Princess Cruises Gains $120M as Harpin AI Routes High-Intent Leads

BLINDSPOT - How Princess Cruises unlocked $120M of Incremental Revenue using Contact Center signal — Ai4

Ai4

Equipping human sellers with real-time customer signals from service calls rather than deflecting inquiries lifted high-intent outreach to 91% and unlocked $120 million in incremental revenue.

AI-assisted summary of BLINDSPOT - How Princess Cruises unlocked $120M of Incremental Revenue using Contact Center signal — original source. Check the original for full context.

Source published: 2026-08-21

Detailed notes

## Background Published on YouTube by Ai4 on August 21, 2026, this presentation features Camille Bailey, Associate Vice President of Consumer Sales at Princess Cruises, and Gregg Hough, Chief Commercial Officer at harpin AI. The video addresses the industry blind spot of viewing contact centers merely as cost centers and demonstrates how analyzing millions of customer conversations unlocked $120 million in incremental revenue.

## TL;DR Facing a 4% direct sales decline and finding that agents spent 77% of their shifts searching through disconnected legacy databases, Princess Cruises partnered with harpin AI to transform raw customer conversations into actionable sales intelligence. By consolidating guest history and tracking real-time digital signals into a unified concierge interface, Princess increased high-intent guest contact rates from 51% to 91% and generated $120 million in incremental revenue. Through a live demonstration of harpin AI's Acuity platform, the speakers showcase how real-time conversation diagnosis exposes the compounding costs of repeat calls, extracts competitor battle cards, and enables macro-level sensing during geopolitical crises to protect bookings and adapt corporate policies.

## Atmosphere & style The presentation carries a sharp, pragmatic, and data-driven corporate conference tone, refreshingly grounded in frontline realities rather than generic AI hype. Camille Bailey provides calm, lived operational authority reflecting two decades in cruise sales, while Gregg Hough delivers a fast-paced, assertive live platform walk-through that cuts straight to bottom-line numbers. Watching the session feels like sitting in an executive war room as the narrative progresses from frontline agent friction to high-stakes crisis response and eight-figure commercial turnarounds.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=8qniFNI-ayE&t=4s) Treating contact centers strictly as operational cost centers rather than rich customer intelligence engines is a major industry blind spot. - [01:41](https://www.youtube.com/watch?v=8qniFNI-ayE&t=101s) Before intervention, Princess Cruises faced a 4% year-over-year decline in direct sales, with agents spending only ~2 hours of an 8-hour shift actively selling. - [02:25](https://www.youtube.com/watch?v=8qniFNI-ayE&t=145s) Princess sellers previously lost 77% of their time (5 of 8.5 hours) manually hunting through five disconnected legacy databases to find customer offers and context. - [02:25](https://www.youtube.com/watch?v=8qniFNI-ayE&t=145s) Consolidating customer intelligence into a single view with real-time digital 'hand raises' increased high-intent guest outreach from 51% to 91% and drove $120 million in incremental revenue. - [06:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=377s) harpin AI's Acuity platform shifts management from lagging outcome metrics to root-cause diagnosis, tracking whether service resolutions earned the right to sell and if closes were attempted. - [10:08](https://www.youtube.com/watch?v=8qniFNI-ayE&t=608s) Customer Effort Score (CES) and real-time interaction sentiment serve as predictive leading indicators of churn and loyalty compared to lagging post-call surveys. - [10:08](https://www.youtube.com/watch?v=8qniFNI-ayE&t=608s) Repeat contacts consume over 50% of contact center budgets, with a customer's second call costing 27% more than their initial interaction. - [11:33](https://www.youtube.com/watch?v=8qniFNI-ayE&t=693s) Cross-contact intelligence allows leaders to address adjacent questions proactively in the first call and automatically generates competitor battle cards from customer mentions. - [12:32](https://www.youtube.com/watch?v=8qniFNI-ayE&t=752s) Macro conversational sensing across 12,000 interactions identified 1,700 guests considering cancellation due to Middle East missile strikes and exposed a critical force majeure policy loophole for stranded travelers. - [15:05](https://www.youtube.com/watch?v=8qniFNI-ayE&t=905s) Closed-loop outbound voicemail analysis reverse-engineers the exact conversational hooks and urgency drivers that lead to callbacks and completed bookings. - [15:05](https://www.youtube.com/watch?v=8qniFNI-ayE&t=905s) The platform synthesizes the tactics of top-performing sellers into auto-generated knowledge base articles ready for supervisor review and ingestion. - [15:05](https://www.youtube.com/watch?v=8qniFNI-ayE&t=905s) Rather than attempting total agentic replacement, the platform categorizes inbound volume to identify which contacts are genuinely safe to deflect to AI agents versus those requiring empowered humans.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=8qniFNI-ayE&t=4s) Introduction and the Industry-Wide Blind Spot

Gregg Hough, Chief Commercial Officer at harpin AI, introduces the session by clarifying the company's core philosophy. Rather than chasing fully autonomous agentic replacement, harpin AI focuses entirely on empowering human contact center agents by giving them the right data at the right moment so they can do their jobs effectively. Gregg introduces his partner on stage, Camille Bailey, Associate Vice President of Consumer Sales at Princess Cruises.

Camille reflects on her twenty-year career leading sales and service operations across major cruise lines—including Carnival, Celebrity Cruises, Crystal Cruises, and now Princess Cruises. She recalls taking her very first call in a cruise contact center two decades ago, sitting on top of a literal goldmine of customer signals without anyone realizing it at the time. Across the industry, the prevailing blind spot has remained stubbornly consistent: organizations treat contact centers merely as cost centers to be minimized rather than recognizing the rich customer intelligence they generate. Princess Cruises partnered with harpin AI to expose that blind spot, understand the underlying signals, and turn missed conversations into substantial revenue.

```mermaid flowchart TD accTitle: Contact Center Signal Transformation Flow subgraph Traditional["Legacy Contact Center Approach"] A["Customer Conversations & Calls"] --> B["Treated as Operational Cost Center"] B --> C["5 Disconnected Tools / Guesswork"] C --> D["Low Productivity & 51% High-Intent Reach"] end

subgraph Harpin["harpin AI Signal Transformation"] E["Real-Time Signals & Digital Hand Raises"] --> F["Acuity Signal Engine & Diagnosis"] F --> G["Single Pane of Context for Agents"] G --> H["91% High-Intent Reach & $120M Incremental Revenue"] end ```

[01:41](https://www.youtube.com/watch?v=8qniFNI-ayE&t=101s) Diagnosing the Gap: Low Agent Productivity and Disconnected Data

Princess Cruises faced a 4% year-over-year decline in direct sales. The organization had no shortage of data, but it lacked clear operational direction. When examining an average eight-hour agent shift, leadership discovered that agents were generating only about two hours of actual selling productivity, giving the appearance of minimal operational output.

To understand the breakdown, harpin AI audited millions of guest interactions to diagnose the contact center funnel. The platform analyzed who the guests were, pinpointed which prospects possessed active intent to book, identified which leads were being completely ignored, priced the monetary value of that missed opportunity, and ranked the operational fixes required to capture it.

- **Sales Deficit:** Direct sales were down 4% year-over-year prior to intervention. - **Productivity Bottleneck:** Out of an 8 to 8.5-hour shift, agents averaged only ~2 hours of active productive conversation. - **Core Diagnostic Steps:** Auditing conversations, identifying high-intent uncontacted guests, pricing the revenue gap, and ranking actionable fixes.

[02:25](https://www.youtube.com/watch?v=8qniFNI-ayE&t=145s) Empowering Sellers: Real-Time Hand Raises and Unified Context

To bridge the gap, Princess Cruises deployed a unified concierge sales application that eliminated the manual guesswork of outbound calling. Instead of agents juggling five disconnected systems to hunt for leads, the platform consolidated all guest intelligence into a single view. The system continuously tracks digital "hand raises"—for example, detecting when a guest who spoke with an agent a month ago suddenly browses a cruise on the website. That trigger immediately flags the lead back to the assigned agent in real time, enabling the seller to call with warm context: *"Camille, it looks like you're ready to talk about that cruise. What other questions do you have? How can I support you?"*

Gregg explains the operational scale: Princess employs roughly 400 sellers, each assigned a portfolio list of approximately 8,000 guests. In a 65-year-old company, legacy data was buried across multiple databases. Previously, agents spent 77% of their time—nearly five hours of an eight-and-a-half-hour shift—manually interrogating systems to figure out who had a casino offer, a Europe promotion, or free airfare. By serving the right guest at the right time with full behavioral history (e.g., past sailing cadences, favorite destinations like Alaska, upcoming wedding anniversaries, or a daughter graduating high school), agents reach prospects with maximum relevance.

This shift altered performance metrics across the sales floor, increasing outreach to high-intent guests from 51% to 91% and directly driving $120 million in incremental revenue.

```mermaid flowchart LR accTitle: Seller Workflow Comparison subgraph Before["Before: 77% Time Searching"] A1["Agent Portfolio (8,000 Guests)"] --> B1["Search 5 Legacy Systems"] B1 --> C1["Guess Offers (Casino, Europe, Flights)"] C1 --> D1["51% High-Intent Contacted"] end

subgraph After["After: Instant Context"] A2["Real-Time Web/Call Signals"] --> B2["Acuity Lead Ranking"] B2 --> C2["Single Screen: Intent + Context + Offer"] C2 --> D2["91% High-Intent Contacted"] end ```

- **Portfolio Scale:** ~400 sellers managing ~8,000 assigned guests each. - **Time Reclaimed:** Eliminated the 77% of shift time (5 of 8.5 hours) previously spent manually searching systems for leads. - **High-Intent Contact Rate:** Surged from 51% to 91%. - **Financial Impact:** $120 million in incremental revenue unlocked from contact center signals.

[06:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=377s) Live Demo: harpin AI's Acuity Platform and Root-Cause Insights

Gregg transitions to a live demonstration of Acuity, harpin AI’s contact center intelligence platform, using a fictional cruise line dataset. He emphasizes that Acuity is not an ordinary dashboard: "The dashboards tell you what happened. This tool tells you why it happened and really what to go do about it."

The platform's home interface starts by quantifying the exact dollar value of uncaptured revenue left on the table across voice, chat, and email channels. It continuously analyzes whether an interaction presented a direct sales or sales-through-service opportunity, evaluates whether the agent earned the right to pitch by resolving the guest's primary issue, checks whether the agent attempted to close, and measures conversion success.

Beyond sales, Acuity monitors agent performance and workforce health. Leaders can stack-rank agents by QA score, inspect individual outcomes, receive AI-generated coaching recommendations, and monitor semantic behavioral shifts indicative of frontline agent burnout. In the sales intelligence view, managers can evaluate performance by product, specific sales triggers (such as cabin upgrades or price inquiries), and conversion rates against each trigger, resolving competing priorities between marketing, finance, and operations.

- **Outcome vs. Diagnostic:** Shifts management from historical lagging metrics to prescriptive root-cause action. - **Full-Channel Audio & Text Analysis:** Evaluates whether service resolution earned the right to sell, whether a close was attempted, and whether it succeeded. - **Workforce Wellness & QA:** Evaluates agent-level QA, delivers prescriptive AI coaching, and detects semantic markers of agent burnout. - **Trigger-Level Visibility:** Tracks discrete conversion rates against specific triggers, such as stateroom cabin upgrades and price inquiries.

[10:08](https://www.youtube.com/watch?v=8qniFNI-ayE&t=608s) Customer Effort and the Compounding Cost of Repeat Calls

Gregg demonstrates how Acuity measures Customer Effort Score (CES) and aligns it with transactional Net Promoter Scores (tNPS) and NPS data. By tracking caller sentiment, language patterns, and explanation clarity in real time, organizations can treat customer effort as a leading indicator of churn and loyalty rather than relying on lagging post-call surveys.

The platform also tackles repeat call volume—an area where many enterprises spend over 50% of their contact center operating budgets. Acuity maps repeat contact patterns across custom time windows (such as one-hour or four-hour increments) and breaks down whether customers call back regarding their original issue or a secondary topic. Gregg highlights that in the live demonstration model, a customer's second call costs 27% more than their initial interaction, illustrating how failure to achieve first-contact resolution compounds the organization's cost basis.

- **Leading vs. Lagging Indicators:** Utilizes CES and real-time conversation sentiment as predictive leading indicators rather than lagging NPS survey results. - **Budget Drain:** Over 50% of typical contact center operating budgets are consumed by repeat contacts. - **Cost Multiplier:** In the demo environment, a customer's second call costs 27% more than the initial contact ($1 + 1.27 \times \text{base cost}$). - **Repeat Call Diagnostics:** Categorizes callbacks by issue type (e.g., recurring pricing questions vs. distinct new issues) and elapsed time increments.

[11:33](https://www.youtube.com/watch?v=8qniFNI-ayE&t=693s) Cross-Contact Intelligence and Competitor Battle Cards

Gregg Hough explains how the platform surfaces deep peripheral insight into repeat customer behavior, helping leaders evaluate whether to grant agents slightly longer handle times to address what is likely to come next rather than letting a customer call back within an hour or a day. Beyond adjacent journey issues, the platform actively parses interactions for every mention of competitor brands, automatically producing structured intelligence briefs and competitor battle cards. These cards outline where the company is winning or losing against specific rivals, giving executive teams granular context on whether they should compete on comfort, pricing, product lineup, or feature sets.

```mermaid flowchart TD accTitle: Competitor and Repeat Contact Signal Workflow subgraph Capture["Interaction Ingestion"] A["Customer Calls & Transcripts"] --> B["harpin AI Signal Engine"] end subgraph Intelligence["Context Extraction"] B --> C["Adjacent Needs & Repeat Drivers"] B --> D["Competitor Mentions & Win/Loss Context"] end subgraph Action["Operational Strategy"] C --> E["Expand Agent Peripheral Vision & Prevent Callbacks"] D --> F["Automated Battle Cards (Price, Comfort, Features)"] end ```

- **Repeat contact mitigation**: Context reveals what adjacent questions follow an initial issue, giving agents the peripheral vision to resolve follow-up questions proactively during the initial interaction. - **Competitor intelligence**: Generates automated battle cards detailing specific competitor strengths and weaknesses based on direct customer mentions. - **Strategic positioning**: Enables commercial leadership to decide whether to win on price, comfort, or feature sets based on real conversation data.

---

[12:32](https://www.youtube.com/watch?v=8qniFNI-ayE&t=752s) Real-World Macro Sensing: Iranian Missiles and Geopolitical Disruption

The harpin AI platform allows leaders to ask open-ended natural language questions about macro-level customer behavior. Hough highlights a high-stakes real-world example when missile strikes involving Iran began, creating urgent questions for executive leadership at Princess Cruises regarding customer cancellations, itinerary adjustments, and whether to divert Mediterranean routes. harpin AI generated executive reports directly for the president of Princess Cruises to quantify sentiment and operational friction across thousands of calls.

```mermaid flowchart TD accTitle: Macro Geopolitical Event Intelligence Flow subgraph Sensing["Macro Signal Detection"] A["Global Event: Missiles & Airspace Closures"] --> B["12,000 Ingested Interactions"] B --> C["harpin AI Event Classifier"] end subgraph Extraction["Context & Policy Audit"] C --> D["72 World Event Mentions / 1,700 Potential Cancellations"] C --> E["Identified Force Majeure Policy Conflicts for Stranded Guests"] end subgraph Strategy["Executive Decision-Making"] D & E --> F["Executive Report to Princess President"] F --> G["Proactive Outreach Strategy (e.g., Kuşadası Concerns)"] end ```

The system ingested roughly 12,000 calls, isolating 72 direct world event references and uncovering that 1,700 guests were actively contemplating cancellation due to the Middle East conflict, UAE airspace closures, and concurrent cartel violence in Mexico. The AI detected a critical policy conflict: the company’s *force majeure* cancellation policy had an unconsidered loophole affecting guests who were already mid-transit on flights or already at sea when the strikes occurred. The platform recommended proactive outreach to guests showing elevated fear around ports like Kuşadası (Turkey) due to geographic proximity to Iran, allowing corporate leaders to adjust fleet policies while guiding contact center agents on tactical handling.

- **Scale of detection**: Out of 12,000 interactions analyzed, the system isolated 72 direct macro-conflict calls, 1,700 potential cancellations, and flagged correlated geopolitical events like UAE airspace closures and cartel flare-ups in Mexico. - **Policy loophole discovery**: Uncovered a *force majeure* policy blindspot for guests who were already stranded or in-flight when hostilities began. - **Proactive intervention**: Recommended specific proactive communication for itineraries calling on Kuşadası due to perceived proximity to conflict zones.

---

[15:05](https://www.youtube.com/watch?v=8qniFNI-ayE&t=905s) Outbound Voicemail Intelligence, Knowledge Base Generation, and Agentic AI Orchestration

Hough details how the platform extends across outbound sales, top-performer synthesis, and AI automation enablement. The system evaluates outbound sales behaviors by tracking whether an agent left a voicemail, whether the guest returned the call, and whether the interaction ended in a closed sale, reverse-engineering the exact conversational hooks and urgency drivers that make an outbound message effective. For internal enablement, the platform monitors top sales agents to identify winning phrasing and auto-generates pre-written knowledge base articles for supervisor review and automatic ingestion into central repositories.

Finally, Hough addresses how harpin AI interfaces with agentic AI and deflection strategies. Rather than replacing agentic chatbots, the platform categorizes all inbound contacts and emails, identifies which interaction layers are safe and ready for agentic deflection or lower-cost tier-one channels, and serves up the corresponding knowledge base articles and intent-driven IVR prompts to guarantee high resolution rates.

- **Voicemail closed-loop analysis**: Measures the entire outbound funnel (voicemail left → callback received → booking closed) to uncover optimal messaging and emotional hooks. - **Automated knowledge harvesting**: Analyzes high-performing agents' tactics and drafts complete knowledge base articles ready for one-click editorial approval. - **Dynamic IVR prompting**: Automates IVR call routing based on real customer intent rather than rigid legacy decision trees. - **Agentic AI enablement**: Classifies contact streams to determine which conversation types are safe to route to autonomous AI agents or offshore teams while attaching the exact context required for resolution.

## Notable quotes > Every vendor in this building will tell you to automate your easy contacts, and they are right, but almost nobody is talking about what happens next — Gregg Hough

> when AI absorbs the simple calls, every conversation left in the human queue is harder, longer, and higher-stakes than the one before it. — Gregg Hough

> I will show what we found when we read millions of real guest conversations at Princess Cruises, including the calls that most deflection strategies in this room would have thrown away, and what happened when we handed those conversations back to humans with better information in the moment. — Gregg Hough

> That work has driven $120 million in revenue, and I will walk through how it was measured, not just what it added up to. — Gregg Hough

> I will be joined on stage by the Princess operator who lived it, including the parts we got wrong. — Gregg Hough

> You will leave knowing how to see what is actually inside your call volume, which contacts are genuinely safe to automate, which ones are quietly worth eight figures, and how to prepare the people who inherit everything else. — Gregg Hough

> The talk includes a live demo, not a slide of one. — Gregg Hough

> [02:25](https://www.youtube.com/watch?v=8qniFNI-ayE&t=145s) Camille, it looks like you're ready to talk about that cruise. What other questions do you have? How can I support you? — Gregg Hough

> [06:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=377s) The dashboards tell you what happened. This tool tells you why it happened and really what to go do about it. — Gregg Hough

## Mentioned Camille Bailey, Gregg Hough, Princess Cruises, harpin AI, Acuity, Ai4, Carnival, Celebrity Cruises, Crystal Cruises, Kuşadası, Iran, Mexico, United Arab Emirates, Alaska, Europe

## Video description Camille Bailey - Associate Vice President, Consumer Sales, Princess Cruises Gregg Hough - Chief Commercial Officer, harpin AI

Every vendor in this building will tell you to automate your easy contacts, and they are right, but almost nobody is talking about what happens next: when AI absorbs the simple calls, every conversation left in the human queue is harder, longer, and higher-stakes than the one before it. I will show what we found when we read millions of real guest conversations at Princess Cruises, including the calls that most deflection strategies in this room would have thrown away, and what happened when we handed those conversations back to humans with better information in the moment. That work has driven $120 million in revenue, and I will walk through how it was measured, not just what it added up to. I will be joined on stage by the Princess operator who lived it, including the parts we got wrong. You will leave knowing how to see what is actually inside your call volume, which contacts are genuinely safe to automate, which ones are quietly worth eight figures, and how to prepare the people who inherit everything else. The talk includes a live demo, not a slide of one.

[Source](https://www.youtube.com/watch?v=8qniFNI-ayE)

中文简报

借助Harpin AI精准推送高意向线索，公主邮轮增收1.2亿美元通过在服务通话中为人工销售人员提供实时客户信号而非转移咨询，高意向客户触达率提升至91%，并释放了1.2亿美元的增量收入。面对直销额同比下降4%以及坐席人员耗费77%的工作时间在断联的传统数据库中搜寻信息的现状，公主邮轮（Princess Cruises）与harpin AI展开合作，将原始客户对话转化为可落地的销售情报。通过将宾客历史记录与实时数字化信号整合进统一的礼宾界面，公主邮轮将高意向宾客触达率从51%提升至91%，并创造了1.2亿美元的增量收入。通过对harpin AI旗下Acuity平台的现场演示，发言人展示了实时对话诊断如何揭示重复来电的复合成本、提炼竞争对手应对策略，并在地缘政治危机期间实现宏观层面的态势感知，以保障预订业务并调整公司政策。将联络中心仅视为运营成本中心，而非丰富的客户情报引擎，是该行业的一大盲点。 (0:04) 在采取干预措施之前，公主邮轮直销额同比下降4%，坐席人员在8小时的班次中仅有约2小时在积极开展销售。 (1:41) 公主邮轮的销售人员此前有77%的时间（8.5小时中有5小时）耗费在五个互不连通的传统数据库中手动搜寻，以查找客户优惠及背景信息。 (2:25) 将客户情报整合至包含实时数字化“举手”信号的统一视图中，使高意向宾客触达率从51%提升至91%，并带动了1.2亿美元的增量收入。 (2:25) harpin AI的Acuity平台将管理重心从滞后的结果指标转向根本原因诊断，追踪服务问题的解决是否赢得了销售机会，以及是否尝试了促成交易。 (6:17)

Reading summary

Princess Cruises partnered with harpin AI to transform its contact center from a perceived operational cost center into a real-time intelligence hub, raising high-intent prospect reach from 51% to 91% and unlocking $120 million in incremental revenue. Unifying Disconnected Systems to Reclaim Selling Time (1:41) Before adopting harpin AI's platform, Princess Cruises experienced a 4% year-over-year decline in direct sales alongside severe agent productivity bottlenecks. An audit revealed that during typical 8 to 8.5-hour shifts, sales agents spent only about two hours engaged in active, productive conversations. Frontline sellers—comprising roughly 400 agents managing portfolios of approximately 8,000 guests each—spent 77% of their shift time manually searching through five disconnected legacy systems to identify customer offers, flight credits, and cruise history. Princess Cruises addressed this shortfall by deploying a unified application that continuously aggregates guest data and tracks digital 'hand raises,' such as an existing contact browsing sailings online. By presenting full behavioral histories and immediate context within a single screen, sellers shifted from manual searching to proactive, relevant outreach. This workflow adjustment expanded contact rates with high-intent prospects from 51% to 91% and generated $120 million in incremental revenue. Diagnostic Analytics and Managing Repeat Call Costs (6:17) Rather than relying on historical, lagging operational dashboards, harpin AI's Acuity platform evaluates interactions across voice, chat, and email to diagnose root causes and uncaptured revenue. The system checks whether service inquiries were adequately resolved to earn the right to pitch, whether an agent attempted a close, and tracks discrete conversion rates against triggers such as cabin upgrades and pricing queries. It also measures agent-level quality assurance and flags conversational indicators of frontline burnout. The platform also targets repeat contacts, an operational failure that typically consumes over 50% of contact center operating budgets. In the live demonstration model, a customer's second call cost 27% more than their initial interaction. By monitoring Customer Effort Scores (CES) and speech patterns in real time as leading indicators of churn, the system equips agents with the peripheral vision to resolve anticipated adjacent questions during the first call, reducing callback frequency. Sensing Macro Disruptions and Resolving Policy Loopholes (12:32) Beyond routine sales and service diagnostics, the platform serves as an enterprise macro-sensing tool during external crises. Following geopolitical escalation involving Iranian missile strikes, executive leadership required rapid insight into guest cancellation risks and Mediterranean itinerary viability. By ingesting 12,000 customer interactions, the system isolated 72 direct world event references and determined that approximately 1,700 callers were actively contemplating cancellation due to flight disruptions and Middle Eastern airspace closures. This conversational analysis surfaced a critical blind spot in Princess Cruises' force majeure policy, which had not accounted for travelers who were already mid-transit on flights or already at sea when the strikes occurred. Executive reports enabled corporate leadership to amend fleet policies, guide agent handling, and initiate proactive outreach to guests booked on itineraries calling on proximate ports such as Kuşadası, Turkey. Automating Battle Cards, Knowledge, and AI Routing (11:33) The intelligence engine automatically parses interaction transcripts for competitor mentions, producing structured battle cards that highlight where the cruise line wins or loses against rivals on pricing, accommodations, and features. For internal enablement, the platform conducts closed-loop analysis on outbound sales behaviors—correlating voicemails with callback and close rates—and extracts successful conversational phrasing from top sellers to generate draft knowledge base articles for supervisor approval. Addressing automation, the speakers noted that rather than fully replacing human agents with autonomous bots, the platform classifies customer contact streams to pinpoint which conversation tiers are safe for agentic AI deflection or lower-cost handling. Inquiries routed to automated agents or dynamic IVRs are supplied with real-time intent context to ensure high resolution rates. Princess Cruises and harpin AI maintain that contact centers house valuable customer signals that are squandered when treated purely as operational cost centers. While empowering human agents with real-time context drove measurable revenue gains, the long-term effectiveness of the model depends on sustained system integration and leadership's ability to act quickly on conversational macro signals.

精炼阅读摘要

Princess Cruises与harpin AI合作，打破将联络中心仅视作被动成本中心的传统误区，通过AI挖掘交互过程中的实时行为与对话信号，克服坐席工具碎片化与效率低下的瓶颈，将高意向客户触达率从51%提升至91%，最终解锁了1.2亿美元的直接增量营收。破除成本中心偏见：坐席诊断与数据孤岛 (0:04) 长期以来，邮轮乃至整个行业普遍将联络中心视为需要压缩的运营成本中心，忽视了日常交互中蕴含的客户意向金矿。Princess Cruises此前直销业务出现同比4%的下滑。管理层在诊断分析中发现，在坐席人员每天8至8.5小时的轮班中，实际用于产生有效销售成果的对话时间仅有约2小时，表面上呈现出极低的运营产出。造成这一生产力瓶颈的核心原因在于系统严重碎片化。公司约400名销售人员每人平均管理多达8,000名客户档案，但长达65年沉淀的历史数据分散在5套独立的遗留系统内。坐席高达77%的工作时间（近5小时）耗费在反复查询系统，手动核实客户是否具备赌场特惠、欧洲航线促销或免费机票资格。由于缺乏自动化线索指引，坐席对高意向潜客的主动触达率仅为51%，造成严重的潜在收益流失。实时意向捕捉与上下文整合实现增收 (2:25) 为了重构销售转化漏斗，Princess Cruises部署了统一礼宾销售系统，实时捕捉客户的数字化“举手”（hand raises）动作。例如，当一位一个月前与坐席沟通过的客户再次在官网浏览特定航线时，系统会立即向原负责坐席发出实时触发提醒，并自动汇聚该客户的历史航行偏好（如偏好阿拉斯加航线）、结婚纪念日或子女毕业等背景细节，使坐席能直接以精准上下文开展针对性沟通。这种具备丰富上下文的单界面操作，彻底消除了盲目外呼和繁冗检索。借助harpin AI的Acuity引擎对未接触的高意向客户进行意图识别与价值排序，销售团队对高意向人群的触达率从51%飙升至91%。这一流程变革帮助企业有效唤醒了被忽视的潜在订单，直接带来了1.2亿美元的增量营收。归因诊断：控制重复进线与坐席赋能 (6:17) 与仅展示历史统计结果的传统仪表盘不同，Acuity平台强调底层归因与行动指导。针对多数企业联络中心预算逾50%耗费在重复进线上的现状，平台演示模型显示，二次呼叫成本相比首次交互激增27%。平台通过即时追踪客户努力度评分（CES）与对话情绪，替代滞后的NPS调研，帮助管理者评估是否应适当延长单次通话时长，以便前置解决客户的潜在次生问题，降低回呼频率。在团队管理与市场策略层面，平台贯穿语音、聊天和邮件全渠道，监测坐席是否在妥善解决服务需求后才尝试促成交易，并通过语义变化识别坐席倦怠迹象，输出针对性的AI培训建议。同时，系统自动解析客户对话中提及的竞品信息，生成包含价格、舒适度及产品维度的结构化竞品对战卡（Battle Cards），为管理层制定市场定位与竞争策略提供支持。宏观突发事件感知与AI分流编排 (12:32) 该智能平台具备理解自然语言非结构化查询的能力，能够在宏观地缘风险事件发生时为决策层提供即时态势感知。例如在涉及伊朗的导弹袭击事件发生后，系统通过分析约12,000次通话记录，迅速隔离出72次直接提及该事件的通话，并识别出1,700名因中东局势、阿联酋领空关闭及墨西哥治安事件而考虑取消行程的旅客。平台协助管理层排查出不可抗力（Force Majeure）退改条款中针对在途或已登船旅客的盲区，并建议对土耳其库萨达斯（Kuşadası）航线旅客展开前置沟通以平息恐慌。在智能化延展方面，系统通过闭环追踪“留留言—客户回电—成功成交”的完整销售链路，萃取顶尖坐席的沟通技巧并自动生成知识库文档。此外，平台基于真实意向对所有进线交互进行分类分流，精准判定哪些低风险场景可以安全交付给自主代理型AI（Agentic AI）或外包初级团队，在保障解决率的前提下实现自动化降本。 Princess Cruises的案例表明，联络中心通过AI从被动客服中心升级为兼顾商业增收、危机感知与组织赋能的数据枢纽。不过，该实践高度依托邮轮行业高客单价、强顾问式销售和长决策周期的业务特征，材料中并未呈现该方案在短决策链或低客单价业务中的适用性与投资回报周期；同时，如何在逐步扩大的代理型AI自动化分流中持续保持高品质的服务体验，仍是企业在降本与体验之间需要长期平衡的问题。

Detailed notes — zh

## Background 本内容由 YouTube 频道 Ai4 于 2026 年 8 月 21 日发布。视频由公主邮轮（Princess Cruises）消费者销售副总裁 Camille Bailey 与 harpin AI 首席商务官 Gregg Hough 共同主讲，重点分享了如何在 AI 过滤简单来电后，通过挖掘深层对话信号与实时上下文赋能人类坐席，最终为企业带来 1.2 亿美元增量营收的实战案例与现场演示。 ## TL;DR 公主邮轮打破了传统将联络中心仅视为“成本中心”的思维定势，直面坐席跨系统查找耗时高、高意向客户触达率低的痛点。通过与 harpin AI 合作引入 Concierge Sales 与 Acuity 分析工具，企业实现了对全量对话意图的精准捕获、坐席工作流的统一以及实时个性化触达，使高意向客户触达率由 51% 跃升至 91%，创造了 1.2 亿美元的增量营收。系统不仅能洞察未成单损失、识别员工职业倦怠、优化客户费力度（CES）与根治重复来电，还在地缘危机等突发事件中为高管层提供了前瞻性的决策支持与政策漏洞预警。 ## Atmosphere & style 整场演讲兼具行业老将的务实反思与前沿科技的敏锐洞察，节奏明快且极富说服力。Camille 结合 20 年邮轮运营经历坦诚复盘历史痛点，Gregg 则通过大屏实时演示 Acuity 系统的硬核功能，从流失金额量化直击业务核心。现场氛围专业而接地气，既没有虚浮的 AI 炒作，而是充满着将联络中心“盲区转化为真金白银”的实战底气与商业洞察力。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=8qniFNI-ayE&t=4s) 企业长期将联络中心视为成本中心，忽视了海量会话数据中蕴含的商业信号与营收潜力。 - [01:41](https://www.youtube.com/watch?v=8qniFNI-ayE&t=101s) 公主邮轮通过 harpin AI 捕捉实时意图信号（Hand Raises）并统一坐席工作流，斩获了 1.2 亿美元的增量营收。 - [01:41](https://www.youtube.com/watch?v=8qniFNI-ayE&t=101s) 高购买意向客户触达率从 51% 跃升至 91%，大幅缩减了坐席在海量名单中盲目搜寻线索的时间浪费。 - [05:01](https://www.youtube.com/watch?v=8qniFNI-ayE&t=301s) 基于 65 年历史数据提炼客户偏好与重要生活事件（如结婚纪念日），让销售坐席能够带着精准上下文主动开展个性化触达。 - [06:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=377s) Acuity 分析工具跳出传统报表“发生了什么”的局限，能够全量监听电话、聊天与邮件，下钻“为什么发生”并量化流失在谈判桌上的未成单收入。 - [08:20](https://www.youtube.com/watch?v=8qniFNI-ayE&t=500s) 系统通过语义与行为多维分析捕捉坐席的职业倦怠迹象，并为管理者自动生成针对性的 AI 辅导建议。 - [09:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=557s) 将客户情绪与客户费力度（CES）转化为领先指标，替代滞后的 NPS，直接指导降低客户费力度的话术与操作策略。 - [10:45](https://www.youtube.com/watch?v=8qniFNI-ayE&t=645s) 针对消耗联络中心超 50% 成本的重复来电，通过拓展坐席“外围视野”在前序通话中预判相邻诉求，从源头阻断客户再次进线。 - [11:33](https://www.youtube.com/watch?v=8qniFNI-ayE&t=693s) 通话数据能自动生成多维竞品“对战卡”，清晰提炼产品、价格和体验维度的优劣势，为企业战略决策提供实证上下文。 - [12:32](https://www.youtube.com/watch?v=8qniFNI-ayE&t=752s) 在中东地缘危机等突发事件中，AI 快速识别退订意向，暴露出不可抗力条款对在途旅客的政策盲区，驱动高管即时调整港口运营与安抚方案。 - [15:05](https://www.youtube.com/watch?v=8qniFNI-ayE&t=905s) 系统提取顶尖销售的最佳留言与沟通模式自动沉淀为知识库草稿，并赋能意图驱动的 IVR 动态路由。 - [15:05](https://www.youtube.com/watch?v=8qniFNI-ayE&t=905s) 通过对全量交互精准分级，明确哪些复杂长尾场景适合由人机协同解决，哪些安全标准场景适合无缝迁移至 Agentic AI 自动化落地。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=8qniFNI-ayE&t=4s) 联络中心的盲区与被忽视的金矿 harpin AI 首席商务官（Chief Commercial Officer）Gregg Hough 首先开场，介绍了 harpin AI 作为一家专注于联络中心 AI 的初创公司定位。他明确强调：“我们做的不是什么自主智能体（agentic anything）”，而是致力于帮助人类坐席更轻松地完成工作，在关键时刻提供所需的数据，使人能够真正高效地运作。随后，他邀请了合作伙伴——公主邮轮（Princess Cruises）消费者销售副总裁 Camille Bailey 登台分享。 ```mermaid flowchart TD accTitle: Princess Cruises Contact Center Blind Spot Transformation subgraph S1["历史痛点 (The Blind Spot)"] A["视为成本中心 忽视数据富矿"] --> B["坐席开着5个系统 8小时仅2小时有效产出"] B --> C["77%时间在盲目找人 高意向客户触达率仅51%"] end subgraph S2["harpin AI 诊断与赋能"] D["全量会话诊断与意图评分"] --> E["实时意图捕捉 (Hand Raises)"] E --> F["单一界面注入上下文 (偏好/纪念日/出游频次)"] end subgraph S3["业务结果 (Value Delivered)"] G["高意向触达率跃升至 91%"] H["斩获 $120M 增量营收"] end S1 --> S2 S2 --> S3 ``` Camille 回顾了自己 20 年前在邮轮联络中心接听第一个电话的经历，坦言当时完全没有意识到自己正“坐在金矿之上”，因为那时行业根本不懂得如何捕捉其中的信号。过去二十年里，她先后领导过嘉年华（Carnival）、精致游轮（Celebrity Cruises）、水晶邮轮（Crystal Cruises）以及如今公主邮轮的销售和服务团队。她指出，联络中心的盲区一直惊人地一致：企业始终将联络中心视为成本中心，而未能充分挖掘和利用其中蕴含的极其丰富的信息资产。现在，他们正在做的事情正是识别这一盲区，并“将盲区转化为真金白银（turn the blind spot into money）”。 ### [01:41](https://www.youtube.com/watch?v=8qniFNI-ayE&t=101s) 破解生产力困境与 1.2 亿美元增量营收的实现在引入新方案之前，公主邮轮面临着销售额同比下降 4% 的严峻挑战。管理层拥有海量数据，却缺乏清晰的行动方向；审视坐席的日常表现时发现，在平均 8 小时的班次中，坐席真正用于有效沟通的生产力时间仅约 2 小时，表面看起来投入度极低。于是，公主邮轮与 harpin AI 展开合作进行全面诊断：分析所有客户会话，找出谁是真正的目标宾客、谁已准备好预订、哪些潜在客户被遗漏，并评估营收差距以对改进方案进行优先级排序。为了解决这一问题，他们上线了 Concierge Sales 工具，实现对潜在客户的智能排序与实时意图捕捉（hand raises）。例如，某位宾客在一个月前曾致电咨询，昨天又在官网上浏览了同款航线，系统便会实时捕获这一“举手”信号并通知坐席。坐席可以在最佳时机主动致电并展开懂客户的对话：“Camille，看起来您准备好聊聊那趟航程了，您还有什么疑问？我能为您提供哪些支持？” - **统一工作流**：过去坐席需要同时打开 5 个不同系统来回切换，现在整合为单一系统，提供所有必要信息与上下文。 - **消除猜测耗时**：以往坐席将低产出的原因归咎于“在 8,000 名客户库中寻找该给谁打电话、谁有赌场优惠或欧洲免费机票”，导致高达 77% 的时间都在搜寻拨打对象；新系统实现了即时精准推客。 - **触达率大幅跃升**：高购买意向宾客的触达率从原先的 51% 飙升至 91%。 - **核心业务回报**：成功消除盲区，直接为公司带来了 1.2 亿美元（$120M）的增量营收。 ### [05:01](https://www.youtube.com/watch?v=8qniFNI-ayE&t=301s) 65 年老牌企业的痛点：从数据泥潭到个性化触达 Gregg Hough 重新接过话筒，向现场观众互动提问是否有管理联络中心的同行，并指出行业通病：“联络中心充满了数据，但事实却极其匮乏（full of data but very few facts）”。他用具体规模阐明了公主邮轮的实际场景：团队大约有 400 名专职销售坐席，每位销售人员名单下分配了多达 8,000 名宾客。坐席每天 8.5 小时的工作时间里，往往有 5 个小时都在跨系统排查和检索数据。对于一家拥有 65 年历史的老牌企业，数据散落在各处却极难调取。通过 AI 提炼过往交互信号后，系统能够自动勾勒出宾客的完整画像——以往的航行历史、航线偏好、居住地、即将到来的结婚纪念日、女儿高中毕业等生活大事件。系统会直接向坐席发出精准建议：“去联系 Camille 吧，她已经 6 个月没出海了，以往每年出行三次；她最喜欢阿拉斯加航线，下个月正好是她的结婚纪念日。现在就打给她，聊聊我们专为她准备的特别优惠。”这种带着精准上下文的触达方式彻底改变了坐席的工作体验。 ### [06:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=377s) 演示 Acuity 工具：洞察“为什么”而不仅是“发生了什么” Gregg 将电脑接入大屏幕，现场演示核心分析工具 Acuity。他强调 Acuity 与市面上传统仪表盘有着本质区别：传统仪表盘只能展示“发生了什么（what happened）”，而 Acuity 能揭示“为什么会发生（why it happened）”以及“接下来该采取什么行动（what to go do about it）”。 ```mermaid flowchart LR accTitle: Contact Center Revenue Leakage Tracking subgraph A1["多渠道交互监听"] Phone["电话通话"] Chat["在线聊天"] Email["电子邮件"] end subgraph A2["Acuity 转化漏斗分析"] D1["识别销售机会 (直接销售 / 服务转销售)"] D2["判断坐席是否 尝试推进成单 (Attempt)"] D3["评估成单结果 (Success vs Missed)"] end subgraph A3["行动与优化输出"] Act1["量化错失营收规模"] Act2["生成 AI 针对性辅导建议"] Act3["识别员工倦怠迹象"] end A1 --> D1 D1 --> D2 D2 --> D3 D3 --> Act1 D3 --> Act2 D3 --> Act3 ``` 演示系统以一家虚构的邮轮公司为例，首先在首页直观量化销售端“流失在谈判桌上的收入金额”。系统全量监听联络中心内的电话、在线聊天和电子邮件，追踪是否存在销售机会（包括直接销售以及服务过程中建立信任后衍生的销售机会），判断坐席是否尝试推进成单，以及尝试后是否成功，最终将未被把握的商业机会转化为清晰的损失金额视图。 ### [08:20](https://www.youtube.com/watch?v=8qniFNI-ayE&t=500s) 坐席绩效管理、员工倦怠识别与 AI 辅导建议在坐席管理与生产力模块中，Acuity 提供了全方位的人员透视能力，能够根据团队需求进行堆叠排名（Stack ranking）、识别顶级表现者以及筛选缺少辅导（missing coaching）的人员。Gregg 特别介绍了工具对员工“职业倦怠（burnout）”的检测能力：系统通过语义分析与行为绩效异动的多维追踪，捕捉坐席是否出现与日常状态明显相符的异常偏差。联络中心坐席是一项高压艰难的工作，一旦发现倦怠信号，管理者就能及时介入，给予关心与支持。管理者可以进一步下钻查看特定坐席（如 Agent A）的详细画像，包括其整体质检（QA）得分、在团队中的相对表现、具体待改进项、销售转化结果，以及系统自动生成的定制化 AI 辅导建议（AI coaching recommendations）。 ### [09:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=557s) 销售触发器分析与客户费力度（CES）的领先指标应用针对管理者在销售指标与运营成本之间的权衡难题，Gregg 演示了“销售情报（Sales Intel）”视图。系统不仅能按产品细分展示潜在营收机会，还能进一步下钻剖析具体的销售触发信号（如客舱升舱需求、价格咨询等）以及不同触发器下的实际转化率，帮助团队在有限预算和多重优先级（销售额、获客成本、营销摩擦等）之间找到最明确的发力点。在客户体验与客户费力度（CES - Customer Effort Score）方面，Acuity 能将客户情绪与努力程度结合： - **前瞻性指标**：将情感分析与费力度转化为领先指标（leading indicators），摆脱传统 NPS/tNPS 等事后滞后指标（lagging indicators）无法即时干预的缺陷。 - **归因与行动指导**：深入剖析导致客户费力增加的具体话术、解释方式及趋势分布，并按坐席、技能组、问题类别和语言细分，直接列出建议采取的重复行动方案。 ### [10:45](https://www.youtube.com/watch?v=8qniFNI-ayE&t=645s) 根治重复来电：压缩高达 50% 的联络中心服务成本 Gregg 指出，许多客户联络中心有超过 50% 的预算耗费在处理“重复来电（repeat calls）”上。Acuity 设立了专门板块帮助企业彻底弄清客户反复致电的根源。系统不仅统计重复来电的频次与时间间隔（如 1 小时内、4 小时内等），还揭示了隐性成本放大效应。以现场示例数据为例，客户因同一问题打入的第二次电话成本比第一次高出 27%（成本基础由 1 变成 1 + 1.27）。系统能够精确下钻到具体业务场景（例如当客户致电咨询价格问题时），分析他们后续再次致电是因为同一原因未得到解决，还是产生了全新的衍生诉求，从而指导管理层从根源上阻断重复来电的发生。 ### [11:33](https://www.youtube.com/watch?v=8qniFNI-ayE&t=693s) 客户回拨预测与多维竞争情报洞察呼叫中心沉淀的对话数据能够精准揭示客户再次联系时的深层原因，帮助管理者看清客户当前诉求的“相邻诉求”（Adjacent Topics）。在传统考核中，联络中心往往一味压缩平均通话处理时长（AHT），但如果系统能明确提示接下来的关联问题，适度拓展一线坐席的“外围视野”（Peripheral Vision）并给予其稍长一点的处理时间，就能当场解决客户的潜在疑问，从而避免客户在一小时或一天后反复致电。系统还会全面捕捉通话中提及的所有竞品信息，自动生成智能简报与竞争“对战卡”（Battle Cards）。平台能够清晰罗列企业在哪些战场上取胜、在哪些维度上失利，帮助管理层做出更具前瞻性的商业决策——明确究竟该在舒适度、价格机制、产品组合还是特色功能上与对手竞争。"Context really is what drives decisions at the end of the day."（归根结底，真正驱动决策的是语境与上下文信息），而这些洞察正是以往企业难以获取的盲区。 - **外围视野拓展**：通过预判客户下一步意图，在首通电话中前置解决关联问题，从源头降低二次进线率。 - **竞品对战卡生成**：动态提取客户对话中的竞品反馈，提炼出价格、舒适度及产品层面的胜负因子。 --- ### [12:32](https://www.youtube.com/watch?v=8qniFNI-ayE&t=752s) 宏观危机应对：从地缘冲突到高管决策与政策闭环除了预设的分析面板，该平台还支持管理人员使用自然语言进行自由问答。演讲者以中东地缘危机爆发时的真实案例展示了其强大的即时分析能力：当伊朗导弹开始发射、阿联酋领空关闭，且墨西哥贩毒集团活动同步加剧时，系统直接向公主邮轮（Princess Cruises）总裁生成并推送了一份宏观影响报告，回答诸如“客户是否在集中退订”、“地中海港口是否该继续开放或改道”等紧迫问题。 ```mermaid flowchart TD accTitle: Crisis Intelligence to Executive Decision Flow subgraph DataEngine["数据捕获与意图分析"] A["12,000 通客户进线通话"] --> B["AI 实时识别 72 通涉及世界事件"] B --> C["锁定 1,700 名潜在取消意向客户"] end subgraph DeepContext["深度语境与冲突挖掘"] C --> D["发现不可抗力条款漏洞 (已在途/已登船滞留游客)"] C --> E["识别特定港口恐慌 (如靠近伊朗的库萨达斯 Kusadasi)"] end subgraph StrategicAction["高管与运营协同响应"] D --> F["修订滞留客人应急保障方案"] E --> G["前置外呼沟通与航线港口动态调整"] end ``` 在调取的 12,000 通电话样本中，系统快速识别出 72 通直接针对国际局势的咨询，并锁定了 1,700 位产生退订意向的客户。更关键的是，AI 敏锐捕捉到底层政策矛盾：邮轮现有的“不可抗力条款”（Force Majeure Policy）与那些已经在旅途中、已登船或正搭乘航班前往出发港的滞留游客权益存在冲突。这一漏洞在以往的应急预案中从未被考虑到。此外，系统检测到游客因地理位置邻近伊朗而对土耳其库萨达斯（Kuşadası）港口产生了巨大恐慌，并据此建议运营团队主动向客户发起外呼沟通。这种能力让一线联络中心的微观信号直接上升为支撑最高领导层战略决策的利器。 - **宏观事件归因**：直观展示地缘冲突与领空关闭引发的客户情绪指数、取消意愿与问题解决状态。 - **政策盲区预警**：在无人工干预的情况下，自动比对政策条款与实际滞留游客诉求，暴露不可抗力执行中的漏洞。 - **主动外呼建议**：针对高危敏感港口恐慌，驱动业务端从“被动接听”转向“前置安抚”。 --- ### [15:05](https://www.youtube.com/watch?v=8qniFNI-ayE&t=905s) 赋能销售、知识库自动化与智能体落地演进除了危机响应，该智能化工具还在销售提升与运营自动化方面展现了多重价值。在外呼销售场景中，系统能够细致分析怎样的语音留言（Voicemail）能够带来最高的回拨率，并在客户回拨后成功促成订单，进而为销售团队总结出具备“情感抓手与紧迫感”的金牌留言模板。同时，系统能横向对比顶级销售顾问（Top Sellers）与普通员工的沟通差异，精确指出优秀销售“说了什么、是怎么表达的”，为团队复制最佳实践。在知识库（Knowledge Base）与自动化体系建设上，该工具同样重塑了传统运维流程： 1. **知识库文章自动生成**：系统持续观察顶尖员工的优秀处置方案，自动提取并撰写知识库草稿，供管理人员审核、补充并一键沉淀至企业知识库中。 2. **基于客户意图的 IVR 智能路由**：根据用户真实意图与上下文动态触发 IVR 提示，替代过去呆板的固定按键树。 3. **Agentic AI（智能体）落地加速器**：尽管 harpin AI 本身不直接提供端到端替代人工的 Agentic 机器人，但它能够对所有通话和邮件进行精准细分与定级，明确指出下一批最适合转入自动化（如 Agentic AI、智能聊天机器人或离岸外包）的联络类型，并打包配套对应的知识库上下文，确保业务在向低成本环境迁移时依然保持极高的问题解决率。演讲最后强调，通过给人类坐席提供更精准的实时上下文，并让自动化建立在真实业务信号之上，这套打法在公主邮轮成功兑现了数千万美元级的商业回报。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=8qniFNI-ayE&t=4s) 我们做的不是什么自主智能体（agentic anything） — Gregg Hough > [00:04](https://www.youtube.com/watch?v=8qniFNI-ayE&t=4s) 将盲区转化为真金白银（turn the blind spot into money） — Camille Bailey > [01:41](https://www.youtube.com/watch?v=8qniFNI-ayE&t=101s) Camille，看起来您准备好聊聊那趟航程了，您还有什么疑问？我能为您提供哪些支持？ — Camille Bailey > [05:01](https://www.youtube.com/watch?v=8qniFNI-ayE&t=301s) 联络中心充满了数据，但事实却极其匮乏（full of data but very few facts） — Gregg Hough > [05:01](https://www.youtube.com/watch?v=8qniFNI-ayE&t=301s) 去联系 Camille 吧，她已经 6 个月没出海了，以往每年出行三次；她最喜欢阿拉斯加航线，下个月正好是她的结婚纪念日。现在就打给她，聊聊我们专为她准备的特别优惠。 — Gregg Hough > [06:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=377s) 发生了什么（what happened） — Gregg Hough > [06:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=377s) 为什么会发生（why it happened） — Gregg Hough > [06:17](https://www.youtube.com/watch?v=8qniFNI-ayE&t=377s) 接下来该采取什么行动（what to go do about it） — Gregg Hough > [11:33](https://www.youtube.com/watch?v=8qniFNI-ayE&t=693s) Context really is what drives decisions at the end of the day. — Gregg Hough ## Mentioned Princess Cruises, harpin AI, Gregg Hough, Camille Bailey, Carnival, Celebrity Cruises, Crystal Cruises, Concierge Sales, Acuity, CES, NPS, tNPS, Kuşadası ## Video description Camille Bailey - Associate Vice President, Consumer Sales, Princess Cruises Gregg Hough - Chief Commercial Officer, harpin AI Every vendor in this building will tell you to automate your easy contacts, and they are right, but almost nobody is talking about what happens next: when AI absorbs the simple calls, every conversation left in the human queue is harder, longer, and higher-stakes than the one before it. I will show what we found when we read millions of real guest conversations at Princess Cruises, including the calls that most deflection strategies in this room would have thrown away, and what happened when we handed those conversations back to humans with better information in the moment. That work has driven $120 million in revenue, and I will walk through how it was measured, not just what it added up to. I will be joined on stage by the Princess operator who lived it, including the parts we got wrong. You will leave knowing how to see what is actually inside your call volume, which contacts are genuinely safe to automate, which ones are quietly worth eight figures, and how to prepare the people who inherit everything else. The talk includes a live demo, not a slide of one. [Source](https://www.youtube.com/watch?v=8qniFNI-ayE)

Source quotes

[0:04] Gregg Hough: 我们做的不是什么自主智能体（agentic anything） [0:04] Camille Bailey: 将盲区转化为真金白银（turn the blind spot into money） [1:41] Camille Bailey: Camille，看起来您准备好聊聊那趟航程了，您还有什么疑问？我能为您提供哪些支持？ [5:01] Gregg Hough: 联络中心充满了数据，但事实却极其匮乏（full of data but very few facts） [5:01] Gregg Hough: 去联系 Camille 吧，她已经 6 个月没出海了，以往每年出行三次；她最喜欢阿拉斯加航线，下个月正好是她的结婚纪念日。现在就打给她，聊聊我们专为她准备的特别优惠。 [6:17] Gregg Hough: 发生了什么（what happened） [6:17] Gregg Hough: 为什么会发生（why it happened） [6:17] Gregg Hough: 接下来该采取什么行动（what to go do about it） [11:33] Gregg Hough: Context really is what drives decisions at the end of the day.

Source quotes

Gregg Hough: Every vendor in this building will tell you to automate your easy contacts, and they are right, but almost nobody is talking about what happens next Gregg Hough: when AI absorbs the simple calls, every conversation left in the human queue is harder, longer, and higher-stakes than the one before it. Gregg Hough: I will show what we found when we read millions of real guest conversations at Princess Cruises, including the calls that most deflection strategies in this room would have thrown away, and what happened when we handed those conversations back to humans with better information in the moment. Gregg Hough: That work has driven $120 million in revenue, and I will walk through how it was measured, not just what it added up to. Gregg Hough: I will be joined on stage by the Princess operator who lived it, including the parts we got wrong. Gregg Hough: You will leave knowing how to see what is actually inside your call volume, which contacts are genuinely safe to automate, which ones are quietly worth eight figures, and how to prepare the people who inherit everything else. Gregg Hough: The talk includes a live demo, not a slide of one. [2:25] Gregg Hough: Camille, it looks like you're ready to talk about that cruise. What other questions do you have? How can I support you? [6:17] Gregg Hough: The dashboards tell you what happened. This tool tells you why it happened and really what to go do about it.

Original transcript

[0:04] All right, good afternoon. My name is Greg with Tarpon AI. Um, startup AI company really focused [0:09] on the AI in the contact center. Um, we're not agentic anything. We're all about how do [0:14] we help the humans do their job easier, how we present the data they need to really be [0:19] effective in whatever it is we're we're trying to do. And I'd love to introduce our partner, [0:24] Camille. Afternoon everyone. How's the conference going? Good. Yeah, right. I was so [0:31] impressed with everyone walking through and more impressed that Greg asked me to come here and be [0:37] his partner to speak today. 20 years ago, I took my first call in a cruise contact center [0:46] and had no idea then that I was sitting on top of goal. Well, really none of us did, [0:50] right? Because we didn't know how to see the signals. So over the last two decades, I've led [0:57] sales and service teams for Carnival, Celebrity Cruises, Crystal Cruises, and now Princess. And [1:06] you know, the blind spot's been pretty uh pretty much the same, right? it's been consistent, [1:12] which is treating our contact centers like cost centers and not really honing in on the richness [1:21] um of information that we get from there and and and that's what we're doing now. We're identifying [1:26] that blind spot and using it to uh turn the blind spot into money. So, I'll go through a few slides [1:33] here and then Greg will do a demo of the tool that we're using today. So, the blind spot, [1:41] we were down 4% in sales year-over-year. And when we looked, we had all the data. We just didn't [1:48] have any direction. We looked at our agents of an average of an eighthour shift. We were getting [1:54] about two hours of productivity. So, really minimal effort is what it seemed like there. So, [2:01] we partnered with our with Harpen and they came in and they were able to do the diagnosis for us. So, [2:09] really looking at all of our conversations, finding out who our guests were, who was ready [2:16] to book and who we were not reaching out to. Um, pricing the gap and then ranking the fix. So, [2:25] what do we do? We implemented concierge sales act, right? which allowed us to rank the leads. [2:32] We it allowed us to also detect hand raises. So maybe we spoke to a guest a month ago [2:40] and then they were on the website yesterday looking at a similar cruise. It allowed us to [2:45] identify in real time and flag that hand raise back to the agent that they could call and say, [2:53] "Camille, it looks like you're ready to talk about that cruise. What other questions do you have? how [2:58] can I support you? So, really helping us to engage. It matched the right seller and gave [3:06] the seller the information before. So, not only are our agents reaching back out to the guests, [3:13] but they're calling and they're calling educated. We know that you were looking at this cruise, [3:18] we also know the propensity to book. So, we're really calling our guests at the best time. So [3:26] we were able to um really give that blind spot a name and a number and uh reach out to them. So [3:34] we handed it back to our team with context. So before our agents are sitting there, [3:41] they have five systems open. We were able to hand back to them. Now they're using one [3:46] system that gives them all the information that they need. time spent on who to call. [3:54] One of the biggest obstacles was our team said, "The reason why our productivity is so low [3:59] is because I'm trying to find who's the best person to call. Do I have a casino offer? [4:05] Do I have a Europe offer?" You know, free flights. So, we took all of that guesswork out. So [4:13] now instead of spending 77% of their time looking for who to call in an instant they had the right [4:20] guest on their line and then guest context. I'm calling but how much do I actually know [4:28] about what we should be talking about today? Well, we removed that. Now they have full visibility of [4:34] who their guest is and what they should be talking about. So, a real gamecher for us [4:40] and and that blind spot that we now fixed drove $120 million in incremental revenue. And really [4:50] high intent guests now are being contacted 91% of the time versus the 51% of the time. So, [5:01] really great partnership. Yeah, we've done incredible things. Anyone here run a call [5:05] center or responsible for a contact center at all? Couple of you. So you'll know when I say [5:10] contact centers are full of data but very few facts. And so what Camille was talking about [5:17] just to make it real was they have about 400 people who spend their day selling, [5:21] calling people, talking to them. Each one of those sellers has about 8,000 guests [5:28] in their list load. And so they literally would spend five of the eight and a half hours a day [5:33] trying to interrogate all these different tools. 65y old country um company imagine the data is [5:40] probably lives there somewhere but how do we get to it? So they're spending all of this time now [5:45] we know when you've cruised before what your preferences are where you live. You have an [5:51] anniversary coming up. Your daughter's getting ready to graduate from high school. We pull [5:55] all this information from previous interactions and really hand it to the Asians to say, "Hey, [6:01] go call Camille. She hasn't cruised in 6 months. She usually cruises three times a year. [6:06] We know she loves going to Alaska and it's her anniversary next month. Let's give her a call [6:10] and have just a conversation about this special offer we have for her." That resonate? Make sense? [6:17] Awesome. So, let me plug in. Um, I'll show you the tool that we were talking about that we used [6:23] to really start the process that they lived in. So, I did see quite a few hands. So, someone [6:29] has to help me and tell me um what what are you trying to solve in your contact center this year? [6:35] Who has something they're trying to solve very specifically? Anybody? Who's going to be brave? [6:41] There are always things to solve. Well, that's the problem. Usually, there's too much to solve, [6:47] right? anyone still blind spots. Um certainly one. So it's all about really productivity. [6:55] I mean I'm sure you have agents that are doing a really good job and then you have the sort of the [6:58] bell curve that you've got a lot of people sitting in the middle not doing such a great job. So [7:03] I'll introduce you to the tool acuity. And so um this is not a dashboard. So I'm sure a lot [7:08] of people you've guys you've guys tried some kind of dashboarding to help you gain insights and the [7:12] dashboards tell you what happened. This tool tells you why it happened and really what to [7:17] go do about it. That's the real difference. And so this is sort of the homepage and um [7:23] this is a fictional cruise company um for obvious reasons. Um but it it starts by telling you how [7:29] much money if you're on sales, it's telling you how much money you're leaving on the table. So [7:32] for every conversation that happens in a contact center, email chat, phone, um we're listening for [7:38] was there a sales a sales call? Could be direct sales, could be sales through service. Um, did [7:42] we earn the right to ask the customer to do more business with us because we gave a great service [7:46] and solved their issue? Um, we then then we're listening for if there was a sales opportunity, [7:53] did our employee, did our agent attempt to close it? And then if they did attempt, were they [7:58] successful? And so this is the manifestation of um all the opportunity that was left on the table [8:05] that the agent didn't actually close. Um, and there's tons of stuff for for those of you [8:10] contact center. a quick view. How's resolution fairing? Um what do you need to know about? Um [8:16] what recent reports are out there? But when you then drill into the the productivity, [8:20] um let's just jump down to agents. And so here we give you a quick view across your agent base. [8:25] So a view into pretty much whatever you feel is important. Stack ranking, top performers, [8:31] by the way, those who have missing coaching. Um we're also looking for things like burnout. [8:36] Um, so we'll look at semantic ways and um, changes in performance [8:40] where our agents are are acting different from how they typically do. Um, a sign of burnout, [8:46] uh, hard job, someone needs to reach out, tap on the shoulder, ask them how they're doing, [8:50] give them some love. Um, they can go in, you can drill in and say, "Okay, here's agent A. [8:56] How are they performing? What's their overall QA score? Um, how do they perform against the team? [9:01] Um, where do they need improvement? Um, if they're doing sales, what are those outcomes? um [9:06] where are they doing a good job and here are some specific AI coaching recommendations that you can [9:10] go to. So lots of context at the individual agent level. Um you can also kind of drill up to say [9:17] look I'm really focused on sales. So I really need to kind of think about sales and so we'll [9:23] go into our sales intel. This view back to what I showed you earlier. This is all about here's [9:28] here's the opportunities that were out there. um here's the revenue opportunity by product [9:35] and then as we drill down to say hey here were the sales triggers [9:39] cabin upgrades price inquiries and then how do we convert against those triggers very specifically. [9:45] So now we're getting a view to hey where do we need to focus our effort? The thing about contact [9:49] centers for those you live in them competing priorities we've got to do everything we've [9:53] got limited budget. Do we focus on sales? Do we focus on cost of sale? um our our our marketing [9:59] team is really interested in customer friction. Our leadership team wants us to save costs. [10:03] Where do we go? This is a tool to help start direct you towards it. Additionally, if you're [10:08] thinking about something like customer effort, we can give you tons of context from, hey, here's [10:14] your average CES. By the way, if you have TMPS or MPS available, we can collate it to where you can [10:18] start using sentiment and customer effort as a leading indicator instead of a lagging indicator, [10:23] which is always fun because you can't influence what already happened. um what's driving effort, [10:28] the language that's used, the explanations. Um by the way, we tell you what to do. Here's repeat [10:33] actions. So now you know what to actually go do. Um here's your trends. Here's by agent, by skill, [10:40] um by issue bucket, language. So there's tons of context if you're trying to focus on something [10:45] like um customer effort. If you're trying to drive calls out, so a lot of our clients spend [10:51] over 50% of their budget on repeat calls, customers who call back. And so we have a [10:55] whole section just focused really around how do we help you understand what's going on. [11:00] So we'll show you here's your calls, your repeat calls. Um they're calling back for the same issue. [11:04] By the way, in this example, your second call cost 27% more than your first one. So it's now 1 plus 1 [11:10] is 1 plus 1.27 is driving your cost basis. Here's when they call back. So hour, four hours, [11:18] whatever increments you're interested in. Most importantly, you come down and say, you know, I [11:22] I really need to get specific. So we start to say okay when a customer calls with a pricing question [11:28] we say do they call back for the same reason or they call back for a different reason um and we [11:33] can give you that insight very clearly on if when they're contacting you back what's happening. [11:37] So this helps you understand hey what's adjacent to what they're calling for and so is it worth me [11:42] taking the time to expand the peripheral vision of that agent give them a little more handle time [11:47] to focus on what's likely to come next versus have that customer call back in in a day or in an hour. [11:52] So, lots of context you can drill into. Um, by the way, if you're interested in competitors, [11:57] um, we're looking for all instances where your competitors are referenced through the [12:00] interactions that happen. Um, we'll give you an intelligent brief on, hey, here's where you're [12:05] winning the battle, here's where you're losing the battle. Um, here's the battle cards for the [12:10] specific um, competitors that are out there, and here's where you win, and here's where you lose. [12:15] So again, he can start to help enable your your business make better decisions around do we want [12:20] to win on comfort versus price versus product set versus features. Um, again, context that you never [12:26] had before because we all know context really is what drives decisions at the end of the day. [12:32] Um, there's a ton in here I'm not going to cover. The last thing I will cover though is [12:36] all this stuff is interesting, but um, this is preset. It's also set up where you can go and [12:41] you can go ask it any question you want to. And I'll give you a real world example. Um, [12:47] why are customers calling about safety reasons? So, when the missiles started flying in Iran, [13:00] we started sending a report into the president of Princess Cruise Lines to help them understand [13:06] at the macro level how that war was influencing their customers. and they were pretty much, hey, [13:14] are they cancelling? Are there concerns? What do we do? Um, we've got a lot of ports in the med. [13:19] Do we keep those ports open? Do we divert? Um, how do we answer these questions? And so, [13:24] here's an example of what that report generated. Um, and so, you know, of the 12,000 calls, [13:31] there were 72 that were focused on world events. 1,700 were were thinking about cancelling. Um, [13:39] some did. It gives you the highest of Middle East conflict and UA airspace closure. This by the way [13:44] was also when the cartels in Mexico were flaring up if you remember that fund. So it flagged that [13:48] as well. So we tell you why it's happening. We give you things like sentiment and resolution [13:53] status. The top issues very clearly labeled out to you and then we kind of tell you hey here's [13:59] some recommendations. Our AI detected that there was a force majour policy that was in conflict [14:05] with guests who are already stranded. the level of context for those of you who run businesses, [14:10] you can never get to without a tool to really enable that. And so that helped um apprentice sort [14:15] of make a decision on how do we want to handle stranded guests even though we have a policy [14:19] um there's this loophole that no one no one thought of because they're already at sea um [14:24] or they were already on a flight into a location um with us and these missiles started flying. [14:29] It also does things like hey you may want to sort of reach out and start proactively calling your [14:34] customers. there's a real fear about Kusadasi as an example due to its proximity to Iran. And so [14:41] this is an example where you take context around what's happening in your business [14:46] and you lift it up to leadership to make strategic decisions around in addition to [14:51] enabling your contact center team to make tactical decisions about what is my team doing really well [14:57] and how do we improve that versus where we need to do more. So a pause there. Um, [15:05] who any questions? Because I can keep I can go all day. I'll keep you guys here until like tomorrow [15:10] morning if we Okay. Um, some other context here, by the way. So, outbound calls, we go as far as [15:17] we will tell you what a good outbound voicemail sounds like. So, we're listening for, hey, did we [15:22] leave a voicemail? If we did, did the customer call back? And if the customer did call back, [15:28] did we close that deal? And so now we can start to give you a view towards, hey, here is what a good [15:34] outbound voicemail sounds like, which prompts action, creates the emotional hook or urgency [15:39] for their customer to call back. So um, lots of of context that, you know, historically were [15:45] impossible to get to that now is very, very easy to get to. Um, how about things like, [15:51] hey, what are my top sellers doing differently that the rest of my sales team is not doing? [15:56] um we can cite for you very specifically um here's what they're doing, here's what they're saying, [16:01] here's how they're saying it. Um a few other things that this tool does. Um [16:06] knowledge bases are always fun for those in the contact center. Managing a knowledge base is [16:10] always fun. Um, this tool also will look at what your top team is doing and it will nom [16:16] knowledgebased articles for you automatically. Pre-written for your team to look at, review [16:21] and append or approve and it automatically will sweep into your knowledge base. Um, [16:27] IVR prompting by customer intent. We can automate the IVR prompting [16:31] by your the intent of your customers um, versus the way it's historically been done. [16:36] Everyone out there, 99% of the people I'm talking to are all about agentic AI and what we're going [16:40] to do. And while we don't do that, we enable it. So, we'll subcategorize all your your your calls [16:46] and your emails and we'll say, "Hey, here's the classification. Here is the likely layer [16:52] of the next contact type that you should consider to automate whether you [16:56] use Aentic Chapi, maybe go offshore, whatever it's going to be. And by the way, [17:00] here's all the knowledgebased articles that goes with them to really orchestrate how you move from [17:08] a live agent into a lowerc cost environment with the right context to ensure that as we heard [17:13] on the last presentation, we're able to resolve the issue. So that's one of our tools that we're [17:20] using um with Princess. And so appreciate your time allowing us to share. Thank you very much.