- **Digital behavioral cues**: Translates physical retail psychology into digital experience design to boost dwell time and cart value. - **LLM-driven contextual descriptions**: Transforms raw clickstream and basket data into descriptive behavioral narratives for advanced segmentation. - **Contextual environmental enrichment**: Factors in real-time attributes like weather and location at the point of sale to uncover new behavioral triggers.

---

[20:18](https://www.youtube.com/watch?v=27f16xgmmx4&t=1218s) Conclusion and Open Exploration with AI

Wrapping up the session, Zack invites attendees to connect on LinkedIn and stop by Tealium’s booth (Booth 642) on the expo floor to continue the discussion and review the behavioral research in detail. Acknowledging the fast pace of the 20-minute presentation, he encourages marketing leaders to embrace emerging technologies without apprehension, borrowing a mindset popularized by *Ted Lasso*: "Be curious, not judgmental. Try AI, explore, have fun."

## Notable quotes > Is your company’s customer data strategy ready for an AI-driven world? — Zack Wenthe

> Traditional strategies and methods are no longer enough, but are you prepared to make the shifts necessary to compete tomorrow and into the future? — Zack Wenthe

> Join me to rethink your approach to customer data before it’s too late. — Zack Wenthe

> [00:04](https://www.youtube.com/watch?v=27f16xgmmx4&t=4s) There's two buttons I never like to push. Panic or snooze. — Zack Wenthe

> [09:03](https://www.youtube.com/watch?v=27f16xgmmx4&t=543s) random acts of engagement — Zack Wenthe

> [11:22](https://www.youtube.com/watch?v=27f16xgmmx4&t=682s) The Person and the Situation — Zack Wenthe

> [11:22](https://www.youtube.com/watch?v=27f16xgmmx4&t=682s) What are you here for today? — Zack Wenthe

> [18:33](https://www.youtube.com/watch?v=27f16xgmmx4&t=1113s) iffy at best — Zack Wenthe

> [20:18](https://www.youtube.com/watch?v=27f16xgmmx4&t=1218s) Be curious, not judgmental. Try AI, explore, have fun. — Zack Wenthe

## Mentioned Zack Wenthe, Tealium, Ai4, Ted Lasso, Google, Perplexity, Comet, OpenAI, ChatGPT, Microsoft Copilot, Apple, Apple Intelligence, Siri, Alexa, Netflix, Hulu, The Person and the Situation

## Video description Zack Wenthe, Director of Product Marketing, Tealium

Abstracts

Is your company’s customer data strategy ready for an AI-driven world? In this talk I will explore five areas that have the potential to blindside most Enterprise marketing organizations. Traditional strategies and methods are no longer enough, but are you prepared to make the shifts necessary to compete tomorrow and into the future? I will walk you through 5 areas that you need to consider, why they're important, and how to adjust your strategies to thrive amid the inevitable disruption. Join me to rethink your approach to customer data before it’s too late.

[Source](https://www.youtube.com/watch?v=27f16xgmmx4)

中文简报

Tealium警告：AI代理削减搜索流量并绕过品牌追踪随着自主AI引擎将消费者的搜索与记忆推向上游，品牌必须放弃静态的人口统计画像，转向实时情境建模与服务端数据架构，以保持转化可见性。 Zack Wenthe解释了解答引擎、代理式浏览以及上游持久性AI记忆如何正在瓦解传统的搜索发现、点击流追踪和线性旅程映射。他指出，营销人员必须摆脱静态的人口画像，转而关注实时情境背景与即时下一步操作的微转化。为了在这一碎片化环境中保持归因有效性与客户互动，企业必须采用服务端追踪、利用数据净室、捕获零方情境数据，并部署LLM以实现行为个性化。 AI 答案引擎与智能体搜索正在重构传统的搜索发现模式，导致出现 AI 摘要时的搜索点击率骤降 50%，品牌需将 AI 助手视为独立渠道并发布便于抓取的结构化事实。 (1:35) 用户偏好正在向上游迁移至个人 AI 助手的持久记忆中并绕过品牌端推荐引擎，企业需要采用支持消费者直接导出偏好的“共同个性化”（Co-Personalization）策略。 (5:32) 高意向碎片化访问使得传统客户端像素追踪与多触点归因失效，企业必须加速过渡至服务端转化 API（CAPI）、数据安全岛以及在结算环节直接捕获零方数据。 (9:03) 实时的即时会话行为与情境模式已彻底超越静态人口画像和历史倾向评分，营销策略需从僵化的线性旅程地图转向优化微时刻的即时转化动作。 (11:22)

Reading summary

In this Ai4 presentation, Tealium Director of Product Marketing Zack Wenthe argues that artificial intelligence and autonomous agents are dismantling legacy customer data pipelines, requiring marketers to adapt their architectures without panicking or ignoring the shift. Rather than relying on continuous clickstreams and static personas, enterprises must rebuild around upstream AI discovery, real-time situational context, and server-side measurement. Upstream AI Discovery and Agentic Memory (1:35) Wenthe observes that consumer discovery is moving from traditional search links to answer engines, agentic browsers like Perplexity's Comet, and OpenAI agent frameworks. When Google displays an AI summary, click-through rates fall by 50%. Although AI-driven referrals grew 357% year-over-year, they represent roughly one billion data points against Google's 119 billion daily volume, a shift that still hurts content-heavy funnels—illustrated by a major marketing enterprise whose stock dropped 30% after Q2 earnings amid softer organic search visibility. Simultaneously, platforms like ChatGPT—processing about 2.5 billion prompts daily—Perplexity, and Microsoft Copilot accumulate persistent user memory. Much as Apple's iOS 14.5 disrupted tracking, private LLMs now pull personalization upstream; consumers increasingly use tools like Perplexity digests to curate entertainment rather than browsing Netflix or Hulu carousels. To adapt, Wenthe recommends reformatting web content into concise, assistant-readable facts and leveraging structured press releases, which Google's answer engines currently index faster and higher than standard SEO blogs. Because upstream agent filtering bypasses legacy on-site recommendation engines, he also proposes "co-personalization," allowing customers to export curated brand preferences and product lists directly into their personal AI agents. Nonlinear Journeys and High-Intent Execution (15:31) Because upstream AI tools resolve early research off-site, visitors who do click through arrive in lower volumes but with substantially higher intent. Wenthe argues that this dynamic, combined with pandemic-era digital adoption that made buyers more brand-agnostic, has caused rigid, linear customer journey maps to collapse. Consumers no longer follow prescribed multi-step funnels and will switch brands immediately if they encounter friction. Instead of forcing visitors through elaborate brand storytelling or static landing pages, Wenthe advises optimizing digital properties for execution velocity. Marketers should strip unnecessary steps from checkout workflows for agent-referred buyers and deploy large language models on-site to handle product objections in real time based on a user's active clicks. Situational Context Over Static Customer Personas (11:22) As fragmented journeys create gaps in identity and historical tracking, Wenthe contends that immediate situational context predicts behavior far better than static demographics or propensity scores. He cites the classic seminary student experiment from literature on The Person and the Situation, where having five minutes versus 60 minutes to reach a presentation overrode students' shared moral traits in deciding whether to help a stranger in distress. In practice, one Tealium enterprise customer found that discarding historical demographic and propensity models after a visitor's first 30 seconds on-site—and personalizing solely around real-time session behavior—significantly improved conversion rates. Wenthe urges teams to shift from predicting a long-term "next best action" to securing the "immediate next action" within a micro-moment, accounting for situational modes like holiday or back-to-school shopping. Marketers can replace long click paths with one-to-three-question micro-surveys to route visitors instantly, while using LLMs to translate raw basket combinations, hospitality interactions, and real-time environmental attributes like weather and season into rich behavioral narratives for segmentation. Server-Side Measurement After the Clickstream Dies (9:03) When AI agents compress discovery into sporadic, high-intent visits, the continuous client-side clickstream breaks down, turning journeys into what Wenthe calls "random acts of engagement." He characterizes client-side JavaScript pixels as "iffy at best," noting they frequently fail to fire during device checkouts and cannot trace upstream agent interactions. Starting immediately, Wenthe recommends tagging and auditing AI-referred visits inside customer data platforms to compare agentic and human traffic monthly. Enterprises should replace brittle client-side pixels with server-side tracking and Conversion APIs (CAPI) to pass transaction data directly to ad networks, stitch cross-channel identities, and suppress wasted retargeting ads on converted buyers. To close remaining attribution gaps, he suggests adding brief checkout questions asking how buyers discovered the brand, running holdout incrementality tests, and partnering with data clean rooms to measure downstream conversions executed by autonomous agents. Wenthe concludes that while client-side tracking and static journey playbooks are losing effectiveness, enterprise data leaders do not need to overhaul their entire martech stacks overnight. By combining foundational server-side infrastructure with open, curious experimentation around AI-driven context and agentic channels, teams can incrementally adapt to nonlinear buyer behavior.

精炼阅读摘要

Tealium产品营销总监Zack Wenthe指出，生成式AI与自主智能体正在重构消费者的搜索、记忆与决策路径，导致传统点击流与线性旅程失效；企业必须将数据策略从依赖静态画像与客户端追踪，转向基于服务端架构、实时情境感知与适配AI引擎的全新模式。 AI回答引擎重塑搜索与内容获客格局 (1:35) 以Perplexity的Comet浏览器及OpenAI智能体框架为代表的回答引擎，正在瓦解沿用数十年的关键词搜索与蓝链跳转模式。AI在搜索界面直接生成答案，导致用户在出现AI摘要时的点击通过率（CTR）锐减50%，直接打击依赖顶部流量和教育性博客的内容出版型企业。尽管AI驱动的推荐流量同比激增357%，但其约10亿次的总量相较于Google每天约1190亿次的搜索量仍属早期。然而，流量转移的冲击已然显现，已有大型营销技术企业在第二季度财报后因自然搜索可见度与流量下滑，股价单日下跌30%。面对这一转变，企业需要将AI助手视作独立的获客渠道并在客户数据平台（CDP）中打标隔离，系统对比智能体与人类访客。同时，营销人员必须调整内容架构，将长篇内容重构为便于AI抓取与索引的要点摘要及事实片段。此外，数据表明结构化的新闻稿在Google回答引擎中的收录与索引速度明显高于常规SEO博客，应成为企业优化内容可见性的重要工具。记忆上移使品牌端个性化推荐面临失效 (5:32) 随着ChatGPT、Perplexity和Microsoft Copilot等平台持续沉淀消费者的长期偏好与上下文约束，个性化层正向上迁移至消费者专属的AI端，从而绕过品牌自建的推荐引擎。Wenthe指出，ChatGPT每日处理约25亿条提示词，其交互规模相当于全球约20%人口的日常交互量。这种在上游沉淀的持久性记忆正在实现早年Siri与Alexa未能兑现的真正个人助理构想。当用户依赖私有大语言模型进行筛选时，品牌自建的站内个性化系统将彻底失去用户前期探索行为的可见度。这种转变要求企业从单向个性化转向“协同个性化”（Co-Personalization）策略。品牌应主动向消费者提供可导出的品牌偏好、精选清单与产品摘要，使其能直接保存到个人AI智能体中。实际应用中，越来越多的消费者不再浏览流媒体平台的常规推荐列表，而是依靠自动化生成的周报定向获取匹配自身喜好的内容，证明了端侧智能体过滤已在实际重塑消费者的最终消费决策。点击流断裂倒逼服务端追踪与净室方案 (9:03) 由于消费者在品牌网站上的前期探索行为被AI大幅压缩，传统长路径的连续点击流正在瓦解为零散且高意向的偶发互动。依赖页面Cookie、跨会话追踪与多次重定向广告培育转化的传统归因模型因此失效。与此同时，客户端JavaScript像素代码在现代浏览器环境中的丢失率居高不下，无法保证交易数据的完整回传。企业必须立即从客户端追踪向服务端转化API（CAPI）全面迁移，通过服务器到服务器的数据管道确保交易属性准确回传给广告网络，防止对已转化用户产生无效广告投放。同时，针对AI智能体端到端代理交易带来的下游可见性缺失，企业需要采用数据净室（Data Clean Rooms）在保护隐私的前提下完成转化行为的多方分析，并在结账或注册环节部署零方数据微型问卷，直接收集用户是否通过AI引擎触达的关键归因信息。实时交互情境替代静态人口统计画像 (11:22) Wenthe援引经典行为学文献《人与情境》中的神学院学生实验指出，内在特质与道德取向在时间紧迫这一即时情境压力面前基本失效；在营销建模中，动态情境对行为的影响同样远超静态人口统计画像与历史倾向评分。Tealium的一家企业客户在生产实践中证实了这一法则：尽管该企业拥有海量历史人口统计数据和购买倾向算法，但其系统在访客落地网站30秒后便会抛弃原有静态模型，完全基于当次会话的实时行为展开动态个性化，显著拉升了转化率。营销策略应从预测数周后成交的“下一步最佳行动”（Next Best Action），转向促成微转化的“即时下一步行动”（Immediate Next Action），通过捕捉前30秒的用户交互保持停留。同时，企业应借助大语言模型解析购物车组合与互动行为背后的情境描述，并融合天气、季节等即时环境变量丰富细分引擎，以此捕捉传统数据模型遗漏的瞬时购买动机。线性旅程瓦解与无摩擦交易架构演进 (15:31) 在疫情促成数字品牌随意切换后，消费者的品牌忠诚度明显减弱，不再遵循企业在旅程编排平台上预设的线性路径。当具备高购买意向的流量由AI智能体直接导入终端时，企业必须舍弃冗长繁琐的品牌叙事，全面转向以执行速度与消除摩擦为核心的交易架构。优化方向在于精简结账流程并移除一切不必要的中间跳转环节。此外，企业应直接部署大语言模型，依据访客当前的实时浏览点击动态解答关于产品规格或服务的核心疑虑，在交互节点即时化解购买抗拒，而不再强求访客按既定路径阅读静态落地页。 Wenthe总结强调，面对AI带来的颠覆，营销技术团队无需推倒重建整套技术栈，而应在下周一立即着手执行关键战术调整：为智能体辅助流量打上追踪标签，并加速部署服务端CAPI与数据净室基础设施。同时，企业应保持对新兴技术的探索心态，以务实且敏捷的数据治理应对智能体时代带来的结构性冲击。

Detailed notes — zh

## Background 该内容由行业峰会频道 Ai4 于 2025 年 8 月 27 日发布在 YouTube 平台，由 Tealium 产品营销总监 Zack Wenthe 主讲。视频简介指出，传统的客户数据策略已无法适应 AI 时代，演讲重点探讨了可能使多数大型企业营销部门陷入被动的五大核心趋势，并指导企业如何调整数据策略以应对必然到来的颠覆。 ## TL;DR 演讲系统剖析了重塑客户数据策略的五大趋势：答案引擎与智能体搜索正在打破传统搜索与长链条点击流，个性化控制权正转移至消费者的端侧智能体记忆，实时的“情境”比静态的历史用户画像更能决定转化，而复杂的线性客户旅程也已走向终结。演讲强调，企业必须告别脆弱的前端 Cookie 追踪与宏观预测模型，转向“会话内即时个性化”与极简的零方数据采集。最后，主讲人提出了包含智能体流量标记、部署服务端转化 API（CAPI）以及利用 LLM 辅助情境化分析在内的“周一行动清单”，帮助企业快速完成架构演进。 ## Atmosphere & style 演讲在拉斯维加斯清晨轻松风趣的开场白中展开，巧妙引用美剧《足球教练》的名言打破了技术布道常见的焦虑感与说教味。主讲人将心理学经典实验、个人育儿生活细节与硬核的数据架构演进融为一体，叙事节奏明快且极具现场感染力。整场分享洋溢着务实而敏锐的专业洞察，在指出行业痛点的同时给出了清晰的破局抓手，传递出理性探索与敏捷试验的积极信念。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=27f16xgmmx4&t=4s) 面对 AI 带来的颠覆浪潮，企业既不能恐慌也不能无视（不恐慌、不小睡），必须主动重构客户数据战略以应对核心范式转移。 - [01:35](https://www.youtube.com/watch?v=27f16xgmmx4&t=95s) AI 概览与答案引擎使传统搜索自然点击率骤降 50%，品牌需将 AI 智能体视为独立流量渠道，并发布易于机器索引的事实片段与新闻通稿。 - [05:32](https://www.youtube.com/watch?v=27f16xgmmx4&t=332s) 随着 AI 智能体沉淀用户长期记忆，个性化控制权从企业网站后台移交至消费者端，品牌需转向“协同个性化”并允许用户导出偏好数据。 - [08:39](https://www.youtube.com/watch?v=27f16xgmmx4&t=519s) 用户决策前置并在站外由智能体完成，导致传统长路径点击流与重定向归因模型失效，触点演变为高度离散的随机互动。 - [08:39](https://www.youtube.com/watch?v=27f16xgmmx4&t=519s) 企业必须加速部署服务端转化 API（CAPI），在后台交易达成的瞬间剔除受众，彻底摆脱脆弱的前端像素并防止广告预算浪费。 - [11:19](https://www.youtube.com/watch?v=27f16xgmmx4&t=679s) “情境胜于人设”：访客进站前 30 秒的实时交互情境远比历史画像与人口统计学数据更能决定当下的转化决策。 - [11:19](https://www.youtube.com/watch?v=27f16xgmmx4&t=679s) 放弃长周期的“下一步最佳行动”（Next Best Action），转而聚焦会话内的“下一步实际行动”，通过极简轻量互动获取高价值零方数据。 - [15:31](https://www.youtube.com/watch?v=27f16xgmmx4&t=931s) 线性客户旅程编排在现实中已形同虚设，消费者的品牌中立与跳跃路径要求营销重心转向会话内即时响应与极简无阻碍结账。 - [15:31](https://www.youtube.com/watch?v=27f16xgmmx4&t=931s) 利用大语言模型（LLM）根据用户浏览轨迹实时生成异议化解内容，无需预制海量静态文案即可敏捷清除下单阻碍。 - [17:49](https://www.youtube.com/watch?v=27f16xgmmx4&t=1069s) 在 CDP 中建立智能体流量属性埋点与月度监测机制，结合跨渠道档案拼接与数据清洁室（Data Clean Room）应对智能体代购隐匿问题。 - [17:49](https://www.youtube.com/watch?v=27f16xgmmx4&t=1069s) 开展情境化行为实验，借助 LLM 自动撰写商品组合洞察，并融入天气、季节等外部即时变量来深度理解消费动机。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=27f16xgmmx4&t=4s) 拉斯维加斯的清晨：在恐慌与小睡之间审视客户数据战略演讲在拉斯维加斯会议第二天清晨的轻松调侃中拉开序幕。演讲者 Zack Wenthe（Tealium 产品营销总监）笑着打趣道，在拉斯维加斯排在第二天一早做开场演讲总是伴随着巨大风险——前一天晚上大家往往在外尽情享受美食、通宵狂欢，次日清晨就必须面临是否能挣扎着爬起来的艰难抉择。作为一家专注于客户数据平台（CDP）、数据集成与数据编排的技术公司，Tealium 每天都在处理海量的客户数据。虽然当前科技界对 AI 带来的无限机遇兴奋不已，但 Zack 明确表示，他今天打算反其道而行之，重点剖析那些可能会让各大企业未来的战略规划或重大项目彻底脱轨的潜在威胁。为了给整场演讲定下基调，Zack 引用了热门美剧《足球教练》（Ted Lasso）里的一句经典台词作为免责声明：“有两个按钮我从来不喜欢按：恐慌（Panic）或小睡（Snooze）。”他强调，这场分享绝不是在贩卖焦虑，也不是渲染“天要塌下来”的危言耸听，但企业同样不能像按下闹钟小睡键一样选择无视。通过 Tealium 与众多企业客户的深度交流，他提炼出了正在重塑数据收集、应用以及客户互动方式的五大前沿趋势，旨在帮助市场营销人员和数据架构决策者在颠覆性浪潮到来之前看清风险并抓住破局机遇。 ### [01:35](https://www.youtube.com/watch?v=27f16xgmmx4&t=95s) 趋势一：答案引擎与智能体浏览颠覆传统搜索发现路径消费者获取信息的方式正在发生根本性的迁移。过去用户依赖标准的 Google 搜索，逐一点击搜索结果链接、浏览网页并最终选定品牌的发现路径正在加速瓦解。从 Perplexity 推出的具备自主搜索与采购能力的 Comet 浏览器，到 OpenAI 开发的自主智能体（Agents），无一不在改写搜索规则。即便是传统搜索巨头 Google，在其展示“AI 概览”（AI Summary）的情况下，自然结果的点击率（CTR）也暴跌了 50%。这意味着，当 AI 直接在搜索结果顶部给出答案时，有一半的用户甚至不再愿意点击进入任何外部品牌网站。 ```mermaid flowchart TD accTitle: 传统搜索与智能体发现路径对比 subgraph Traditional["传统发现路径"] A["Google 关键词搜索"] --> B["逐个点击蓝色链接"] B --> C["深入网站内容/博客"] C --> D["形成品牌认知与转化"] end subgraph Agentic["智能体/零点击路径"] E["向 AI / 答案引擎提问"] --> F["AI 生成摘要与直接解答"] F --> G["零点击流失 / 极少数高意向跳转"] end ``` 尽管数据显示，从各类 AI 搜索跳转至品牌站点的流量迎来了高达 357% 的同比暴增，但在体量上，这部分流量每天仅产生约 10 亿个数据点，与 Google 每日约 1190 亿次查询的海量基数相比仍显微小。然而，这种指数级增长的趋势不可逆转。对于那些重度依赖内容营销和长篇博客来教育受众的品牌而言，“零点击搜索”无异于灭顶之灾。Zack 提到，在刚刚过去的第二季度财报季中，就有一家以营销见长的大型知名品牌因为自然搜索流量大幅走软而在财报公布后股价重挫 30%。针对这一颠覆，企业需要迅速调整策略： - **将智能体与 AI 助手视为独立流量渠道**：开始在 CDP 和数据管道中专门捕获并标记由智能体带来的推荐流量，将其与普通的人类直接访问流量进行精细化对比分析。 - **发布适合机器阅读的事实片段**：AI 偏好简明扼要的信息片段。品牌在创作内容时应增加摘要、核心结论和列表要点，确保能够轻松被 AI 索引与抓取。 - **利用新闻通稿（Press Releases）拉升权重**：Zack 分享了一个实操小技巧，在 Google 目前的抓取逻辑中，官方新闻通稿被 AI 概览引用的频率显著高于普通的 SEO 优化文章。 ### [05:32](https://www.youtube.com/watch?v=27f16xgmmx4&t=332s) 趋势二：智能体记忆将个性化控制权移交至用户端伴随着用户在日常生活中高频使用 ChatGPT、Perplexity 或微软 Copilot，这些系统正在持续沉淀关于用户的长期记忆（Agentic Memory）。这意味着个性化层（Personalization Layer）正在发生历史性的位移——它不再由各个企业的网站后台决定，而是彻底转移到了消费者自己的智能终端与 AI 智能体手中。用户主动向 ChatGPT 透露自己的偏好、思维习惯、购物意向和钟爱的品牌，而各大科技巨头为了提升月活跃用户（MAU）和留存率，正在展开激烈角逐以图占领这一核心阵地。 Zack 指出，这就是人们多年来一直期盼 Siri 或 Alexa 能够蜕变成为的“真正懂你”的个人专属助手。目前，全球每天在 ChatGPT 上产生的 Prompt 高达 25 亿次，几乎相当于全球 20% 的人口每天都在与 AI 交互。虽然苹果目前的 Apple Intelligence 仍处于起步阶段，但从其宣传策略可以看出，苹果绝不会缺席这场争夺。营销人员对当年苹果推出 iOS 14.5 应用追踪透明度（ATT）政策给广告投放带来的毁灭性打击依旧记忆犹新，因此苹果深度整合的智能体记忆同样是一记不可小觑的重拳。为了应对个性化能力的上移，品牌必须转向“协同个性化”（Co-personalization），主动配合系统而非与之对抗： - **提供可导出的偏好数据**：允许用户将自己在品牌处积累的偏好、心愿单和常购清单轻松导出并注入其个人 AI 助手中，让外部助手在未来检索时能准确召回品牌。 - **重构推荐逻辑**：Zack 以个人经历为例，他每周五都会让 Perplexity 自动整理一份涵盖新上线流媒体、脱口秀专场和播客的清单。虽然作为一名父亲，他忙碌的生活往往抽不出时间观看，但他通过 Perplexity 掌握全局，彻底告别了在 Netflix 或 Hulu 上手动翻找内容的方式——“我不再需要流媒体平台自带的推荐引擎，因为我已经有了属于我自己的引擎。” ### [08:39](https://www.youtube.com/watch?v=27f16xgmmx4&t=519s) 趋势三：随机触点时代的归因重构与测量变革当消费者在品牌官网上耗费的早期探索时间急剧缩水，企业以往引以为傲的测量体系便面临崩塌。传统营销策略依赖于一条清晰平滑的点击流（Clickstream）：一个匿名访客登录网站并四处点击，品牌随之建立受众画像，接着通过发放九折优惠券引导其注册成为已知用户，随后在全网用广告进行“疯狂重定向（retarget them to death）”，直至最终达成转化。然而，在 AI 时代，消费者的触点变成了不可预测的“随机互动行为”（random acts of engagement），他们往往无声无息地在外部由智能体完成大部分决策，然后突然降临在品牌网站上。这种行为模式的变化意味着长路径的点击流数据正走向终结，企业原有的归因模型将彻底失真。为此，技术团队与营销团队必须加快两项基础设施建设： ```mermaid flowchart LR accTitle: 测量体系与数据捕获重构方案 subgraph ClientSide["传统前端依赖 (面临失效)"] A["浏览器 JavaScript 像素"] --> B["客户端重定向追踪"] end subgraph ModernStack["现代数据与测量方案"] C["服务端转化 API (CAPI)"] --> D["实时受众剔除 / 节省预算"] E["零方数据 (直接问询)"] --> F["AI 影响度对照组测试"] end ``` - **全面落地服务端转化 API（CAPI）**：如果企业的广告投放和用户旅程触发依旧依赖浏览器端的前端 JavaScript 像素，随着点击链条被打断，企业极易陷入持续向已购买用户推送无效广告的误区。必须借助服务端接口，在后台交易达成的瞬间立即将用户移出重定向受众包，避免宝贵广告预算的严重浪费。 - **强化零方数据（Zero-Party Data）直接采集**：面对高意向直接落地付款的陌生访客，品牌应当像应对第三方 Cookie 退场时那样，在结账流程中主动设问进行溯源（例如询问“您是如何获知我们的？是否使用了 AI 搜索工具辅助决策？”），并通过设立留置对照组（Holdout groups），严谨评估 AI 对全流程购买决策的真实影响。 ### [11:19](https://www.youtube.com/watch?v=27f16xgmmx4&t=679s) 趋势四：情境胜于人设（Situation vs. Person）长久以来，营销人员极度依赖受众画像（Personas）和人口统计学数据（Demographics）作为核心定向手段，习惯按照年龄段、地理区域或历史倾向分（Propensity Scoring）来构建用户细分，并理所当然地假设“特征相似的人未来也会做出相同的行为”。然而在 AI 智能体（Agent）时代，当访问数据变得愈发碎片化且毫无规律、后台根本无法判断某个代理行为究竟归属于用户 A 还是用户 B 时，传统的预测模型便彻底失灵了。对此，演讲者援引了经典心理学著作《人与情境》（*The Person and the Situation*）中著名的神学院学生实验来阐释情境的决定性力量：研究人员将神学院学生分为两组去完成决定期末成绩的紧急演讲，一组拥有充裕的 1 小时路程时间，另一组则被告知考场临时更换、只有匆忙的 5 分钟；途中，所有学生都会偶遇一名急需救助的无家可归者。结果表明，道德操守极高的神学生是否施以援手，几乎完全取决于他们“是否赶时间”，而非个人品格。这在数字营销中同样成立：用户的当前情境远比历史画像更重要。Tealium 的一家客户在行业大会上分享道，尽管他们拥有海量的历史数据与购买意向预测模型，但实践证明，客户进站 30 秒内的实时行为足以让所有历史模型黯然失色——仅依据访客当前的实时交互展开动态个性化，其转化效果就远超基于历史画像的干预。消费者的采购模式会因情境剧烈波动，返校季、圣诞节与普通周二的购买心态截然不同。面对数据碎片化导致的细分断层，品牌必须放弃虚无缥缈的长周期“下一步最佳行动”（Next Best Action），转而全力优化会话内的“下一步实际行动”（Next Action）——哪怕只是让用户在微时刻（Micro-moments）中多停留一秒、多完成一次点击或回答一个简单提问。通过轻量级的 3 题互动问卷获取零方数据（Zero-Party Data），直接将用户引导至所需内容，反而能沉淀出最真实的高价值细分。 ```mermaid flowchart LR accTitle: 从静态用户画像转向实时情境驱动 subgraph S1["传统画像驱动 (失效)"] A["人口统计与历史标签"] --> B["静态细分模型"] B --> C["长周期 Next Best Action"] end subgraph S2["会话情境驱动 (新范式)"] D["站内前 30 秒行为"] --> E["即时意图与微时刻"] E --> F["会话内 Next Action 推进"] end ``` - **神学院学生行为实验**：验证了个体内在道德与画像在紧迫情境面前影响极微，“赶时间”的情境因素直接决定了交互行为。 - **前 30 秒黄金法则**：Tealium 客户实践表明，进站 30 秒内的会话行为价值远超积累数年的历史倾向分模型。 - **零方数据捕获机制**：用户讨厌冗杂无效的检索，通过极简的主动调研（如“今天您想找什么？”）即时交付目标内容并完成动态分群。 --- ### [15:31](https://www.youtube.com/watch?v=27f16xgmmx4&t=931s) 趋势五：传统客户旅程的终结与重构客户旅程编排（Journey Orchestration）正在迎来根本性剧变，品牌试图单方面规划线性购买路径的时代已经一去不复返。如今的消费者掌握了前所未有的自主权，行为路径极度跳跃，而 AI 的普及更是成为加速这种离散行为的强效催化剂。回顾疫情期间，大量从不网购的群体被迫转向线上并习惯了生鲜配送，当供应链短缺或出现更优价格时，消费者迅速打破原本的品牌忠诚，变得对品牌完全中立（Brand Agnostic）。这种对多元替代选项的敏锐认知在后疫情与 AI 时代进一步巩固：消费者既苛求极致的个性化体验，又极其决绝，一旦遭遇糟糕的交互就会毫不犹豫地离开。演讲者坦言，复杂的客户旅程图在营销策划案上虽然“看起来酷炫、让人兴奋”，但在现实执行中早已形同虚设。营销团队需要将重心从宏观旅程转向“会话内、当下时刻”的即时个性化（In-session, in-moment personalization），聚焦于促成细微的微转化（Micro conversions）。尽管由 AI 和搜索引擎变革引流至站内的绝对访问量可能会下降，但留存流量的购买意向往往更高。面对高意图访客，品牌需要全面优化响应速度：“如果他们正快速走向结账，那就给他们所需的一切，然后赶紧让开路，别做多余打扰；如果他们仍在犹疑，再主动提供支持与答疑。”借助大语言模型（LLM），企业可以在用户浏览时实时生成针对性的异议消除内容（Objection Handling），无需预先人工撰写成百上千篇静态文案，直接化解转化阻碍。 - **线性旅程失效**：复杂的旅程编排图无法映射消费者现实中随机跳跃的决策过程。 - **流量漏斗形态演变**：顶层粗放流量总体收缩，但进站访客的购买意向显著集中，要求站点向高敏捷、零阻碍的结账链路演进。 - **LLM 实时异议化解**：依托访客点击轨迹，由大模型动态生成针对特定顾虑的答疑内容，敏捷清除下单前的犹豫。 --- ### [17:49](https://www.youtube.com/watch?v=27f16xgmmx4&t=1069s) “周一行动清单”与未来展望面对迫在眉睫的变革，演讲者风趣地调侃“今晚肯定没人打算加班”，并为营销团队整理出了一套下周一即可着手推进的实战行动清单。首先，立即建立 AI 智能体流量追踪机制：全面埋点标记各类流量事件，识别具有 Agent 辅助特征的访问属性，在客户数据平台（CDP）中为使用智能体协助交易的用户打上标签，并建立月度复盘机制以捕捉流量趋势的走向。其次，坚决落地服务端追踪与转化 API（Conversion APIs, CAPI）：抛开 AI 因素不谈，传统的客户端 Cookie 与 Pixel 追踪在跨设备购买中本身就极易丢失信号，转向服务端才能确保交易数据的完整回传；在此基础上推进跨渠道的档案拼接（Profile Stitching），并积极探索数据清洁室（Data Clean Room），以应对未来智能体代客闭环下单导致的用户身份隐匿问题。 ```mermaid flowchart TD accTitle: 下周一落地行动框架 subgraph Track["1. 流量识别与追踪"] T1["埋点捕获 Agent 流量属性"] T2["在 CDP 中打标并月度复盘"] end subgraph Data["2. 数据基础设施加固"] D1["部署服务端追踪与 CAPI"] D2["接入数据清洁室 (Data Clean Room)"] end subgraph Action["3. 行为个性化与情境扩展"] A1["试点会话内行为实时个性化"] A2["利用 LLM 生成购物车情境描述"] A3["注入天气、季节等即时环境变量"] end Track --> Data --> Action ``` 最后，应立即启动行为个性化试点。演讲者指出，Tealium 在线下购物中心的研究发现，微小的行为心理刺激足以促使消费者延长逗留时间并增加支出，而这一逻辑在数字端同样成立。营销人员可以借助大语言模型深度理解用户的交互组合——分析访客加购商品的组合特征或在酒店文旅场景中的动作偏好，让 LLM 自动生成富含情境的文字洞察，进而反哺更精准的即时细分。此外，团队应积极引入过去未曾纳入考量的外部环境属性，如用户下单当下的天气、节气与突发事件，这些情境变量将为理解消费动机开辟全新维度。在演讲结尾，演讲者公布了展台互动信息，并引用美剧《足球教练》（*Ted Lasso*）的名言激励全场：“保持好奇，不要妄下断语（Be curious, not judgmental）”，鼓励大家积极探索 AI 的无限可能。 - **周一核心任务清单**： 1. **捕获智能体痕迹**：在全站部署流量事件属性，在 CDP 内标记使用 AI 代理的用户群并按月监测其比例走势。 2. **升级服务端架构**：摆脱脆弱的客户端代码，通过 Server-Side Tracking 与 CAPI 稳固转化归因，联合数据清洁室伙伴应对匿名代购挑战。 3. **落地情境化行为实验**：借鉴线下商场行为学，利用 LLM 对多商品组合、浏览路径撰写情境描述，并融合天气等外部变量丰富用户行为语境。 - **结语寄语**：践行“Be curious, not judgmental”的探索精神，在 AI 浪潮的快速迭代中打破陈规、敏捷试验。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=27f16xgmmx4&t=4s) 有两个按钮我从来不喜欢按：恐慌（Panic）或小睡（Snooze）。 — Zack Wenthe > [01:35](https://www.youtube.com/watch?v=27f16xgmmx4&t=95s) 将智能体与 AI 助手视为独立流量渠道 — Zack Wenthe > [05:32](https://www.youtube.com/watch?v=27f16xgmmx4&t=332s) 我不再需要流媒体平台自带的推荐引擎，因为我已经有了属于我自己的引擎。 — Zack Wenthe > [08:39](https://www.youtube.com/watch?v=27f16xgmmx4&t=519s) 疯狂重定向（retarget them to death） — Zack Wenthe > [08:39](https://www.youtube.com/watch?v=27f16xgmmx4&t=519s) 随机互动行为（random acts of engagement） — Zack Wenthe > [11:19](https://www.youtube.com/watch?v=27f16xgmmx4&t=679s) 情境胜于人设（Situation vs. Person） — Zack Wenthe > [15:31](https://www.youtube.com/watch?v=27f16xgmmx4&t=931s) 如果他们正快速走向结账，那就给他们所需的一切，然后赶紧让开路，别做多余打扰；如果他们仍在犹疑，再主动提供支持与答疑。 — Zack Wenthe > [17:49](https://www.youtube.com/watch?v=27f16xgmmx4&t=1069s) 保持好奇，不要妄下断语（Be curious, not judgmental） — Zack Wenthe ## Mentioned Zack Wenthe, Tealium, Ted Lasso, Perplexity, Comet, OpenAI, Google, ChatGPT, Microsoft Copilot, Apple, Apple Intelligence, Siri, Alexa, Netflix, Hulu, The Person and the Situation, CAPI (Conversion API), CDP (Customer Data Platform), Data Clean Room ## Video description Zack Wenthe, Director of Product Marketing, Tealium Abstracts Is your company’s customer data strategy ready for an AI-driven world? In this talk I will explore five areas that have the potential to blindside most Enterprise marketing organizations. Traditional strategies and methods are no longer enough, but are you prepared to make the shifts necessary to compete tomorrow and into the future? I will walk you through 5 areas that you need to consider, why they're important, and how to adjust your strategies to thrive amid the inevitable disruption. Join me to rethink your approach to customer data before it’s too late. [Source](https://www.youtube.com/watch?v=27f16xgmmx4)

Source quotes

[0:04] Zack Wenthe: 有两个按钮我从来不喜欢按：恐慌（Panic）或小睡（Snooze）。 [1:35] Zack Wenthe: 将智能体与 AI 助手视为独立流量渠道 [5:32] Zack Wenthe: 我不再需要流媒体平台自带的推荐引擎，因为我已经有了属于我自己的引擎。 [8:39] Zack Wenthe: 疯狂重定向（retarget them to death） [8:39] Zack Wenthe: 随机互动行为（random acts of engagement） [11:19] Zack Wenthe: 情境胜于人设（Situation vs. Person） [15:31] Zack Wenthe: 如果他们正快速走向结账，那就给他们所需的一切，然后赶紧让开路，别做多余打扰；如果他们仍在犹疑，再主动提供支持与答疑。 [17:49] Zack Wenthe: 保持好奇，不要妄下断语（Be curious, not judgmental）

Source quotes

Zack Wenthe: Is your company’s customer data strategy ready for an AI-driven world? Zack Wenthe: Traditional strategies and methods are no longer enough, but are you prepared to make the shifts necessary to compete tomorrow and into the future? Zack Wenthe: Join me to rethink your approach to customer data before it’s too late. [0:04] Zack Wenthe: There's two buttons I never like to push. Panic or snooze. [9:03] Zack Wenthe: random acts of engagement [11:22] Zack Wenthe: The Person and the Situation [11:22] Zack Wenthe: What are you here for today? [18:33] Zack Wenthe: iffy at best [20:18] Zack Wenthe: Be curious, not judgmental. Try AI, explore, have fun.

Original transcript

[0:04] going first on uh second day in Vegas is [0:06] always is always a risk uh because I [0:08] think people go out and have have a good [0:10] night and and uh you know dinner and [0:12] then have to decide if they want to get [0:14] up early. So, thank you for joining me [0:15] this morning. Uh I am Zach Wy, director [0:17] of product marketing for uh a company [0:19] called Telium. uh we are a uh customer [0:22] data platform uh data integration and [0:25] data orchestration platform uh for [0:27] customer data and so aptly so I'm [0:30] talking today about customer data [0:31] strategy and the reality is you know AI [0:35] has a lot of exciting opportunities but [0:38] today I'm going to take slightly [0:40] different approach and talk about one of [0:41] the some of the things that may [0:44] derail some of your projects or [0:47] strategies in the coming future But I [0:50] will start with a disclaimer because I [0:52] have to use a Ted Lasso quote. There's [0:54] two buttons I never like to push. Panic [0:56] or snooze. So this is not a panic [0:59] presentation. Uh this is not a sky is [1:01] falling. This is a here are some trends [1:03] that we're seeing that we've talked with [1:05] our customers about that people are [1:07] starting to question that may change how [1:09] you collect data, use data, uh and [1:12] interact with your customers. Uh so [1:14] let's walk through. I've got five [1:16] trends. I'll talk a little bit about [1:17] each one. uh why they why they may be uh [1:20] important and then ultimately we'll talk [1:22] a little bit about the risks that they [1:24] may have and the opportunities uh as as [1:27] marketers or as as data leaders you guys [1:31] have to potentially mitigate or or [1:33] change some of those. So let's dive in. [1:35] Number one uh trend or one trend on the [1:38] horizon that is definitely going to you [1:40] know impact how we use data is answer [1:43] engines or agentic browsing. And I I'll [1:46] combine those together for for this [1:48] purpose. But ultimately, the path of [1:50] discovery is changing. How consumers are [1:53] getting their information [1:56] is is slowly moving away from your [1:59] traditional Google search, click through [2:02] a bunch of links, get some answers, and [2:05] you know, ultimately, you know, pick a [2:07] pick a brand. If you uh if you went to [2:10] the perplexity session, obviously they [2:13] talk about their their browser uh their [2:15] new browser comet uh which allows [2:17] agentic search and and you know [2:19] purchasing open AI has their agents uh [2:22] but even just in Google Google summaries [2:25] alone um are seeing 50% less [2:30] click-through rates when an AI summary [2:32] is presented. Meaning people are leaving [2:35] the search results and going to a link [2:39] 50% less if a summary is shown. Now the [2:43] overall side of that is everybody's [2:45] like, "Yeah, that's great, but we're [2:46] seeing all this new AI search, so we're [2:48] driving traffic. We're people are [2:50] clicking on links." That is true. [2:54] 357% year-over-year growth in people [2:58] going from some AI search into a a [3:02] brand. But it still is only counting [3:04] accounting for about one billion data [3:08] points versus the I think it's 119 data [3:12] uh billion data points a day on Google. [3:14] So it's still it's it's a small piece [3:16] like we get it but it's going to change [3:19] how your traffic comes. It's only going [3:21] to exponentially grow. So uh so think [3:25] about it this way. AI summaries [3:28] resolve queries without click. people [3:31] are not doing [3:34] they're not doing investigation on your [3:36] website especially if you are a [3:38] contentheavy [3:41] brand. If you are expecting consumers to [3:45] come to your site and do education in [3:47] your blog content uh read your [3:49] information that is going to continue to [3:52] diminish. Um I won't I won't name names [3:55] on the stage, but I just saw uh you know [3:57] Q2 earnings report uh from a very large [4:02] uh marketing focused brand who I think [4:05] took a 30% hit in the stock market [4:08] because they said they're their search [4:10] results are soft. They're showing up and [4:12] getting way less organic traffic. [4:15] It's a lot of money. So the opportunity [4:19] start treating assistants and agentic [4:22] browsing as channels. [4:24] Track it when when agents are referring [4:27] traffic. Capture that visit and add it [4:31] to your CDP. Add it to your data [4:33] pipeline so you understand what your [4:35] impact is. I'm still surprised how many [4:38] brands are not comparing agent traffic [4:41] to let's call it human traffic. [4:45] So quick opportunity and start [4:48] publishing assistant readable facts. AI [4:51] likes short sweet snippets. So even if [4:54] you are a publishing heavy brand, think [4:58] summaries, think quick kind of, you [5:02] know, bullets at the end of the article. [5:04] You don't have to make it unreadable, [5:06] but if you can make it scannable for an [5:08] AI, you're more likely to end up [5:10] influencing those those summary results. [5:13] Um, and quick tip, I've noticed on [5:16] Google that press releases um are [5:19] getting indexed significantly higher [5:21] than average um SEO content. So, if [5:26] you're trying to get higher in an [5:29] answer, maybe do a press release. [5:32] All right. Now, adding on to that idea, [5:35] aentic memory. So gentic memory is the [5:37] idea that as you interact with chat GPT [5:40] or as you interact with Perplexity or [5:42] Microsoft Copilot or others they are [5:45] beginning to store memories about you [5:48] and that means the personalization layer [5:50] is moving to the user side. [5:54] It is moving into the hands of the [5:56] consumer because I'm giving Chad GBT [5:59] this is what I like. This is how I think [6:01] this is what I want to buy. These are [6:02] the brands I like. All of that [6:04] information is starting to get gathered. [6:06] And if you don't think AI companies are [6:08] racing to own that category, [6:12] you're wrong. That is a huge opportunity [6:15] to drive monthly users, repeat visitors, [6:18] the more the agent is able to help you, [6:21] right? So, think about uh what we all [6:24] wanted Siri or Alexa to be, right? that [6:29] that that agentic or assistant who who [6:32] knows you. [6:34] That's where we are going. And the [6:36] reality is right now in chat GBT the [6:39] landscape is 2.5 billion prompts every [6:42] single day. [6:45] That's 20% of the world's population [6:47] prompting potentially every day. And on [6:50] the iPhone, they're not there yet. [6:53] They've got a lot of work to do on their [6:54] on their intelligence, but I guarantee [6:58] just looking at their their marketing [6:59] materials, this is a huge opportunity [7:01] for them to drive that personalization [7:04] and own that memory experience. [7:08] And let's not count Apple out. We all, [7:11] if you're a marketer in the room, you [7:13] remember what happened with when iOS [7:14] 14.5 came out and it changed our world [7:17] when they gave people the opportunity to [7:19] not be tracked. and all of a sudden it [7:22] became a whole lot harder to do things. [7:24] So Apple intelligence has has a risk. So [7:26] what do we do about that? All right. [7:28] Start understanding that AI is forcing [7:30] personalization to move upstream. [7:32] Co-personalize by providing exportable [7:35] preferences as a as a uh a customer can [7:38] have to their agent, right? Start giving [7:40] them opportunities to add, you know, oh, [7:43] here's your favorite products. These are [7:45] the things you like. This is what you're [7:46] interested in. Save that in your chat so [7:48] you get better results next time. start [7:50] moving towards that that that idea of [7:53] working with these systems as as opposed [7:55] to against them. Um, you know, I [7:59] personal story. I mean, I' I've seen [8:02] this like as I've started to kind of [8:03] play around with, uh, you know, [8:05] especially Perplexity. I have a task [8:07] that comes every every, uh, Friday that [8:10] tells me about all the new streaming [8:11] shows that have been released, uh, [8:13] standup comedy specials, um, you know, [8:15] interesting things, and some podcasts [8:16] that I want to watch. and I've given it [8:18] kind of what I like and what I want to [8:19] watch. Uh because I'm a dad and I don't [8:23] actually ever get to watch any of them, [8:24] but I like to know what's out there, [8:26] right? And so now I am not going to Hulu [8:29] or or Netflix and searching through the [8:32] list. Perplexity is bringing it to me [8:35] and I don't need their recommendation [8:36] engines. I've got my own. [8:39] Okay. Measurement. Now, as we start to [8:42] move and we kind of tackle these [8:44] together, right? We we think about okay [8:46] if users are spending less time on [8:48] websites if users are spending less uh [8:51] uh time you know in what we'll call top [8:54] offunnel mode discovery mode interacting [8:57] with brands. Now your measurement data [9:00] is going to be all over the place. We're [9:03] moving from this idea of having a click [9:04] stream where somebody lands on the [9:06] website they're an unknown visitor and [9:08] they click around. We start to build a [9:10] profile and ultimately, you know, we [9:13] convert them into some known thing. Um, [9:16] you know, sign up for a coupon or get [9:18] 10% off your next purchase and then [9:20] eventually we retarget them to death and [9:22] then they buy, right? All of this is [9:25] going to become random acts of [9:28] engagement. People are going to come and [9:30] go. They're going to do different [9:32] pieces. you're not always going to know [9:34] that that agentic [9:37] experience or search influence the [9:39] behavior. Now, the opportunity for that [9:43] um you know is is is you have to shift [9:46] your thinking because clickstream data [9:48] is going to potentially go away. Not [9:52] fully, but this idea of the long click [9:54] path, right? So couple things absolutely [9:59] crucial implement server side [10:02] capabilities like conversion API. So if [10:05] you are running um retargeting data to [10:08] ad networks, if you are running anything [10:11] pixelbased, if you are looking at [10:14] triggering um journeys or or next [10:18] actions based on client side [10:20] information, meaning JavaScript on their [10:23] browser, you know, in their interaction, [10:25] if you don't have the server side [10:27] transaction setup to, for example, [10:29] remove them from an ad group, you're [10:32] potentially going to be missing [10:35] that that connect from from a data [10:37] perspective, right? So, conversion APIs, [10:39] make sure you're not wasting your ad [10:40] your ad spent. And then zero party and [10:43] first-party data capture. We talked [10:45] about this with when cookies started [10:46] going away and when it was getting [10:48] harder to know how we could how we could [10:50] target people honestly. The same it's [10:52] the same strategy here. Capture data, [10:55] bring it in. Ask questions. When [10:58] somebody lands on your website as [10:59] they're purchasing, especially if they [11:01] all of a sudden land out of nowhere as a [11:02] high intent, like they're making a [11:05] purchase, ask along the way. How did you [11:07] get, you know, how did you get here? Did [11:09] you use um did you use AI to search? [11:12] Start doing hold out groups and asking [11:14] to see what kind of impact your your [11:17] purchase journey uh is is being affected [11:19] by by AI. [11:22] All right, this one is a little bit more [11:26] I'll call it esoteric um for for lack of [11:29] a a better thing. So, um situation [11:33] versus the person in in marketing for [11:36] years, we have used personas and [11:39] demographics as a targeting mechanism. [11:42] We have gone through and said, "Okay, I [11:44] need to go find a person who is, you [11:47] know, in this age range, in this [11:49] geography, in [11:51] Right? And we have built our segments [11:54] and we have built our audiences based on [11:56] this logic. [11:58] And then on top of that, we have built [12:01] models, propensity scoring, all based on [12:04] this data that once they get there, all [12:06] of these people who look like this will [12:08] continue to do this in the future. [12:10] Now the problem is if we have less data [12:13] to begin with and people are coming in [12:15] in a very sporadic manner and you don't [12:18] know if the agent counts to person A or [12:23] person B, you start to lose the ability [12:25] to do those kind of models. [12:27] But there's there's there's hope. [12:29] There's hope, right? So think about [12:32] context, the situation. There was a [12:36] study years ago um that [12:40] and this is a great there's actually a [12:42] book called uh the person versus the [12:43] situation. I just flipped it for the [12:45] purpose of this. Um but that they took [12:48] seminary students so theology students [12:50] and uh put them through this little [12:52] behavioral study. They half the group [12:55] was given instructions to go across [12:57] campus for their final presentation. It [13:00] was important. It was urgent but they [13:02] had an hour to get there. [13:04] The other group, [13:06] same instructions except for they had [13:08] five minutes. They had to get across [13:10] campus in five minutes. The room had [13:12] moved and now they were forced to be in [13:14] a hurry. Now along the way, every one of [13:18] them interacted with a or or encountered [13:20] a homeless person asking for help. [13:24] Now these are theology students [13:26] generally of good moral compass, right? [13:30] the audience that was in a hurry [13:33] significantly [13:34] less interacted with that person versus [13:37] the people who had time. So the person [13:39] mattered very little in that equation. [13:42] The context and situation that they are [13:45] in mattered way more. We have a customer [13:49] um who who has spoke at our conference [13:51] in the past where they said, you know, [13:53] they've got reams of historical data, [13:56] propensity models, likelihood to buy [14:00] information that you, you know, you [14:02] would use to model. And yet once a [14:05] customer's on the site within 30 seconds [14:07] they can throw all of those away and [14:10] just focus on what that customer is [14:12] doing and personalize based on that and [14:14] have significantly higher outcomes. [14:17] Understand that buyers change modes, [14:20] change behavior. If it's back to school [14:22] time, if it's Christmas, you might be in [14:25] a very different buying behavior than [14:27] you are just on a regular Tuesday. Focus [14:29] on the context. So what do we do with [14:31] this? Because data fragmentation is [14:33] going to lead to targeting and [14:34] segmentation and predictive model gaps. [14:37] Start optimizing and personalizing for [14:39] situational context behavior in session [14:42] human interactions. [14:44] Drive next action on the website. So not [14:48] next best action, which is, oh, this [14:51] person's probably going to buy from me [14:52] sometime down the road. No. How do I get [14:54] them to make that next click on the [14:56] website? How do I get them to stay a [14:58] little bit longer? How do I get them to [15:00] answer one question? Micro moments. [15:03] Focus on that personalization on the [15:04] website and then use that to start [15:07] segmenting zero party data. Ask people [15:10] don't want to end up looking at junk [15:14] content, right? They don't want to have [15:15] to click around to find. It's okay to [15:17] ask them, "What are you here for today? [15:19] How can I help you find it?" Get a [15:21] little three question survey, one [15:24] answer, you know, one answer. Boom. Show [15:25] them and take them right to that [15:27] content. All of a sudden, you've got a [15:28] better segment of your audience. [15:31] And number five, all right, customer [15:34] journey. The customer journey world is [15:37] going to change, especially when you [15:39] think about tools like journey [15:41] orchestration because we are bouncing [15:44] all around. Brands don't get to control [15:47] what they believe the linear path of [15:49] purchase is going to be anymore. [15:51] Consumers are enabled and willing to do [15:55] it themselves. And we've seen this not [15:58] just AI driven, but AI is just going to [16:00] be a forcing function that accelerates [16:02] it. During the pandemic, people who [16:04] never shopped online all of a sudden [16:05] started shopping online. They had [16:07] groceries delivered to their house. They [16:08] became brand agnostic because they were [16:11] willing to switch brands for supply [16:12] chain issues, for oh, just because they [16:15] had a better price. People who were [16:17] locked into a method were forced into a [16:19] new way. And that is just continuing to [16:22] grow. The world has gotten smaller. [16:24] People have realized there are options [16:25] and alternatives. Consumers want [16:28] personalized experiences, but they're [16:31] also very willing to leave a brand over [16:33] a bad experience. So, what can we do [16:36] with that? So, understand that journey [16:38] orchestrations, the investment in these [16:40] long complex journey maps, they look [16:43] cool, they're fun. I'm a marketer. I [16:46] like them. I still get excited by them, [16:47] but yet they they they don't work as [16:50] well as they used to. So focus on in [16:52] session in- moment personalization. [16:54] Drive that immediate next step. Drive [16:57] you know interactions drive small micro [17:00] conversions and understand that your [17:03] traffic coming may be significantly less [17:06] because of the things that we talked [17:07] about leading up to this but it will [17:08] likely be higher intent traffic. So [17:12] optimize for speed. Are they moving [17:15] quickly to purchase? Get out of their [17:17] way. Give them what they need and get [17:19] out of their way. Right? If they are [17:21] still questioning, [17:24] offer support, offer help, answer [17:25] questions, proactively, [17:28] remove objections. We have the ability [17:30] to drive objection handling content on [17:34] the website like we've never had before [17:35] with LLM. Being able to answer questions [17:38] based on what somebody's clicking on, [17:40] you don't have to generate new content. [17:41] I'm just going to answer a question in [17:43] in there, something that I I I can [17:45] assume that that person is looking at [17:46] and and needs answered. [17:49] Um, all right. So, let's wrap this up. [17:51] What to do on Monday? Because obviously [17:52] nobody's going to work on this tonight. [17:55] Um, so what can you do on on Monday? [17:58] Again, number one, start tracking your [18:00] agent traffic. Tag all your traffic [18:02] events. Understand what agent assisted [18:05] attributes uh, you know, are are coming [18:08] to your site and compare that uh, you [18:10] know, flag flag your customers in CDP. [18:12] who's using agents to to assist um you [18:15] know assist their their transactions and [18:18] review it monthly. See where it's going. [18:21] Watch for directionality. Number two, [18:23] server side tracking conversion [18:24] conversion APIs. If you're not doing [18:26] this, this is a huge opportunity. Take I [18:30] mean throw AI away for a second. Just [18:33] the fact that we we when people make a [18:35] purchase on a device, there's a good [18:38] likelihood that that pixel didn't fire [18:40] always, right? So understanding that [18:42] client side tracking, pixel tracking is [18:46] iffy at best. So server side gives you [18:48] the ability to to do a lot of that um [18:51] transactional profile with your ad [18:52] networks. Stitch profiles for [18:54] consistency. Test data clean room. You [18:57] may not get to understand who all of [18:59] these conversion customers are, [19:01] especially if the the agents are are are [19:03] you know buying the full cycle through [19:05] it. So look at data clean room part [19:07] partners and opportunities. And last, [19:09] personalized content. So, start piloting [19:12] that behavioral personalization. Start [19:14] to see what you can influence um to [19:16] drive people on site. There's a really [19:18] cool study uh uh if you come to uh our [19:21] booth, Telium's booth is is 642. We're [19:23] I'll be there all day. Um there's a cool [19:25] study that that was actually done in [19:26] shopping malls that talked about the [19:28] little behavioral impacts that would [19:30] drive people to spend more and stay [19:31] longer at a mall. The reality is it all [19:34] applies to digital, so it doesn't have [19:36] to just be in person. Um, use LLM to add [19:39] context to that behavioral data. Start [19:41] looking at what people are buying, what [19:43] they've clicked on, and the combination [19:44] of things that they're putting in their [19:45] basket or the combinations of things [19:47] that they're doing at the hospitality [19:48] and use LLM to write a contextual [19:51] description of that. So, you can use [19:53] that for more robust segmentation. And [19:56] last one, look at additional attribute [19:59] points that that AI can give you. [20:01] weather, seasonal information, um, you [20:04] know, context on on, you know, what [20:06] happened at that moment, where they [20:08] were, what the weather was like when [20:09] they were buying, because that could [20:11] give you clues into your consumer's [20:13] behavior that you never had, and you [20:14] never had a data point before. [20:18] So, with that, that's my LinkedIn. If uh [20:22] if anybody wants to connect, scan that. [20:24] I'm I'm happy to answer any questions [20:25] because I know we only have 20 minutes, [20:27] so it's a short short window of time. [20:29] But again, my name is Zach Wthy. I'm uh [20:32] I'm with Telium. We're at our booth. Uh [20:34] you know, feel free to stop by and and [20:36] talk about this. Um I'm happy to answer [20:39] answer any questions. And I will leave [20:41] you with the other I have to wrap the [20:44] Tet Lasso. So be curious, not [20:46] judgmental. Try AI, explore, have fun. [20:50] Thank you.