JD Power's Vermillion Sees AI Agents Cutting CSAT Scores 20 Points
Ai4
Failed AI deployments drag customer satisfaction down 20 points, requiring companies to design mandatory human escape hatches and focus on first-contact resolution to protect brand trust.
## Background
Published on YouTube by Ai4 on August 19, 2026, this presentation features Michael Vermillion, Head of Customer Solutions in the FSAAS Division at J.D. Power. According to the page description, the session explores why current AI agent deployments often fail to improve customer satisfaction and provides data-backed frameworks for building high-trust, seamless handoffs between AI and human teams.
## TL;DR
Michael Vermillion argues that successful autonomous agent deployment relies on customer-centric design rather than technology alone. Drawing on extensive J.D. Power survey data across banking, insurance, and telecom, he demonstrates that trust is the primary driver of customer satisfaction, whereas forcing customers into dead-end AI interactions severely damages brand loyalty. To avoid high-profile failures, organizations must implement clear 'escape hatches' to human agents, start with low-stakes interactions, align metrics around First Contact Resolution rather than deflection, and explicitly demonstrate time savings to users.
## Atmosphere & style
The presentation carries a fast-paced, pragmatic, and data-grounded tone as Vermillion powers through 30 slides in under 20 minutes in a hot Las Vegas conference venue. He anchors the session's emotional arc around a vivid contrast between a panic-stricken executive 'bad news room' and a celebrating 'good news room,' creating an urgent yet optimistic atmosphere. The delivery combines authoritative industry expertise with approachable real-world corporate anecdotes, keeping the audience engaged through sharp, actionable advice.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=4s) The success or failure of an autonomous AI agent deployment depends on customer experience design rather than the technology itself.
- [01:48](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=108s) J.D. Power defines customer trust as a consumer's belief that a brand will fulfill its brand promise, making trust the leading driver of customer satisfaction across industries.
- [04:47](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=287s) Highly trusted brands differentiate themselves through seamless cross-channel experiences, clear communication without fine print, proactive updates, and plain-language communication.
- [10:28](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=628s) Consumer AI adoption faces key barriers including ambiguity around AI technology, privacy concerns, and general skepticism regarding its helpfulness.
- [10:28](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=628s) Current J.D. Power telecom data shows that AI interactions succeed only 48% of the time, causing a 20-point drop in CSAT when they fail.
- [13:42](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=822s) Perceived time savings significantly boost consumer comfort with AI, making users five times more likely to feel comfortable using AI agents when time is saved.
- [14:52](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=892s) Organizations should always design an accessible escape hatch to human agents before deploying AI and use escalation data to train their systems.
- [14:52](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=892s) AI agent metrics must shift away from cost-cutting KPIs like deflection or average handle time toward First Contact Resolution (FCR) to maintain long-term loyalty.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=4s) A Tale of Two Meeting Rooms
Opening his talk in a sweltering August in Las Vegas, Michael Vermillion sets a fast-paced tone, warning the audience that he has 30 slides to power through in just 20 minutes. He kicks off with a thought experiment: within the next 12 months, every executive in the room will likely be called into a high-stakes meeting about their company's autonomous agent deployment. That meeting will go one of two ways.
In the "bad news room," the deployment has failed catastrophically. Customers hate the experience, customer satisfaction (CSAT) scores are plummeting, financial goals are missed, and the room is filled with stressed executives either panicked on their phones or sitting in a corner with their head in their hands. Conversely, in the "good news room," the deployment went flawlessly. Customers love the experience, financial goals are exceeded, and everyone is high-fiving and cheering.
Vermillion emphasizes that the difference between these two outcomes is not the technology itself—as tech is a level playing field available to everyone—but rather the design of the customer experience. If companies design AI agents centered around customer needs, they will end up in the good news room; if they fail to do so, they face the bad news room.
- Speaker: Michael Vermillion (Head of Customer Solutions, FSAAS Division at J.D. Power).
- Session context: Ai4 Conference in Las Vegas (August).
[01:48](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=108s) Measuring Customer Sentiment and Defining Trust
To ground his guidance in hard evidence rather than opinion, Vermillion breaks down J.D. Power’s data collection capabilities. J.D. Power interviews over 6 million consumers annually across 150 benchmark studies, covering 1,700 brands in more than 20 industries—a methodology honed over 50 years. Additionally, a newer daily research stream gathers data from 3,000 consumers every day regarding shopping intent and behaviors, ensuring all insights reflect the true "voice of the customer."
Turning to the core topic of trust, Vermillion recalls a 2008 New York City meetup with Alan Adamson (author of *Brand Digital*), who defined a brand with a simple, powerful sentence: "A brand is a promise." Brands like Disney, Apple, Nike, and Spirit Airlines each carry distinct brand promises. J.D. Power defines customer trust as the belief consumers hold that a company will actually deliver on its brand promise.
After J.D. Power recalibrated its customer satisfaction studies to explicitly measure trust, the data revealed that trust is now the single most critical driver of customer satisfaction across nearly every industry covered. High trust directly correlates with overall satisfaction, which in turn drives critical business outcome metrics such as market share, revenue growth, profitability, and positive word-of-mouth.
- J.D. Power Benchmark Data: 6M+ consumers surveyed annually across 150 studies, 1,700 brands, and 20+ industries.
- J.D. Power Daily Data: 3,000 consumers surveyed daily regarding shopping intent.
[04:47](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=287s) What Differentiates Highly Trusted Brands
By analyzing top-performing companies across benchmark studies, J.D. Power identified specific operational habits that set highly trusted brands apart from their competitors:
- **Seamless, Cross-Channel Experiences:** In auto insurance and banking studies, making interactions effortless is the leading driver of trust. A common frustration occurs when a customer moves from a website to a web chat, then to a phone call, and finally to a transferred agent—only to be forced to repeat their information at every single step. Highly trusted brands eliminate this friction by maintaining context across channels.
- **Clarity Without Fine Print:** Brands build trust when customers easily understand what they are purchasing. Top auto insurers excel at clearly explaining policy coverage without hiding caveats behind "17 footnotes."
- **Proactive Communication:** Top-tier brands reach out before the customer feels forced to ask. For instance, when an auto insurance claim is filed, trusted brands push status updates to the customer automatically rather than requiring them to call or check a portal.
- **Eliminating Jargon:** Trusted brands communicate in plain language, avoiding internal acronyms, technical terms, and industry jargon that alienate and confuse everyday customers.
[08:55](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=535s) Real-World Trust Signals in Banking and Insurance
Vermillion highlights concrete examples of companies actively signaling trust to their customer base:
- **Huntington Bank:** The Columbus, Ohio-based regional bank consistently ranks at the top of satisfaction studies due to products like its "Asterisk-Free Checking." The account features no monthly fees, no hidden costs, and a 24-hour grace period to clear negative balances before overdraft fees are applied. The explicit absence of fine-print caveats serves as a powerful trust signal.
- **AMICA Insurance:** The Rhode Island-based insurer features a dedicated page on its website containing a signed letter from the CEO titled "Our Commitment to Ethics." This visible executive pledge creates strong consumer alignment and transparency.
[10:28](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=628s) Current AI Sentiment and the State of Agent Deployments
Addressing consumer perceptions surrounding AI, Vermillion points out that key adoption barriers include uncertainty over what AI actually is, general privacy concerns, and pre-existing skepticism about its helpfulness. Fortunately, clear communication and trust-building can overcome all three obstacles.
A major driver of trust breakdown occurs when virtual assistants act as a wall, preventing customers from reaching a human representative. Drawing on decades of call center and Interactive Voice Response (IVR) research, Vermillion notes that customers need an "escape hatch" to hit zero and speak with a human. Even if they never use it, knowing the option exists provides necessary psychological reassurance.
Data from J.D. Power’s telecom research highlights current friction in AI adoption:
- **Perceptions of AI Usage:** 45% of telecom consumers believe their interactions currently involve AI.
- **Success Rate:** AI interactions currently succeed only 48% of the time, failing more often than they succeed.
- **Satisfaction Impact:** Overall customer satisfaction drops by 20 points when AI is involved in a failed or frustrating interaction. However, when AI works successfully, it generates a dramatic boost in customer satisfaction.
[13:42](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=822s) Driving AI Adoption Through Comfort and Time Savings
Preliminary findings from J.D. Power's upcoming insurance study focus on consumer "comfort" with AI as a mechanism to drive adoption:
- **Time-Saving Multiplier:** When consumers perceive that an AI interaction saves them time, they are **5 times more likely** to feel comfortable using AI.
- **Low-Stakes Gateway:** Positive, low-stakes interactions with AI build confidence, making consumers significantly more willing to adopt AI for higher-stakes or more complex tasks in the future.
[14:52](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=892s) Strategic Principles for Autonomous Agent Design
To help organizations stay in the "good news room," Vermillion outlines five core principles for designing AI agents:
1. **Design the Escape Hatch First:** Always keep a human in the loop. While CFOs may push back to maximize cost savings, removing human backup entirely causes long-term brand damage. *Pro Tip:* Track why customers use the escape hatch to identify gaps and train your AI system.
2. **Start with Existing Trust:** Launch AI agents on low-stakes, simple interactions first to build user familiarity before introducing higher-stakes workflows.
3. **Know Your Agent's Ceiling:** Do not push customers into digital or AI channels for complex problems that the technology cannot resolve—a key failure mode of early digital transformations.
4. **Align Key Performance Indicators (KPIs):** Avoid tracking cost-centric metrics like deflection or average handle time (AHT), which sacrifice long-term loyalty for short-term savings. Instead, center operational goals on **First Contact Resolution (FCR)**, the ultimate driver of satisfaction and trust.
5. **Demonstrate Time Savings Explicitly:** Ensure the agent saves the user time, and actively remind them of it. *Pro Tip:* Take a lesson from airlines where pilots announce early landings over the PA system—explicitly reinforcing efficiency trains customers to associate the brand with speed and convenience.
[18:00](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=1080s) Conclusion and Key Takeaways
Vermillion concludes by urging teams to print out one critical question and tape it to their desks throughout the design process: *"Who are we designing this for?"* If the answer remains genuinely focused on the customer rather than internal cost-cutting, the AI deployment will succeed.
- Upcoming Research: J.D. Power Insurance AI Study publishing at the end of August 2026, followed by a free webinar on September 1, 2026.
## Notable quotes
> [01:48](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=108s) A brand is a promise. — Michael Vermillion
> [18:00](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=1080s) Who are we designing this for? — Michael Vermillion
## Mentioned
Michael Vermillion, JD Power, Ai4, Alan Adamson, Brand Digital, Disney, Apple, Nike, Spirit Airlines, Huntington Bank, AMICA Insurance
## Video description
Michael Vermillion - Head of Customer Solutions, FSAAS Division, JD Power
Are we about to repeat the mistakes of early digital transformations by forcing customers down dead-end AI paths? As enterprises race to deploy autonomous agents, many early rollouts are actually causing customer satisfaction to plummet rather than improve. In this session, I will share J.D. Power data illustrating why current agent deployments fail and how customer readiness must dictate your technical roadmap. I will walk through frameworks for building high-trust, seamless handoffs between AI and human teams that turn agentic workflows into massive satisfaction drivers. Join me to discover how putting user trust at the core of your strategy ensures your AI investments build long-term brand loyalty.
J.D. Power's Michael Vermillion argues that the success of autonomous AI agent deployments depends not on the underlying technology, but on whether the customer experience is intentionally engineered to build consumer trust.
Trust as the Primary Driver of Satisfaction (1:48)
Drawing on J.D. Power research covering more than 6 million consumers annually across 150 studies and 1,700 brands, Vermillion defines consumer trust as the expectation that a company will deliver on its brand promise. After recalibrating its customer satisfaction benchmarks to quantify trust, J.D. Power observed that trust is the single most critical determinant of customer satisfaction across nearly all tracked industries. High satisfaction driven by trust directly correlates with critical commercial metrics, including market share, revenue expansion, profitability, and organic advocacy.
Operational Behaviors that Signal Brand Trust (4:47)
Analysis of benchmark leaders reveals consistent operational habits that establish and sustain trust. In banking and auto insurance, seamless cross-channel continuity—sparing customers from repeating information across web portals, chat tools, and live phone agents—serves as the primary driver of consumer trust. Top performers also eliminate technical jargon and provide upfront transparency without burying terms in footnotes. Real-world illustrations include Huntington Bank's 'Asterisk-Free Checking,' which eliminates hidden fees and offers a 24-hour grace period for negative balances, and AMICA Insurance's transparent executive pledge on ethics.
AI Performance Gaps and the Comfort Multiplier (10:28)
Despite aggressive AI rollouts, J.D. Power telecom data indicates significant friction: while 45% of telecom consumers believe their interactions involve AI, these interactions succeed only 48% of the time. When an AI interaction fails or causes frustration, overall customer satisfaction falls by 20 points, whereas successful interactions produce substantial satisfaction gains. Addressing adoption barriers like privacy concerns and skepticism requires delivering tangible utility; preliminary J.D. Power insurance research shows that consumers who perceive an AI interaction as saving them time are five times more likely to report feeling comfortable using the technology.
Strategic Design Rules for Autonomous Agents (14:52)
To prevent catastrophic implementation failures, Vermillion outlines five design requirements for autonomous systems. First, developers must construct a reliable 'escape hatch' back to human representatives, using escape-hatch exit data to train models. Second, deployments should begin with low-stakes, simple workflows where existing trust already exists. Third, systems must respect technological ceilings rather than routing intricate issues to incapable bots. Fourth, operational metrics should shift away from cost-centric indicators like deflection or average handle time toward First Contact Resolution. Finally, systems should explicitly communicate time savings to reinforce customer perceived value.
Vermillion concludes that organizations must anchor their deployment strategies around the fundamental question of who the system is designed for, cautioning that prioritizing short-term cost cuts over user experience inevitably triggers severe customer dissatisfaction. J.D. Power intends to publish further findings in its insurance AI study at the end of August 2026.
## Background
本视频由 Ai4 频道于 2026 年 8 月 19 日在 YouTube 平台发布，主讲人为 J.D. Power 金融与软件即服务（FSAAS）部门客户解决方案负责人 Michael Vermillion。视频简介指出，当前企业在竞相部署自主智能体时，许多早期尝试因陷入死胡同而导致客户满意度暴跌；演讲结合 J.D. Power 的权威调研数据，分享了智能体部署失败的根源以及如何以用户信任为核心构建人机无缝协作框架。
## TL;DR
演讲指出决定 AI 智能体成败的关键在于客户体验（CX）设计而非底层技术，因为“信任”才是驱动客户满意度与商业回报的核心杠杆。当前电信等行业的 AI 交互成功率仅为 48%，导致用户满意度平均下滑 20 分，其主因在于强行切断人工通道和落入能力盲区。为此，J.D. Power 提出了高信任度智能体设计的五大原则：优先设计人机转接的“逃生舱”、从低风险场景渐进切入、摸清智能体能力天花板、以首次联系解决率（FCR）替代单纯的拦截率，以及通过强化省时优势来快速建立用户舒适度。企业必须时刻围绕“我们在为谁而设计”这一核心拷问，才能避免走入客户体验崩溃的死局。
## Atmosphere & style
整场演讲节奏紧凑凌厉，演讲者在 20 分钟内高密度拆解了 30 张数据幻灯片，展现出极强的专业信服力与实战紧迫感。开篇以极具戏剧张力的“两类会议室”思想实验迅速抓住全场注意力，将枯燥的行业数据转化为生动直观的商业警示。叙事风格兼具宏观数据洞察与犀利针砭，在揭示企业追求成本拦截（Deflection）盲区的严峻现实之余，给出了清晰落地的五大设计框架，令整场分享既发人深省又充满行动导向。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=4s) AI 智能体部署成败的分水岭不在于底层技术的成熟度，而在于是否从一开始就以客户体验（CX）为核心进行工作流设计。
- [01:48](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=108s) 品牌是向客户做出的承诺，而“信任”则是相信企业能兑现该承诺的信念；J.D. Power 数据显示信任是驱动客户满意度、市场份额与盈利能力的最核心基石。
- [04:47](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=287s) 赢得高信任度的品牌普遍具备三大特质：多渠道流转时无缝衔接且无需重复提供信息、条款清晰透明无隐形费用、以及未问先答的主动沟通。
- [05:50](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=350s) 与客户沟通时必须摒弃内部黑话与专业缩写（Jargon），牢记客户并非内部员工，晦涩的用词会直接削弱用户信任。
- [10:32](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=632s) 缺乏人工兜底的死路设计严重伤害体验：电信业中被感知的 AI 交互成功率仅 48%，直接导致客户满意度平均暴跌 20 分。
- [12:00](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=720s) 省时是建立 AI 信任的最强催化剂：当 AI 成功帮助用户节省时间时，用户对 AI 的舒适度会飙升 5 倍；低风险场景下的正面体验能为未来高复杂度交互铺平道路。
- [14:52](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=892s) 必须在开发智能体前优先设计人机协作的“逃生舱”（Escape Hatch），并密切监控用户选择逃生的原因，将其作为系统迭代的重要日志。
- [15:50](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=950s) 清楚掌握智能体的能力天花板，切勿重蹈早期数字化转型覆辙，严禁把客户逼入 AI 注定无法解决的自动化死胡同。
- [16:50](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=1010s) KPI 考核应坚决从追求降低成本的“拦截率”（Deflection）与处理时长转向“首次联系解决率”（FCR），以保障长期满意度。
- [17:30](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=1050s) 在产品体验中不仅要加快处理速度，更要像航班提前抵达进行机上广播一样，主动向用户强化“已为你节省时间”的心理感知。
- [18:00](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=1080s) 企业在推进 AI 战略时应时刻拷问“我们到底在为谁而设计”，唯有真正为客户利益设计，才能赢得长期的品牌忠诚与商业回报。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=4s) 思想实验：一年后的两类会议室与 CX 设计决胜论
在 8 月炙热的拉斯维加斯，J.D. Power 金融与软件即服务（FSAAS）部门客户解决方案负责人 Michael Vermillion 开启了一场节奏极快的演讲。为了在 20 分钟内讲完 30 张幻灯片，他首先抛出了一个令在场管理者深思的思想实验：在接下来的 12 个月内，你必然会被召集参加一场关于 AI 智能体（AI Agent）部署情况的总结会议，而这场会议的氛围只可能有两种截然不同的走向。
第一种是“坏消息会议室”。在这间会议室里，智能体的部署彻底砸了，用户体验一塌糊涂，客户甚至对品牌产生了强烈的敌意。这种糟糕的表现直接拉低了客户满意度（CSAT），财务指标全面告急，未达既定目标。房间里的每个人要么陷入恐慌不停地刷手机，要么蜷缩在角落里双手抱头。而第二种则是“好消息会议室”：智能体部署大获成功，创造了极佳的客户体验，客户对品牌好评如潮，业务指标大幅超额完成，全场都在击掌欢呼、热烈庆祝。
究竟是什么决定了这两种截然相反的命运？演讲者斩钉截铁地指出，**差异绝不在于技术本身**。如今的技术已经非常成熟，普及到了各个角落，大家处于同等的起跑线上，相信企业都能选对技术栈。真正的分水岭在于**客户体验（Customer Experience, CX）的设计**。如果从一开始就以客户为中心来设计智能体工作流，企业就会走进“好消息会议室”；反之，若忽略了用户的真实诉求，就只能沦落至“坏消息会议室”。
- 两种会议室情境：坏消息会议室（体验崩塌、CSAT/财务未达标、团队恐慌焦虑）与好消息会议室（体验极佳、指标超额完成、团队欢呼击掌）。
- 关键结论：技术本身已是平等的竞争场地，最终决定部署成败的是**客户体验的设计**而非技术选择。
---
### [01:48](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=108s) J.D. Power 数据基础与“品牌即承诺、信任即基石”
为了支撑后续的观点，Michael 说明了 J.D. Power 数据的权威性与客观性。J.D. Power 绝非输出个人或专家主观意见，而是通过极其庞大的数据来真实折射消费者的声音：每年深度访谈超过 600 万名消费者，发布 150 多项涵盖 20 多个行业、1,700 多个品牌的标杆研究。此外，除了持续 50 多年的传统标杆研究外，他们还推出了日更的数据流，每天对 3,000 名消费者的购物与意图进行追踪。
在探讨 AI 前，演讲者引出了对“信任”的深度定义。他回忆起 2008 年在纽约的一次聚会中，听取了《Brand Digital》作者 Alan Adamson 对品牌的诠释——**“品牌就是一种承诺”（A brand is a promise）**。无论是迪士尼、苹果、耐克，还是廉价航空精神航空（Spirit Airlines），都有各自独特的品牌承诺。在此基础上，**“信任”（Trust）则是客户相信你能够兑现该品牌承诺的信念**。
几年前，J.D. Power 在重新校准客户满意度研究时，将“信任”引入作为核心评估维度。研究结果显示，在几乎所有行业和研究中，“信任”已成为驱动客户满意度最为关键的因子。信任直接提升客户满意度，进而带动市场份额、营收增长、盈利能力以及口碑推荐等关键财务与业务结果。因此，信任正是企业所有商业活动的基石所在。
```mermaid
flowchart LR
accTitle: 信任驱动商业成果的逻辑链条
subgraph S1["品牌与信任"]
A["品牌承诺 Brand Promise"] --> B["客户信任 Trust"]
end
subgraph S2["业务成果"]
B --> C["客户满意度 CSAT"]
C --> D["财务与市场表现 (营收/份额/口碑)"]
end
```
- 数据体量：每年访谈 600 万+ 消费者，涵盖 20+ 行业、1,700+ 品牌；每日新增 3,000 名消费者购物意图追踪。
- 品牌与信任的定义：品牌即“承诺”，信任即“客户相信企业能兑现承诺的信念”。
- 核心发现：信任已成为驱动满意度的第一大杠杆，也是商业成果的绝对底层逻辑。
---
### [04:47](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=287s) 高信任度品牌的典型特质与商业信号
J.D. Power 对市场上获得高信任度的品牌进行了对比研究，总结出这些优秀品牌与普通品牌之间的三项核心差异，这些特质也为 AI 智能体的设计提供了重要借鉴：
1. **无缝与便捷（Seamless & Easy）**：在银行业，能够做到“与我们做生意很轻松”的银行，其信任度评分极高；在汽车保险业，“无缝的渠道体验”已成为解释信任度的第一大驱动力。所谓“无缝”，是指当客户从官网转移到在线客服、再转接至电话坐席乃至二次转接时，**无需重复提供个人信息与背景数据**。强制客户反复说明情况是造成体验挫败、破坏信任的主因。
2. **清晰易懂、绝无隐形门槛（Easy to Understand）**：高信任品牌不会用密密麻麻的附注和免责声明来误导用户。例如在汽车保险领域，能让客户轻松搞懂“政策到底保障了什么”的公司满意度最高——毕竟在现实中，很少有车主能准确说清自己保单的具体条款。
3. **未问先答的主动沟通（Proactive Communication）**：高信任品牌会在客户提出疑问前主动推送进展。比如在车险理赔流程中，客户无需频繁打电话、发邮件或登录网站查询进度，保险公司会自动将最新的理赔动态推送给客户。
此外，演讲者提醒，品牌在与客户交流时极易陷入内部术语和行业缩写（Jargon）的陷阱。企业必须时刻记住“客户并不是你的员工”，使用晦涩难懂的专业术语只会让用户感到困惑与挫败。
在行业标杆案例中，俄亥俄州哥伦布市的亨廷顿银行（Huntington Bank）推出了名为“无星号支票账户”（Asterisk-Free Checking）的产品。该账户不设月费和隐形费用，并在用户透支时提供 24 小时宽限期，以便其补足余额免受罚款。这种“没有密集脚注文档”的直观设计极大地增强了信任。另一个例子是罗德岛州小型保险公司 AMA，其 CEO 直接在官网首页发表并签署了一封题为《我们的道德承诺》（Our Commitment to Ethics）公开信，向客户作出庄严保证，构成了极为强烈的信任信号。
- 驱动信任的三大关键因素：无缝切换且不重复信息、条款清晰透明、主动推播进展。
- 标杆案例 1：Huntington Bank 的“Asterisk-Free Checking”，免月费隐形费并提供 24 小时透支宽限期。
- 标杆案例 2：AMA 保险公司 CEO 署名的《我们的道德承诺》网页。
---
### [10:32](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=632s) AI 部署现状：低成功率带来的“20 分沉没代价”与舒适度突破口
随着 J.D. Power 在各类调查中增加对 AI 的考量，数据透露出了当前企业 AI 部署的严峻现状。阻碍消费者拥抱 AI 的主要因素包括对概念的不理解、对数据隐私的顾虑，以及预先存在的偏见（“这对我没用”）。但好消息是，这些阻碍完全可以通过透明的沟通和信任建立来解决。
真正导致信任崩塌的是体验中的绝路设计。例如，银行业的虚拟助手（Virtual Assistant）若坚决拒绝将客户转接到人工客服，会引发极大的反感。如同传统的交互式语音应答（IVR）电话系统一样，即便用户可能用不上，但只要系统中保留一个“按 0 键直达人工”的逃生舱（Escape Hatch），就能给用户带来巨大的安全感。强行封死人工通道的品牌，其客户满意度表现极其惨淡。
从用户感知的统计数据来看（非后台实际操作统计，而是用户主观感知），在电信行业中，已经有 45% 的客户互动被用户感知为有 AI 参与。然而令人尴尬的是，**这些 AI 交互的成功率仅有 48%**，失败率高于成功率。这种糟糕的体验导致有 AI 参与的交互中，**客户满意度平均暴跌了 20 分**。
不过，黑暗中依然存在曙光：**一旦 AI 成功解决了问题，客户满意度就会出现极其显著的提升**。在 J.D. Power 即将发布的保险业 AI 专题研究中，探讨了用户对 AI 的“舒适度”（Comfort）问题，并得出了两项核心发现：
1. **省时是立竿见影的催化剂**：如果 AI 能成功为客户节省时间，客户对 AI 的接受与舒适度会**飙升 5 倍**。
2. **小步快跑建立信任**：用户在低风险、低门槛的互动中获得微小而积极的 AI 体验后，会显著增加其在未来尝试高风险、高复杂性 AI 交互的意愿。
- 电信行业 AI 感知渗透率：45% 的客户交互被感知到有 AI 介入。
- 现状瓶颈：AI 交互成功率仅为 48%，导致整体客户满意度暴跌 20 分。
- 突破口：AI 成功解决问题时满意度剧增；AI 成功省时可使舒适度提升 5 倍；低风险小成功可带动高风险场景的信任延伸。
---
### [14:52](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=892s) 设计高信任度 AI 智能体的五大核心原则
结合 J.D. Power 的长期数据，Michael 总结出了企业在设计 AI 智能体时必须遵循的五项黄金原则：
1. **提前设计“逃生舱”（Design the Escape Hatch First）**：在设计智能体之前，就必须把人机协作（Human-in-the-loop）的逃生机制设计好。尽管财务首席官（CFO）可能会出于成本考虑提出反对，但长远来看，彻底斩断人工转接只会带来灾难。专家给出的高级技巧（Pro Tip）是：**密切监控并分析用户为何选择逃生（Escape）**，将逃生行为作为反馈日志，以学习系统的短板和用户的核心诉求。
2. **依托既有信任，从低风险场景切入（Start with Existing Trust & Low Stakes）**：利用用户现有的信任，先部署低风险交互，用高成功率建立舒适感，再逐步引入复杂的高风险场景。
3. **在客户撞墙前，摸清智能体的能力天花板（Know Your Agent's Ceiling）**：吸取早期数字化转型的教训，切忌把用户逼入一个绝对不可能成功的 AI 渠道。如果某个场景目前 AI 无法解决，就应当将其排除在自动化流程之外，切勿强行上线。
4. **慎重选择 KPI，以首次联系解决率为核心（Focus on First Contact Resolution）**：过度追求“偏转率/拦截率”（Deflection）或传统呼叫中心的“平均处理时长”（AHT），虽然能带来短期成本节省，但长期会严重损害品牌。企业应将**首次联系解决率（First Contact Resolution, FCR）**作为核心指标，FCR 表现优异的品牌在信任度和长期满意度上均遥遥领先。
5. **把“节省时间”作为建立舒适度的最快路径（Saving Time as the Fastest Path）**：在体验设计中突出省时优势。专业建议是：**不仅要加快速度，更要主动提醒用户“这为你节省了时间”**。这就好比航班提前降落时，机长会专门广播提醒“我们提前抵达了”一样，这种心理强化能大幅提升用户的正面感知。
```mermaid
flowchart TD
accTitle: AI 智能体设计五大原则框架
subgraph P1["安全与边界"]
A["1. 优先设计逃生舱 Escape Hatch"]
C["3. 明确智能体能力天花板 Ceiling"]
end
subgraph P2["渐进与指标"]
B["2. 从低风险低门槛场景切入 Low Stakes"]
D["4. 聚焦首次联系解决率 FCR"]
E["5. 强化省时体验与心理告知 Save Time"]
end
```
- 原则 1：优先设计 escape hatch，并将转接原因作为系统优化的迭代数据。
- 原则 2：从低风险场景切入，逐步培养舒适度。
- 原则 3：清楚掌握能力边界，严禁将用户推入必败的自动化渠道。
- 原则 4：拒绝盲目追求拦截率（Deflection），以首次联系解决率（FCR）为核心指标。
- 原则 5：主打省时优势，并像航空公司一样向用户进行明确的体验强化广播。
---
### [18:00](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=1080s) 结语：始终拷问“我们在为谁而设计？”
在演讲的尾声，Michael 再次强调了选择权就在企业自己手中——究竟是走入相互指责的坏消息会议室，还是走向击掌庆祝的好消息会议室，决定权取决于当下的智能体设计理念。
同时，他预告了 J.D. Power 即将在 8 月底发布关于保险业 AI 的最新研究报告，并于 9 月 1 日举办免费研讨会，欢迎参会者联系索取邀请。
最后，演讲者留给了全场一句需要打印出来贴在桌上的灵魂拷问：**“我们到底是在为谁设计这个产品？”（Who are we designing this for?）**。如果你的团队回答始终是“为我们的客户而设计”，那么你的 AI 战略必将赢得长期的品牌忠诚与商业成功。
- 行业活动：J.D. Power 将于 8 月底发布保险 AI 研究，9 月 1 日举办免费线上研讨会。
- 核心灵魂拷问：办公桌上应贴上的标语——“Who are we designing this for?”，时刻坚持以客户为中心。
<ElicitationsGroup message="如果您希望对这份深度演说进行更具体的延伸探讨，可以参考以下方向：">
<Elicitation label="生成一份‘AI 智能体逃生舱与人机转接机制’设计 checklist" query="请根据视频中提到的人机协作与 escape hatch 原则，为我生成一份详细的 AI 智能体逃生舱与人机转接机制设计 checklist。"/>
<Elicitation label="对比解析 Deflection（拦截率）与 FCR（首次解决率）对智能体架构的影响" query="请深入对比解析为什么以 Deflection 为 KPI 会损害品牌，以及如何基于 FCR（首次联系解决率）重新设计 AI 智能体的评估指标体系。"/>
</ElicitationsGroup>
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=4s) 差异绝不在于技术本身 — Michael Vermillion
> [01:48](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=108s) 品牌就是一种承诺 — Alan Adamson
> [01:48](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=108s) 信任则是客户相信你能够兑现该品牌承诺的信念 — Michael Vermillion
> [04:47](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=287s) 无缝的渠道体验 — Michael Vermillion
> [04:47](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=287s) 客户并不是你的员工 — Michael Vermillion
> [10:32](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=632s) 这些 AI 交互的成功率仅有 48% — Michael Vermillion
> [10:32](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=632s) 客户满意度平均暴跌了 20 分 — Michael Vermillion
> [10:32](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=632s) 如果 AI 能成功为客户节省时间，客户对 AI 的接受与舒适度会飙升 5 倍 — Michael Vermillion
> [14:52](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=892s) 密切监控并分析用户为何选择逃生 — Michael Vermillion
> [14:52](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=892s) 不仅要加快速度，更要主动提醒用户“这为你节省了时间” — Michael Vermillion
> [18:00](https://www.youtube.com/watch?v=OhMm7LfZIEw&t=1080s) 我们到底是在为谁设计这个产品？ — Michael Vermillion
## Mentioned
Michael Vermillion, J.D. Power, FSAAS, Alan Adamson, Brand Digital, Disney, Apple, Nike, Spirit Airlines, Huntington Bank, Asterisk-Free Checking, AMA, Our Commitment to Ethics, Ai4
## Video description
Michael Vermillion - Head of Customer Solutions, FSAAS Division, JD Power
Are we about to repeat the mistakes of early digital transformations by forcing customers down dead-end AI paths? As enterprises race to deploy autonomous agents, many early rollouts are actually causing customer satisfaction to plummet rather than improve. In this session, I will share J.D. Power data illustrating why current agent deployments fail and how customer readiness must dictate your technical roadmap. I will walk through frameworks for building high-trust, seamless handoffs between AI and human teams that turn agentic workflows into massive satisfaction drivers. Join me to discover how putting user trust at the core of your strategy ensures your AI investments build long-term brand loyalty.
[Source](https://www.youtube.com/watch?v=OhMm7LfZIEw)
Source quotes
[0:04] Michael Vermillion: 差异绝不在于技术本身
[1:48] Alan Adamson: 品牌就是一种承诺
[1:48] Michael Vermillion: 信任则是客户相信你能够兑现该品牌承诺的信念
[4:47] Michael Vermillion: 无缝的渠道体验
[4:47] Michael Vermillion: 客户并不是你的员工
[10:32] Michael Vermillion: 这些 AI 交互的成功率仅有 48%
[10:32] Michael Vermillion: 客户满意度平均暴跌了 20 分
[10:32] Michael Vermillion: 如果 AI 能成功为客户节省时间，客户对 AI 的接受与舒适度会飙升 5 倍
[14:52] Michael Vermillion: 密切监控并分析用户为何选择逃生
[14:52] Michael Vermillion: 不仅要加快速度，更要主动提醒用户“这为你节省了时间”
[18:00] Michael Vermillion: 我们到底是在为谁设计这个产品？
Source quotes
[1:48] Michael Vermillion: A brand is a promise.
[18:00] Michael Vermillion: Who are we designing this for?
Original transcript
[0:04] Uh so welcome to uh Las Vegas in August everybody.
Uh I've got like 30 slides to get through in 20
[0:09] minutes. We're going to go pretty quick. I thought
we would start with a uh a thought experiment.
[0:14] And uh that would be you are likely
going to be called into a a meeting
[0:19] within the next 12 months. And that meeting is
going to go one of two ways. Uh the first way
[0:26] uh let's call this the bad newsroom. Uh this is
um your agent deployment just didn't go well. Uh
[0:34] your consumers are not having a great experience.
As a matter of fact, your your customers hate you.
[0:39] Uh that's that's showing up in your seat
results. That's also showing up in your
[0:44] financials. You're missing your goals. Everybody
in the room is either panicked and on their phone
[0:49] or they're sitting in the corner with their their
head in their hands. Uh the other way it can go,
[0:54] uh let's call this the good news room. Your agent
deployment went very well. uh you're delivering a
[0:59] great customer experience. Your customers love
you. That's showing up in your results. You're
[1:04] you're exceeding your goals and everybody is
either high-fiving or cheering uh in the room.
[1:12] So, the question is what's the
difference between those two outcomes?
[1:16] And we would argue that it's not the
technology. The technology is here.
[1:21] It's in the building. It's all around you.
The technology is really a level playing
[1:24] field. We're very confident that you will choose
the right technology. uh the difference between
[1:29] those two outcomes, it's really going to be in the
design of the customer experience. If you design
[1:35] the experience with your customer in mind, design
it around the customer, uh you're more likely to
[1:41] end up in the good news room and if you don't,
more likely to end up in the bad news room.
[1:48] Uh so I thought I'd share some um some of our
research here at JD Power around AI and trust. Uh
[1:57] before I do that, uh just wanted to let you know
kind of where this data is coming from. So we um
[2:03] we interview uh over 6 million consumers a
year, publish over 150 studies covering 1,700
[2:09] brands across more than 20 industries.
That's our traditional benchmark uh
[2:14] study research that we've been doing for over 50
years. More recently, we've introduced another
[2:20] research stream where we're interviewing 3,000
consumers every day uh around shopping and intent
[2:26] and uh we publish that data daily. So
the p the key point here, the takeaway
[2:31] is that what I'm sharing with you is based on data
and facts and and holding up a mirror from what
[2:36] consumers are saying, the voice of the customer,
and it's not our our opinion or expert opinion.
[2:45] Um, so let's let's talk I thought I thought what
we would do is talk a bit about trust and then
[2:50] talk a bit about what we know about AI and then
we'll kind of bring everything together and um
[2:55] and wrap that up into some principles that you
can use for designing your your AI agents. Uh
[3:00] before we dive into the the research around
trust, I think it's important to define it.
[3:06] And to define trust, let's start with defining
brand. So, I was in a uh a meetup in New York
[3:12] City in 2008 with a guy named Alan Adamson. He was
on a book tour for his book called Brand Digital.
[3:18] And when asked the question, how do you
define brand? Uh Allan described it this way.
[3:24] A brand is a promise. And I think that's such a
powerful way to think about brand. It's really
[3:29] kind of stuck with me all these years. Just think
about Disney. What's the Disney brand promise?
[3:36] Um, Apple, what's the Apple brand promise?
Uh, Nike, what's that brand promise?
[3:43] But other brands, you know, Spear Airlines,
they had a brand promise as well. Uh, so, uh,
[3:49] think about what your brand is, what your brand
promises. Uh, and then let's define trust as
[3:56] the belief that your customers have that
you're going to deliver on that brand promise.
[4:02] A couple years ago, we rec-calibrated
our customer satisfaction studies
[4:05] and introduced trust as a dimension. And what we
found over the past couple of years is that trust
[4:11] is now uh a very important if not the most
dimension or driver of customer satisfaction
[4:18] across almost every single study that we
do uh every single industry that we cover.
[4:24] Uh and so trust and satisfaction are highly
correlated. Of course, satisfaction ends up
[4:29] driving your business uh results. Uh things
like uh market share and revenue growth and
[4:35] and uh and profitability and of
course um uh word of mouth and and
[4:39] reference. So think about trust is really kind
of the foundation uh of everything that you do.
[4:47] Uh so what do we what have we learned about uh
trust and um the way we answer that question is
[4:55] we look at brands that are highly trusted uh and
then we just ask the question what are they doing
[5:02] differently than the other brands? Why why do they
have a higher level of trust than the other brands
[5:06] in their industry? And uh the first one is this
idea of make it seamless and easy. So a key trust
[5:13] dimension in our banking study uh the brand uh
the banks that rate very high in the question of
[5:21] uh we make it easy to do business with us also
rate very high in terms of trust. In our uh auto
[5:28] insurance study a seamless channel experience
is now the number one driver or or dimension
[5:35] uh that explains uh trust. What we mean by
seamless is that uh when a a customer contacts you
[5:43] to ask a uh a question or get a problem resolved
uh that um if if they go on to your website and
[5:50] then have to move into a chat uh and then move
from the chat into a call with an agent and
[5:55] then that agent then transfers them to another
agent. Uh the question is are are you forcing
[6:01] your customer to repeat all of their information
every time they change channels? And if you do,
[6:09] that's that is not a seamless experience. It's
a highly frustrating experience. And it's a um
[6:15] a true differentiator between brands that have
high levels of trust and and ones that uh don't.
[6:23] uh a second uh driver of trust and and
differentiator between brands that are highly
[6:27] trusted and brands that that have lower levels of
trust is just making it easy for your customers to
[6:33] understand what they're getting. Uh so in banking
uh that would be a a credit card or a checking
[6:41] account where there's not tons of footnotes on the
page in terms of here's uh what we're promising
[6:47] you but you have to read the 17 footnotes in
terms of all the caveats. Uh when it comes to um
[6:54] insurance, car insurance, uh car insurance
companies that do very well with trust and
[7:00] satisfaction do a good job of helping their
customers completely understand what is covered
[7:06] by their policy. And that's not an easy thing to
do. You know, if I had did a poll in the room of
[7:12] can anybody in this room actually explain
what's covered by their car insurance,
[7:15] I I I very few people could actually, you know,
offer uh that up. So helping people uh understand
[7:23] um show up before they have to ask. And so
another differentiator we see between highly
[7:31] trusted brands and ones that are not uh are brands
that are proactive about their communications
[7:37] with their customers. So, for example, if you
file a claim uh with your car insurance company,
[7:43] uh you shouldn't have to pick up the phone or send
an email or go onto the web and try and figure out
[7:49] what is happening, what is going on. The brands
that do really well are actually pushing that uh
[7:54] out to the the consumer without them asking.
They're they're being proactive about it.
[7:59] So another key driver uh so I think you know all
three of these things uh translate pretty well
[8:06] as you start to think about the design of your
AI agents. Uh one um key point I I can't actually
[8:15] see these slides guys I don't if there's another
way to change the monitor or not. Uh but uh
[8:20] the key point in the middle I know is this term
uh jargon. Uh so understand and one mistake
[8:27] that we see brands make is that when they're
interacting with customers, they tend to slip into
[8:33] acronyms or terms that you use within your company
or terms that you use within your industry.
[8:38] But you got to remember that your customers aren't
you. They're they don't work at your company. They
[8:43] don't work in your industry. And so if you're uh
if you're interacting with them in a way where
[8:47] you're um speaking in jargon, you're going to
confuse your customers and uh and frustrate them.
[8:55] Uh so what does uh trust look like uh in in
terms of highly trusted brands communicating
[9:01] or signaling trust? Uh so the first example,
Huntington Bank uh based in Columbus, Ohio
[9:07] Regional Bank, they do super well in our studies
year after year and they have a product called
[9:13] asteriskfree checking and it's a really
simple idea. It's a checking account that
[9:18] doesn't have a monthly fee. Uh it doesn't have
hidden fees. uh if you overdraw your account,
[9:24] they give you a 24-hour grace period so that you
can uh bring the balance back up before they they
[9:29] hit they do hit you with the overdraft fee. And uh
and so it's a it's this kind of the idea that it's
[9:35] asterisk free. There aren't a bunch of footnotes
associated with it. Uh that's a uh a way to signal
[9:41] trust and Huntington is a highly trusted brand. Uh
another example from our insurance uh work is AMA.
[9:48] Uh AMA is a small insurance company based in Rhode
Island and they do very well. Their customers,
[9:54] AMA customers love them and uh if you go to the
AMA website, what you're going to find is a page
[10:00] where the CEO has written a letter uh titled
our commitment to ethics. So this is a uh
[10:07] a promise to customers signed by the CEO. It's on
their on their web page. very strong signal of um
[10:15] of trust us and I'm not sure how many we
[snorts] don't see this very often and so
[10:21] I'm not sure if if your companies have this kind
of commitment to ethics uh page on your website
[10:28] okay so what do we know about um AI so we we've
been we've been sprinkling AI questions across
[10:37] our studies for a while we're doing more and more
concentrated work uh around AI in our uh research.
[10:44] Uh so I just want to share a few uh results
from that uh that work that we're doing.
[10:50] Uh so uh one question uh that we're looking at
trying to understand is what is keeping people
[10:57] back from adopting uh adopting AI. And it's
it's it's it's I don't think it's a big surprise
[11:04] here, but it's things like um you not really sure
what it is. uh some concern around data privacy.
[11:11] Uh some people have already made up their mind
like just doesn't seem like that's something
[11:15] that would work for me. But the good news is
that you can address all of these concerns
[11:21] with uh trust and and communication. Uh
so another um reason to to focus on trust.
[11:32] Uh we're also looking for um
[11:35] instances where trust is breaking down in the
experience. And so a good example of this is uh
[11:42] a virtual assistant in banking that refuses to
let the customer have access to a human being.
[11:50] Uh so we we know from our prior work in call
center and with IVR and that kind of thing.
[11:56] Uh that's just a very frustrating experience for
customers if there's no way to hit zero and get
[12:01] out of the IVR for example uh to get to to to get
to a human people. Even if people don't need that,
[12:09] they like having the comfort of knowing that they
can have that um escape hatch, if you will, uh
[12:14] if they need it. So, uh, so brands that, um, that
force customers in a channel with no access to
[12:22] humans tend to do worse than, uh, ones that where
you can, uh, punch out and get to a human being.
[12:31] Um, how uh what kind of incidence rate are
we seeing in the in the field today in terms
[12:36] of experience with AI? Uh, so um, these are
not operational statistics. This is based on
[12:43] consumer perception and recall. And so the
consumer may or customer may or may not be
[12:49] actually interacting with AI, but they think
they are. And if they think they are, that that
[12:54] perception is, you know, your reality. And so
in our telecom work, we're finding now that 45%
[13:01] of um of customer interaction, the consumers
perceive that is um involving AI somehow.
[13:08] Uh the bad news is that it only works 48% of the
time. Uh so right now it's not a great experience.
[13:15] Uh the uh AI solution is failing more often than
it's succeeding. So it's not a surprise that
[13:23] um if if AI is involved, we're seeing overall
a 20 point reduction in customer satisfaction.
[13:32] Uh but the light at the end of the tunnel
is that if AI is involved and it works,
[13:36] we actually see a pretty dramatic increase in
customer satisfaction. So that's the good news.
[13:42] uh we have an insurance uh study uh around
AI and insurance that's publishing at the
[13:46] end of this month and uh we're looking
through some of the data early data
[13:50] early results on that looking for some
nuggets. So we're exploring this term
[13:54] uh called comfort and and so we're trying to
understand consumer or customers comfort with AI
[14:00] uh with the with the idea of like how do we
do a better job of getting people to adopt AI.
[14:05] And one thing we found was that if you can
find a way to save a customer time with AI,
[14:13] they're five times more likely to be comfortable
with AI. So that's a um something they may want
[14:20] to dig into and explore uh in your own designs.
kind of a second data point from the same study
[14:27] is that uh we're finding that if consumers have
some kind of a small positive experience with AI,
[14:34] even if it's a low stakes um uh interaction,
they're more likely to be open to adopting
[14:42] higher stakes u um kind of more dangerous if you
will type interactions with AI. Um so let's um
[14:52] take all that data and and map it into principles.
So principle number one is uh design the escape
[14:59] hatch uh while you're designing the agent
before you design the agent. Uh so remember
[15:05] you got to keep the human in the loop um at um at
some point. And your CFO is going to fight you on
[15:12] this. Uh that's okay. uh you can have him or her
call me and I'm happy to talk them through it. Uh
[15:19] what we see is that you know if if you if you take
the human you know out of the loop permanently or
[15:24] or make it impossible to reach a human uh that
may work in the short run. It doesn't end well
[15:29] in the the long run. Uh and then the pro tip here
is that let people escape but but figure out you
[15:36] watch and learn from why they're escaping.
That's going to teach you about your own system.
[15:41] is going to teach you about the problems they're
trying to solve questions they're trying to ask.
[15:45] Uh principle number two, um so start with uh
trust that already exists. So again, the idea
[15:53] start with low stakes interactions. Get
people comfortable with it. Hey, it works.
[15:57] I'm now ready to to introduce uh higher stakes
interactions. Uh three, know your uh agents um
[16:06] ceiling before your customer finds it. So this
is a mistake we found with the digital rollout.
[16:12] People were for brands were forcing consumers into
the digital channel uh in certain situations where
[16:18] there was zero possibility that they would be
successful in the digital channel. And that's
[16:23] a uh uh that was a highly frustrating experience
for people. So you need to understand if you're
[16:29] if you're putting people in a channel, can they be
successful? If if there's no possibility of that,
[16:33] then set that off to the side and introduce it
later. Uh next um principle is to um be careful
[16:41] with what you're measuring your KPIs. Uh so if
you're measuring things like um deflection uh in
[16:47] the in the call center days was called average
handle time. Uh again that delivers short-term
[16:52] benefit but um it's going to really hurt you uh
in the long run. So the if if we're really looking
[16:58] for a metric to to focus on and center on, it's
first contact resolution. brands that do really
[17:04] well with first contact resolution do super well
with trust and and satisfaction down the road.
[17:12] Uh and then um finally uh saving time is
your fastest path to comfort that the idea
[17:19] of getting people to be more comfortable with
adopting AI. So um I'd like to that's thisly
[17:27] still kind of an early research nugget for us. uh
if you try this and have good experience with it,
[17:32] you know, please uh let me know. And really
the pro tip on this one is to not just make
[17:38] it faster, but to let remind people that it
is faster. So if you've ever been on a flight
[17:43] that arrived early, and then the captain of the
plane comes on and says, "Hey, just as a reminder,
[17:48] we arrived uh early on this flight." Uh that's
because of our research that we're we're training
[17:53] the airlines to to train their uh their crews to
do that just to reinforce it. So, so the the take
[18:00] the the choice is yours. Are you going to be in
that bad news room or that good news room? And the
[18:05] and the choice really needs to be now early
in the design of your agents. Uh a quick ad,
[18:11] we have a u our next research published at the end
of August for the insurance industry and we have a
[18:17] webinar coming up uh September 1st. More than
welcome to join that. There's no cost for it.
[18:26] And um I had a QR code. Send me an email if you
want an invite to that. And then uh the the thing
[18:33] I'd like you to take out of the room today uh is
this idea that um the question you need to keep
[18:38] asking yourself, asking your team, asking your
organization is who are we designing this for?
[18:44] So something to print out and and tape on your
desk. And uh if you if you if the answer to that
[18:49] question is you're designing it for your customers
uh you're going to do very well. So that's it.
