前進保險 (PGR) 2024 Q4 法說會逐字稿

Good morning, and thank you for joining us today for Progressive's Fourth Quarter Investor Event. I'm Doug Constantine, Director of Investor Relations, and I will be a moderator for today's event. The company will not make detailed comments related to its results in addition to those provided in its annual report on Form 10-K and the letter to shareholders, which have been posted to the company's website.

早安，感謝您今天參加 Progressive 的第四季投資人活動。我是投資者關係總監 Doug Constantine，我將擔任今天活動的主持人。除了已發佈在公司網站上的 10-K 表年度報告和致股東信中提供的評論外，本公司不會對其業績發表任何詳細評論。

This quarter includes presentation on a specific portion of our business, followed by a question-and-answer session with members of our leadership team. The introductory comments and the presentation were previously recorded. Upon completion of the previously recorded remarks, we will use the balance of the 90 minutes scheduled for this event for live questions and answers with the leaders features in our recorded remarks as well as other members of our management team.

本季度包括對我們業務特定部分的介紹，然後與我們的領導團隊成員進行問答環節。介紹性評論和演講之前均已錄製。完成先前錄製的發言後，我們將利用本次活動剩餘的 90 分鐘時間，與錄製發言中的領導人以及管理團隊的其他成員進行現場問答。

As always, discussions in this event may include forward-looking statements. These statements are based on management's current expectations and are subject to many risks and uncertainties that could cause actual events and results to differ materially from those discussed during today's event. Additional information concerning those risks and uncertainties is available in our annual report on Form 10-K for the year ended December 31, 2024, where you will find discussions of the risk factors affecting our businesses, safe harbor statements related to forward-looking statements and discussions of the challenges we face. These documents can be found via the Investor Relations section of our website at investors.progressive.com.

與往常一樣，本次活動的討論可能包括前瞻性陳述。這些聲明是基於管理層目前的預期，並受許多風險和不確定性的影響，可能導致實際事件和結果與今天活動中討論的結果有重大差異。有關這些風險和不確定性的更多信息，請參閱我們截至 2024 年 12 月 31 日的 10-K 表年度報告，您可以在其中找到影響我們業務的風險因素的討論、與前瞻性陳述相關的安全港聲明以及我們面臨的挑戰的討論。這些文件可透過我們網站 investors.progressive.com 的投資者關係部分找到。

To begin today, I'm pleased to introduce our CEO, Tricia Griffith, who will kick us off with some introductory comments. Tricia?

今天首先，我很高興介紹我們的執行長 Tricia Griffith，她將為我們做一些介紹性的評論。特里西亞？

Good morning, and thank you for joining us today. As we usually do for our calls after even number of quarters, today, we will have a presentation that dives into a certain aspect of our business. Today, we'll speak to two of our strategic pillars: people and culture, where everything starts for us and competitive prices with a focus on technology investments in our claims process.

早安，感謝您今天加入我們。正如我們通常在偶數個季度後的電話會議上所做的那樣，今天我們將進行一次深入探討我們業務某個方面的演示。今天，我們將討論我們的兩大策略支柱：人才和文化，我們的一切從人才和文化開始，以及具有競爭力的價格，並專注於索賠流程中的技術投資。

To present these topics today, we've enlisted two experts in Progressive. First, we have John Murphy, Claims President. John assumed the role of Claims President in December 2021 after spending nearly seven years as Progressive's Customer Relationship Management President. In this role, he was responsible for all aspects of claims strategy and resolution across all product lines. John joined Progressive as a claims representative in 1992. During his initial tenure in claims, he held various strategy, process and operations roles in the country, including regional claims manager, national implementation leader, claims business leader and physical damage process business leader.

為了今天介紹這些主題，我們邀請了 Progressive 的兩位專家。首先，我們有請索賠總裁約翰墨菲 (John Murphy)。在擔任 Progressive 客戶關係管理總裁近七年後，約翰於 2021 年 12 月擔任理賠總裁。在這個職位上，他負責所有產品線的索賠策略和解決方案的各個方面。約翰於 1992 年加入 Progressive 擔任索賠代表。在最初擔任理賠工作期間，他曾在該國擔任過各種策略、流程和營運職務，包括區域理賠經理、國家實施負責人、理賠業務負責人和物理損壞流程業務負責人。

After John, we have Matt White wide joining us today. Matt has been our claims business leader for data and analytics since May of 2022, after spending six years in Personal Lines managing our hologram innovation team in direct acquisition. In this role, he's responsible for our claims data science, business intelligence and data strategy to build solutions for all aspects of claims handling. Matt joined Progressive in 2014 as an IT manager after serving the CoStar. His military career took him from various leadership positions on ships petroleum in the Caribbean and Eastern Pacific to Washington, D.C., where he worked on Capitol Hill in Budget and Program Management. He was the Director of Operations in Key West Florida and ultimately to Cleveland, where he oversaw several business lines across the great Lakes region.

繼約翰之後，今天我們迎來了馬特懷特 (Matt White)。馬特 (Matt) 在個人險領域工作了六年，負責管理直接收購的全像圖創新團隊，之後自 2022 年 5 月起擔任我們的數據和分析理賠業務負責人。在這個職位上，他負責我們的索賠資料科學、商業智慧和資料策略，為索賠處理的各個方面建立解決方案。在為 CoStar 服務後，Matt 於 2014 年加入 Progressive 擔任 IT 經理。他的軍事生涯使他從加勒比海和東太平洋的石油船舶擔任各種領導職務，到華盛頓特區，他在國會山莊從事預算和專案管理工作。他曾擔任佛羅裡達州基韋斯特的營運總監，最後調任克里夫蘭，負責監督大湖地區的多條業務線。

Again, thank you for joining us this morning, and I will now pass it on to John. John?

再次感謝您今天早上加入我們，現在我將把發言權轉交給約翰。約翰？

Thank you, Tricia. Good morning, and thank you for joining us today. Over the next 40 minutes, you're going to hear a quick recap of 2024, which was a tremendous year for Progressive. We'll dive into two of our strategic pillars: First, people and culture, which is where our long history of success starts and then competitive pricing. We will focus on the critical role that claims plays in creating a competitive advantage that helps fuel business growth. And finally, we'll share just one of the many examples where the effective use of data and technology at Progressive enhances our claims team's accuracy and efficiency.

謝謝你，特里西亞。早安，感謝您今天加入我們。在接下來的 40 分鐘內，您將聽到 2024 年的簡要回顧，這一年對 Progressive 來說是意義非凡的一年。我們將深入探討我們的兩大策略支柱：首先是人才和文化，這是我們悠久成功的開始；其次是具有競爭力的價格。我們將重點關注索賠在創造有助於推動業務成長的競爭優勢方面發揮的關鍵作用。最後，我們將分享眾多案例中的一個，這些案例表明 Progressive 透過有效使用數據和技術提高了我們索賠團隊的準確性和效率。

Now before I get to the topics at hand, I wanted to acknowledge the devastating wildfires that ravaged the Los Angeles area injury. Our thoughts and prayers remain with those affected by this event. In that vein, we know our customers want and expect action to support their recovery. Our catastrophe services team responded quickly with empathy and respect to meet their needs. While the majority of the claims we received are now resolved, our team continues to assist our customers however they can.

現在，在我開始討論當前話題之前，我想先承認肆虐洛杉磯地區的毀滅性野火。我們的思念和祈禱與那些受到這事件影響的人們同在。從這個角度來看，我們知道我們的客戶希望並期待採取行動來支持他們的復甦。我們的災難服務團隊迅速做出反應，以同理心和尊重的態度滿足他們的需求。雖然我們收到的大部分索賠現已解決，但我們的團隊仍將繼續盡力為客戶提供協助。

I also wanted to acknowledge our human resources and local leadership team who responded quickly to support our employees who are impacted by these fires. As a senior leader in this organization, the urgency, empathy and care that our collective team has shown here is an excellent example of our core values in action and fills me with incredible pride.

我還要感謝我們的人力資源和當地領導團隊，他們迅速做出反應，為受火災影響的員工提供支援。身為組織的高階領導，我們團隊所展現的緊迫感、同理心和關懷是我們核心價值付諸行動的絕佳例證，讓我感到無比自豪。

Quickly recapping 2024. In short, this was arguably the greatest year in the 87-year history of Progressive. Net premiums written grew approximately 21% year-over-year, finishing the year at $74.4 billion. In absolute dollars, premium grew by nearly $13 billion in this single calendar year. So let me try and put that in perspective.

快速回顧 2024 年。簡言之，這可以說是進步黨 87 年歷史上最偉大的一年。淨保費收入年增約 21%，全年達到 744 億美元。以絕對美元計算，光是這一年保費就增加了近 130 億美元。因此，讓我嘗試客觀地看待這個問題。

That nearly $13 billion net premium increase in '24 is equivalent to adding the premium of the eighth largest auto insurer in the calendar year of 2023. For those who have been following us for a while, the premium growth in '24 is about $1 billion more than we wrote for the entire calendar year of 2003. Average written premium increases provided a little bit of a tailwind here, but the majority of this premium growth comes from customer growth, which is our preferred measure, and 2024 was a record year here as well. We increased active policy by more than 5 million this year, which is more than twice the previous highest annual rate of policy growth in our history.

24 年淨保費增加近 130 億美元，相當於 2023 日曆年第八大汽車保險公司的保費總額。對於一直關注我們的人來說，24 年的保費成長比我們 2003 年全年的保費成長高出約 10 億美元。平均承保保費的成長帶來了一點推動作用，但保費成長的大部分來自客戶成長，這是我們優先考慮的指標，而 2024 年也是創紀錄的一年。今年，我們的活躍保單增加了 500 多萬份，這是我們歷史上最高年保單成長率的兩倍多。

Growth is one important objective for us and profitability is the other. You're well aware of our stated calendar year goal of a 96 combined ratio or below. And in 2024, we produced a CR of 88.8, well below the 96 and also about 6 points lower than 2023. We often say that growing or generating a profit individually is fairly straightforward. To do both simultaneously and at the performance levels that we did in 2024 is significantly more challenging.

成長是我們的一個重要目標，獲利是另一個重要目標。您很清楚我們設定的年度目標是綜合比率達到 96 或以下。而在 2024 年，我們的 CR 為 88.8，遠低於 96，也比 2023 年低約 6 個百分點。我們常說，個人的成長或獲利是相當簡單的。要同時完成這兩項任務並達到我們在 2024 年所達到的績效水平，難度要大得多。

In true Progressive fashion, more than 65,000 of us worked diligently together to deliver what may very well be the best combination of growth and profitability in the industry. While we celebrate these successes, our excellence core value challenges us to continually improve for our teammates, our customers, partners and investors. I believe strategic investments in our people, data, processes and technology position us to perhaps have an even better 2025.

以真正的進步方式，我們 65,000 多名員工齊心協力，努力實現行業內可能是最佳的成長和盈利組合。在我們慶祝這些成功的同時，我們的卓越核心價值也激勵著我們不斷為我們的隊友、客戶、合作夥伴和投資者做出改進。我相信，對人才、數據、流程和技術的策略性投資或許能讓我們在 2025 年取得更好的成績。

The fourth cornerstones is a construct that we've shared in the past. It's how we think about competitive advantage and defines how we win and win in the right way. We start with our core values or who we are. These are unifying statements that provide the very foundation of our company culture since the late 1980s.

第四個基石是我們過去共同建構的。這是我們思考競爭優勢的方式，也是定義我們如何取勝並以正確的方式取勝的方式。我們從我們的核心價值或我們是誰開始。這些統一的聲明構成了我們自 20 世紀 80 年代末期以來公司文化的基礎。

Next is our purpose or why we're here. We evolved this statement in 2022 to build on the legacy of our core values, and our history of challenging the status quo to accelerate progress. Then we have our vision of where we're headed. This company-wide objective highlights the constituents that we feel so fortunate to serve and our never-ending pursuit of becoming the number 1 destination.

接下來是我們的目的或是我們為什麼在這裡。我們在 2022 年制定了這一聲明，以繼承我們的核心價值以及我們挑戰現狀以加速進步的歷史。然後我們就會對未來的方向有自己的願景。這個全公司目標凸顯了我們非常榮幸能夠服務的客戶群，以及我們永無止境地追求成為第一目的地的決心。

Finally, is our strategy or how we'll get there. These are the four important pillars of this strategy, and two of them will be the focus of today's discussion. People and culture, which are collectively our most powerful source of competitive advantage and competitive pricing is driven by industry-leading segmentation, claims accuracy and operational efficiency.

最後，是我們的策略或說我們如何實現目標。這是該策略的四大重要支柱，其中兩個將是今天討論的重點。人才和文化是我們最強大的競爭優勢和競爭性定價的源泉，而這源自於業界領先的細分、索賠準確性和營運效率。

Progressive people and culture are collectively our most powerful source of competitive advantage, a very well-known and respected brand, a routinely recognized top work environment and sustained business success makes us an attractive option for folks in market looking for jobs.

不斷進步的人才和文化是我們最強大的競爭優勢源泉，一個非常知名和受人尊敬的品牌、一個經常被認可的頂級工作環境和持續的業務成功使我們成為市場上尋找工作的人們的一個有吸引力的選擇。

While our growth rate in staff over the past 10 years lags PIF growth and premium, it has been substantial. Our outstanding talent acquisition group generates robust Canada pools for our roles from which we then choose top talent with personal values that align with our core values. Due to the very significant number of people interested in working at Progressive, we're able to be highly selective. In fact, we hire less than 3% of people who apply for jobs at Progressive. This scale advantage allows us to bring additive skills and experiences to our team, ensuring that we're always growing and always evolving.

雖然過去 10 年我們的員工成長率落後於 PIF 的成長和保費的成長，但這一成長率仍然相當可觀。我們優秀的人才招募團隊為我們的職位建立了強大的加拿大人才庫，然後我們從中挑選出與我們的核心價值相符的個人價值的頂尖人才。由於有非常多的人對在 Progressive 工作感興趣，因此我們能夠嚴格篩選。事實上，我們僱用的 Progressive 求職者不到 3%。這種規模優勢使我們能夠為團隊帶來附加技能和經驗，確保我們始終成長、不斷發展。

Our culture is built on the foundation of our core values, and it has evolved beautifully with our people growth over several years. A transparent, inclusive and forward-thinking environment ensures our people feel empowered, welcomed, valued and respected. Development focus allows our people to build careers have become talented business leaders who can drive enduring success for decades to come.

我們的文化建立在我們的核心價值的基礎之上，隨著我們員工的成長，多年來它也得到了完美的演變。透明、包容和前瞻性的環境確保我們的員工感受到授權、歡迎、重視和尊重。發展重點使我們的員工能夠發展事業，成為能夠在未來幾十年推動持久成功的才華橫溢的商業領袖。

Gallup is a multinational analytics and advisory company well known for its public opinion polls and surveys conducted worldwide. In an article published on January 14, 2025, Gallup reported that US employee engagement sank to a 10-year low. That is not the case at Progressive. At Progressive 2024 marked a significant year of achievement where engagement has surpassed previous records. Progressive ranks in the 98th percentile of engagement and the 99th percentile in overall satisfaction. These results have us in the top 10 and top 5, respectively, for US companies in Gallup's client database.

蓋洛普是一家跨國分析和諮詢公司，以其在世界各地進行的民意調查和調查而聞名。蓋洛普在2025年1月14日發表的一篇文章中報告稱，美國員工敬業度跌至10年來的最低水準。但 Progressive 的情況並非如此。對 Progressive 來說，2024 年是取得重大成就的一年，參與度超越了以往的紀錄。Progressive 的參與度排名排名第 98 個百分位，整體滿意度排名排名第 99 個百分位。這些結果使我們在蓋洛普客戶資料庫中的美國公司中分別位列前十名和前五名。

As we look a level deeper into our 2024 survey results, we see year-over-year increases in every single question and in every tenure bucket. As we explore the tens of thousands of comments from our team, responses to the question, the one thing I like most, generates themes of a positive work environment, care for customers and colleagues, strong leadership and support, career growth and development, flexibility and values. You can see just a few of the comments here on this slide.

當我們更深入研究 2024 年的調查結果時，我們發現每個問題和每個任期範圍都呈現同比增長。當我們瀏覽團隊數以萬計的評論時，對這個問題的回答是，我最喜歡的一件事，產生了積極的工作環境、對客戶和同事的關心、強有力的領導和支持、職業成長和發展、靈活性和價值觀等主題。您可以在這張投影片上看到一些評論。

Once again, consistent with our excellence core value and our desire to consistently improve, we also ask about things that should change. The data that we collect from that question, really all others are rich and guide our efforts to further nurture our special culture. We believe firmly that the investments we've made in our people and culture have fueled our growth and success for several years now. We'll continue to care for both to position us to be e-better in the future.

再次，為了與我們的卓越核心價值和不斷改進的願望保持一致，我們也詢問了應該改變的事情。我們從這個問題中收集的數據實際上所有其他數據都是豐富的，並指導我們進一步培養我們特殊文化的努力。我們堅信，我們對員工和文化的投資多年來推動了我們的成長和成功。我們將繼續關注這兩者，以便我們在未來能夠更好地進行電子化。

In prior presentations, we've discussed our ability to rapidly deploy risk selection and new product models, our segmentation excellence as well as efficient media and agency compensation each of which has had a positive impact on rate competitiveness. Claims plays a very important role here as well. The combination of loss costs and loss adjustment expenses typically represents between 70% and 75% of our total company expenditures.

在先前的演示中，我們討論了快速部署風險選擇和新產品模型的能力、我們卓越的細分以及高效的媒體和代理補償，這些都對利率競爭力產生了積極的影響。索賠在這裡也扮演著非常重要的角色。損失成本和損失調整費用的總和通常占我們公司總支出的 70% 到 75%。

So we focus heavily on producing accurate claim outcomes while optimizing efficiency, customer experiences and employee engagement. We refer to these as our four guiding principles, and they have helped inform our daily efforts, investments and priorities in claims for more than two decades. On the left and right, you see the two key constituencies that we are privileged to serve. In the middle are the two items that have the most linear connection to our combined ratio and ultimately the rates that we then have in market. If we don't deliver here, we simply can't compete at the level we become accustomed to and continue to aspire to.

因此，我們非常注重提供準確的索賠結果，同時優化效率、客戶體驗和員工參與度。我們將這些稱為我們的四個指導原則，二十多年來，這些原則指導了我們在索賠方面的日常工作、投資和優先事項。左邊和右邊，你們可以看到我們有幸服務的兩個關鍵選區。中間的兩個項目與我們的綜合比率以及最終我們在市場上的利率具有最線性的關聯。如果我們不能實現這一目標，我們就無法達到我們習慣並繼續追求的水平。

This chart is our theoretical cost curve and gives incident to how we approach our work. The x-axis represents loss adjustment expenses, or LAE, essentially what we spend to operate the claims organization. The Y-axis represents loss costs, both claims payments and reserves plus LAE. Much like my commentary earlier about growth and profitability, being accurate or being efficient individually are straightforward. Doing both at the same time is a far greater challenge, and we believe we do it better than anyone in the industry.

這張圖表是我們的理論成本曲線，並給出了我們如何處理工作的情況。x 軸代表損失調整費用，或 LAE，本質上是我們營運索賠組織的支出。Y 軸代表損失成本，包括索賠支付和準備金加上 LAE。就像我之前關於成長和獲利能力的評論一樣，準確或高效是簡單的。同時做到這兩點是一個巨大的挑戰，但我們相信我們比業內任何人都做得更好。

Now there are downsides to focusing too much on either side of this curve. If sole objective was to be the lowest cost, we couldn't make the investments in people, data and technology that enable us to pay the right amount. Quality would suffer, cycle time would increase, rentals would extend and ultimately, loss costs would rise.

過度關注這條曲線的兩側都會帶來一些負面影響。如果唯一目標是降低成本，我們就無法在人員、數據和技術方面進行投資，因此無法支付正確的金額。品質會受到影響，週期時間會增加，租金會延長，最終損失成本會上升。

Likewise, if accuracy was the sole objective, you would need to overspend to eliminate the potential for errors, likely spend on non-value-added activities, and end up with a cost structure that would be uncompetitive. We leverage operational excellence to drive us down this curve. This is where having the best people in the industry who are willing to get into the details, investigate tirelessly and effectively and make the right decisions around coverage, liability and damages, really pays dividends and moves us down the left side of this curve. Empowering them with the right data, processes and tools, increases their efficiency and throughput, moving us down the right side of this curve.

同樣，如果準確性是唯一目標，您將需要超支以消除潛在的錯誤，可能花費在非增值活動上，最終導致成本結構缺乏競爭力。我們利用卓越的營運來推動我們實現這一目標。正是在這裡，我們擁有業內最優秀的人才，他們願意深入細節，孜孜不倦地進行有效調查，並就保險範圍、責任和損害賠償做出正確的決定，這確實帶來了回報，並使我們走在了這條曲線的左側。為他們提供正確的數據、流程和工具，可以提高他們的效率和產量，使我們朝著曲線的右側前進。

Now this view of driving down total cost is more about the journey and the incremental gains that come with it than any theoretical destination because no matter how good we get, we will always seek to be better. In the claims organization, we constantly measure accuracy and efficiency, so we always know how we're doing. Objective assessments of the work provide valuable insights into how and where we can be even better.

現在，這種降低總成本的觀點更多的是關於旅程和隨之而來的增量收益，而不是任何理論目的地，因為無論我們變得多麼好，我們總是會追求更好。在索賠組織中，我們不斷衡量準確性和效率，因此我們始終知道自己做得如何。對工作的客觀評估為我們如何以及在哪些方面可以做得更好提供了寶貴的見解。

And while we're always trying to find the optimal spot on this curve, we also strategically target efforts to shift the curve to a better position for us and for our customers. You'll hear from Matt White shortly about how we invest in new technology, data and processes to shift this curve. And this is what driving down and shifting the curve looks like in practical application, using perhaps the best measure of success for claims, which total cost.

我們不僅一直在努力尋找這條曲線上的最佳位置，而且還在策略上努力將曲線轉移到對我們和我們的客戶更有利的位置。您很快就會聽到 Matt White 講述我們如何投資新技術、數據和流程來改變這一趨勢。這就是在實際應用中降低和移動曲線的情況，使用也許是索賠成功的最佳衡量標準，即總成本。

While claims supports all manufactured products across auto, special lines, commercial lines and home, I use private passenger auto data here because it offers the cleanest comparison relative to industry. This graph is a visual representation of the competitive advantage our claims organization in conjunction with all other business areas brings to Progressive and shows why we're able to grow faster and with better margins than the industry as a whole.

雖然索賠支持汽車、專用線、商業線和家用的所有製造產品，但我在這裡使用私人乘用車數據，因為它提供了相對於行業最清晰的比較。這張圖表直觀地展示了我們的理賠部門與所有其他業務領域相結合為 Progressive 帶來的競爭優勢，並展示了我們為何能夠比整個行業更快增長並獲得更好的利潤率。

Total cost is a combination of loss costs, which tied directly to accuracy and LAE, which is our primary efficiency metric. This slide compares Progressive reflected by the to the industry ex Progressive, which is in orange, and shows that we've maintained an advantage for more than a decade. More recently, though, you can see the gap widen. And while we don't have complete 2024 statutory data for the industry yet, we have loss ratio data through the first three quarters of '24, and we maintain a 7-point advantage in loss ratio alone. As we compare our own results for 2024 to 2023 for total indemnity, we see a near 9-point improvement and finish the most recent year, sub-70%.

總成本是損失成本的組合，它與準確性和 LAE 直接相關，而 LAE 是我們的主要效率指標。這張投影片將 Progressive 所反映的情況與業界前 Progressive（橘色）進行了比較，顯示我們十多年來一直保持著優勢。然而，最近你可以看到差距正在擴大。雖然我們還沒有該行業完整的 2024 年法定數據，但我們擁有 2024 年前三個季度的損失率數據，僅在損失率方面，我們就保持了 7 個百分點的優勢。當我們將 2024 年和 2023 年的總賠償結果進行比較時，我們發現賠償總額提高了近 9 個百分點，而最近一年的賠償總額低於 70%。

Back to our full set of guiding principles. I'm proud to say that in 2024, our claims organization had the highest engagement in our history. The lowest LAE ratio in our history, improved accuracy, and generated the best customer satisfaction and retention that we have in several years. These results don't just happen and they don't happen overnight. This culture of execution excellence has been developed over years, and isn't easily replicable, but we believe it can be improved upon.

回到我們的全套指導原則。我很自豪地說，2024 年，我們的索賠組織的參與度達到了歷史上最高的。我們歷史上最低的 LAE 比率、更高的準確性以及多年來最佳的客戶滿意度和保留率。這些結果不會憑空而來，也不會在一夜之間發生。這種卓越執行的文化已經發展多年，並且不易複製，但我們相信它可以改進。

And now Matt White, our claims business leader of data and analytics is going to share with you how we use data and technology to enable and enhance our human capital. While the efficiency and cost wins you'll hear about are very significant, it's our intentionality around also increasing accuracy that makes these efforts really valuable.

現在，我們的數據和分析索賠業務負責人 Matt White 將與您分享我們如何利用數據和技術來實現和增強我們的人力資本。雖然您所聽到的效率和成本方面的勝利非常重要，但我們對提高準確性的意圖也使得這些努力真正有價值。

Matt?

馬特？

Thanks very much, John. Empowering our employees is indeed at the heart of our approach and claims. As emphasized by John and as Tricia states often, along with our culture, our people are our greatest competitive advantage. We focus on enhancing, not replacing their roles through technology. Similarly, we want to enable our customers with user-friendly digital options so they can interact with us when, where and how they choose. We'll share some of those concepts and how our progressive DNA around digital experiences, segmentation and data translate to claims. And we'll dive into a specific use case around photo estimating, highlighting the Progressive made and our commitment to achieve the optimal balance of accuracy and efficiency.

非常感謝，約翰。賦予員工權力確實是我們的方法和主張的核心。正如約翰所強調的以及特里西亞經常說的，除了我們的文化之外，我們的員工也是我們最大的競爭優勢。我們專注於透過技術增強而不是取代他們的角色。同樣，我們希望為客戶提供用戶友好的數位選項，以便他們可以在他們選擇的時間、地點和方式與我們互動。我們將分享其中的一些概念，以及我們圍繞數位體驗、細分和數據的進步 DNA 如何轉化為主張。我們將深入研究照片估算的具體用例，重點介紹 Progressive 所做的努力以及我們致力於實現準確性和效率的最佳平衡。

John showed how we stack up from a total loss LAE perspective. Here is an alternative view of efficiency. An indexed comparison of auto policy growth and the number of staff aligned to auto claims handling. We continue to grow and hire. But our scale and strategic investments have allowed us to grow staff at a slower rate than policies, shown in the gray shaded area indicating the growing gap. And while we acknowledge that frequency declines in this period are a tailwind, we're very pleased with the efficiency improvements we've been able to achieve.

約翰從 LAE 總損失的角度展示了我們的表現。這是對效率的另一種看法。汽車保單成長與負責處理汽車索賠的員工數量的指數比較。我們將繼續發展並招募人才。但我們的規模和策略投資使我們的員工成長速度低於政策規定的速度，灰色陰影區域顯示差距正在擴大。雖然我們承認這段時期頻率的下降是一個順風，但我們對已經實現的效率提升感到非常滿意。

Market conditions may have us make specific calendar year staffing decisions to ensure we're positioned for the future. But over the long term, we believe we can continue to become more efficient and selective with the human capital required to service our customers while maintaining the competitive advantage we believe our culture provides.

市場條件可能會要求我們做出具體的年度人員配置決策，以確保我們為未來做好準備。但從長遠來看，我們相信，我們可以繼續利用服務客戶所需的人力資本變得更有效率和選擇性，同時保持我們認為我們的文化所提供的競爭優勢。

In prior calls, you've heard us talk about our large and growing media spend and how critical it is for us to manage it efficiently. Efficiency is also top of mind with our claims technology spend. The vertical axis shows the indexed gap between auto policies claims staff, the same gray area I highlighted in the previous slide. In general, we want to be higher on the vertical axis, indicating more efficiency in our staffing. The horizontal axis is our claims IT spend, a subcomponent of LAE. In this case, both indexed to 2007.

在先前的電話會議中，您已經聽到我們談論我們龐大且不斷增長的媒體支出，以及有效管理這些支出對我們的重要性。我們在索賠技術支出方面也最注重效率。縱軸顯示的是汽車保險理賠人員之間的指數差距，與我在上一張投影片中突出顯示的灰色區域相同。整體來說，我們希望縱軸越高，表示我們的人員配備效率越高。橫軸是我們的索賠 IT 支出，它是 LAE 的一個子組成部分。在這種情況下，兩者都以 2007 年為索引。

You'll note that while we made some progress growing that policy of the staff gap from 2008 to 2016, when our technology spend was flat to down slightly, it grew notably faster with some strategic investment decisions after that point. The policy to staff gap grew by 50% from the index values of 1 to nearly 1.5 in the 9-year period from 2007 to 2016. But we've been able to grow it by 130% from 1.5 to 3.4 in the seven years since 2017. Our investments in technology to improve the claims process have shown through and clear efficiency gain. And not unlike our media spend, we're willing to invest here as long as we can do so efficiently.

您會注意到，雖然我們在 2008 年至 2016 年期間在擴大員工缺口政策方面取得了一些進展，當時我們的技術支出持平或略有下降，但在此之後通過一些戰略投資決策，其增長速度明顯加快。從2007年到2016年的9年間，政策與人員的差距增加了50%，指數值從1增加到近1.5。但自 2017 年以來的七年間，我們已將其從 1.5 增長至 3.4，增幅達 130%。我們為改善索賠流程而對技術的投資已明顯體現並提高了效率。與我們的媒體支出不同，只要能夠有效進行，我們願意在這裡投資。

You've heard us talk before about our digital capabilities and acquisition and policy servicing, and the obvious value we believe advanced product segmentation provides for our company results. That progressive DNA and experience applies just as well to claims. In fact, when it comes to first notice of loss or the moment of truth when a customer contacts us to report a claim, we think about it similarly to our acquisition funnels. That said, it is a bit more nuanced as it's a funnel that then feeds our claims handling funnel from the point of assignment to resolution. And that path can be non-linear depending on the circumstances of the claim or what facts might emerge a initial assignment.

您之前聽過我們談論我們的數位能力以及收購和政策服務，以及我們認為先進的產品細分為我們的公司業績帶來的明顯價值。這種進步的 DNA 和經驗同樣適用於索賠。事實上，當第一次收到損失通知或客戶聯繫我們報告索賠的關鍵時刻時，我們會以類似的方式來考慮我們的收購管道。也就是說，它更加微妙，因為它是一個漏斗，然後從分配點到解決點為我們的索賠處理漏斗提供資訊。並且該路徑可能是非線性的，這取決於索賠的情況或初始任務可能出現的事實。

Nonetheless, there are similar trade-offs at first notice of loss as in our acquisition funnels. Asking a lot of questions might aid initial triage and segmentation, but it might also frustrate customers, drive digital abandonment and increase call times and expense. What's very important, we gather as much accurate information as we can, as efficiently as we can, but limited to only what we truly need. As John alluded to previously, it's straightforward to one, it's much more complicated to do both at the same time. To be clear, segmentation in this context is about the facts of the claim and not customer segments. We are not talking about Robinsons or Sams here, but not every claim or accident is the same, knowing the number of vehicles and parties involved, where their vehicles are likely total losses and whether there are injuries and how severe are just a few of the many important facts that allow us to segment income and claims and make assignment decisions to get the claim in the right hands faster.

儘管如此，在第一次注意到損失時，也存在著與我們的收購管道類似的權衡。提出大量問題可能有助於初步分類和細分，但也可能會讓客戶感到沮喪，導致數字放棄並增加通話時間和費用。非常重要的是，我們要盡可能有效率地收集盡可能多的準確信息，但僅限於我們真正需要的資訊。正如約翰之前提到的，對一個人來說這很簡單，但同時做這兩件事就複雜得多。需要明確的是，此處的細分是關於索賠的事實，而不是客戶細分。我們在這裡不是在談論羅賓遜或薩姆斯，但並非每個索賠或事故都是一樣的，了解涉及的車輛和當事人的數量、他們的車輛可能完全損毀的位置以及是否有人受傷以及受傷程度如何只是許多重要事實中的一部分，這些事實使我們能夠細分收入和索賠，並做出分配決策，以便更快地將索賠交到正確的人手中。

Getting it wrong can negatively impact cycle time, which can increase expenses such as rental and storage fees, if we must spend time retriaging claims and/or vehicles. It also resulted in a suboptimal customer experience and even hurt retention. We believe there remains additional upside on the digital side of our first notice of loss funnel and have been investing accordingly. Industry data from Bain suggests that digital adoption in the auto claim submission process averages about 25%. While we are meaningfully above that industry data point, we think there remains additional opportunity given the long-term trends we've seen in acquisition and policy servicing.

如果我們必須花時間重新分類索賠和/或車輛，那麼錯誤的做法會對週期時間產生負面影響，從而增加租賃和倉儲費等費用。這也導致了客戶體驗不佳，甚至損害了客戶保留率。我們相信，我們首次損失通知的數字方面仍有額外的上升空間，並且一直在進行相應的投資。貝恩的行業數據顯示，汽車索賠提交流程的數位化採用率平均約為 25%。儘管我們遠高於該行業數據點，但我們認為，鑑於我們在收購和政策服務方面看到的長期趨勢，仍有更多機會。

Many customers prefer interacting digitally, and we'll continue investing to ensure we have those options available. That said, an accident is certainly a very different kind of life event than purchasing or modifying a policy, and we will always be available for human-to-human contact with customers that prefer or need that channel. One of the reasons we continue to invest in digital experiences for our customers and our employees is it further accelerates a virtuous data cycle we've referenced a variety of context over the years.

許多客戶喜歡透過數位方式進行互動，我們將繼續投資以確保提供這些選項。話雖如此，事故肯定是一種與購買或修改保單截然不同的人生大事，我們將始終為喜歡或需要這種管道的客戶進行人與人之間的聯繫。我們繼續為客戶和員工投資數位體驗的原因之一是，它進一步加速了我們多年來在各種背景下參考的良性數據循環。

As we work to resolve claims, our customers and employees generate a variety of interactions with each other and other involved parties. Historically, across the industry, these have been predominantly analog in their nature with lots of phone calls, manual claim and relatively unstructured data. At Progressive, we've invested significantly in monitoring digital experiences. Those enable us to turn many of those unstructured processes and associated data into more structured information. Additionally, the very nature of digital interactions creates even more data that's not exclusive to a fine set of events or outcomes. They also provide experiential insights to how our customers and employees are using those systems.

在我們努力解決索賠的過程中，我們的客戶和員工之間以及與其他相關方之間會產生各種互動。從歷史上看，在整個行業中，這些業務本質上主要是模擬的，包括大量的電話呼叫、手動索賠和相對非結構化的數據。在 Progressive，我們在監控數位體驗方面投入了大量資金。這些使我們能夠將許多非結構化流程和相關數據轉化為更結構化的資訊。此外，數位互動的本質創造了更多的數據，這些數據並不局限於一組精細的事件或結果。他們還提供了有關我們的客戶和員工如何使用這些系統的經驗見解。

This data helps us identify areas for additional claims process and technology investment connecting back to John's description of how claims contributes to competitive pricing. Our operational efficiency improved helped lower LAE, which enables lower rates, powering additional business growth and scale, generating even more data and enabling additional targeted investments.

這些數據幫助我們確定需要額外索賠流程和技術投資的領域，這與約翰關於索賠如何促進競爭性定價的描述相呼應。我們提高的營運效率有助於降低 LAE，從而降低費率，推動額外的業務成長和規模，產生更多數據並實現額外的有針對性的投資。

One example of these cycles in action has been photo estimating. Photo estimating is a digital experience whose adoption has grown significantly. It can be a convenient one for customers and claimants when they're uncertain as to the repair cost relative to any deductible or they're not sure when or where they'll get the vehicle repaired. From the comfort of their own home or workplace, they can submit photos via guided digital experience and get a repair estimate back quickly, often within a day, if not faster. If you recall the conceptual curves John discussed, photo estimating is a good example of a shift in the curve. And as we'll walk through even when we do a -- a shift, we continue to optimize so we can operate at the bottom of the curve.

這些循環實際運作的一個例子是照片估算。照片估價是一種數位體驗，其應用範圍已顯著擴大。當客戶和索賠人不確定與免賠額相關的維修費用，或不確定何時何地修復車輛時，它可以為他們提供便利。他們可以在自己舒適的家中或工作場所透過引導式數位體驗提交照片，並快速獲得維修估價，通常在一天之內，甚至更快。如果您還記得約翰討論過的概念曲線，那麼照片估算就是曲線轉變的一個很好的例子。正如我們將要介紹的那樣，即使我們進行了轉變，我們仍會繼續進行最佳化，以便能夠在曲線底部運行。

The chart on the left shows annual photo estimates of 2016, over which time we've enjoyed an 82% growth rate. While we're not showing actual numbers on the prime axis, know that in total, this represents millions of photo-based estimates written over the last eight years. And with it, of course, tens of millions of customer-submitted photos.

左側的圖表顯示了 2016 年的年度照片估計數，在此期間我們享受了 82% 的成長率。雖然我們沒有在主軸上顯示實際數字，但要知道，總的來說，這代表了過去八年中基於照片的數百萬個估計值。當然，還有數千萬張顧客提交的照片。

Secondary vertical axis is an indexed view of our Estimate Quality Index plotted on the orange line. These objective assessments of the work that John mentioned earlier are a core benchmark for quality and accuracy, and an invaluable source of ground truth. Despite the growing volume, there's been no material degradation in accuracy. So while we're pleased with the growth of photo estimating as a viable and efficient vehicle inspection channel, we are consistently measuring accuracy alongside the growth in volume to ensure the experience we're offering and outcomes we're generating aren't compromising the claims guiding principles John described earlier.

次要縱軸是繪製在橙色線上的估計質量指數的索引視圖。約翰之前提到的這些對工作的客觀評估是品質和準確性的核心基準，也是寶貴的事實來源。儘管數量不斷增加，但準確度卻沒有顯著下降。因此，雖然我們對照片估價作為一種可行且高效的車輛檢查管道的增長感到高興，但我們仍在不斷衡量準確性以及數量的增長，以確保我們提供的體驗和產生的結果不會損害約翰先前描述的索賠指導原則。

The chart on the right indexes daily estimates completed compared to traditional in-person estimate completion. In total, we've seen we can complete 2.5 times the number of estimates each day, making it our most efficient vehicle inspection channel.

右側的圖表顯示了每日完成的估算與傳統的現場估算完成情況的比較。總體而言，我們發現每天可以完成 2.5 倍的估價數量，這使其成為我們最高效的車輛檢查管道。

Once we achieve viable scale, refined our processes, organizational and data structures, we began to apply additional technology so that we could potentially drive further down the cost curve. In this case, we're using machine vision models alongside a variety of other technologies to help us automate parts of the photo estimating process. These photos illustrate the process without disclosing actual customer images. At a high level, we're using deep learning models or neural networks, alongside more traditional machine learning peaks to look at customer-submitted photos like the example on the left, identify the correct parts through segmentation masks and then identify the location and type of damage by part.

一旦我們達到可行的規模，改進了我們的流程、組織和資料結構，我們就開始應用額外的技術，以便我們可以進一步降低成本曲線。在這種情況下，我們使用機器視覺模型以及各種其他技術來幫助我們實現照片估算過程的部分自動化。這些照片說明了整個過程，但並未透露實際客戶的圖像。在高層次上，我們使用深度學習模型或神經網絡，以及更傳統的機器學習峰值來查看客戶提交的照片（如左側的示例），透過分割蒙版識別正確的部分，然後按部分識別損壞的位置和類型。

Generating an accurate estimate requires several discrete predictions. For example, we need to know not just the location of the damage by specific part, but also the type of damage. Is it a dent, scratch, et cetera. From there, we need to decide whether we should repair or replace the part. And in the case of repair, how many labor hours do we believe it will take. These models use transformer architectures you may have heard about, quite literally the second tea and ChatGPT, but they are trained very differently than some of the models making headlines more recently. These are supervised models that we train on our photos and our own curated data to ensure accuracy consistent with progressive estimating standards.

產生準確的估計需要幾個離散的預測。例如，我們不僅需要知道特定零件損壞的位置，還需要知道損壞的類型。是否有凹痕、刮痕等等。從那裡，我們需要決定是否應該修理或更換該零件。如果進行修復，我們認為需要花費多少工時。這些模型使用了您可能聽說過的轉換器架構，確切地說是第二茶和 ChatGPT，但它們的訓練方式與最近成為頭條新聞的一些模型非常不同。這些是我們根據自己的照片和自己整理的資料進行訓練的監督模型，以確保準確性符合漸進式估算標準。

Such models all benefit from very large, clean and diverse data sets. We're fortunate to have all three in-house without feeling a need to look externally. And again, that ground truth data from objective assessments and our own people's disciplined quality assurance processes and willingness to do the hard work to get it right give us the benefit of a very clean, trustworthy and we believe, differentiating historical data set. It's my experience that it's really the combination of technology and Progressive's talented people that make the brands. On this solution, our data science team works daily with our physical damage process team that combined brings over 400 years of experience. Most of them have worked at every level of our claims organization. They have estimates themselves, worked in and alongside shops and have a deep understanding of all facets of vehicle repair. It's that difficult to replicate expertise alongside more advanced technology that really helps us deliver competitive advantage.

這些模型都受益於非常龐大、乾淨且多樣化的資料集。我們很幸運能夠在公司內部擁有這三種資源，而不需要從外部尋找。再次強調，來自客觀評估的真實數據、我們自己人員嚴謹的品質保證流程以及為確保正確結果而付出艱苦努力的意願，為我們帶來了非常乾淨、可信且我們相信具有差異化的歷史數據集。根據我的經驗，真正打造品牌的是科技與 Progressive 的優秀人才的結合。為了解決這個問題，我們的資料科學團隊每天都會與物理損壞處理團隊合作，他們的總經驗超過 400 年。他們中的大多數人都曾在我們索賠組織的各個層面工作過。他們自己進行過估算，在車間內和車間旁邊工作過，對車輛維修的各個方面都有深入的了解。複製專業知識和更先進的技術是非常困難的，而這些技術確實能幫助我們獲得競爭優勢。

As one quick example of why the details and subject matter expertise matters, let's consider a bumper. It's not enough to be able to identify a bumper and the existence of damage on it, you need to be able to distinguish between such things as upper and lower bumper covers depending on the year, make and model. And in this case, the presence of bumper sensors.

為了簡單說明為什麼細節和主題專業知識如此重要，讓我們考慮一下保險桿。僅僅能夠識別保險桿及其是否存在損壞是不夠的，您還需要能夠根據年份、品牌和型號區分上保險桿蓋和下保險桿蓋等。在這種情況下，存在保險桿感測器。

At left is a rendering of an example with some of the things our solution does behind the scenes in relative real time. In this case, cropping the bumper from the original image and then making a prediction as to whether it sees the smaller outlines that indicate the presence of sensors in the bumper of a particular car. These can be very hard for even a human eye to detect from a photo. But we must get such things right to ensure we get the accurate part of the estimate in the first place, and account for the incremental labor required as shown in the estimate ample on the right.

左邊是一個範例的渲染，其中展示了我們的解決方案在幕後相對即時地執行的一些操作。在這種情況下，從原始影像中裁剪保險槓，然後預測是否能看到較小的輪廓，這些輪廓表明特定汽車的保險桿中存在感測器。即使是人眼也很難從照片中發現這些。但我們必須正確處理這些事情，以確保我們首先獲得估算的準確部分，並考慮到右側估算中所示的所需的增量勞動力。

Of course, while machine vision is a very interesting technology, we are not in the business of winning data science competitions. We are in the business of insurance, so it's important we prove the value. The orange line here represents the index value of traditional in-person inspections completed per day, the same value as before. You can see the light blue line as we introduced and then scaled photo estimating get into that original 2.5x increase in productivity that I mentioned previously.

當然，雖然機器視覺是一項非常有趣的技術，但我們的業務並不是贏得資料科學競賽。我們從事保險業務，因此證明其價值非常重要。這裡的橙色線代表每天完成的傳統現場檢查的指數值，與之前的值相同。您可以看到淺藍色的線，因為我們引入了縮放照片估算，所以生產力得到了我之前提到的 2.5 倍的提高。

We began this journey in 2019, well before the current hype cycle, as an R&D effort with a small group of our photo estimating representatives. We continue to iterate and refine our models and approach until such time that we were convinced one of the efficiency gains and two, that we were able to equal or better accuracy results.

我們在 2019 年就開始了這一旅程，遠早於當前的炒作週期，當時我們與一小群照片估算代表一起進行了一項研發工作。我們不斷地迭代和改進我們的模型和方法，直到我們確信效率得到了提高，並且我們能夠獲得相同或更好的準確度結果。

Now the darker blue line shows our estimates per day when enhanced by machine vision. In 2022, we began expanding these capabilities throughout our photo estimating organization and fully rolled them out in 2023, having achieved an incremental 2x increase in productivity. To put that in some context, without these advances, we would have needed a 200 additional staff in 2024. While it has been a multiyear journey, the combination of data science and other automation techniques have now doubled the productivity of what, as I mentioned, was already our most efficient means of vehicle inspection and estimating. 100% of our photo estimates are now initially drafted by these solutions and then validated and/or corrected by our estimating professionals.

現在，深藍色的線顯示了通過機器視覺增強後我們每天的估算值。2022 年，我們開始在整個照片估算組織中擴展這些功能，並於 2023 年全面推出，實現了生產力的 2 倍增量提升。具體來說，如果沒有這些進步，到 2024 年我們將需要增加 200 名員工。雖然這是一個多年的歷程，但數據科學和其他自動化技術的結合現在已經使我們最有效的車輛檢查和估算方法的生產率提高了一倍。現在，我們 100% 的照片估算都是透過這些解決方案初步起草的，然後由我們的估算專業人員進行驗證和/或更正。

To be clear, this is not full automation or straight through processing that you may see referenced across the industry. We have people involved. We do think full automation is possible at some point, but we won't do that until we're sure we can do it consistent with our guiding principles and, of course, applicable regulations. For now, we're quite happy to continue empowering our people to be even more and more productive. And we continue to see gains with nothing suggesting to us, we've even hit us feeling yet.

需要明確的是，這並不是您可能在整個行業中看到的完全自動化或直通式處理。我們有人參與。我們確實認為完全自動化在某個時候是可能的，但我們不會這樣做，直到我們確信我們能夠按照我們的指導原則和適用法規做到這一點。目前，我們很高興能夠持續幫助我們的員工提高工作效率。我們繼續看到收益，但沒有任何跡象表明，我們甚至還沒有感受到收益。

I'll cover just two examples on how we're continuing to further refine our solution that we think will generate incremental gains in accuracy and efficiency beyond even what we've already achieved. Beyond our people's efficiency, we also work to make our model training more efficient. At our scale, we must keep up with a very diverse set of vehicles and an OEM market that is always moving. One of the ways we do that is with a semi-supervised training approach called pseudo-labeling. This is the practice of training a model on a small set of label data to label the rest of the data. Traditional machine vision use cases rely on a very large set of manually labeled photos, as I mentioned. But as we've scaled our solution, we sought to improve the speed at which we can deliver value across a broader and ever-evolving list of car parts.

我將只介紹兩個例子，說明我們如何繼續進一步完善我們的解決方案，我們認為這將在準確性和效率方面產生漸進式的提升，甚至超越我們已經取得的成就。除了提高員工的效率之外，我們也致力於提高模型訓練的效率。在我們的規模下，我們必須跟上多樣化的車輛和不斷變化的 OEM 市場。我們實現這一目標的方法之一是採用一種稱為偽標籤的半監督訓練方法。這是在一小組標籤資料上訓練模型以標記其餘資料的做法。正如我所提到的，傳統的機器視覺用例依賴大量手動標記的照片。但隨著我們擴大解決方案的規模，我們力求提高在更廣泛且不斷發展的汽車零件中提供價值的速度。

For context, and while it certainly varies by manufacturer, there are an average around 125 distinct external parts on most common vehicle types. Our process starts with our original segmentation model capable of identifying major parts such as doors and quarter panels. From this base model trained on subject matter experts labels, pseudo labels are generated on a much larger data set to create an initial base model that's even better at those major parts. This improved base model is then fine-tuned with even more detailed part labels from a diverse set of vehicles. This allows us to get more details about very specific parts in the vehicle, like portions of the grill and emblems that can all vary in shape, location or even existence depending on the specific vehicle type. This new part segmentation model is a general template model that works reasonably well on all vehicles.

就背景而言，雖然因製造商而異，但最常見的車輛類型平均有大約 125 個不同的外部零件。我們的流程從原始分割模型開始，該模型能夠識別門和四分之一面板等主要部件。從這個在主題專家標籤上訓練的基礎模型開始，在更大的資料集上產生偽標籤，以創建一個在主要部分錶現更好的初始基礎模型。然後使用來自不同車輛的更詳細的零件標籤對這個改進的基礎模型進行微調。這使我們能夠獲得有關車輛中特定部件的更多詳細信息，例如格柵和徽標的各個部分，它們的形狀、位置甚至存在都可能因具體車輛類型的不同而有所不同。這種新的零件分割模型是一種通用模板模型，適用於所有車輛。

But to ensure the most accurate outcome, we customize the model further for specific vehicle makes and models. With only around 100 detailed label examples, we can apply these vehicle-specific pseudo labels to hundreds of thousands more examples of that vehicle using a bootstrap technique to randomly sample additional internal data.

但為了確保最準確的結果，我們針對特定的車輛品牌和型號進一步客製化模型。只需大約 100 個詳細標籤範例，我們就可以使用引導技術將這些特定於車輛的偽標籤應用於該車輛的數十萬個範例，以隨機抽取額外的內部資料。

For computational efficiency, we then fine-tune the model with a technique called low-rank adaptation or LORA. LORA allows us to efficiently adapt large models to specific tasks enhancing precision without needing as much training data or training time. The point of all this is shown in the images on the right. While you might look at the top right image and suggests the damage is probably on the bumper or maybe the right front vendor, the accurate view is the damage extends across three parts: the bumper, front fender, and molding, the details of which we achieved by much more fine-grained part segmentation capability. Our approach ensures both broad applicability and precise customization across vehicle reducing the need for extensive label data sets. It has enabled us to achieve a 10x speed improvement in training while providing more accurate identification of all relevant external parts. It's important to recognize that such techniques are not possible without very trustworthy and accurate labels and a large data set upon which to extend them. Our subject matter experts and tens of millions of available photos give us both.

為了提高運算效率，我們使用一種稱為低秩自適應或 LORA 的技術對模型進行微調。LORA 使我們能夠有效地將大型模型應用於特定任務，從而提高精確度，而無需太多的訓練資料或訓練時間。所有這些的要點都顯示在右側的圖像中。雖然您可能會看著右上角的圖像並認為損壞可能在保險桿上或可能是右前方供應商，但準確的看法是損壞延伸到三個部分：保險槓，前擋泥板和裝飾條，我們通過更細粒度的零件分割功能實現了這些細節。我們的方法確保了廣泛的適用性和跨車輛的精確定制，減少了對大量標籤資料集的需求。它使我們能夠實現 10 倍的訓練速度提升，同時更準確地識別所有相關的外部部件。重要的是要認識到，如果沒有非常可靠和準確的標籤以及可以擴展它們的大量數據集，這種技術是不可能的。我們的主題專家和數千萬張可用照片為我們提供了這兩點。

This is obviously a very fast-moving space. We continue to keep our eyes on. This animation illustrates a newer approach called 3D Gaussian Splatting. It allows us to transform 2-dimensional images into dynamic 3-dimensional models. Doing this well requires a lot of photos. But back to those trade-offs of accuracy, efficiency and customer experience, we can't reasonably ask customers to take hundreds of photos. But we do ask them to take a video in addition to around 8 to 12 photos on average. A one-minute video at 30 frames per second of the car yields us around 1,800 frames, and it's that data we are increasingly leveraging as part of our solution.

這顯然是一個快速發展的領域。我們持續關注。該動畫展示了一種稱為 3D Gaussian Splatting 的新方法。它使我們能夠將二維圖像轉換為動態三維模型。要做好這件事需要大量的照片。但回到準確性、效率和客戶體驗之間的權衡，我們不能合理地要求客戶拍攝數百張照片。但我們確實要求他們除了平均拍攝 8 到 12 張照片外，還要拍攝一段影片。一段長達一分鐘、每秒 30 幀的汽車影片可以為我們帶來大約 1,800 幀的數據，而這些數據正是我們解決方案的一部分。

Some research scientists have suggested that what large language models are to text, Gaussian Splatting is to graphics. The latter is what's interesting to us. Being able to reconstruct the vehicle as if we were in person to understand the full extent of damage. 3D Gaussian Splatting builds on previous work in the field of reconstruction such as neural radiance field, but it's much more computationally efficient and as a solution that meets our operational needs. Remember, we must be efficient and accurate.

一些研究科學家認為，大型語言模型之於文本，就像高斯分佈之於圖形一樣。後者才是我們感興趣的。我們能夠像親臨現場一樣重建車輛，以了解損壞的全部程度。3D Gaussian Splatting 建立在神經輻射場等重建領域的先前工作的基礎上，但它的計算效率更高，並且是一種滿足我們操作需求的解決方案。記住，我們必須有效率、準確。

While we won't walk through the math here, there are four primary steps to this technique. First, we use the underlying data, again, thousands of frames to understand the position of the Cara and construct a 3-dimensional coordinate grid to understand the relative position of every pixel in the frame. In parallel with understanding camera position, algorithmic techniques such as motion create a point cloud or mesh of the vehicle. This essentially gives us the center point or mean upon which to locate the gaussian splats.

雖然我們不會在這裡詳細討論數學，但這項技術主要有四個步驟。首先，我們使用底層資料（同樣是數千幀）來了解 Cara 的位置，並建立一個三維座標網格來了解幀中每個像素的相對位置。在了解攝影機位置的同時，運動等演算法技術可以創建車輛的點雲或網格。這實際上為我們提供了定位高斯條紋的中心點或平均值。

On those points, we gaussian functions onto each pixel to simulate the physical characteristics of the vehicle surface. It can capture intricate details, providing a more comprehensive reconstruction of the vehicle surface and geometry. Those 3-dimensional splits or blobs help us render the point cloud, representing every point as a gaussian or normal distribution in 3-dimensional space. From there, we optimize using traditional graphic rasterization techniques and adjusting the density of the gaussians until we get results representative of the original training data.

在這些點上，我們將高斯函數應用到每個像素上，以模擬車輛表面的物理特性。它可以捕捉複雜的細節，提供更全面的車輛表面和幾何重建。這些三維分割或斑點幫助我們渲染點雲，將每個點表示為三維空間中的高斯分佈或常態分佈。從那裡，我們使用傳統的圖形光柵化技術進行最佳化並調整高斯的密度，直到我們得到代表原始訓練資料的結果。

What's ultimately produced is a fully interactive 3D model of the vehicle. This is important because it allows you to view the vehicle and damaged regions from novel angles and views that may not have been fully captured in the original data and also provide for more accurate scale and distance measurement enabled by full understanding the relative camera angles. Such approaches can enable more accurate labor hour predictions where the service and depth are very relevant to how long it will take to repair a part. Longer term, we think it could enable virtual inspections of additional damage types. Vehicle conditioning in the case of total loss and perhaps enable LEAP decisions by anticipating damage to interior parts that aren't visible from external pictures alone because we have a more detailed understanding damage depth.

最終產生的是車輛的完全互動式 3D 模型。這很重要，因為它允許您從原始資料中可能未完全捕獲的新角度和視圖查看車輛和受損區域，並且還可以透過充分了解相對攝影機角度提供更準確的比例和距離測量。當服務和深度與修理零件所需的時間密切相關時，這種方法可以實現更準確的工時預測。從長遠來看，我們認為它可以實現對其他損壞類型的虛擬檢查。在車輛完全損毀的情況下進行車輛調節，也許可以透過預測僅從外部圖片無法看到的內裝零件損壞情況來做出 LEAP 決策，因為我們對損壞深度有了更詳細的了解。

Lastly, I do want to reiterate that at Progressive, this isn't about cool science or gaussian math for its own sake. It's about our excellence core view striving for continuous improvement and better business outcomes, for our customers, employees and investors. And as John mentioned previously, we are willing to dig deep and work hard to achieve those outcomes consistent with our core values.

最後，我確實想重申，在 Progressive，這並不是為了酷炫的科學或高斯數學本身。它是關於我們卓越的核心觀點，為我們的客戶、員工和投資者不斷改進並取得更好的業務成果。正如約翰之前提到的，我們願意深入挖掘並努力實現與我們的核心價值一致的成果。

Before wrapping up, I wanted to share our general approach when we consider build and buy decisions. First, we think those are complementary and not competitive sides of the same coin. For example, our vision solution for photo estimating is powered by things we have built, but also relies on very tight integrations with our EX platform partner to turn those predictions into the correct part numbers and actual estimates. When building or buying, we prefer to keep things decoupled. Rather than monolithic solutions, good for one purpose, but perhaps difficult to change or modify for other purposes, we prefer to build or buy systems that are open or can be decoupled so that some of the pieces and parts can be reassembled and is something entirely new.

在結束之前，我想分享我們在考慮建造和購買決策時的一般方法。首先，我們認為它們是同一枚硬幣的互補面，而不是競爭面。例如，我們用於照片估算的視覺解決方案由我們建造的東西提供支持，但也依賴與我們的 EX 平台合作夥伴的緊密整合，以將這些預測轉化為正確的零件編號和實際估算。當建造或購買時，我們傾向於保持事物分離。我們不喜歡單一的解決方案，因為單一的解決方案只適用於一個目的，但可能難以改變或修改用於其他目的，我們更喜歡建造或購買開放或可以解耦的系統，這樣一些部件和零件就可以重新組裝成全新的東西。

This gives us the potential to earn outsized returns on the initial technology investment and provides flexibility when new technology solutions inevitably arise. We can use those decouple pieces to drive near-term value, but it also allows us to be agile in adopting new solutions like pseudo labeling, gaussian splatting and whatever may be next. We can also start to apply parts of the original solution to build something entirely new that we may not have even considered at the time.

這使我們有可能從初始技術投資中獲得超額回報，並在不可避免地出現新技術解決方案時提供靈活性。我們可以使用這些解耦部分來推動近期價值，但它也使我們能夠靈活地採用新的解決方案，例如偽標籤、高斯濺射以及接下來的任何解決方案。我們也可以開始應用原始解決方案的部分內容來建立我們當時可能甚至沒有考慮過的全新事物。

On the right is an example of that potential emerging. When we began a firm commitment to our claims Machine Vision journey with photo estimating back in 2019, we were focused on photos that were submitted after first notice of loss from customers that had selected photo estimating as their vehicle inspection option. But as we've iterated on that experience or funnel, we're now receiving hundreds of thousands of photos per year at first notice of loss, and that is an accelerating rate as we optimize. This is before a customer may have decided to opt in for photo estimating.

右邊就是這種潛力正在顯現的例子。當我們於 2019 年開始堅定致力於透過照片估算進行索賠的機器視覺之旅時，我們專注於選擇照片估算作為車輛檢查選項的客戶在首次收到損失通知後提交的照片。但隨著我們不斷重複這一經驗或管道，我們現在每年在第一次收到丟失通知時都會收到數十萬張照片，而且隨著我們不斷優化，這一速度還在不斷加快。這是在客戶決定選擇照片估價之前。

We can and do use those for photo estimating, but we can also apply our existing machine vision capabilities in that initial triage and segmentation decision and not just photo estimating. It could allow us to make even more fine grand triage decisions based on a much more accurate damage estimate. We can also use those photos to enable more sophisticated review processes of estimates that weren't written through the photo estimating channel.

我們可以而且確實將它們用於照片評估，但我們也可以將我們現有的機器視覺功能應用於初步的分類和分割決策，而不僅僅是照片評估。它可以讓我們根據更準確的損害估計做出更精細的分類決策。我們也可以使用這些照片來對未透過照片估價管道撰寫的估價進行更複雜的審查流程。

We hope a few things came through today as we conclude. First, I'll remind again that the 2024 company-wide results John shared and our efforts in claims to help enable competitive pricing starts with our people and culture. They and it are what makes such things possible. Secondly, we remain tirelessly committed and energized to chase that ever-elusive perfect cost curve balance while simultaneously investing strategically to shift it further downward, and we see further opportunities ahead to continue doing just that. And as shared in the photo estimating example, we continue to exploit advanced technology by putting it in the hands of our people because we know they will generate outcomes fully aligned with our Progressive core value and claims guiding principles and keep those virtuous cycles turning.

我們希望今天結束時能夠取得一些進展。首先，我要再次提醒，約翰分享的 2024 年全公司業績以及我們在幫助實現競爭性定價方面的努力都始於我們的員工和文化。它們和它使得這些事情成為可能。其次，我們將繼續不懈地致力於並全力以赴地追求難以捉摸的完美成本曲線平衡，同時進行策略性投資以進一步降低成本曲線，並且我們看到未來還有更多機會繼續這樣做。正如照片估算範例中所分享的，我們繼續利用先進技術，將其交到我們的員工手中，因為我們知道他們將產生完全符合我們進步核心價值和主張指導原則的成果，並保持這些良性循環的運作。

On behalf of John and I and 65,000-plus Progressive teammates, thank you for your attention.

我代表約翰、我以及 65,000 多名進步隊友感謝您的關注。

This concludes the previously recorded portion of today's event. We now have members of our management team available live to answer questions, including presenters, John Murphy and Matt White, who can answer questions about the presentation. (Event Instructions)

今天活動的先前錄製部分到此結束。現在，我們的管理團隊成員可以現場回答問題，包括演講者 John Murphy 和 Matt White，他們可以回答有關演示的問題。（活動須知）

Michael Zaremski, BMO.

邁克爾·扎雷姆斯基，BMO。

First question, in John Murphy's prepared remarks, and he set up perhaps having even better 2025. I guess -- good to hear that, but it surprising, given that the tailwind from pricing increases has looks like it's meaningfully moderated. I don't know if you want to comment on sorting is roughly, but it looks like low single digits at best. So I thought it was fair to assume the combined ratio mean reverts here and also the frequency tailwind you guys received as well in the industry in 2024. It seems like that was much better than a normal year.

第一個問題，在約翰墨菲準備好的發言中，他提出也許 2025 年會更好。我想——聽到這個消息很高興，但也令人驚訝，因為價格上漲帶來的順風似乎已經明顯減弱了。我不知道您是否想對排序進行粗略評論，但它看起來最多只是一個數字。因此，我認為可以公平地假設這裡的綜合比率平均值會回歸，而且你們在 2024 年也會在行業中獲得頻率順風。看起來這比正常年份好得多。

Yes. I mean I can comment overall on pricing, if you want to add anything, Pat, you can. I feel like we are in a really great position pricing-wise. And we're in more of the position that we have wanted to be in the last several years when I've talked frequently about small bites of the apple. So we've taken rates up slightly in a handful of states. We've taken rates down slightly in a handful of states. And that's really where we want to be to be able to allow our product managers to tweak as needed to make sure that we reach our target profit margins. And then we obviously want to grow and grow as fast as we can. So we always are balancing those. So we feel really good about our pricing, and we'll continue to watch it very closely, and we believe -- and we believe we can grow more that we want to let out some of that in terms of rate decreases like we did in Florida will do so. But we feel like we are in a really, really good position. And starting 2025, I think, just in a position of strength with all that we have done. And I have to take this moment to thank the 66,000-plus Progressive people of what we accomplished last year. It was a volatile year before and as we headed into 2024, but the fact. And John referenced it, the fact that we grew premium 21% to nearly $75 billion, added 18% in PIFs, 5.3 million additional PIFs, and an [AV88] is really nothing short of phenomenal. But more importantly, we did it with great results of our culture and engagement scores. And -- all those together are really difficult. It really takes a team in unison, reaching for the same thing as everything is based on our core values.

是的。我的意思是我可以對定價進行總體評論，如果你想添加任何內容，帕特，你可以添加。我覺得從定價角度來看，我們處於非常有利的地位。我們現在所處的位置更符合我們過去幾年所希望的位置，當時我經常談論小規模的行動。因此，我們在少數幾個州略微提高了利率。我們已在少數幾個州略微降低了利率。這正是我們想要的，以便我們的產品經理能夠根據需要進行調整，以確保我們達到目標利潤率。然後我們顯然希望盡快成長。所以我們一直在平衡這些。因此，我們對我們的定價感到非常滿意，我們將繼續密切關注，我們相信 - 我們相信我們可以實現更多增長，我們希望通過降低利率來釋放其中的一部分，就像我們在佛羅裡達州所做的那樣。但我們感覺我們處於一個非常非常好的位置。我認為，從 2025 年開始，憑藉我們所做的一切，我們將處於強勢地位。我要藉此機會感謝 66,000 多名進步人士，感謝他們去年為我們所取得的成就。2024 年即將到來，而今年也是動盪的一年，但事實是。約翰提到，我們的保費成長了 21%，達到近 750 億美元，PIF 增加了 18%，增加了 530 萬份 PIF，[AV88] 這真是太了不起了。但更重要的是，我們在文化和參與度方面取得了優異的成績。而且——所有這些加在一起真的很難。這確實需要一個團隊齊心協力，為同一個目標而努力，因為一切都基於我們的核心價值。

Thanks, Mike. Mike, also, I wanted to comment, I appreciated your in-depth Florida piece, that was good reading. Do you have anything?

謝謝，麥克。麥克，另外，我想評論一下，我很欣賞你對佛羅裡達的深入分析，讀起來很有意思。你有什麼嗎？

Can I have one quick follow-up?

我可以快速跟進嗎？

Sure, can.

當然可以。

On policy with seasonality, historically, this time of the year, exhibit faster growth. Any comments on whether that would or wouldn't be the case in '25 just given I feel like the dynamics of the cycle is they're always a bit different?

就政策的季節性而言，從歷史上看，每年的這個時候都會表現出更快的成長。鑑於我覺得週期的動態總是有點不同，對於 25 年是否會出現這種情況，您有何評論？

I think typically, first quarter is higher. You've got tax payments dropping, so people shop. You saw the 18% PIF increase in our January results. So we feel good about our growth -- our intended growth in the first quarter. Spent a lot on advertising last year, especially in the second half. And a lot of that was because we saw an opportunity to gain share, especially when our competitors did not quite priced in -- we had more quickly, I think, than others. And so -- basically, we'll have to let that play out. We're going to grow as fast as we can and do so to get to our target profit margins. But yes, I would say typically, first quarter is a pretty big shopping season, especially in the private passenger auto part of the business.

我認為通常情況下，第一季會更高。稅收減少了，所以人們開始購物。您可以在我們一月份的業績中看到 PIF 成長了 18%。因此，我們對我們的成長——第一季的預期成長——感到滿意。去年在廣告上花了很多，尤其是下半年。這在很大程度上是因為我們看到了獲得市場份額的機會，特別是當我們的競爭對手還沒有完全考慮到這一點時——我認為，我們比其他公司更快地獲得了市場份額。所以 — — 基本上，我們必須讓它發揮作用。我們將盡可能快速地發展，以達到我們的目標利潤率。但是的，我想說通常情況下，第一季是一個非常大的購物季，尤其是在私人乘用車業務方面。

David Motemaden, Evercore ISI.

大衛·莫特馬登，Evercore ISI。

Tricia, I just wanted to talk a little bit about your appetite for continuing to ramp the advertising spend. And I think in August, you had noted that the cost per sale was about -- I think it was like 25% below your targeted acquisition costs. Wondering where that is today? And what sort of runway you think you can continue to increase ad spend to capitalize on the growth opportunity?

特里西婭，我只是想談談你對繼續增加廣告支出的興趣。我認為在八月份，您已經注意到每次銷售的成本大約是 - 我認為它比您的目標收購成本低 25％。想知道今天那裡在哪裡嗎？您認為可以透過什麼樣的方式繼續增加廣告支出以利用成長機會？

Yes. Our CPS and TAC are much closer than that now because of what we have spent. And we also did invest in Q4 and some delayed response ads. So we did something that we called Progressive's overnight. And so those are things that we think about that really increase our brand and our presence in terms of hopefully engine. So those are a little delayed response. So we're a little bit tighter now.

是的。由於我們投入的資金，我們的 CPS 和 TAC 現在比那時更接近了。我們確實也投資了第四季度的一些延遲響應廣告。因此，我們連夜做了一些我們稱之為「進步」的事情。因此，這些都是我們所考慮的事情，希望能夠真正提升我們的品牌和在引擎方面的影響力。所以這些反應有點延遲。所以我們現在的關係更緊密了。

We were actually talking about this before the call. We look at our budget for media at the beginning of the year, we work with Pat's team on upfront buys and things that we know we need to put into the system that we have to be very thoughtful about for the year. And then we have a lot of ability to flex both ways. And you've seen it both ways, and we butt it up against our 96, and you saw it last year when we were able to spend more. So that's how we're going to look at it this year. We're going to continue to spend to the efficiency. We obviously want to keep CPS below TAC. It's a little bit closer now. And a lot of it depends on what competitors do. And if they're -- in the auctions, if they're spending more, they're spending less, can react really quickly to all of those things like we have in the past.

我們實際上在通話之前就討論過這個問題。我們在年初查看媒體預算，與 Pat 的團隊合作進行前期購買以及我們知道需要納入系統的事情，我們必須對今年的事情深思熟慮。然後我們就有很大的能力來雙向靈活變通。你已經看到了這兩種情況，我們與 96 競爭，去年你也看到了，我們能夠花更多的錢。這就是我們今年看待這個問題的方式。我們將繼續投入資金以提高效率。我們顯然希望將 CPS 保持在 TAC 之下。現在距離更近了一點。這很大程度上取決於競爭對手的做法。如果他們在拍賣中花費更多，他們會花費更少，他們可以像過去一樣對所有這些事情做出快速反應。

Got it. That's helpful. And then for my follow-up, I noticed the policy life expectancy continued to tick down this quarter sort of has been moderating a bit. But I think if I look at the comps, they're pretty high comps a year ago. Could you just talk through how you guys are feeling on the retention side of things because it feels like the new business is definitely robust, but some of the retention numbers have been moderating a bit.

知道了。這很有幫助。然後，在我的後續行動中，我注意到本季的保單預期壽命持續下降，並且有所緩和。但我認為，如果我看一下比較結果，我會發現它們與一年前相比相當高。您能否談談您對保留方面的看法，因為感覺新業務確實很強勁，但一些保留數字有所下降。

Yes. You're spot on with that. And obviously retention is such a big piece of our growth and really sort of the holy grail. We're not surprised because of the rate increase we've taken over the years that retention has dipped. We're hopeful that turn it started to turn on the trailing 30 commercial auto. So yes, we're -- it's never great to have our PLEs go down, either on a T3 or T12. We're working diligently, and you heard about that from John Murphy in terms of just making sure that we have this near-perfect balance of our claims guiding principles. And what we know is when people have a claim, they're more likely to stay because of the service we give them, we look at it from an NPS perspective.

是的。你說得完全正確。顯然，留住客戶是我們業務成長中非常重要的一部分，也是我們真正的目標。我們並不感到驚訝，因為多年來我們的費率一直在增加，但保留率卻有所下降。我們希望它能對後面 30 輛商用車產生影響。所以是的，我們的 PLE 故障絕對不是什麼好事，無論是在 T3 還是 T12 上。我們正在努力工作，正如約翰墨菲所說，我們只是確保我們的索賠指導原則達到近乎完美的平衡。我們知道，當人們提出要求時，他們更有可能因為我們提供的服務而留下來，我們從 NPS 的角度來看這一點。

So what I would say is, we're going to work diligently on turning that around. I think one of the biggest things you can do is have stable rates for your customers. With all the things that have happened from an inflationary perspective, people are shopping. It's easy to shop, and there's a lot of shopping going on still. So if we can keep these prices stable, I think that is really a key when you get that renewal rate and you don't have to shop because it's stable or, in some cases, even going down. So that's really where we're at now. We obviously want that to improve our overall PIF growth.

所以我想說的是，我們將努力扭轉這種局面。我認為您能做的最重要的事情之一就是為您的客戶提供穩定的費率。從通貨膨脹的角度來看，隨著發生的所有事情，人們正在購物。購物很方便，而且仍有許多購物活動。因此，如果我們能夠保持這些價格穩定，我認為這真的是獲得續訂率的關鍵，而且您不必購物，因為它是穩定的，或者在某些情況下甚至會下降。這就是我們現在的處境。我們顯然希望這能改善我們的整體 PIF 成長。

Elyse Greenspan, Wells Fargo.

富國銀行的伊莉絲‧格林斯潘。

So it's been very topical just to talk about like the tariffs, and I think your call is obviously lined out at kind of going through a today. So have you guys sized the potential impact on margins? It seems like it's going to be more of a second half story. But I guess, how are you kind of balancing growing as fast as you can with something that's going to potentially go cash cost up in the second half? I guess where do you find the balance between the two?

因此，談論關稅等話題非常具有現實意義，我認為您的呼籲顯然與今天的討論有關。那麼你們是否已經評估過對利潤的潛在影響？這看起來更像是一個後半段的故事。但我想，您如何在盡可能快速的成長與下半年現金成本可能上升之間取得平衡？我猜你在哪裡找到兩者之間的平衡點？

Yes, super relevant and timely question. And clearly, typically tariffs are a one-sided risk to our loss cost. So yes, we've been thinking about that a lot. And in fact, our pricing team is working with our economics team at PCM, and they have been for a while to understand the implication of tariffs. And so we have a lot of -- I'm not going to share with you the raw data or the assumption we use, but I will tell you, about three weeks ago, I was reading through some of the materials they're putting out and it's evolving every day. And I called Pat to say what incredible work these two groups were doing and just really diligent. So yes, we actually have what we think at this time, percentages would be on certain -- if certain tariffs happen. Now those will ultimately change, and we'll be able to kind of flex our models.

是的，非常相關且及時的問題。顯然，關稅通常會為我們的損失成本帶來單方面的風險。是的，我們一直在思考這個問題。事實上，我們的定價團隊正在與 PCM 的經濟團隊合作，他們已經合作了一段時間，以了解關稅的影響。因此我們有很多——我不會與你們分享我們使用的原始數據或假設，但我可以告訴你們，大約三週前，我閱讀了他們發布的一些材料，而且它每天都在不斷發展。我打電話給帕特，告訴他這兩個團隊的工作非常出色，而且非常勤奮。是的，我們目前確實有這樣的想法，如果某些關稅發生，百分比將是確定的。現在這些最終都會改變，我們將能夠靈活調整我們的模型。

It will depend on the countries, the products, the magnitude of severity, but ultimately, we into our indications. There's a lot of, I think, puts into as things have inflow. So with the tariffs that went to effect today, we'll have to think about new car prices. Are those prices -- do those prices get passed on from the OEs to our customers? What does that mean to values? Obviously, from both Mexico and Canada, we get a lot of our parts to repair cars. And so when will those play out? And I do agree with you. I think if things go as planned, I think it will be more second half and into 2026.

這將取決於國家、產品和嚴重程度，但最終，我們會根據情況做出調整。我認為，隨著事物的流入，會有很多投入。因此，隨著今天生效的關稅，我們必須考慮新車價格。這些價格——這些價格是否從 OE 轉嫁給我們的客戶？這對價值觀意味著什麼？顯然，我們從墨西哥和加拿大獲得了許多汽車維修零件。那麼這些什麼時候會發生呢？我確實同意你的看法。我認為，如果一切按計劃進行，那將更多的是下半年和 2026 年。

And then there are some other things we're thinking about that we're modeling and that is if there tariffs on oil, does that increase gas prices, does that make people drive less, does that change frequency, even with -- there a talent shortage in the body shop industry that happened a couple of years ago. This morning, it looks like there might be some additional tariffs on lumber for Canada at a minimum, will that due to a home price? Of course, home repair prices and we've modeled that in as well as what I talked about with fixing cars. So we're modeling all of that together, and a lot of it will depend on how much inventory is out there.

然後，我們正在考慮並建模其他一些事情，那就是如果對石油徵收關稅，這是否會提高汽油價格，是否會讓人們減少駕駛，是否會改變駕駛頻率，即使幾年前汽車修理廠行業出現了人才短缺。今天早上，看起來至少可能會對加拿大的木材徵收一些額外的關稅，這會因為房價嗎？當然，我們已經對房屋維修價格以及我談到的汽車維修價格進行了建模。因此，我們將所有這些因素綜合起來進行建模，其中很大一部分取決於有多少庫存。

And then if you think about the future, I'm always trying to think about second and third order effects. If you think about the future, and I was watching CNBC this morning, I thought, well, if we want to have more of our lumber that starts here in the United States, then we're going to need to build more sawmills. Those are going to -- that can't happen overnight, but what does that do to trucking and laggers. And so it could be a tailwind in terms of our Commercial Lines organization because that's very much based on macroeconomic data. But here's where we're at. We have a bunch of models. We were able to flex those models every time we get a new piece of data or the data changes. David and his team are reading through all the executive orders and understanding how that affects us.

如果你考慮未來，我總是試著考慮二階和三階效應。如果你考慮未來，我今天早上看了 CNBC，我想，如果我們想要更多的木材產自美國，那麼我們就需要建造更多的鋸木廠。這些都不會在一夜之間發生，但這會對卡車運輸和滯後運輸產生什麼影響呢？因此，對於我們的商業線路組織而言，這可能是一個順風，因為這很大程度上是基於宏觀經濟數據。但我們現在的處境就是這樣。我們有很多模型。每次我們獲得新的數據或數據變更時，我們都能靈活調整這些模型。大衛和他的團隊正在閱讀所有行政命令並了解它們對我們的影響。

And there's disruption in pricing, we are really good to react really quickly and you've seen that. And probably the last thing I'll say on this is that we are sitting in a good position because right now, our margins are below our 96. So we're sitting on some margins, so we can kind of see this out as things evolve.

當定價出現混亂時，我們能夠迅速做出反應，這一點您已經看到了。我最後想說的是，我們處於有利地位，因為目前我們的利潤率低於 96。因此，我們處於某種邊緣位置，因此我們可以隨著事態的發展看到這一點。

Great. Thank you for the holistic response. And then for my follow-up for the ad spend. So -- it kind of sounds like it will be maybe not to the same absolute dollar amount, but it sounds like the seasonality on the because you typically spend a little bit more in the first half. It sounds like it will continue to be kind of like spread out as we go through the year. And I understand there's some uncertainty with the tariffs so that could change drastically. But is it kind of right to think about in terms of the seasonality and ad spends seasonality kind of continue to be huge just given the higher customer shopping?

偉大的。感謝您的全面回覆。然後我會跟進廣告支出。所以——聽起來可能不是相同的絕對金額，但這聽起來像是季節性的，因為你通常在上半年會花多一點的錢。聽起來，隨著​​時間的流逝，這種趨勢仍會繼續蔓延。我知道關稅存在一些不確定性，因此可能會發生巨大變化。但是，考慮到顧客購物量增加，從季節性和廣告支出季節性持續成長的角度來考慮是否正確？

I mean, yes, for the most part, we will spend when people are shopping and people have typically the shops in the quarter. But we're going to be really flexible to see what plays out. A lot of it isn't just what we're doing, it's what the competitors are doing. So like last year, we had a budget and then we increased it based on the opportunity to gain share when we could. So we -- that's a great part about the flexibility. We don't have a budget where we say we have to spend this much or it's gone the next year. It's flowing. It's -- and it changes depending on the needs of our growth and watching what the competitors are doing and watching what's happening in pricing. Overall, for both media spend as well as just overall pricing. So we're flexible is what I'd say as we talk about this all the time, what we need and why we need it and we're going to try to reach our main objective in that is grow as fast as we can at or below a 96.

我的意思是，是的，大多數情況下，我們會在人們購物時消費，人們通常會在街區裡的商店裡消費。但我們會非常靈活地觀察結果。很多事情不只是我們在做的，也是競爭對手在做的。因此，就像去年一樣，我們有預算，然後我們根據獲得份額的機會增加了預算。所以，我們——這是靈活性的很大一部分。我們沒有預算，沒有規定說必須花這麼多，否則明年就花光了。它在流動。它會根據我們成長的需求、關注競爭對手的行為以及關注定價的變化而變化。整體而言，這既包括媒體支出，也包括整體定價。因此，我想說的是，我們總是在談論這個問題，我們需要什麼以及為什麼需要它，我們將努力實現我們的主要目標，即在 96 或以下盡可能快地增長。

Joshua Shanker, Bank of America.

美國銀行的 Joshua Shanker。

I'm going to ask a similar question maybe in a different way. If I look at the ad spending, it seems like you spent more in 4Q than you spent in 3Q. And the PIF growth slowed down in November a little bit and then much in December as we expect it's normal. That's the seasonality of it. But -- you talked about this, I guess, delayed response advertising in some ways. How does it work exactly? And if we look at how should we compare ad spend to PIF growth? What's the relationship there?

我將以不同的方式提出類似的問題。如果我看一下廣告支出，似乎您在第四季度的支出比第三季的支出要多。正如我們預期的那樣，PIF 成長在 11 月略有放緩，12 月則大幅放緩，這是正常的。這就是它的季節性。但是——我想，您談到了這一點，在某種程度上延遲了響應廣告。它究竟是如何運作的？那麼，我們該如何比較廣告支出和 PIF 成長呢？這裡面有什麼關係？

Well, we look at -- when we spend, we look at a measure -- take the delayed response that. We look at a measure called MP6 -- the new prospects that are shopping new in the last six months. And we look at MP6 and how that increases. So we know people are shopping, and we know they're active. So that's how we look at that. And of course, we look at our targeted acquisition cost and our cost per sale. So we look at all that together, and that's for the immediate response.

好吧，我們看一下——當我們花錢時，我們會看一個措施——採取延遲回應。我們檢視了一項稱為 MP6 的指標——過去六個月內新購物的潛在客戶。讓我們看看 MP6 以及它是如何增加的。所以我們知道人們正在購物，我們知道他們很活躍。這就是我們對此的看法。當然，我們也會考慮目標收購成本和每次銷售的成本。因此，我們將所有這些放在一起考慮，並立即做出反應。

For the delayed response, and this is something slightly new. And John Murphy talked about it during his slides, we have a new purpose statement that we developed a few years ago. And we started what we call our Anthem and that is progresses overnight, and we spent some money on that in the fourth quarter, which we it's going to be harder to measure and so we have some different measurements in the longer term. I'd probably be able to share this with you once we have them because they'll play out. But that's how we think about those two differently. Do you want to add anything, Pat, are you -- Thanks, Josh.

對於延遲響應，這是一個稍微新鮮的事情。約翰·墨菲在他的幻燈片中談到了這一點，我們有一份幾年前製定的新宗旨聲明。我們開始了所謂的 Anthem 項目，這個項目在一夜之間取得了進展，我們在第四季度為此投入了一些資金，這將更難衡量，因此從長遠來看，我們會有一些不同的衡量標準。一旦我們有了它們，我可能就能與你們分享，因為它們會發揮作用。但這就是我們對這兩者的不同看法。帕特，你想補充什麼嗎？謝謝喬希。

Then on claims, is there any evidence or numbers you give that about cost per Claim that the efficiencies you're building in are delivering a cheaper cost per Claim than it would have been a year earlier under the same circumstances? I think back 15 years ago when you added the claims centers. This is a goal that the same Claim could be satisfied more cheaply. Are there any statistics or confidence you have around your ability to resolve the same Claim cheaper with the same input costs?

那麼關於索賠，您是否提供任何證據或數字來表明每項索賠的成本，表明您正在建立的效率使每項索賠的成本比一年前在相同情況下更低？我回想起 15 年前您新增索賠中心的時候。目標是以更便宜的方式滿足相同的索賠要求。您是否有任何統計數據或信心表明您能夠以相同的投入成本以更低的成本解決相同的索賠？

Yes. We have a lot of data and care deeply, obviously, about our LAE, but we look at cost per feature for claims. We look at features per day per FTE. So when you're thinking about what goes into all the different Claim features, it's really getting it to the right person at the right time. And if we can do that throughput, that's going to be much more efficient. And as long as accuracy is still great, then that's what we care about.

是的。顯然，我們擁有大量數據，並且非常關心我們的 LAE，但我們會考慮索賠的每個功能的成本。我們查看每個 FTE 每天的功能。因此，當您考慮所有不同的索賠功能時，實際上是在正確的時間將其交給正確的人。如果我們能夠達到這樣的吞吐量，效率就會更高。只要準確度仍然很高，這就是我們所關心的。

So we look at a lot of -- actually a lot of data across the board, my team and I talk a lot about efficiency. And when I think about that, I think about it from a non-acquisition expense ratio. But how can we continue to get efficient because competitive prices is one of our strategic pillars. And so we have to continue to push that cost down. And of course, the curve that John showed, we want to continue to push that down to the left.

因此，我們查看了大量數據——實際上是大量的全面數據，我和我的團隊討論了很多關於效率的問題。當我考慮這個問題時，我會從非收購費用率的角度來考慮。但我們如何繼續保持高效，因為有競爭力的價格是我們的策略支柱之一。因此我們必須繼續降低成本。當然，我們希望繼續將約翰所展示的曲線向左推。

(Event Instructions)

（活動須知）

Andrew Andersen, Jefferies.

安德魯·安德森（Andrew Andersen），傑富瑞集團（Jefferies）。

Could you maybe talk about any changes you're seeing in bundle rates? And perhaps it would be helpful to break it down between in areas where you are growing the property book and areas where you're derisking?

您能否談談您所看到的捆綁價格的任何變化？或許，將其劃分為增加房地產帳簿的領域和降低風險的領域會有所幫助？

Yes, that's a great question. We've been talking a lot about this. And we refer to this in the annual report documents when we talk about our blueprint for the future. And so -- it's a good time, I guess, to step back a little bit. And I'm going to read you our mission statement for property because this has really been evolving and it's really important and all is a big piece of it. So our mission statement is Progressive Home primarily offers products for owner-occupied property bundles with Progressive Auto. So that is going to be where permitted. The main thing we're going to have a -- we will restrict business coming in where we can to have a B home bundle, whether it's condo or home. We will build a diverse portfolio of products with partner agents who understand our underwriting strategy and proactively support it.

是的，這是一個很好的問題。我們已經多次討論過這個問題。當我們談論未來藍圖時，我們在年度報告文件中提到了這一點。所以——我想，現在是稍微退一步的好時機。我將向你們宣讀我們的房地產使命宣言，因為它一直在不斷發展，而且非常重要，而且是其中的重要組成部分。因此，我們的使命宣言是 Progressive Home 主要與 Progressive Auto 一起提供自住房產捆綁產品。所以這將是在允許的情況下。我們要做的最重要的事情是——我們將盡可能限制 B 類住房套餐的業務，無論是公寓還是住宅。我們將與了解我們的核保策略並積極支持的合作代理商一起建立多樣化的產品組合。

When capacity is limited, we use that capacity for bundles written with our partners in the agency channel. So that's the overarching mission statement. So we'll be referring to that over the next few years as we continue to have our blueprint for the future. Of course, first in source, we have to get the rate we need. So we had right into the system in 2024, and that will continue into 2025. Meanwhile, balancing that out with investments we're making in people, processes, IT, segmentation.

當容量有限時，我們會將該容量用於與代理商通路中的合作夥伴一起編寫的捆綁包。這就是總體使命宣言。因此，在未來幾年中，我們將繼續參考這一點，制定未來的藍圖。當然，首先從源頭上，我們必須獲得我們需要的利率。因此，我們在 2024 年就進入了該系統，並將持續到 2025 年。同時，我們在人員、流程、IT、細分方面的投資也得到了平衡。

And then we have the blueprint that talks about bundling, which I just talked about. We want to accept new business with auto, home bundle were permitted. We've talked, I think, a couple of times about exiting the DP3. So we've in 44 states and that is, I think, of rental properties.

然後我們有了關於捆綁的藍圖，我剛才談到了。我們希望接受允許汽車、家庭捆綁銷售的新業務。我想，我們已經討論過幾次退出 DP3 的事情了。所以我認為我們在 44 個州都有租賃物業。

And then really aged alignment, holding our agent partners to volume and quality bundles, and we have very specific expectations for our agents. And then we have to cost share with our insurers. Insurance, home insurance shouldn't be a maintenance on products. So we cost share with mandatory wind inhaled deductibles and then what we call roofing materials payment schedule, think of sort of a sliding scale on roof depreciation as the roof ages.

然後進行真正的年齡調整，讓我們的代理商合作夥伴遵守數量和品質捆綁，我們對我們的代理商有非常具體的期望。然後我們必須與保險公司分攤成本。保險、房屋保險不應該是產品維護。因此，我們透過強制性風吸入免賠額來分攤成本，然後我們稱之為屋頂材料付款計劃，想想隨著屋頂老化，屋頂折舊的滑動比例。

And then, of course, we've talked a lot in the last couple of years about derisking and making sure that our portfolio across the board. So I'm not going to share specifics of bundles. The majority of our homes are bundled with our auto. I will tell you that much. And I think probably another data that I won't share either, but it's a very high percentage. And very much a majority is how many of our rental policies are bundled with auto. So think of what we call future Robinson. So these are people that are already bundled with us. And as they do buy a home, the next step would get to be Progressive home in their auto. So that's a reason to believe of the future that we'll continue to bundle more auto and home, and there's a lot in the pipeline for us as we grow.

當然，過去幾年我們多次談論降低風險和確保我們的投資組合全面合規。所以我不會分享捆綁包的具體細節。我們大多數的房屋都與汽車捆綁在一起。我就告訴你這麼多。我想可能還有其他數據我不會分享，但這個比例非常高。絕大多數情況下，我們的租賃政策都與汽車捆綁在一起。想想我們所說的未來的魯濱遜。這些人已經與我們捆綁在一起了。當他們購買了房屋後，下一步就是在他們的汽車中安裝漸進式住房。因此，我們有理由相信，未來我們將繼續捆綁更多的汽車和家居產品，而且隨著我們的發展，我們還有許多計劃要做。

Maybe just going back to tariffs. I suppose you can't prospectively reflect it in pricing, but the extent to which you start to see it, how many quarters or periods would it have to take before you would kind of be allowed to reflect that in a rate filing and have it be approved?

也許只是回到關稅問題。我想您無法預先在定價中反映它，但是您開始看到它的程度，需要多少季度或時期才能在費率申報中反映它並獲得批准？

Well, it's a little bit different in different states. But as soon as we see the data, we try to put that into our pricing indications, and our -- and that, of course, is state by state. And then we work there's different regulations in different departments. We work with each department to get that in there as soon as possible.

嗯，不同州的情況有點不同。但一旦我們看到數據，我們就會嘗試將其納入我們的定價指示中，當然，這是因州而異的。而且我們工作時不同部門有不同的規定。我們與各部門合作，盡快完成這項工作。

So it could be relatively quickly, I suppose, once there is an impact --

所以我認為，一旦產生影響，它可能會相對較快--

Yes. As soon as we see it in our data, we're able to -- we'll be able to present it to the Department of Insurance to try to get the necessary rate.

是的。一旦我們在數據中看到它，我們就可以 - 我們將能夠將其提交給保險部門以嘗試獲得必要的費率。

That appears to have been our final question. So that concludes our event. Josh, I will hand the call back over to you for the closing scripts.

這似乎是我們的最後一個問題。我們的活動到此結束。喬希，我會把電話轉回給你，讓你負責最後的劇本。

That concludes the Progressive Corporation's Fourth Quarter Investor Event. Information about a replay of the event will be available on the Investor Relations section of Progressive's website for the next year. You may now disconnect.

進步公司第四季投資者活動到此結束。有關該活動重播的資訊將於明年在 Progressive 網站的「投資者關係」部分提供。您現在可以斷開連線。