Upstart Holdings, Inc. (UPST) 2026 Q3 法說會逐字稿

Good morning, and welcome to the Upstart third quarter 2026 earnings call. (Operator Instructions) As a reminder, this conference call is being recorded. I would now like to turn the conference over to Sonya Banerjee, Head of Investor Relations. Sonya, please go ahead.

早安，歡迎參加 Upstart 2026 年第三季財報電話會議。(接線員指示) 提醒各位，本次電話會議將被錄音。現在我想將會議交給投資人關係主管 Sonya Banerjee。Sonya，請開始。

Thank you. Welcome to the Upstart earnings call for the third quarter of 2026. Joining me today are Paul Gu, our Co-Founder and CEO; and Andrea Blankmeyer, our CFO. During today's discussion, we will make forward-looking statements, which include statements about our outlook and business strategy. These statements are based on our expectations and beliefs as of today, which are subject to a variety of risks, uncertainties and assumptions and should not be viewed as a guarantee of future performance.

謝謝。歡迎參加 Upstart 2026 年第三季財報電話會議。今天與我一同出席的有共同創辦人暨執行長 Paul Gu，以及財務長 Andrea Blankmeyer。在今天的討論中，我們將作出前瞻性陳述，其中包括關於我們展望與業務策略的陳述。這些陳述係基於我們截至今日的預期與信念，並受多項風險、不確定性與假設影響，不應被視為對未來表現的保證。

Actual results may differ materially as a result of various risk factors that have been described in our SEC filings. We assume no obligation to update any forward-looking statements as a result of new information or future events, except as required by law. Our discussion will include non-GAAP financial measures which are not a substitute for our GAAP results. Reconciliations of our historical GAAP to non-GAAP results can be found in our earnings materials, which are available on our IR website.

由於我們在向美國證券交易委員會（SEC）提交的文件中所描述的各項風險因素，實際結果可能出現重大差異。除法律要求外，我們不承擔因新資訊或未來事件而更新任何前瞻性陳述的義務。我們的討論將包含非 GAAP 財務衡量指標，該等指標不能替代我們的 GAAP 結果。我們歷史 GAAP 與非 GAAP 結果之間的調節表可在我們的財報資料中找到，相關資料可於我們的投資人關係網站取得。

With that, Paul, over to you.

接下來，Paul，交給你。

Thanks, Sonya. Thank you, everyone, for joining us today. I'd like to use the expression that the growth of our business is equal to alpha plus beta times macro. The idea behind this classic formula is relevant to every business, but especially important for understanding our results and outlook today. Since laying out our 2026 focus areas earlier this year, I couldn't have asked for better execution.

謝謝你，Sonya。也謝謝各位今天加入我們。我想用一句話來表達：我們業務的成長等於 alpha 加上 beta 乘以宏觀環境。這個經典公式背後的概念對所有企業都適用，但對理解我們今天的業績與展望尤其重要。自今年稍早提出我們 2026 年的重點領域以來，我對執行成果再滿意不過。

The team has delivered durable compounding growth from technology and marketing advances at an exceptional rate. At the same time, our business is sensitive to macro conditions, particularly borrower health and interest rates. Those became a sizable headwind in Q3 and have gotten worse to start Q4. Looking back over the past few months, I've landed on three conclusions I want to share with you today. First, I'm proud of the team and our delivery in Q3.

團隊以極高的速度，透過技術與行銷的進步，帶來可持續的複利式成長。同時，我們的業務對宏觀條件相當敏感，特別是借款人健康狀況與利率。這些因素在第三季成為相當大的逆風，且在第四季初期進一步惡化。回顧過去幾個月，我整理出三點結論，今天想與各位分享。第一，我為團隊以及我們在第三季的交付成果感到自豪。

Our results were strong by just about any measure and exceptional when accounting for macro. Second, we expect a mixed finish to the full year relative to our original guidance. We see fee revenue and contribution profit finishing higher, but total revenue and adjusted EBITDA finishing lower. That's because macro has an outsized impact on revenue derived from our balance sheet. Third, I am more confident than ever in Upstart's long-term trajectory.

以幾乎任何衡量標準來看，我們的結果都很強勁；若將宏觀因素納入考量，更是表現出色。第二，相較於我們原先的指引，我們預期全年收官將呈現好壞參半。我們看到手續費收入與貢獻利潤將高於預期，但總營收與調整後 EBITDA 將低於預期。原因在於，宏觀因素對來自我們資產負債表的營收具有不成比例的影響。第三，我對 Upstart 的長期發展軌跡比以往任何時候都更有信心。

That's not just obligatory CEO optimism. Macro comes and goes, but alpha compounds, and our results are proving we can deliver it. Let's talk about each conclusion, starting with Q3 results. Contribution profit, our best single measure of profitable growth in new originations, reached $218 million in Q3. That's up 48% year-over-year, a new record, breaking the one we set just last quarter.

這不只是執行長例行公事式的樂觀。宏觀環境來來去去，但 alpha 會複利累積，而我們的結果正在證明我們能夠交付它。讓我們逐一談談這三點結論，先從第三季業績開始。貢獻利潤——我們衡量新放款中獲利性成長的最佳單一指標——在第三季達到 2.18 億美元。年增 48%，創下新紀錄，打破了我們就在上一季才創下的紀錄。

This result is strong, but it understates the underlying progress in the business. That's because our Q3 numbers, starting with origination volumes and cascading down the P&L, were held back by a sizable macroeconomic headwind. The Upstart Macro Index, or UMI, is our estimate of how macroeconomic conditions affect defaults on our personal loans. It averaged 1.5 in Q3, up from 1.41 in Q2, and we underwrote proportionately more conservatively. The explanation is likely pretty simple.

這個結果很強勁，但它低估了業務的內在進展。原因在於，我們第三季的數字——從放款量開始並一路反映到損益表——都受到顯著的宏觀經濟逆風所拖累。Upstart 宏觀指數（UMI）是我們對宏觀經濟條件如何影響個人貸款違約的估計。第三季 UMI 平均為 1.5，高於第二季的 1.41，因此我們相應地採取了更保守的承作標準。其原因可能相當簡單。

Annual inflation reaccelerated to 3.7% with prices rising twice as fast as wages over the past quarter. Not surprisingly, consumer delinquencies have risen across a range of third-party measures. The share of all Americans at least 30 days late on any credit obligation jumped to over 19% in September, the highest level in years. Without that 9-point increase in UMI, we estimate Q3 originations would have been at least $370 million higher. In that scenario, year-on-year origination growth would have been 57% and contribution profit growth 61%.

年通膨率回升至 3.7%，且在過去一季中，物價上漲速度是薪資的兩倍。不令人意外地，根據多項第三方指標，消費者逾期情況在各面向均有所上升。9 月份，所有美國人中任一信用義務逾期至少 30 天的比例躍升至超過 19%，為多年來最高水準。若沒有 UMI 增加 9 個點的影響，我們估計第三季放款額至少會高出 3.7 億美元。在該情境下，放款額年增率將為 57%，貢獻利潤年增率將為 61%。

We earned that alpha in two places. In core personal loans, we improved the models, user experience and borrower ecosystem. In secured, we improved automation, conversion and monetization in pursuit of profitability. In Q3, we added over 140 new features to our core underwriting model, including numerous proprietary signals of how consumers engage with our platform before they take a loan. Combined with improvements to our algorithms and training process, our model's accuracy lead over a traditional credit scoring benchmark widened again. Our model is now 2.77 times as accurate, and 87.27% of the inaccuracy gap is still ahead of us. That's our runway.

我們的 alpha 主要來自兩個方面。在核心個人貸款方面，我們改善了模型、使用者體驗與借款人生態系。在擔保型產品方面，我們提升自動化、轉換率與變現能力，以追求獲利能力。第三季，我們為核心承作模型新增了超過 140 項功能，其中包括多項專有訊號，用以反映消費者在借款前如何與我們的平台互動。結合我們對演算法與訓練流程的改進，我們模型相較於傳統信用評分基準的準確度領先幅度再次擴大。我們的模型目前準確度為其 2.77 倍，而不準確差距中的 87.27% 仍在我們前方。那就是我們的跑道。

We also use better models to reach more people. Our marketing programs and user experiences drove more than 1 million people to apply for an Upstart loan for the first time this quarter, bringing the cumulative number of unique Americans who have applied for an Upstart loan to more than 25 million, all while lowering our cost of customer acquisition. More than 555,000 unique Americans took out loans in Q3, a new record. And they are more likely than ever to come back. Over 250,000 of them were returning customers who had already borrowed with us before.

我們也運用更好的模型觸及更多人。本季，我們的行銷方案與使用者體驗推動超過 100 萬人首次申請 Upstart 貸款，使累計曾申請 Upstart 貸款的美國獨立個人數超過 2,500 萬，同時還降低了我們的客戶取得成本。第三季有超過 555,000 名美國獨立個人成功借款，創下新紀錄。而且，他們回流的可能性比以往任何時候都更高。其中超過 250,000 人為回頭客，先前已曾向我們借款。

Now to Secured products. We told you we wanted to be contribution margin positive by the end of the year. The team went all in and got it done in Q3. Secured contribution margin was a positive 17%, an improvement of more than 50-percentage-points sequentially.

接下來談擔保型產品。我們曾告訴各位，我們希望在年底前達到貢獻毛利為正。團隊全力以赴，並在第三季就完成了目標。擔保型產品的貢獻毛利為正 17%，較前一季改善超過 50 個百分點。

I'll give a few examples of how we got there. First, we sold more of our secured loans to third-party capital partners and at better terms. Second, we increased the level of automation across our secured products. In home, nearly 40% of funded loans were fully automated with no human intervention by Upstart, up from less than 5% a year ago. In auto-secured personal loans, we significantly increased the fully automated rate over the last year from 4% to 21%. Third, in Auto Retail, we began optimizing dealer compensation around where we create the most differentiated value, and we rebuilt our credit application, increasing end-to-end conversion by 31%.

我舉幾個例子說明我們如何達成。第一，我們將更多擔保貸款出售給第三方資金合作夥伴，且條件更佳。第二，我們提高了擔保型產品的自動化程度。在房屋相關產品中，近 40% 的撥款貸款由 Upstart 在無任何人工介入下完成全自動化，較一年前不到 5% 大幅提升。在汽車擔保個人貸款中，我們在過去一年將全自動化比例從 4% 顯著提高至 21%。第三，在汽車零售（Auto Retail）方面，我們開始圍繞我們能創造最具差異化價值之處來優化經銷商補償，並重建了我們的信用申請流程，使端到端轉換率提升 31%。

These are young products in enormous markets. After locking in our margin gains this year, we'll gradually shift our focus back to growth in 2027. As we do that, we intend to hold secured contribution margin consistently positive between breakeven and 25% quarter-to-quarter as we refine our approach to growth in each category. Longer term, we expect those margins to resume their march upward as we create increasing differentiation and value for our customers, just as they did historically in core personal loans. Our other major focus this year has been capital efficiency.

這些都是處於早期階段、但面向龐大市場的產品。在今年鎖定我們的毛利改善成果之後，我們將在 2027 年逐步把重心轉回成長。在此過程中，隨著我們在各類別中精進成長策略，我們打算讓擔保型產品的貢獻毛利在每季之間穩定維持為正，介於損益兩平到 25% 之間。長期而言，隨著我們為客戶創造更高的差異化與價值，我們預期這些毛利將恢復向上攀升的趨勢，正如其在核心個人貸款業務中歷來所呈現的那樣。我們今年另一個主要重點是資本效率。

Loans held directly on our balance sheet declined to 4.8% of total loans outstanding in Q3. More than 95% of all Upstart-powered loans outstanding are now third-party funded with the bulk coming from committed partnerships. In exchange for commitments of up to two years, we co-invest in sharing the performance of those loan cohorts. Those partnerships are designed to make our funding base resilient through market cycles.

我們資產負債表上直接持有的貸款在第三季降至未償貸款總額的 4.8%。目前，超過 95% 的所有 Upstart 驅動之未償貸款皆由第三方出資，其中大部分來自承諾式合作夥伴關係。作為交換，在最長兩年的承諾期下，我們共同投資並分享這些貸款批次的績效表現。這些合作夥伴關係旨在讓我們的資金來源在市場循環中更具韌性。

Our priority now is reducing the share of equity capital our loan funding requires from us. Over the past year, we grew outstanding loans across the platform by 38% while increasing equity used only 7%. That's a good start, and we intend to keep going. Another project relevant to our capital efficiency is Upstart Bank. We are on track to open Upstart Bank in early 2027, subject to the remaining regulatory approvals.

我們目前的優先事項是降低貸款資金來源對我們自有股權資本的需求占比。過去一年，我們平台上的未償貸款成長 38%，而使用的股權僅增加 7%。這是個良好的開始，我們打算持續推進。另一個與資本效率相關的專案是 Upstart Bank。我們正按計畫在 2027 年初開設 Upstart Bank，但仍須取得剩餘的監管核准。

When that happens, we plan for most of the loans on Upstart's balance sheet today to move to the bank and become deposit funded, which will unlock net cash rather than consume it. As such, we expect capitalizing the bank will require no incremental equity capital. Once open, Upstart Bank will provide the many benefits we've previously shared, expanding our market reach, reducing operational complexity and providing more efficient financing for certain loans. As we've said before, it will not change our strategy of funding the vast majority of loans via third-party capital. Now let's turn to our outlook for the full year.

屆時，我們計畫將目前 Upstart 資產負債表上的大部分貸款移轉至銀行，並改由存款資金支持，這將釋放淨現金而非消耗現金。因此，我們預期銀行資本化不需要任何額外的股權資本。一旦開業，Upstart Bank 將帶來我們先前分享過的多項效益，包括擴大市場觸及、降低營運複雜度，並為特定貸款提供更高效率的融資。如同我們先前所說，這不會改變我們以第三方資本為絕大多數貸款提供資金的策略。接下來我們談談全年展望。

In February, we guided to total revenue of $1.4 billion and adjusted EBITDA of $294 million. This assumed UMI went no higher than 1.5 and interest rates were flat or down. Since then, UMI has climbed 14 points from 1.39 to 1.53, while the two-year Treasury yield has jumped by more than 115 basis points as of the end of October, with around half of that increase happening in September alone. Those are large moves, and I want to put them into context.

在 2 月，我們指引全年總營收為 14 億美元，調整後 EBITDA 為 2.94 億美元。此假設 UMI 不會高於 1.5，且利率持平或下行。自那之後，UMI 已上升 14 個點，從 1.39 升至 1.53；同時，截至 10 月底，兩年期美國公債殖利率已跳升超過 115 個基點，其中約一半的升幅僅發生在 9 月。這些都是很大的變動，我想把它們放在脈絡中說明。

We have previously shared our estimate that a 5-point increase in UMI, for example, from 1.4 to 1.45 reduces our daily originations by roughly 5% to 10%. Higher funding costs have a similar effect. Put simply, for each 1 point increase in UMI or 10 basis point increase in funding costs, we'd expect daily originations to fall by roughly 1% to 2%. This heuristic applies to our primary business model of earning fee revenue from new originations. But there is an additional way that macro impacts our P&L, and that is via the loans and co-investments on our balance sheet.

我們先前分享過估計：例如 UMI 上升 5 個點（從 1.4 到 1.45）會使我們的每日放款量約下降 5% 至 10%。資金成本上升也有類似影響。簡而言之，UMI 每上升 1 個點或資金成本每上升 10 個基點，我們預期每日放款量約下降 1% 至 2%。這個經驗法則適用於我們主要的商業模式，也就是從新增放款中賺取手續費收入。但宏觀環境影響我們損益表（P&L）的方式還有另一種，即透過我們資產負債表上的貸款與共同投資。

Because these positions were accumulated over many quarters, they are relatively large. As a result, an increase in UMI causes a significant reduction in fair values and net interest income. The impact to total revenue and profit from this is actually larger than the impact from reducing new originations. Higher interest rates also lower fair values, albeit to a much lesser extent.

由於這些部位是在多個季度累積而成，因此規模相對較大。因此，UMI 上升會導致公允價值與淨利息收入顯著下降。其對總營收與獲利的影響，實際上大於新增放款減少所帶來的影響。利率上升也會降低公允價值，但程度小得多。

Combined, these effects have created a substantial macro headwind for us. Without that pressure, we believe the full year will be trending well ahead of our original guidance on every line. As it stands, we've outrun the macro on the metrics tied to new originations and now expect full-year fee revenue and contribution profit to modestly exceed our original guidance. However, income from the balance sheet is tracking well below our initial outlook. In addition to being more sensitive from macro, as I just explained, this line item, unlike fee revenue, does not benefit from most of our advances in technology and marketing.

綜合而言，這些效應為我們帶來了顯著的宏觀逆風。若沒有這些壓力，我們相信全年各項指標都將明顯優於原先指引。就目前而言，在與新增放款相關的指標上，我們已跑贏宏觀環境，並預期全年手續費收入與貢獻利潤將小幅超越原先指引。然而，來自資產負債表的收入表現則遠低於我們最初的展望。除了如我剛才所解釋的，這一項目對宏觀更敏感之外，該項目也不同於手續費收入，無法從我們在技術與行銷上的多數進展中受益。

As a result, we now expect full year total revenue of approximately $1.35 billion and adjusted EBITDA of approximately $260 million. This updated outlook assumes UMI does not exceed 1.55 through year-end and no material change to two-year Treasury yields from current levels. Finally, I said I'm more optimistic than ever about the long term. It may sound like an odd conclusion to draw when macro is worsening and we're missing on profit, but it's the truth.

因此，我們目前預期全年總營收約為 13.5 億美元，調整後 EBITDA 約為 2.6 億美元。此更新後的展望假設 UMI 在年底前不超過 1.55，且兩年期美國公債殖利率相較目前水準沒有重大變化。最後，我說過我對長期前景比以往任何時候都更樂觀。當宏觀環境惡化、我們獲利未達預期時，這樣的結論聽起來或許有些奇怪，但這是事實。

I'll say it again, growth equals alpha plus beta times macro. Mathematically, alpha is the difference between actual results and the results you'd expect from macro alone. Our nominal contribution profit accelerated to 48% year-over-year in Q3, well above what is implied in our multiyear guide. Without the increase in UMI, we estimate it would have been at least 61%. On a full year basis, it is remarkable, frankly, that despite the 14-point increase in UMI and over 115 basis point increase in interest rates, we still expect to generate more fee revenue and profits from new originations than we expected at the start of this year.

我再說一次，成長等於 alpha 加上 beta 乘以宏觀。從數學上來看，alpha 是實際結果與僅由宏觀因素所能預期的結果之差。我們第三季的名目貢獻利潤年增加速至 48%，遠高於我們多年期指引所暗示的水準。若沒有 UMI 的上升，我們估計至少會達到 61%。以全年來看，坦白說相當令人驚訝的是，儘管 UMI 上升了 14 個點、利率上升超過 115 個基點，我們仍預期能從新增放款中創造比今年年初預期更多的手續費收入與利潤。

Beta. Beta is the sensitivity of our balance sheet-derived income to macroeconomic conditions, and it is high today. Fortunately, we're already bringing it down as we reduced the ratio of Upstart equity dollars required to total loans funded. We expect that will continue. Macro bounces around.

Beta。Beta 是我們資產負債表衍生收入對宏觀經濟狀況的敏感度，而目前它很高。幸運的是，隨著我們降低 Upstart 需要投入的股權資金相對於總放款資金的比例，我們已在把它降下來。我們預期這個趨勢將會持續。宏觀環境會上下波動。

Sometimes it's a plus, sometimes it's a minus. UMI has gone down in exactly 15 of the last 30 quarters. Same for interest rates. My own view on macro reminds me a little of Warren Buffett's New York Times piece in 2008 called buy American I Am. American borrowers are hurting this year because inflation flipped real wage growth negative.

有時是正面，有時是負面。在過去 30 個季度中，UMI 有 15 個季度是下降的。利率也是如此。我對宏觀的看法，讓我想起華倫・巴菲特在 2008 年於《紐約時報》發表的一篇文章〈Buy American. I Am.〉今年美國借款人之所以承受壓力，是因為通膨使實質薪資成長轉為負值。

But this is the United States. People get back on their feet, real wages rise over time. The macro will take care of itself. Most importantly, we're proving the Upstart engine. Execution against our years-deep technology and marketing road map delivers alpha at an exceptional rate. Keep compounding that, and I believe no matter what, Upstart becomes an extremely large and profitable company.

但這裡是美國。人們會重新站起來，實質薪資也會隨時間上升。宏觀終將自我調整。最重要的是，我們正在驗證 Upstart 引擎。對我們多年深耕的技術與行銷路線圖的執行，正以卓越的速度帶來 alpha。持續複利累積下去，我相信無論如何，Upstart 都會成為一家規模極大且獲利能力很強的公司。

With that, I'll turn the call over to Andrea.

接下來，我把電話交給 Andrea。

Thanks, Paul, and good morning, everyone. In Q3, our focus on core personal loans, improving secured unit economics drove record contribution profit. Combined with expense discipline, we delivered higher GAAP net income and improved capital efficiency against a harder macro backdrop. I'll walk through the quarter, then discuss our revised full year outlook. Total originations were $4.1 billion, up 44% year-over-year.

謝謝你，Paul，大家早安。在第三季，我們聚焦於核心個人貸款、改善擔保型產品的單位經濟效益，推動貢獻利潤創下新高。再加上費用紀律，我們在更具挑戰的宏觀背景下，實現更高的 GAAP 淨利並提升資本效率。我將先回顧本季表現，接著說明我們修訂後的全年展望。總放款額為 41 億美元，年增 44%。

Within that, Unsecured originations were $3.5 billion, up 33% versus last year, including Core at approximately $2.45 billion. Secured originations were $587 million, up 194%, with auto loan volume roughly tripling and Home roughly doubling relative to Q3 of last year. Total revenue was approximately $371 million, up 34% year-over-year. Revenue from fees was a record $364 million, up 41% year-over-year. Unsecured lending contributed $330 million in revenue from fees, up 30% year-over-year. Secured products contributed $34 million, up 538% year-over-year.

其中，無擔保放款為 35 億美元，較去年成長 33%，包含核心（Core）約 24.5 億美元。有擔保放款為 5.87 億美元，年增 194%，其中汽車貸款量相較去年第三季約成長至三倍，房屋（Home）約翻倍。總營收約為 3.71 億美元，年增 34%。手續費收入創新高，達 3.64 億美元，年增 41%。無擔保放款貢獻手續費收入 3.30 億美元，年增 30%。有擔保產品貢獻 3,400 萬美元，年增 538%。

Take rate, which we define as revenue from fees as a percentage of total originations, was 8.8%, up about 60 basis points from 8.2% in Q2, with both product categories improving sequentially. Net interest income and fair value adjustments totaled approximately $6 million, down $12 million or 65% year-over-year. Two main things drive this line, the average balance of loans we hold and the expected credit performance of those loans and the loans underlying our beneficial interest. Sequentially, this line declined about $10 million. Roughly half of that decline was fair value write-downs and charge-offs as average UMI increased from 1.41 in Q2 to 1.5 in Q3.

抽成率（Take rate），我們將其定義為手續費收入占總放款額的百分比，為 8.8%，較第二季的 8.2% 上升約 60 個基點，且兩個產品類別皆較前一季改善。淨利息收入與公允價值調整合計約 600 萬美元，較去年同期減少 1,200 萬美元或 65%。此項目主要由兩個因素驅動：我們持有之貸款的平均餘額，以及這些貸款與構成我們受益權益（beneficial interest）之基礎貸款的預期信用表現。按季比較，此項目下降約 1,000 萬美元。其中約一半的下降來自公允價值減記與沖銷（charge-offs），因平均 UMI 自第二季的 1.41 上升至第三季的 1.5。

The other half reflected the debt sale in Q2 that did not repeat. Rising benchmark rates had a minimal impact on our fair value position in Q3. Next, Contribution Profit, which is our non-GAAP metric defined as revenue from fees minus variable costs for borrower acquisition, verification and servicing. Contribution profit reached a record $218 million in Q3, up 48% year-over-year. Contribution margin was 60% versus 57% in Q3 2025 and 55% in Q2 2026. The 5-percentage-point sequential increase was driven by margin improvements in both our unsecured and secured categories.

另一半則反映第二季的債務出售未在本季重複發生。第三季基準利率上升對我們的公允價值部位影響甚微。接下來是貢獻利潤（Contribution Profit），這是我們的非 GAAP 指標，定義為手續費收入扣除借款人獲取、驗證與服務的變動成本。第三季貢獻利潤創下 2.18 億美元新高，年增 48%。貢獻毛利率為 60%，相較 2025 年第三季為 57%、2026 年第二季為 55%。按季增加 5 個百分點，主要由我們無擔保與有擔保兩大類別的毛利率改善所帶動。

Unsecured contribution margin was 64%, up 2-percentage-points from 62% in Q2, reflecting a higher take rate, along with lower borrower acquisition cost as a percentage of originations. Secured contribution margin was a positive 17% compared with a negative 35% in Q2 and roughly 200-percentage-points better than Q3 of last year. A higher take rate and lower borrower acquisition costs helped us reach a positive contribution margin a quarter ahead of our target. As Paul mentioned, we expect more quarter-to-quarter variability in this margin than in unsecured as we scale these businesses, and we intend to hold it to a band of breakeven to 25%. GAAP operating expenses were approximately $335 million in Q3, up 32% year-over-year and down 4% sequentially.

無擔保貢獻毛利率為 64%，較第二季的 62% 上升 2 個百分點，反映抽成率提高，以及借款人獲取成本占放款額的比例下降。有擔保貢獻毛利率為正 17%，相較第二季為負 35%，也較去年第三季改善約 200 個百分點。較高的抽成率與較低的借款人獲取成本，協助我們比目標提前一季達到正的貢獻毛利率。如 Paul 所提到，隨著我們擴大這些業務規模，我們預期此毛利率的季對季波動將高於無擔保類別，並打算將其維持在損益兩平至 25% 的區間。第三季 GAAP 營業費用約 3.35 億美元，年增 32%，季減 4%。

Within that, variable expenses -- borrower acquisition, verification and servicing costs -- declined 6% sequentially with borrower acquisition costs alone down 11% to $91 million. Fixed expenses, defined as total operating expenses minus borrower acquisition, verification and servicing costs declined 3% sequentially to approximately $188 million, better than the low single-digit growth we guided to in August. Within that, stock-based compensation was $41 million, down from $46 million in Q2, reflecting seasonality in how we recognize this expense. Excluding stock-based compensation, fixed expenses were roughly flat sequentially.

其中，變動費用——借款人獲取、驗證與服務成本——按季下降 6%，僅借款人獲取成本就下降 11% 至 9,100 萬美元。固定費用（定義為總營業費用扣除借款人獲取、驗證與服務成本）按季下降 3% 至約 1.88 億美元，優於我們 8 月所指引的低個位數成長。其中，股票基礎薪酬為 4,100 萬美元，較第二季的 4,600 萬美元下降，反映我們認列此費用的季節性。排除股票基礎薪酬後，固定費用按季大致持平。

Net income was approximately $37 million, up 15% year-over-year with a 10% margin. Diluted EPS was $0.33 based on a weighted average diluted share count of 125 million and reflects the assumed conversion of dilutive convertible senior notes, consistent with GAAP requirements. Adjusted EBITDA was approximately $89 million, up 26% year-over-year with a 24% margin. Turning to funding and capital efficiency. As Paul discussed, we're focused on supporting platform growth with proportionately less equity capital, bringing down the beta in his equation.

淨利約 3,700 萬美元，年增 15%，淨利率 10%。稀釋後每股盈餘（EPS）為 0.33 美元，係以加權平均稀釋後股數 1.25 億股計算，並反映具稀釋性的可轉換優先票據之假設轉換，符合 GAAP 規定。調整後 EBITDA 約 8,900 萬美元，年增 26%，利潤率 24%。接下來談資金來源與資本效率。如 Paul 所述，我們專注於以相對更少的股權資本支持平台成長，降低他方程式中的 beta。

To measure our overall equity capital use, we combined the equity supporting: one, the loans on our balance sheet; two, our co-investment exposure alongside committed capital partners; and three, our retained interest in securitizations, all net of related financing.

為衡量我們整體股權資本使用情況，我們合併計算用以支撐以下項目的股權：一、資產負債表上的貸款；二、與承諾資本合作夥伴並行的共同投資曝險；三、我們在證券化中的保留權益（retained interest），以上皆扣除相關融資後的淨額。

Taken together, the overall equity efficiency of our platform has been improving. Total equity capital used represented 9% of our total platform loan balance outstanding in Q3, down from 11.6% a year ago. That improvement reflects the declining share of loans held directly on our balance sheet, even as we increased our co-investments alongside committed capital partners. Capital ultimately flows to the assets with the best risk-adjusted returns. We've been able to grow our capital partnerships on better terms over time because of our strong credit performance.

綜合來看，我們平台的整體股權效率持續改善。第三季使用的股權資本占我們平台未償放款總餘額的 9%，低於一年前的 11.6%。此改善反映我們直接持有於資產負債表上的貸款占比下降，即使我們同時增加了與承諾資本合作夥伴並行的共同投資。資本最終會流向風險調整後報酬最佳的資產。由於我們強勁的信用表現，我們得以隨時間以更佳條件擴大資本合作關係。

The average return of our last 12 quarterly loan vintages exceeds US Treasuries by approximately 660 basis points. Next, Upstart Bank. I'll explain how we expect to fund this launch and transition loans from the parent company, subject to the remaining approvals. First, we'll capitalize the bank with approximately $404.1 million in cash and servicing assets.

我們最近 12 個季度貸款分期（vintages）的平均報酬率，較美國公債高約 660 個基點。接下來談 Upstart Bank。我將說明我們預期如何為此次啟動提供資金，並在取得剩餘核准的前提下，將貸款自母公司移轉。首先，我們將以約 4.041 億美元的現金與服務資產（servicing assets）為銀行注資。

Second, Upstart Bank will raise brokered deposits through deposit partners. We have agreements with two major partners, giving us the ability to raise ample deposits shortly after the bank launches. Using deposit funding, Upstart Bank will buy most of the existing loan book and begin funding new originations on our network. Third, the equity funding underlying the existing loan book will be released as cash and we'll take orderly steps to pay down the related warehouse facilities. We expect our co-investments in committed capital partnerships will remain outside of the bank.

第二，Upstart Bank 將透過存款合作夥伴募集經紀存款（brokered deposits）。我們已與兩家主要合作夥伴簽署協議，使我們能在銀行啟動後不久募集充足存款。運用存款資金，Upstart Bank 將買入現有貸款組合的大部分，並開始為我們網路上的新放款提供資金。第三，支撐現有貸款組合的股權資金將以現金形式釋出，我們也將有序採取措施償還相關的倉儲融資（warehouse facilities）。我們預期在承諾資本合作關係中的共同投資將維持在銀行之外。

The net effect of these transactions is expected to be an increase in our consolidated cash balance. We expect these transactions to take place over several weeks following the initial bank capitalization. Illustratively, using our actual Q3 balance sheet, these transactions would result in pro forma consolidated unrestricted cash of over $1 billion, and the bank would have a Tier 1 leverage ratio and a total risk-based capital ratio well in excess of our required minimums. Turning to October and our full year outlook. October originations were approximately $1.25 billion, with daily volume down 9% versus September and up 26% year-over-year.

這些交易的淨效果預期將增加我們的合併現金餘額。我們預期這些交易將在銀行初始注資後的數週內完成。舉例而言，若以我們第三季實際資產負債表計算，這些交易將使備考（pro forma）合併不受限制現金（unrestricted cash）超過 10 億美元，且銀行的一級槓桿比率（Tier 1 leverage ratio）與總風險基礎資本比率（total risk-based capital ratio）將遠高於我們的最低要求。接下來談 10 月與全年展望。10 月放款額約 12.5 億美元，日均量較 9 月下降 9%，較去年同期上升 26%。

The sequential decline reflects two macro headwinds, UMI rising over the course of the month and benchmark-driven increases in funding costs in September flowing through to borrowers and pricing in October. Those pressures weighed on volume even as our models continue to improve. October is a useful starting point for our outlook as benchmark-driven increases in funding costs are already reflected in October's pricing and volume. For the full year, we now expect approximately revenue from fees of $1.32 billion and total revenue of $1.35 billion, contribution profit of $750 million, up roughly 41% year-over-year. Adjusted EBITDA of $260 million, representing a 19% margin.

按季下滑反映兩項總體逆風：UMI 在月內走升，以及 9 月由基準利率驅動的資金成本上升在 10 月傳導至借款人與定價。即便我們的模型持續改善，這些壓力仍對放款量造成影響。10 月可作為我們展望的有用起點，因為由基準利率驅動的資金成本上升已反映在 10 月的定價與放款量中。就全年而言，我們目前預期手續費收入約 13.2 億美元、總營收 13.5 億美元，貢獻利潤 7.5 億美元，年增約 41%。調整後 EBITDA 為 2.6 億美元，代表 19% 的利潤率。

Relative to our original outlook, we now expect higher fee revenue and contribution profit, but lower total revenue and adjusted EBITDA. Our execution has more than offset macro pressure on the metrics tied to new originations. The reduction in our outlook reflects lower net interest income and fair value adjustments, primarily driven by higher UMIs.

相較我們原先的展望，我們目前預期手續費收入與貢獻利潤更高，但總營收與調整後 EBITDA 較低。我們的執行表現已超過總體環境對與新放款相關指標的壓力。展望下修反映淨利息收入與公允價值調整較低，主要由較高的 UMI 所驅動。

Our latest UMI reading is 1.53, above the average we saw in October. Our updated guidance assumes UMI remains at or below 1.55 through year-end and no material change in two-year Treasury yields. With October originations known and only two months remaining, we have greater visibility into full year fee revenue. Fair value adjustments remain more sensitive to changes in macroeconomic conditions across our existing loan portfolio. To close, we entered the final months of the year with record contribution profit, positive secured contribution margin and improved capital efficiency. Our focus remains on building on that product while maintaining expense discipline.

我們最新的 UMI 讀數為 1.53，高於我們在 10 月看到的平均值。我們更新後的指引假設 UMI 在年底前維持在 1.55 或以下，且兩年期美國公債殖利率沒有重大變化。在已知 10 月放款額且僅剩兩個月的情況下，我們對全年手續費收入有更高的能見度。公允價值調整仍對我們既有貸款組合中總體經濟條件的變化更為敏感。最後總結：我們在今年最後幾個月以創紀錄的貢獻利潤、轉正的有擔保貢獻毛利率，以及改善的資本效率進入收官階段。我們的重點仍是持續打造該產品，同時維持費用紀律。

With that, operator, we're ready for Q&A.

那麼，接線員，我們準備開始問答。

(Operator Instructions)

(接線員指示)

Dan Dolev, Mizuho.

Dan Dolev，瑞穗。

Nice results. Can we ask you a little bit about the growth in home loans? We thought this was like a really nice thing. There is a lot of strength in the home loan growth. How do you feel about selling those versus some of the other loans? And how you feel about this market overall?

業績很不錯。我們能否請教一下房貸成長的情況？我們覺得這是一個非常好的亮點。房貸成長展現出很強的動能。你們如何看待銷售這些房貸相較於其他類型貸款？以及你們對整體市場的看法如何？

Dan, thanks for the question. Yes, we're really excited about our home loans business as we are really about, at this point, every product that we're in, every business line that we've got is one we have a lot of conviction in. So we're really, really focused as a company. And in this quarter, we were so -- we were hyper-focused on getting the secured products to profitability. What we shared over the last couple of quarters is that, that was our real number one priority for the secured product teams.

Dan，謝謝你的提問。是的，我們對房貸業務非常振奮；事實上，就目前而言，我們所涉足的每一項產品、每一條業務線，都是我們高度有信心的。因此公司在策略上非常、非常聚焦。而在本季，我們——我們極度專注於讓擔保型產品達到獲利。我們在過去幾季分享過，這一直是擔保型產品團隊的首要優先事項。

And I'm really proud of what those teams pulled off is they did a really hard pivot to just focus on margin, getting above breakeven. And you can see in this quarter, we did it a whole quarter early before the end of the year. We're at -- did a 17% secured contribution margin.

我也為這些團隊的成果感到非常自豪：他們做了一個非常艱難的轉向，專注在利差，讓業務跨越損益兩平。你可以看到在本季，我們比年底目標提前整整一季達成。我們的擔保型貢獻毛利率達到17%。

And yes, we absolutely expect a lot more growth out of these products. But to me, the sort of first order of business was getting to profitable. We have a little bit more work to finish up some of the efforts there that will really be what we do this year. And then as I said, in 2027, our attention is going to turn back to growth, and we certainly expect this to be a really big part of our future.

是的，我們絕對預期這些產品還會有更多成長。但對我而言，第一要務是先做到獲利。我們還有一些工作要完成，收尾部分會是我們今年的重點。然後如我所說，到了2027年，我們的注意力會再回到成長上，而我們也確實預期這會是我們未來非常重要的一部分。

Kyle Peterson, Needham.

Kyle Peterson，Needham。

I guess I wanted to start out on interest rates and kind of the moving pieces kind of impact on the model. So, Paul, heard your comment that, obviously, I think there's a bit of kind of volume and approval choke-off with the higher rates. But I guess, is that the full impact that you guys expect? Or are there any kind of knock-on impacts from funding partners? I'm assuming their cost of capital probably ticked up a little bit in October as well. So I just want to see if there are any other kind of second derivative impacts we should be mindful of with regards to rates.

我想先從利率以及各種變動因素對模型的影響談起。Paul，我聽到你提到，顯然在利率較高的情況下，量能與核准率會有些被掐住。但我想問，這就是你們預期的全部影響嗎？還是資金合作夥伴那邊會有一些連鎖效應？我猜他們的資金成本在10月也可能上升了一點。所以我想確認一下，關於利率，我們還需要留意哪些其他的二階影響。

Thanks for the question, Kyle. Yes, for everything we sort of know as of today, the impact of rate increase that we saw in September and where rates were in October is fully reflected in the pricing on our platform in October. So our October originations are reflective of the pass-through of that cost of capital to our platform and our pricing. Just kind of stepping back maybe to say a little bit more about how rates kind of show up on our platform and in our funding costs. At a high level, you can think about most of our institutional capital as kind of being most levered to two-year Treasuries on average, and our depository and bank capital kind of being benchmarked to Fed funds. So that's kind of the mix of benchmarks kind of on average underlying our platform.

謝謝你的提問，Kyle。是的，就我們截至今天所掌握的一切資訊來看，我們在9月看到的升息影響，以及10月的利率水準，都已完整反映在10月平台的定價上。因此，我們10月的放款承作量反映了資金成本向平台與定價的轉嫁。稍微退一步，補充一下利率如何在我們的平台與資金成本中呈現。概括而言，你可以把我們大部分的機構資金視為平均而言與2年期美國公債連動性最高，而我們的存款機構與銀行資金則多以聯邦基金利率作為基準。這大致就是我們平台底層基準利率的平均組合。

And as a marketplace, what we're typically doing is updating pricing on the marketplace on about a monthly lag. So to the extent that we see rising rates in the month of September, as we did, for the most part, that is flowing through into pricing on our platform and funding costs in the month of October. The final thing I'll just maybe comment on is that while we've seen Treasuries up 120 basis points over the course of this year, that hasn't necessarily been the effective funding cost passed through to our platform. Over the course of this year, we have seen across our cost of capital improved spreads and improved cost of funding that has partially though not fully offset that increase in rates.

而作為一個市場平台，我們通常會以大約一個月的落後期來更新市場上的定價。因此，若我們在9月看到利率上升（如同我們所見），大多會在10月反映到平台定價與資金成本上。最後我再補充一點：雖然今年以來公債殖利率上升了120個基點，但這不一定等同於轉嫁到我們平台的有效資金成本。今年以來，我們在資金成本方面看到利差改善與融資成本改善，部分（但並非完全）抵銷了利率上升的影響。

Okay. That's helpful. And then I guess just as a follow-up, I really appreciate all the color you just gave on Upstart Bank. I thought that was great. But I wanted to see if you guys could provide any color on where take rates should go for loans originated by the bank once it's up and running.

好的，這很有幫助。接著我想追問一下，我很感謝你剛才對 Upstart Bank 提供的所有細節，我覺得很棒。但我想請你們談談，一旦銀行開始營運後，由該銀行承作的貸款，其 take rate 應該會走向何處？

I guess I would assume it's directionally better if you can originate directly. But any color there on kind of what the unit economics would look like for a loan originated through the bank versus through a conduit bank would be very helpful.

我想如果能直接承作，方向上應該會更好。但若能說明一下，透過銀行承作的貸款與透過通道銀行（conduit bank）承作的貸款，在單位經濟（unit economics）上大概會是什麼樣子，會非常有幫助。

Yes, that's a great question. And so one of the benefits of the bank charter will be improved economics for Upstart in that we will no longer need to pay the fees and accrued interest to our bank partners that are originating loans today. So we will retain those economics. And so in expectation, we would expect to see modest improvement in take rates as we transition to Upstart Bank as we will be retaining fees that we currently pay to our originating partners today as well as a modest improvement in net interest income as some of the economics to those parties are flowing through interest income line.

是的，這是個很好的問題。銀行執照的一項好處是 Upstart 的經濟效益會改善，因為我們將不再需要支付目前為我們承作貸款的銀行合作夥伴之手續費與應計利息。因此我們會保留這部分經濟利益。所以我們預期，隨著我們轉向 Upstart Bank，take rate 會有溫和改善，因為我們將保留目前支付給承作合作夥伴的費用；同時，淨利息收入也會有小幅改善，因為原本流向那些合作方的一部分經濟利益，會透過利息收入科目反映出來。

Peter Christiansen, Citi.

Peter Christiansen，花旗。

Paul, I guess, how should we think about UMI staying around these kind of levels, 1.55, 1.53-ish throughout next year? And I'm curious how do you think about Upstart still growing fee revenue, contribution profit at attractive rates? Or does the longer-term framework require some improvement in consumer credit conditions from here?

Paul，我想問，UMI 明年是否會大致維持在這種水準，例如1.55、1.53左右？另外我也好奇，你如何看待 Upstart 仍能以具吸引力的速度成長手續費收入與貢獻利潤？或者，長期框架是否需要消費者信用環境從目前水準有所改善？

Yes, Pete. Well, first of all, I think the thing I always say is we don't claim to predict future changes to macro conditions. We don't have any special view on what's going on in Iran or the war, oil prices. I mean I think everyone's just got to have their own models for that and where interest rates and inflation are going. But we do pride ourselves on being the fastest and most precise to respond to that, at least as it relates to consumer credit, and that's what we do with the UMI.

是的，Pete。首先，我一直強調的是，我們不自稱能預測未來總體環境的變化。我們對伊朗局勢或戰爭、油價等沒有什麼特別的看法。我想每個人都得有自己的模型，去判斷利率與通膨會走向何處。但我們以能最快、最精準地做出反應為傲，至少在消費者信用相關的層面，而這正是我們透過 UMI 所做的事。

I would say in terms of our longer-term outlook, I think the results today, what we basically said was, hey, in the near term, the macro has a pretty large impact on our results, particularly via the sort of balance sheet component of our business. But actually, on a longer-term basis, we're feeling really optimistic about the state of the company. And that basically comes from this framework around how much of the performance we've seen is alpha, how much of it is really just the macro.

就我們的長期展望而言，我認為今天的結果基本上是在說：短期內，總體環境會對我們的業績造成相當大的影響，特別是透過我們業務中的資產負債表（balance sheet）組成部分。但其實從更長期的角度，我們對公司的狀態感到非常樂觀。而這主要來自一個框架：我們所看到的表現中，有多少是 alpha（超額表現），又有多少其實只是總體環境所致。

And in the kind of medium term, I think if the macro kind of takes care of itself, that will mostly just means it kind of stays around where it is, then we're really optimistic that the rate at which we are delivering alpha, the rate at which we're generating growth that is tied to our advances in technology and marketing is going to be plenty to sort of sustain the kind of multiyear growth that we've talked about and that we've guided to.

在中期來看，我認為如果總體環境能自行消化——也就是大致維持在目前水準——那我們就非常有信心：我們交付 alpha 的速度、以及由技術與行銷進步所帶動的成長速度，將足以支撐我們所談到、也所給出的那種多年期成長。

So we actually don't think that we need a tailwind from macro, but certainly, a headwind is going to subtract against whatever efforts we're putting in. But right now, we're feeling really optimistic that the rate of progress we're generating is going to be very good in a flat -- or in a flat environment.

所以我們其實不認為需要宏觀環境帶來順風，但當然，逆風會抵消我們正在投入的各項努力。不過就目前而言，我們非常樂觀，認為我們所帶來的進展速度在一個持平——或在一個平坦的環境中——會非常好。

That's helpful. And then I do want to ask about the secured contribution margin. Great to see that now positive 17%, quite a flip in one quarter. Just trying to get a sense of how sustainable that is. Is the improvement coming from, I guess, more structural operational improvements or scale versus just favorable loan sale execution and funding market conditions?

這很有幫助。接著我確實想問一下擔保業務的貢獻毛利率。很高興看到現在轉為正的17%，單季就有很大的翻轉。我想了解這樣的水準有多可持續。改善主要來自於更結構性的營運改善或規模效應，還是只是因為貸款出售執行得宜以及資金市場條件有利？

Yes, it's all of the above. As I said, we put a real focus on getting these secured products to profitable and really every team touching those products. What they had in their mind was how do I improve the margins of this product. And so certainly, loan sales is a part of it. There's sort of both a mechanical and accounting part of it. There's a real part, which is about getting the deal terms right so that we're earning the right take rates on those products, the right fee revenues.

是的，以上皆是。如我所說，我們非常專注於讓這些擔保產品達到獲利，而且幾乎每個接觸到這些產品的團隊，腦中想的都是：我要如何提升這個產品的利潤率。因此，貸款出售當然是其中一部分。這裡面同時有一些機械性與會計層面的因素。也有更實質的部分，也就是把交易條款談對，確保我們在這些產品上取得正確的抽成率（take rates）與正確的費用收入。

There's sort of real improvements in automation, which I talked about. That's true in both the home and auto products. Just more automation is both better for conversion and better for cost. And then there's some wins in monetization and figuring out where we generate the most value and capturing that in an intelligent way using models there. And so all across the board, I would say really everything got worked on.

另外也有一些自動化方面的實質改善，我之前提過。這在房屋與汽車產品上都成立。更多自動化不僅有助於轉換率，也有助於成本控制。此外，在變現方面也有一些成果：我們找出在哪裡能創造最大價值，並用模型以更聰明的方式把它捕捉下來。所以我會說，幾乎各個面向都做了改善。

And I said that we need to lock in some of these gains the rest of this year, and that's just because we did so much in such a short period of time that we want to really make sure everything is ready to scale before we go back to focusing on growth.

我也提到，我們需要在今年剩下的時間把其中一些成果鎖定下來，原因是我們在很短的時間內做了很多事情，因此在回到以成長為重心之前，我們想確保一切都已準備好可以擴張。

Patrick Moley, Piper Sandler.

Patrick Moley，Piper Sandler。

So I wanted to ask on expenses. Year-over-year, engineering and product was up about 45%. Total OpEx was up around 32% in the quarter and 38% year-to-date. So with the implied 4Q EBITDA margin in the high teens, I was wondering how you're thinking about fixed cost growth into next year. Is the plan to let margins keep or compress and keep investing in the models? Or should we expect headcount growth to slow from here? And then how does that square maybe with the 25% longer-term margin target for 2028?

我想問一下費用。年對年來看，工程與產品部門大約成長了45%。本季總營運費用（OpEx）約成長32%，年初至今成長38%。因此，考量到第4季隱含的EBITDA利潤率在十幾個百分點的高段，我想了解你們如何看待明年固定成本的成長。計畫是讓利潤率維持或被壓縮、並持續投資模型嗎？還是我們應該預期從現在開始人員成長會放緩？另外，這與你們2028年長期25%利潤率目標如何相符？

Yes. Thanks for the question. So implied in our guide, what we expect from Q4 is that expenses -- fixed expenses will be about flat from Q3 to Q4 after a modest sequential decline here in Q3. Stepping back as we look at the sort of total arc and trajectory of 2026, this has been a bit of an investment year for the business as we've invested forward in laying the groundwork for our newer secured products, which have just slipped to contribution margin positive this quarter as well as making forward investments in the bank.

是的，謝謝你的問題。在我們的指引中，對第4季的預期是：費用——固定費用在第3季小幅季減之後，從第3季到第4季大致持平。退一步看，從2026年的整體弧線與走勢來看，今年對公司而言算是投資年；我們提前投入，為較新的擔保產品打基礎，而這些產品也在本季剛轉為貢獻毛利率為正，同時也對銀行業務做了前瞻性投資。

And so as we look ahead to 2027 and beyond and certainly as implied in our 25% guide in 2028, our expectation is that we'll see improved operating leverage in 2027 and 2028 than we saw here in 2026 as the fruits and the products of those investments start to come forward.

因此，展望2027年及之後，並且如我們在2028年25%指引所隱含的，我們預期在2027與2028年會看到比2026年更好的營運槓桿，因為這些投資的成果與產品將開始顯現。

Simon Clinch, Rothschild & Company.

Simon Clinch，Rothschild & Company。

I wanted to go back to the prior question around how to think about growth when the UMI stays around this kind of level, the 1.53 to 1.55 level. I was wondering if you could talk about the -- I guess, the ability for the company to calibrate to a high level of UMI and what kind of structural growth can therefore be delivered and how long that might take for you to calibrate to those levels?

我想回到前一個關於如何看待成長的問題：當UMI維持在這種水準、約1.53到1.55的區間時。我想請你談談——我想是——公司校準到高UMI水準的能力；在此情況下可以帶來什麼樣的結構性成長，以及你們校準到那些水準大概需要多久？

Yes. We think it bakes in pretty fast. We don't, as a company, count on macro getting better. We're not hoping for a macro tailwind to really push the business forward. And this really comes back to this idea that there's an alpha component to growth and there's a beta component to growth.

是的，我們認為適應得很快。作為一家公司，我們不會指望宏觀環境變好。我們不會寄望宏觀順風來真正推動業務向前。這其實回到一個概念：成長有alpha成分，也有beta成分。

And I think there are some truly cyclical businesses out there that are just sort of hoping that the macro gets better, and that's what powers their growth. And that's not us in any way, none of our numbers, none of our forecasts, none of our guidance is built on the idea that we're waiting for interest rates to come down or waiting for UMI to come down. Really, the growth driver in our business is kind of this Upstart road map around technology and marketing.

我認為市場上確實有一些高度循環性的企業，只是希望宏觀環境變好，然後那就驅動它們的成長。但我們完全不是那樣；我們的任何數字、任何預測、任何指引，都不是建立在「等利率下降」或「等UMI下降」的想法上。我們業務的成長驅動力，主要是Upstart在技術與行銷方面的路線圖。

And we think there's a multiyear long road map where we can keep delivering alpha at an extremely high pace like we've been coming back to this idea that in Q3, in spite of this 9-point average increase in UMI, we did 48% year-over-year contribution profit growth. And if you took that 9 points of UMI increase out of the equation, it would have been like 61% in contribution profit growth. That's what we think about as sort of the rate at which we're delivering alpha, the sort of secular rate of growth when you sort of denoise the graph and take out the macro part of it.

我們認為這是一個跨多年的長期路線圖，我們可以持續以極高的速度交付alpha；就像我們一直回到的例子：在第3季，儘管UMI平均上升了9個點，我們的貢獻利潤仍年對年成長48%。如果把UMI上升的那9個點因素排除，貢獻利潤成長會接近61%。這就是我們所說的交付alpha的速度——當你把圖表去噪、把宏觀因素拿掉後，那種更偏結構性的成長率。

And so obviously, if the macro is just flat, even if it's flat at a high number, and we keep delivering a significant rate of sort of alpha-driven growth, then I think the business is going to do really, really well. And so that's kind of how we make our forecast, how we build our business plan. And because of that, we're extremely optimistic about the long term here.

因此很明顯，如果宏觀只是持平，即使是維持在較高的數值，而我們仍持續交付相當可觀、由alpha驅動的成長率，那我認為公司表現會非常、非常好。這就是我們做預測、建立營運計畫的方式。也因為如此，我們對長期非常樂觀。

Great. And just a follow-up as well. Paul, you mentioned the -- I think it was 250,000-plus repeat customers. I was wondering if you could just give some color around the characteristics of those? How many products are they buying? How much of that kind of repeat penetration opportunity there is with each -- just within those repeat customers alone?

很好。另外再追問一下。Paul，你提到——我記得是——超過25萬名回流客戶。我想請你補充一些這些客戶特徵的資訊。他們購買了多少種產品？僅就這些回流客戶本身而言，每位客戶的回購滲透機會大概有多大？

Yes. That is a combined metric. It is a metric that covers both people who are taking out repeat using the same product as well as people switching to a different product. And obviously, this is a pretty new dynamic in our business and one we're really excited about. We used to just have essentially one product, personal loans, and there wasn't a lot else for people to do.

是的。那是一個合併指標。它同時涵蓋了重複使用同一產品的人，以及轉換到不同產品的人。而且很明顯，這在我們的業務中是相當新的動態，我們也非常興奮。過去我們基本上只有一個產品——個人貸款——客戶也沒有太多其他選擇。

And now we have this whole suite of products around home and small dollar and Cash Line and we have the HELOC product. We have the auto secured personal loans. So there's a lot of different products that a consumer could engage with.

而現在我們有一整套圍繞房屋、小額、Cash Line的產品，還有HELOC產品。我們也有汽車擔保的個人貸款。因此，消費者可以使用的產品種類很多。

And I think as we complete building out the consumer product suite, we expect this to be a more and more important dynamic. And that's why we started sharing some of these metrics around there are 25 million unique Americans that have applied for credit with Upstart. I mean that is a huge number. And I don't think there are a lot of other platforms that have that kind of reach.

我認為，隨著我們完成消費者產品組合的建置，這將成為越來越重要的動態。這也是為什麼我們開始分享一些指標，例如有2,500萬名不重複的美國人曾向Upstart申請信貸。我的意思是，那是一個非常龐大的數字。而我不認為有很多其他平台能有那樣的觸及範圍。

And as we keep growing that number, I think it's going to become increasingly important for us to just sort of do marketing within the sort of ecosystem of people who already know about Upstart, who already have data shared with Upstart, who already have a kind of relationship and trust in Upstart. And I think as we just have more products, each of those products becomes better and people get to know and trust us. I think that's going to be a really powerful customer ecosystem.

而隨著我們持續擴大那個數字，我認為對我們而言，在一個由已經認識 Upstart、已經與 Upstart 分享資料、並且已經對 Upstart 建立某種關係與信任的人所構成的生態系內進行行銷，將會變得愈來愈重要。而我認為，隨著我們擁有更多產品，每一項產品都會變得更好，人們也會更了解並信任我們。我認為那將會是一個非常強大的客戶生態系。

John Hecht, Jefferies.

John Hecht，Jefferies。

Andrea, you sounded like the take rate was up in the quarter as you passed on some fee increases through the channels. I'm wondering, was that across all products? Was there any of the products that had a steeper increase in take rate? And are there -- is there any ceiling like from a competitive factor on your ability to increase your fee structures?

Andrea，聽起來你們本季的抽成率（take rate）上升，因為你們透過通路轉嫁了一些費用上調。我想請問，這是涵蓋所有產品嗎？是否有任何產品的抽成率上升幅度更大？另外，從競爭因素來看，你們提高費率結構的能力是否存在某種上限？

Thanks, John. Yes. So we did see our take rate improve from Q2 to Q3 by about 60 basis points, and that was driven by improved take rate across both unsecured and secured. So I wouldn't necessarily characterize the main driver as kind of increasing fees through to the borrower. So on the unsecured side, the main driver of the increase in take was modest mix shift towards core personal loans, which inherently has a higher take rate versus prime.

謝謝你，John。是的。我們確實看到抽成率從第二季到第三季提升了約 60 個基點，而這是由無擔保與有擔保兩邊的抽成率改善所帶動。因此，我不一定會把主要驅動因素描述為把費用上調轉嫁給借款人。在無擔保端，抽成率上升的主要原因是產品組合略微轉向核心個人貸款，而核心個人貸款相較於優質（prime）客群，本來就具有較高的抽成率。

And on the secured side, we saw a very meaningful improvement in take rate really driven by: one, increased sell-through of loans to third parties; two, improved execution on those deals; and three, some degree of the monetization that you're referring to. So continuing to optimize how we think about origination and other fees charged across those newer products. So those are really the sort of two drivers. There's a mix dynamic and then the sort of the broad improvement we saw in secured as we're selling more loans through at improved deal terms.

而在有擔保端，我們看到抽成率有非常顯著的改善，主要由以下因素驅動：第一，向第三方出售貸款的售出率（sell-through）提高；第二，這些交易的執行表現更佳；第三，也有一定程度來自你所提到的變現（monetization）。也就是持續優化我們在這些較新產品上，對於放款（origination）及其他費用的收取方式。因此，這基本上是兩個驅動因素。一個是組合效應，另一個則是我們在有擔保端因為以更好的交易條件售出更多貸款，而帶來的整體改善。

I was just going to jump in and just to remind everyone that the take rate number really isn't a number that we manage the business to. It's just -- it's kind of like an output of the system, but we aren't every day thinking about, hey, what's the right take rate for us. Really, we think about contribution profit, and that's sort of the North Star for us.

我補充一下，提醒大家，抽成率這個數字其實不是我們用來管理業務的指標。它只是——有點像系統的輸出結果，但我們不會每天都在想，嘿，對我們而言正確的抽成率是多少。我們真正關注的是貢獻利潤（contribution profit），那才是我們的北極星指標。

Okay. That's helpful. And then turning to funding. I mean, I think a majority of your loans are invested in the forward flow type contracts. As we get into next year, what -- do you think that mix will change? Or will you still have a focus on that, particularly given your desire to kind of optimize or minimize the capital requirements? How do we think about your goals for funding next year?

好的，這很有幫助。接著談到資金來源（funding）。我的理解是，你們大多數貸款是透過遠期流（forward flow）類型的合約來投資。進入明年後，你認為這個組合會改變嗎？或者你們仍會聚焦在這一塊，特別是考量到你們希望優化或降低資本需求？我們該如何看待你們明年的資金來源目標？

Overall, I'd say that we are -- so one, sort of growing the committed capital partners as a proportion of our funding has been an important strategic objective for us to help improve the resilience of our platform through the cycle. And what I'd say is that we sort of like the level that it's at today. So as of Q3, about 68% of our platform was funded from committed capital partnerships. That's about the right level from our perspective on a go forward. And so our expectation is that, obviously, with some plus or minuses over time, we would expect that level to stay about constant over the next few quarters. And that translates and helps translate through to as we think about improving the equity usage of our overall platform.

整體而言，我會說——第一，讓承諾資本合作夥伴（committed capital partners）在我們資金來源中的占比提高，一直是我們重要的策略目標，以協助提升平台在景氣循環中的韌性。而我想說的是，我們其實滿喜歡目前的水位。截至第三季，大約 68% 的平台資金來自承諾資本合作關係。從我們的角度看，這大致是未來前進方向上合適的水準。因此我們的預期是，當然隨時間可能有些正負波動，但在接下來幾季，我們預期這個水準大致會維持不變。而這也會轉化並有助於我們在思考如何提升整體平台的股權使用效率（equity usage）時的表現。

And so as we look to improve the equity usage of the overall platform, one level will be sort of that line item stabilizing versus continuing to grow. Two, we expect that the committed capital partnerships will increasingly start to cash flow out to us. There's some lag to cash flow there, which will reduce equity usage associated with prior vintages. And third, we continue to expect that as we renew our committed capital partnerships, as we've seen in the past, we'll drive better deal terms against those, including on the amount of capital co-invested.

因此，當我們著眼於提升整體平台的股權使用效率時，第一個層面會是這個項目穩定下來，而不是持續成長。第二，我們預期這些承諾資本合作關係將會愈來愈開始對我們產生現金流入（cash flow out to us）。現金流有一定的時間落後（lag），這將降低與先前批次（prior vintages）相關的股權使用。第三，我們仍預期在續約承諾資本合作關係時，如同過去所見，我們將能爭取更好的交易條款，包括共同投資（co-invested）的資本金額。

So overall, sort of stepping back across the platform, very important to have a proportion that is in these committed capital partnerships for overall stability and resilience of our platform. And we think the level that we're at today is about right. And so holding that level constant along with the two other drivers, we do expect that to contribute to reduced equity capital usage on the platform on average over the next multi-quarter period.

所以整體來看，從平台層面退一步看，讓一定比例落在這些承諾資本合作關係中，對平台整體的穩定性與韌性非常重要。而我們認為目前的水準大致是合適的。因此，在維持該水準不變並搭配另外兩個驅動因素的情況下，我們確實預期在接下來跨越多個季度的期間，這將有助於平台平均股權資本使用量（equity capital usage）的下降。

Mihir Bhatia, Bank of America.

Mihir Bhatia，美國銀行（Bank of America）。

I wanted to ask -- I wanted to start with UMI and just go back to the impact on the business. I think you've quantified the impact on originations. But what about on the balance sheet? I think like to put a finer point on it, you've assumed UMI stays at or below 1.55. But if it moves to 1.6, can you just help us quantify the incremental balance sheet impact -- like how much of the credit deterioration is already reflected in fair value and how much could potentially come in more as UMI moves?

我想問——我想先從 UMI 談起，回到它對業務的影響。我認為你們已經量化了對放款量（originations）的影響。但資產負債表方面呢？更精確地說，你們假設 UMI 維持在 1.55 或以下。但如果它升到 1.6，你們能否協助我們量化額外的資產負債表影響——例如有多少信用惡化已經反映在公允價值（fair value）中，又有多少可能會在 UMI 上升時進一步反映進來？

Yes, it's a great question. And you can sort of see in our P&L and our balance sheet that even over the last couple of quarters in Q2 and Q3, we've been reflecting negative fair value write-downs associated with our risk capital position, so the beneficial interest. And so that line item moved from a positive $13 million in Q1 to negative $7 million, $8 million in each of Q2 and Q3, respectively, which is reflective of sort of credit and default pressure on those populations largely driven by the increase in UMI that we are seeing broadly.

是的，這是個很好的問題。你可以從我們的損益表與資產負債表看到，即使在最近幾季的第二季與第三季，我們已經反映了與我們風險資本部位（risk capital position）相關的負向公允價值減記（write-down），也就是受益權益（beneficial interest）。因此，該項目從第一季正的 1,300 萬美元，變成第二季與第三季分別為負的 700 萬、800 萬美元，這反映了在我們廣泛看到 UMI 上升的情況下，這些族群所承受的信用與違約壓力。

And so as we've guided toward an expectation in Q4, as UMI has risen to 1.53 over the course of October, our expectation is that in hitting that $260 million of adjusted EBITDA, UMI at 1.55 or below would allow us to kind of deliver that. We are sort of looking ahead, allowing for some degree of variability in the fair value mark as of the end of Q4. The flow-through from UMI to fair value is not necessarily a direct one, though, of course, it is highly correlated as UMI is a measure of the macro-driven default risk of the population. And so it has a real impact on the fair values and hard to quantify that exact amount specifically. And what we will say again is that in that $260 million of adjusted EBITDA for the full year, assuming UMI of 1.55 or below should be able to deliver that.

因此，當我們指引第四季的預期時，隨著 UMI 在 10 月期間上升至 1.53，我們的預期是：要達成 2.6 億美元的調整後 EBITDA（adjusted EBITDA），UMI 在 1.55 或以下將能讓我們大致交付該目標。我們也在往前看，並允許在第四季末的公允價值評價（fair value mark）上存在一定程度的變動性。UMI 對公允價值的傳導並不一定是直接的，當然它高度相關，因為 UMI 是衡量由總體因素驅動之族群違約風險的指標。因此它確實會對公允價值造成影響，但很難精確量化具體金額。我們再次要說的是，在全年 2.6 億美元的調整後 EBITDA 這個數字中，假設 UMI 為 1.55 或以下，應該能夠達成。

Got it. Okay. And then maybe just turning to credit and credit performance. I think you highlighted 660 basis point premium to Treasuries over the last 12 vintages. But maybe talk about the most recent vintages against your loss expectations as UMI has moved higher. Are they all currently running above expectations?

了解。好的。接著也許轉到信用與信用表現。我認為你們強調過，過去 12 個批次（vintages）相對於美國公債（Treasuries）有 660 個基點的溢價。但能否談談在 UMI 走高之際，最近幾個批次相對於你們的損失預期（loss expectations）表現如何？它們目前是否都高於預期在運行？

Thanks for the question on credit, Mihir. There's some variability in the credit performance vintage to vintage. And so we don't have like vintage-specific disclosures that we release. But you are generally right that all else equal, when UMI is shifting up, it means that our -- what we are observing is that the same borrower is more likely to be delinquent or default compared to an environment where UMI is lower. And so mechanically, that does tend to mean that loans that were underwritten in a lower UMI environment are going to have more delinquency or more default in a higher UMI environment, just sort of, I think, kind of mathematically how that mechanic works.

Mihir，感謝你關於信用的提問。信用表現會因不同放款批次（vintage）而有所差異。因此我們不會發布按批次劃分的特定揭露。但你一般而言是對的：在其他條件相同下，當 UMI 上升時，我們觀察到同一位借款人在相較於 UMI 較低的環境下，更可能發生逾期或違約。因此從機制上來看，在較低 UMI 環境下承作的貸款，進入較高 UMI 環境時，確實會出現更多逾期或更多違約；我想這在數學上大致就是這個機制的運作方式。

But having said that, I think there's a couple of important things to keep in mind. One is that we really think about the accuracy of our models as being relative to UMI. And that's because, ultimately, again, we don't control or predict what's going on in the background macro conditions. What we can do is be the fastest and most precise in responding to macro, whereas kind of traditional credit metrics give you a bit of a lagged or a compounded view of what's going on in true macro as opposed to maybe what our underlying mix effects or timing effects or cohort effects. The UMI system is really good at giving us a clear read of what is really attributable to macro.

不過話雖如此，我認為有幾個重要事項需要記住。第一，我們看待模型準確度時，會以相對於 UMI 的表現來衡量。因為歸根究柢，我們無法控制或預測背景的總體經濟狀況。我們能做的是以最快、最精準的方式回應總經變化；相較之下，傳統信用指標往往會讓你看到較滯後或被多重因素疊加後的總經狀況，而不是我們底層的組合效應、時間效應或同批次效應。UMI 系統非常擅長讓我們清楚判讀哪些變化真正可歸因於總經。

And so we can get that faster and more precise response that means newer vintages can get underwritten correctly on a much faster time line. And then we can really judge how our models performed relative to what's gone on in the UMI. So yes, if UMI stays elevated at 1.53 or 1.55 for the rest of the life of these loans, and they're going to have higher delinquencies than if it had been at 1.5 or 1.4 for sure. But we think separation accuracy, the thing that really our models are best-in-class, that's something that matters in any kind of macro environment.

因此我們能更快、更精準地回應，讓較新的放款批次能在更短的時間軸上被正確承作。接著我們也能真正評估模型相對於 UMI 變化的表現。所以是的，如果 UMI 在這些貸款的剩餘存續期間都維持在 1.53 或 1.55 的高位，那它們的逾期率一定會高於 UMI 若維持在 1.5 或 1.4 的情況。但我們認為「區隔準確度」（separation accuracy）——也就是我們模型同級最佳的核心能力——在任何總經環境下都很重要。

And then the second thing I would say is we talk about returns relative to Treasuries as a benchmark because really, it's that spread, that's the number that investors are really earning. That's really the sort of thing you get by partnering with Upstart and buying our loans. And really, by any measure, I think those spreads are really healthy, really good. And as you can see ultimately that investors and capital partners are still really excited to partner with us and buy our loans, and we're still continuing to grow and improve those programs, and that's the ultimate testament that we're delivering good risk-adjusted returns.

第二點我想說的是，我們會以美國公債殖利率作為基準來談報酬，因為真正投資人賺到的是相對於公債的利差（spread）。這基本上就是與 Upstart 合作、購買我們貸款所能獲得的報酬來源。而且我認為以任何衡量方式來看，這些利差都相當健康、相當不錯。最終你也可以看到，投資人與資金合作夥伴仍然非常期待與我們合作並購買我們的貸款；我們也持續擴大並改善這些計畫。這就是我們正在提供良好風險調整後報酬的最佳證明。

Arvind Ramnani, Truist.

Arvind Ramnani，Truist。

I just wanted to ask like as these AI models are getting bigger and kind of more complex, how should we think about kind of compute cost per loan? Does it really kind of scale with your originations? Or does it really come down with volume?

我想問的是，隨著這些 AI 模型變得更大、也更複雜，我們應該如何看待每筆貸款的運算成本（compute cost）？它真的會隨著你們的放款量（originations）一起擴張嗎？還是會因為量變大而下降？

Yes. It's actually a great question. And it's interesting that you asked because I think typically in lending businesses, compute costs are not a real thing to worry about because mostly if you're using kind of a traditional credit model, it's just not going to be that expensive to compute and do inference. In our case, inference is a real cost, training is a real cost. And these are things we think about, and these are things that have shown up in some of our financials in 2026, and it's something that we actively put a lot of effort into optimizing.

會的。這其實是個很好的問題。你問到這點很有意思，因為我認為在典型的放貸業務中，運算成本通常不是需要擔心的事情；因為如果你用的是傳統信用模型，計算與推論（inference）通常不會太昂貴。但在我們的情況下，推論是一項實際成本，訓練（training）也是一項實際成本。這些都是我們會思考的事情，也確實在我們 2026 年的一些財務數字中有所反映；而且我們正投入大量精力積極優化。

And so that's -- there's a bunch of efforts on the training side. There's a bunch of efforts on the inference side. Some of that is really kind of like low-level stuff in making the actual computations more efficient, and we just -- we do a lot of computation. So it ends up getting pretty expensive. I think we actually did make a lot of progress on this over the course of the year.

因此在訓練端有一系列的工作。在推論端也有一系列的工作。其中一部分是偏底層的工作，例如讓實際計算更有效率；而我們確實做了大量計算。所以最後成本會變得相當高。我認為我們在今年的過程中，確實在這方面取得了很大進展。

It's been a real focus area for us. And I think that's one of the things that when Andrea talks about us getting more operating leverage relative to sort of the OpEx growth that we expect to see. This is one of the things that is getting better and we feel better about. So it's actually, I think, going to be increasingly an area of competitive advantage for us. It's something we've invested in that I think most others in the lending world just haven't had too much reason to think about yet.

這一直是我們非常聚焦的領域。我認為這也是 Andrea 談到我們相對於預期的營運費用（OpEx）成長，能取得更多營運槓桿（operating leverage）的原因之一。這是正在改善、而且我們感覺更有把握的一項。所以我認為這其實會愈來愈成為我們的競爭優勢。這是我們已經投入的領域，而我認為放貸世界中的多數其他業者，至今還沒有太多理由去思考這件事。

Yes. No, that's great. And I guess as far as the investment, do you own some of the infrastructure here? Or are you kind of using outside vendors, particularly on the compute side?

好的，這很棒。那我想再問一下，關於這些投資，你們是否擁有部分基礎設施？還是主要使用外部供應商，特別是在運算資源這一塊？

We use cloud vendors. Yes, we don't have our own on-prem servers.

我們使用雲端供應商。是的，我們沒有自建的本地端（on-prem）伺服器。

Reggie Smith, Loop Capital.

Reggie Smith，Loop Capital。

Really good numbers on the fee take rate. I know you don't drive the business that way, Paul. But just curious, thinking about kind of recent origination trends. And I'm curious about the duration of the loans and whether or not you've seen a lengthening of duration because I believe that, that can impact kind of upfront economics for you guys.

費用抽成率（fee take rate）的數字非常好。我知道你們不是用那種方式來經營業務的，Paul。但我想問一下，從近期放款趨勢來看。我想了解貸款的存續期間（duration），以及你們是否看到存續期間拉長；因為我相信這可能會影響你們的前端經濟效益（upfront economics）。

And also curious about thinking about like where, I guess, your weighted average coupon is for some of these core loans today and how much headroom you have to raise rates if treasury yields continue to rise and being mindful of kind of APR caps and things like that. So anything you could tell us about the duration mix of new originations and the weighted average capital coupon rather would be helpful. And then I have one UMI follow-up.

另外也想請教，目前你們一些核心貸款（core loans）的加權平均票息（weighted average coupon）大概在哪裡？如果公債殖利率持續上升，在留意 APR 上限（APR caps）等因素下，你們還有多少調升利率的空間？所以如果你們能分享新承作放款的存續期間組合，以及加權平均資金票息（更準確地說是資金端 coupon）的資訊，會很有幫助。然後我還有一個關於 UMI 的追問。

Thanks, Reggie. Yes, on the overall duration, it's been relatively consistent. So our personal loans have about a 1.8-year weighted average life, auto, about two, home about three years. So maybe a little bit of modest mix shift to secured has extended the duration very modestly, but it's not particularly visible on an aggregate basis. And then your question on -- can you remind me the second part of your question?

謝謝你，Reggie。是的，整體存續期間方面，一直相對一致。所以我們的個人貸款加權平均存續期（weighted average life）約 1.8 年，車貸約 2 年，房貸約 3 年。因此，或許因為組合略微轉向有擔保（secured），使存續期間非常小幅地延長，但在整體彙總層面並不特別明顯。然後你第二個問題是——可以提醒我一下第二部分問的是什麼嗎？

Weighted average -- real quick, sorry -- just on duration, I'm really focused on the core loan product. And so I'm curious whether that duration in terms of new originations has lengthened at all as you've seen better fee revenue.

加權平均——很快補充一下，抱歉——在存續期間方面，我主要關注的是核心貸款產品。所以我想知道，就新承作放款而言，當你們看到更好的費用收入時，存續期間是否有任何拉長？

No, no, that's been relatively consistent. And then on the weighted average coupon front, you are right that we are originating given the borrower that we are often focused on in the core at higher APRs that often approach a 36% interest rate cap. And so kind of the sensitivity to our origination volume related to that is all embedded in the UMI and funding cost heuristic that Paul provided. So about a 10 basis point change in funding cost that flows through the platform or about a 1 point change in UMI would be expected to drive a 1% to 2% change in originations. And that sort of sensitivity and relationship includes several things, but also includes the impact of where we sit from an APR perspective relative to the cap.

沒有，沒有，那一直相對一致。至於加權平均票息方面，你說得對：就我們在核心業務中經常聚焦的借款人而言，我們承作的 APR 較高，且常常接近 36% 的利率上限。因此，與此相關的放款量敏感度，都已包含在 Paul 提供的 UMI 與資金成本啟發式（heuristic）之中。也就是說，若平台端資金成本變動約 10 個基點，或 UMI 變動約 1 點，預期會帶動放款量變動約 1% 到 2%。而這種敏感度與關係包含多項因素，也包含我們在 APR 角度相對於上限的位置所帶來的影響。

Okay. That makes sense. And if I could sneak one last one in on the UMI. And just thinking about unemployment levels, they're still fairly low from a historic perspective and UMI is already at kind of 1.53. How should we think about that sensitivity if you were to see an increase in unemployment, like how might that impact UMI?

好的。這很合理。另外如果我可以在 UMI 上再塞最後一個問題。從失業水準來看，從歷史角度仍然相當低，而 UMI 已經大約在 1.53。若未來失業率上升，我們應該如何看待這種敏感度？例如失業率上升可能會如何影響 UMI？

Yes, it's a great question. Whenever we talk about macro, I think there's always some confusion around what macro is for us and which parts of macro really matter. People always say things like, oh, well, the American consumer seems to be really healthy. People are still spending lots of money and buying lots of stuff. And I'm always like, well, we don't like it when people are buying lots of stuff. That's actually not the macro that is good for us.

是的，這是個很好的問題。每當我們談到總體環境（macro）時，我覺得大家常常會混淆：對我們而言 macro 是什麼，以及 macro 的哪些部分才真正重要。大家總會說，像是，美國消費者看起來非常健康。人們仍在花很多錢、買很多東西。而我總會說，嗯，我們其實不喜歡人們買很多東西。那其實不是對我們有利的那種總體環境。

And I would say recessions and employment, sort of the typical measures that people talk about are probably somewhere in between. They certainly do matter to us. But they by themselves probably aren't the whole story or the most important story. What we really have been very focused on are things like real wages, how much money is coming in versus how much is going out. Interest rates obviously matter.

我會說，衰退與就業這些人們常談的典型衡量指標，大概落在中間。它們確實對我們有影響。但單靠它們本身，可能不是全貌，也可能不是最重要的故事。我們真正非常關注的是像實質薪資這類因素：進來的錢相對於出去的錢有多少。利率當然也很重要。

We talked about that today. And so I think typically, in a recession, if you see unemployment go up a couple of percent, that's certainly all else equal, a bad thing, but then you'll typically also see inflation lower, you'll see interest rates lower, and those things will be a real positive. And depending on the magnitude of those things, it could end up being sort of positive or negative to the overall business. And that's just because if unemployment affects narrowly only, say, you get -- if you get unemployment going from 4% to 6%, for example, that's 2-percentage-points of the population that are negatively affected.

我們今天也談到了這點。因此我認為通常在衰退中，如果你看到失業率上升幾個百分點，當然在其他條件不變下是壞事，但同時你通常也會看到通膨下降、利率下降，而這些會是很大的正面因素。而視這些因素的幅度而定，對整體業務最後可能是正面也可能是負面。原因在於，如果失業的影響相對狹窄——例如失業率從 4% 升到 6%，那就是有 2 個百分點的人口受到負面影響。

But then the other 98% are getting the benefit of maybe lower inflation and lower interest rates, you could easily see how the math could go either way on that. And so we don't tend to be hyper focused on that as a metric. We do spend a lot of time thinking about things like real wages, the sort of personal savings rate, interest rates. And I think those things have been historically a much bigger driver of what kind of macro matters to us.

但另外 98% 的人可能會受惠於較低的通膨與較低的利率，你很容易理解這樣的數學結果可能兩邊都說得通。因此我們通常不會把它當作一個過度聚焦的指標。我們確實花很多時間思考像實質薪資、個人儲蓄率、利率等因素。而我認為從歷史上看，這些因素對於什麼樣的總體環境會影響我們，是更大的驅動因子。

James Faucette, Morgan Stanley.

James Faucette，摩根士丹利。

It's Michael Infante on for James. Just circling back on the fair value marks again, I think when I look at the underlying discount rate disclosure in terms of the loans that are actually held in consolidated securitizations, I think it's closer to 11.75%, still roughly, call it, 200 basis points below the late '23 peak. So how should we be thinking about the incremental earnings pressure from further discount rate increases, particularly as we think about constructing our models out to '27?

我是 Michael Infante，代 James 提問。再回到公允價值評價（fair value marks）這件事，我想當我看你們披露的底層折現率（discount rate），就實際持有在合併證券化（consolidated securitizations）中的貸款而言，我認為大約更接近 11.75%，仍大概比 2023 年底的高點低約 200 個基點。那麼我們應該如何看待折現率進一步上升所帶來的增量獲利壓力，特別是當我們把模型往 2027 年建構時？

And then secondly, just on the beneficial interest dynamics, like can you just comment on how your cash contribution or maximum loss -- maximum loss exposure per dollar has sort of changed when we look at this gradual increase in beneficial assets as a percentage of associated loan balances?

第二個問題，關於受益權益（beneficial interest）的動態：你們能否評論一下，當我們看到受益資產占相關貸款餘額的比例逐步上升時，你們每投入 1 美元的現金貢獻（cash contribution）或最大損失——每 1 美元的最大損失曝險（maximum loss exposure）——是如何變化的？

Yes, absolutely. I'll take that one. So on the fair value sensitivity to discount rates, a good place to look is in our Q, we sort of share the sensitivity to about 100 basis point move in discount rates. In aggregate across both our whole loan book and the beneficial interest, that is about $16 million or so. So in the scheme of things, relatively smaller impact to our fair value are changes in discount rates and a relatively larger impact comes obviously from changes in expectations around credit performance. But you can kind of look to that heuristic as helpful to help you kind of size potential impacts on a go forward.

好的，當然可以。我來回答這題。關於折現率變動對公允價值的敏感度，一個很好的參考點是在我們的 10-Q 裡，我們有分享折現率變動約 100 個基點時的敏感度。把整體（包含整個全額貸款組合與受益權益）加總起來，大約是 1,600 萬美元左右。因此就整體而言，折現率變動對我們公允價值的影響相對較小；相對更大的影響顯然來自於對信用表現預期的變化。但你可以把這個經驗法則當作一個有用的參考，幫助你估算未來可能的影響幅度。

In terms of the cash contribution as a percent of originations, so that number, if you look at on a year-to-date basis, so through the first three quarters of the year, it's been about 3% on the total origination volume. So we've acquired about $400 million in beneficial interest assets about just south of that against just south of $12 billion or so of origination volume. And so you can sort of see that marginal use is lower than the overall aggregate platform equity use and expect to -- especially as the older deals start to cash flow more to contribute to the reduction in equity capital used on average across the total platform.

至於現金貢獻占承作量（originations）的比例，如果你看年初至今的數字——也就是今年前三季——大約是總承作量的 3%。我們取得了大約 4 億美元的受益權益資產（略低於 4 億），對應的承作量略低於 120 億美元左右。因此你可以看到，邊際上的資金使用低於平台整體的股權使用水準；並且預期——特別是隨著較早期的交易開始產生更多現金流——將有助於降低整體平台平均使用的股權資本。

That's really helpful. And then, Paul, just a quick follow-up on the conversion rate. I saw that you guys intend to discontinue that conversion rate disclosure. Can you just elaborate on why? Because I just -- I'm curious, just given it seems to at least reinforce like the selectivity of your model and the available growth that is still out there as you drive incremental separation. So I'm curious if you could elaborate on that specifically.

這非常有幫助。然後 Paul，再追問一個關於轉換率（conversion rate）的問題。我看到你們打算停止披露轉換率。你能否多說明一下原因？因為我只是好奇，畢竟它看起來至少能強化你們模型的選擇性，以及在你們推動進一步差異化（incremental separation）時，市場上仍存在的成長空間。所以我想請你特別針對這點多做說明。

Yes, sure. I think the conversion rate was really a good metric back when we were kind of like a single product, single-use company. And really like what we did was we acquired new people for personal loans. And really, they were kind of personal loans in one segment, the segment we now call core. And I think it was a pretty useful metric back then. Increasingly, as we've had new products enter the mix, new segments of borrowers enter the mix, both on the prime and the less prime end of products that just have really like different conversion dynamics.

好的，當然。我認為轉換率在我們還算是單一產品、單一用途的公司時，確實是一個很好的指標。當時我們做的事情基本上就是為個人貸款獲取新客。而且主要是個人貸款的一個細分市場，也就是我們現在稱為 core 的那個區隔。我認為那時候它是相當有用的指標。但隨著我們的產品組合加入新產品、借款人區隔也加入新的族群，無論是優質（prime）或較非優質（less prime）端，不同產品的轉換動態其實差異很大。

And then we have this -- as we talked about today, we've got 250,000 people in the quarter that came back for a loan that was their second or third or fourth or nth loan. And those dynamics really kind of start confusing what this conversion rate thing really measures. And it no longer kind of obviously is like up is better or down is better. It just sort of became a metric that unless you controlled for all of these different factors was not going to tell you a clear story.

另外我們也有——如同今天提到的——本季有 25 萬人回來借款，那是他們的第二筆、第三筆、第四筆或第 N 筆貸款。這些動態會讓「轉換率」這個指標到底在衡量什麼變得更混亂。它也不再那麼明顯是「越高越好」或「越低越好」。除非你能控制所有這些不同因素，否則它就不會是一個能講出清楚故事的指標。

And so that's why we're kind of focusing the shifting to some of these different metrics, especially metrics that tell you about this kind of repeat borrower ecosystem, how many people are -- how many new people are coming, how many returning people are coming and then, of course, sort of ultimately, you'll see the kind of product-specific growth metrics.

因此這就是為什麼我們把重點轉向一些不同的指標，特別是能反映這種重複借款人（repeat borrower）生態系的指標：有多少人是新客、有多少人是回流客；然後當然，最終你也會看到各產品別的成長指標。

Kyle Joseph, Stephens.

Kyle Joseph，Stephens。

Appreciate all the color you guys provided on the bank, and I know it's too early, but from a high level, how are you guys thinking about the potential margin expansion possible there once that deal is done and how quickly that will ramp? And then a follow-up there is, obviously, really looks like it strengthens the balance sheet and your liquidity. Any sort of updates on capital allocation?

感謝你們提供關於銀行的各種細節。我知道現在還太早，但從高層次來看，一旦交易完成，你們如何看待那邊可能帶來的利差擴張（margin expansion），以及提升速度會有多快？接著一個追問是，很明顯這看起來會強化資產負債表與流動性。在資本配置（capital allocation）方面有沒有任何更新？

Great. Thanks for that question. Yes, I think without sharing -- we're not sharing anything specific about the exact impact to margins, but -- and we are very excited about the opportunity -- strategic opportunity that the bank presents to us to kind of improve our position across a number of dimensions. So one, we expect the bank to improve the overall TAM for our business that we can underwrite all products in every state at the same rate where we're limited today in some places in how we do that.

很好。謝謝你的問題。是的，我想在不分享——我們不會分享任何關於對利差影響的具體數字——但我們對這個機會非常興奮；這家銀行為我們帶來的策略性機會，能在多個面向改善我們的整體位置。第一，我們預期這家銀行能提升我們業務的整體 TAM（可服務市場總量），因為我們可以在每個州以相同的方式承作所有產品；而目前在某些地區，我們在承作方式上仍受到限制。

Two, as we alluded to before, we do expect to see modest uptick in take rates and improvement in net interest income driven by the fact that we'll be capturing economics that today we're currently paying out to the banks that are originating partners for us. Third, we do expect to see lower cost of capital with deposit funding versus warehouse funding. And then overall, we expect to see a reduction in sort of overall complexity, as we operate under a single regulatory regime versus via the dozens to hundreds of banking partners that we're operating with today. So all of that in aggregate are the benefit that we expect to see from the bank. Some of that will come in relatively quickly.

第二，如同我們先前提到的，我們確實預期承作費率（take rates）會小幅上升，且淨利息收入將有所改善，原因在於我們將能夠攫取目前我們支付給作為我們合作放款起始方之銀行的經濟利益。第三，相較於倉儲融資（warehouse funding），以存款資金（deposit funding）作為資金來源，我們確實預期資本成本會更低。此外，整體而言，我們預期整體複雜度會降低，因為我們將在單一監管制度下運作，而不是像現在一樣透過數十到數百家銀行合作夥伴來運作。因此，綜合以上，這些都是我們預期從銀行帶來的效益。其中一部分將會相對快速地實現。

So as we alluded to, we do expect that very shortly after the launch of the bank, Upstart Bank will start originating substantially all of the loan volume on our platform. And so that component will be realized more quickly. And then some of these other components will be realized over a longer period of time as we access that greater TAM and drive the operational efficiencies that come with operating as a bank.

因此，如同我們所提到的，我們確實預期在銀行推出後不久，Upstart Bank 就會開始承作我們平台上幾乎所有的放款量。因此，那一部分將會更快實現。而其他一些部分，則會在較長時間內逐步實現，因為我們將能觸及更大的可服務總市場（TAM），並推動作為銀行運作所帶來的營運效率。

In aggregate, our three-year guide contemplates us getting the bank charter, and that's incorporated in. And fundamentally, our core business model with respect to acting as a marketplace and selling the majority of our loans through to third parties, that also is not intended to change as we become a bank.

整體而言，我們的三年指引已將取得銀行特許執照（bank charter）納入考量，並已反映在其中。而從根本上來看，就我們作為市場平台運作、並將大多數貸款出售給第三方的核心商業模式而言，即使我們成為銀行，也不打算改變。

Giuliano Bologna, Compass Point.

Giuliano Bologna，Compass Point。

One thing I'd be curious about, as a quick question, it seems like the revenue from fees on the secured side jumped close to 50% quarter-over-quarter. It seems like there's maybe some timing impacts of loan sales. But just putting that over volume comes out to, I believe, 5.8%. Is there a rough way to think about how that should trend? And then how long it should take before it kind of comes back to the older kind of rough outlook that you guys have put out between the upfront fees and then the servicing strip?

我有件事想快速請教一下：看起來擔保端（secured side）的手續費收入季增接近 50%。看起來可能有一些放款出售時點（timing）影響。但如果把它除以放款量，我認為大約是 5.8%。有沒有一個大致的方式可以理解這個比率應該如何走勢？另外，大概需要多久才會回到你們先前提出的、介於前端費用（upfront fees）與後續服務分成（servicing strip）之間的那種較早期的粗略展望？

Yes. I would say, we've been really focused on getting the secured loans to profitable, and that move is a very real move. But there definitely are some -- there are some sort of like timing elements to it, and we've alluded to needing to spend a little more time kind of locking in all the margin gains here because some of them are a function of things that haven't been totally systematized. That's some stuff on the loan sales side, some stuff on the monetization side. And so there's still work to do, but the sort of -- the gains are very much real. We do think that these products have reached contribution profitable in a sort of sustainable way, and we expect that to continue through the rest of this year and into 2027.

是的。我會說，我們一直非常專注於讓擔保貸款達到獲利，而這個轉變是非常真實的轉變。但確實也有一些——有一些類似時點因素的影響；而且我們也提到過，為了把這些利潤率提升真正「鎖定」下來，我們還需要多花一些時間，因為其中一部分是來自尚未完全系統化的事項。這包含放款出售端的一些事情，也包含變現（monetization）端的一些事情。因此仍有工作要做，但這些——這些改善是非常真實的。我們確實認為這些產品已以可持續的方式達到貢獻獲利（contribution profitable），並預期這種狀況會在今年剩餘時間延續並延伸到 2027 年。

We will expect to see a little bit of sort of movement around that margin. We're not so focused on maximizing that margin today. So I said in my remarks earlier, anywhere between like 0% and 25% to us is contribution positive. It's a good number. It's where we think these products should be at this stage in their life cycle. And what we'll want to see them do starting in 2027 is increasingly return their focus to growing the volumes at a profitable margin.

我們預期那個利潤率會有一些小幅波動。我們目前並沒有那麼專注於把利潤率最大化。所以我在先前的談話中提到，對我們而言，大約 0% 到 25% 的區間都屬於貢獻為正（contribution positive）。這是一個不錯的數字。也符合我們認為這些產品在其生命週期此階段應有的水準。而我們希望從 2027 年開始看到的是，它們會逐步把重心轉回在可獲利的利潤率下擴大放款量。

Rob Wildhack, Autonomous Research.

Rob Wildhack，Autonomous Research。

I saw that auto volumes were down quarter-over-quarter. The product had been scaling really quickly prior to that. I would not really think of the third quarter as being a seasonal trough for auto. So just wondering if there's anything that you would call out or anything to read in there?

我看到汽車貸款量季減。在此之前，這個產品的規模擴張得非常快。我不太會把第三季視為汽車貸款的季節性谷底。所以想請問你們是否有任何特別想指出的原因，或這裡面有沒有什麼值得解讀的訊號？

80% of this answer is like we told everyone just focused on profitability. And I said that last quarter, that's just the one thing we're going to focus all of our energy on. And that's what the team did, and I'm really proud of how they executed on that. So we have no complaints about where that landed. Maybe there's a 20% part of the answer, which is that we talked a lot about macro and people tend to think about that in the context of our core business, but it's certainly relevant across all of the loan categories.

這個問題的 80% 答案就是：我們跟大家說過，我們只專注在獲利。而我上季也說過，這就是我們要把所有精力投入的唯一一件事。團隊也確實這麼做了，我對他們的執行成果非常自豪。所以我們對目前的結果沒有任何抱怨。另外可能有 20% 的部分是：我們談了很多總體環境（macro），大家往往會在我們核心業務的脈絡下思考它，但它當然也與所有貸款類別都相關。

Interest rates are a factor and sort of borrower health is a factor. And so those have an effect, too. But I think in a category like this, that's pretty young for us, sort of the most important thing is going to be where our focus is and our focus has been on profitability. And as we get into '27, we'll start turning it back the other way.

利率是一個因素，借款人健康狀況也是一個因素。所以這些也會產生影響。但我認為在像這樣對我們而言還相當新的類別中，最重要的會是我們的重點放在哪裡，而我們的重點一直放在獲利。等到進入 2027 年，我們會開始把方向再轉回去。

Okay. And then just one more on that earlier dialogue between the UMI and fair value. If I just look at the fourth quarter, the guidance implies a pretty healthy negative number for fair value. But if I heard you correctly, you're saying the UMI is only going to be up maybe a couple of hundredths of a point, right, max 1.55. So it just seems like a much more dramatic swing in fair value for the fourth quarter versus a much smaller increase in the UMI. I'm wondering if there's any more color you can give on the sensitivity and the mechanics there. And then is the bigger driver of like realized charge-offs for on-balance sheet loans or fair value markdowns?

好的。那再追問一個關於先前 UMI 與公允價值（fair value）之間對話的問題。如果我只看第四季，你們的指引似乎暗示公允價值會是一個相當明顯的負數。但如果我理解正確，你的意思是 UMI 可能只會上升幾個百分點的小數位，對吧，最多到 1.55。這看起來像是第四季公允價值的波動幅度遠大於 UMI 的小幅上升。我想了解你是否能再補充一些敏感度與運作機制（mechanics）的細節。另外，更大的驅動因素是資產負債表內貸款的已實現核銷（realized charge-offs），還是公允價值減記（fair value markdowns）？

Yes, it's a great question, Rob. And so the fair value that we're implying for Q4 reflects both an expectation of UMI about 1.55 and allows -- we're allowing for some variability around that because there are -- when it comes to the fair value line, there's inherently a little more variability around it than our ability to predict kind of fee revenue and contribution profit. Ultimately, the rising UMI environment that we are seeing here in Q4, but also the higher UMI environment we've seen over the course of most of 2026 is contributing to kind of building pressure with macro-driven defaults on our borrower population.

是的，這是個很好的問題，Rob。因此，我們對第四季所暗示的公允價值，同時反映了對 UMI 約 1.55 的預期，並且——我們也為其保留了一些波動空間，因為在公允價值這一行項目上，本質上就比我們對手續費收入與貢獻利潤的預測能力更具變動性。最終而言，我們在第四季看到的 UMI 上升環境，以及 2026 年大部分時間所處的較高 UMI 環境，正在透過總體因素驅動的違約（macro-driven defaults），對我們的借款人族群形成逐步累積的壓力。

And so the fair value marks start to reflect that pressure as we see delinquencies and charge-offs realized in our overall portfolio. And so that's kind of the, I would say, the indirect mechanism by which UMI flows through into our visibility and understanding of the default risk of our population, which then is the direct flow-through to the fair value marks. As we look at the composition of the fair value -- NII and fair value number that we are implying for Q4, most of that is related to fair value write-downs on the beneficial interest assets, so the loans underlying those assets, that would be the primary expected driver as of right now.

因此，當我們在整體投資組合中看到逾期（delinquencies）與核銷（charge-offs）逐步實現時，公允價值評價（fair value marks）就開始反映這種壓力。所以我會說，這就是 UMI 透過一種間接機制，影響我們對族群違約風險的可見度與理解，而後者再直接傳導到公允價值評價。就我們對第四季所暗示的公允價值——淨利息收入（NII）與公允價值數字的組成來看，其中大部分與受益權資產（beneficial interest assets）的公允價值減記（write-downs）有關，也就是這些資產所對應的底層貸款；就目前而言，這會是主要的預期驅動因素。

With no other questions from the phone lines, I'd like to turn the call back to Sonya Banerjee.

電話線上沒有其他問題的話，我想把電話會議交回給 Sonya Banerjee。

Thank you. Just as we did last quarter, we invited retail investors to submit questions through X. We received a number of thoughtful submissions, and we've selected a few to answer today. A number of those questions were also addressed during the analyst Q&A portion. But certainly, we welcome follow-ups if people are curious.

謝謝。就像上個季度一樣，我們邀請散戶投資人透過 X 提交問題。我們收到了許多深思熟慮的提問，並挑選了幾個在今天回答。其中不少問題也在分析師問答環節中被回答過。不過，如果大家還有好奇之處，我們當然也歡迎後續追問。

I'll start with Andrea. As earlier co-investment vintages begin returning cash, is the net new capital required to support forward flow commitments declining? And when do you expect those returns to fund a majority of new co-investments?

我先從 Andrea 開始。隨著較早期的共同投資批次（vintages）開始回收現金，用於支持遠期流量（forward flow）承諾所需的淨新增資本是否正在下降？你們預期何時這些回收款能夠資助大多數新的共同投資？

Yes, it's a great question. Just kind of going back to a few things that we touched on during this call. Over a multi-quarter time horizon, with obviously some variability quarter-on-quarter, we do expect the overall equity usage to support the platform to decline. And one of the key drivers of that will be related to our co-investment positions in committed capital relationships. And so the three drivers that we see to reduce the equity required or used from those positions are sort of, one, holding the funding mix about constant. So we don't expect to materially increase the proportion of our platform that is funded from committed capital from where it is today.

是的，這是個很好的問題。回到我們在本次電話會議中提到的幾點。在跨多個季度的時間範圍內（當然各季度之間會有一些波動），我們確實預期用於支持平台的整體股權使用量會下降。其中一個關鍵驅動因素，將與我們在承諾資本（committed capital）合作關係中的共同投資部位有關。因此，我們看到能降低這些部位所需或所使用股權的三個驅動因素：第一，將資金來源組合大致維持不變。也就是說，我們不預期會在現有基礎上，實質提高由承諾資本為平台提供資金的比例。

Two, as the question alludes to, continue to expect that sort of earlier vintages of co-invests as they release cash will help reduce the equity usage. And three, over time, would expect to continue to improve the deal terms as we renew our committed capital partnerships. So all of those things in combination are expected to be key drivers of bringing both the co-invest percent of the overall platform down, but importantly, the overall equities required to support the platform down.

第二，如問題所暗示的，我們仍預期較早期的共同投資批次在釋放現金時，將有助於降低股權使用量。第三，隨著我們續約承諾資本合作夥伴關係，長期而言我們預期會持續改善交易條款。上述因素綜合作用，預期將成為關鍵驅動力：一方面降低共同投資占整體平台的比例，更重要的是，降低支持平台所需的整體股權。

Paul, as consumers start comparing and applying for loans through AI assistants like ChatGPT and Claude, does Upstart plan to let these external AI agents directly access your pricing and underwriting services? Will you build your own AI agents to proactively grow the business and integrate with other companies?

Paul，當消費者開始透過 ChatGPT、Claude 等 AI 助理比較並申請貸款時，Upstart 是否計畫讓這些外部 AI 代理（agents）直接存取你們的定價與承保（underwriting）服務？你們會建立自己的 AI 代理來主動推動業務成長，並與其他公司整合嗎？

Yes. I love this category of question because I think there's just so many companies right now that are kind of like looking at what people are doing with AI, with personal agents and trying to build sort of these walled gardens and say like, oh, man, like if people could just freely shop for the best product, they might stop being loyal to us. And I think that just says something terrible about companies that think that way is they're not confident in the real value add that their product gives to their customers. And I think one thing that will always be really, really centrally important to us is we just want to build the best product. We think we have the best product in a lot of cases. We want to make it even better.

會的。我很喜歡這類問題，因為我認為現在有太多公司在觀察人們如何使用 AI、個人代理，然後試圖打造某種「圍牆花園」，並說：哎呀，如果人們可以自由地挑選最好的產品，他們可能就不再忠誠於我們了。我覺得這種想法其實反映出一件很糟糕的事：抱持這種想法的公司，對自家產品能為客戶帶來的真正附加價值並沒有信心。而對我們來說，永遠、永遠最核心重要的一點是：我們只想打造最好的產品。在很多情況下，我們認為我們已經有最好的產品。我們也希望把它做得更好。

We want to build the best product in a lot of our emerging categories. And so we want to just have no fear around like how the customer ultimately finds our product, whether it's on our surface or whether it's via a Muse or a Grok or whatever sort of they're using out there, we want our product to be everywhere, and we want to compete and win on having the objectively best product.

我們也希望在許多新興類別中打造最好的產品。因此，對於客戶最終如何找到我們的產品，我們完全不會感到害怕——不論是在我們自己的介面上，或是透過 Muse、Grok 或其他他們正在使用的工具；我們希望我們的產品無所不在，並且以客觀上最好的產品來競爭並勝出。

And so we think that the era of personal agents and personal AI is going to usher in a world where people just have much more ability to shop for the very best product, and we think that's going to help whoever has the best product to offer the customer, and we think we're going to win in that world.

因此，我們認為個人代理與個人 AI 的時代，將帶來一個世界：人們將更有能力去挑選真正最好的產品；而這會幫助那些能為客戶提供最佳產品的公司。我們相信在那樣的世界裡，我們會勝出。

And then the final question is the market still values Upstart like a cyclical lender rather than an AI platform. What milestones over the next two to three years do you think will force investors to change that view?

最後一個問題是：市場仍然把 Upstart 當作循環性放款機構來估值，而不是 AI 平台。你認為未來兩到三年有哪些里程碑，會迫使投資人改變這種看法？

There's so many things I'm excited about if we're talking a multiyear time frame, two to three years. There's all these things we've talked about, refocusing on the core and the sort of the strength that's already driving in our contribution profits, but I think that's really going to prove out how much alpha there is in this business, prove out the technology differentiation because that's really where we shine. There's this whole piece we've talked about today about lowering the equity usage in the business, lowering the sensitivity of the balance sheet P&L lines. I think that's really going to help investors. I think there's this whole kind of change in the company going on right now to go from just really a one-product company to having multiple products that are now contributing profits.

如果我們談的是兩到三年的多年期時間框架，有太多事情讓我感到興奮。我們談過的各種事項，包括重新聚焦核心業務，以及已經在推動我們貢獻利潤（contribution profits）的既有優勢；但我認為這將真正驗證這個業務能創造多少超額報酬（alpha），並驗證技術差異化，因為那正是我們最突出的地方。我們今天也談到一整塊內容：降低業務中的股權使用量，降低資產負債表損益（P&L）科目的敏感度。我認為這會對投資人很有幫助。此外，公司目前也正在發生一種轉變：從本質上是一家單一產品公司，轉向擁有多項產品，而這些產品現在都在貢獻利潤。

There's this customer ecosystem and sort of the growing importance of how many people already have a relationship with Upstart. I mean there are so many pieces of it that I think investors will start to appreciate over a couple of years and kind of see how rich the story around this company is. But actually, I could say all that, but I actually just don't think altogether, it's that complicated the answer to this question because in some sense, businesses are just ultimately valued on their expectations for future profits.

還有客戶生態系，以及越來越重要的一點：已經有多少人與 Upstart 建立了關係。我意思是，這裡面有太多面向，我認為投資人會在幾年內逐步開始欣賞，並看到這家公司故事的豐富性。不過我可以說很多，但其實我也不覺得這題的答案有那麼複雜，因為某種程度上，企業最終的估值就是基於對未來利潤的預期。

And I think especially in lending, lending is arguably the main profit center of financial services. And someone just today, I think, looking at us would say, you've been around for 14 years, and you haven't really made any meaningful profits yet. Like what's the point of all this fancy AI if it doesn't make money.

我認為尤其在放款領域，放款可以說是金融服務的主要利潤中心。而今天如果有人看我們，可能會說：你們已經存在 14 年了，但還沒有真正產生任何有意義的利潤。如果不能賺錢，那這些花俏的 AI 到底有什麼意義？

And so today, I just think -- I don't think we get to complain about how investors are valuing us yet. And so even though I think investors are -- to be clear, I think investors are wrong about the stock. I think they're undervaluing the stock, but I kind of think they have every right to do that because they can't tell if and when we're finally going to make some real money. And ultimately, it's our job to change that.

所以我認為就今天而言——我不覺得我們現在就有資格抱怨投資人如何給我們估值。因此，即使我認為投資人——講清楚一點，我認為投資人對這檔股票的看法是錯的。我認為他們低估了這檔股票，但我也某種程度上覺得他們完全有權這麼做，因為他們看不出我們是否、以及何時，終於能真正賺到錢。而最終，改變這一點是我們的工作。

Thanks, Paul. Operator, we can now conclude the call.

謝謝你，Paul。接線員，我們現在可以結束本次電話會議。

Ladies and gentlemen, that does conclude today's call. We thank you for your participation. You may disconnect at this time.

各位女士、先生，今天的電話會議到此結束。感謝各位的參與。您現在可以掛線。