Box (BOX) 2027财年第二季度业绩电话会：AI应用推动开票额增长并上调业绩指引

核心要点

• Box公布2027财年第二季度营收为3.21亿美元，同比增长9%，按固定汇率计算增长11%，这标志着其按固定汇率计算的营收增速已连续第五个季度提速。
• 账单开立额增长17%至3.10亿美元，剩余履约义务增长15%至17亿美元。管理层将这一表现主要归因于第二季度创纪录的签约额以及 Enterprise Advanced 方案的采用。
• 在定价提升、席位扩展以及 Enterprise Advanced 客户更强留存率的支撑下，净客户留存率从上年同期的103%提升至106%。
• 尽管面临100个基点的外汇逆风，营业利润率仍扩大90个基点至29.4%。自由现金流增长67%至6000万美元。
• Box将2027财年营收预期上调1000万美元至约12.9亿美元，按报告数据计算增长10%，按固定汇率计算增长11%。
• 管理层表示，市场需求日益与安全的AI Agent接入、智能工作流自动化以及从传统企业内容管理系统迁移紧密相关。

核心财务数据

年消费金额至少10万美元的客户数同比增长10%。Suites 套餐客户占营收的69%，而一年前为63%。Box预计将在未来12个月内确认约55%的剩余履约义务。

Box在本季度以约6600万美元回购了260万股股票。截至2026年7月31日，当前回购计划下的剩余授权额度约为3.78亿美元。

业务与运营表现

Enterprise Advanced 仍是主要的商业驱动力。Box提到，某家投资银行从 Enterprise Plus 升级并扩展到全公司范围的部署；某大型联邦机构在替换传统合同生命周期管理和协作系统的同时，将席位数量翻了四倍。

管理层表示，Enterprise Advanced 客户带来的净留存率高于全公司106%的水平。跨部门的AI和内容工作流也在支撑席位扩展，特别是在销售赋能、合同管理和文档处理跨越多个业务部门的领域。

Box继续开发模型中立的AI平台，将企业内容与 Claude、ChatGPT、Gemini、Microsoft Copilot 和 Salesforce Agentforce 等服务连接起来。该公司计划通过 AI 单位（AI units）和 API 调用量来实现使用量变现。管理层表示，在基数相对较小的情况下，AI消费正在快速增长，目前大部分活动来自现存客户，且往往能带来更大的追加销售机会。

产品开发重点关注针对 Box 及第三方 Agent 的护栏机制、提示词注入检测、活动监控和访问策略。新的 MCP 集成包括 Anthropic 的法律解决方案和 Databricks，允许客户在不将非结构化 Box 内容移出 Box 治理边界的情况下，结合结构化数据进行使用。

Box还在投资 Box Automate、Box Extract 和 Box Apps，以支持 Agent 化工作流自动化。目标应用场景包括客户入职流程、合同审查、文档分类、供应链流程和元数据提取。

管理层报告称，在美国、日本以及欧洲、中东和非洲地区（EMEA）的需求广泛。该公司正在扩大前向部署工程（Forward Deployed Engineering）及系统集成商合作关系，以帮助客户选择模型、评估性能并微调AI工作流。

管理层业绩指引

2027财年营收预期上调了1000万美元。管理层还预计，在2027财年末离场时，净客户留存率将达到106%。

第三季度账单指引包含约70个基点的汇率顺风。全年账单预期包含约150个基点的汇率逆风。第三季度和全年的营业利润率指引均计入了估计80个基点的汇率逆风。

管理层预计第四季度毛利率约为80%。由于 BoxWorks 峰会将在该季度举行，以及公司扩展了红木城（Redwood City）总部的租赁协议，预计下半年的费用支出将更多集中在第四季度。

风险与关注要点

• Box AI 使用率的增长以及更繁重的 Agent 工作负载增加了基础设施使用量，并对毛利率构成压力。
• 获取某些公有云基础设施组件受到限制，导致计划中的效率提升项目受阻。管理层表示，随着这些限制缓解，将能够释放额外的效率。
• 外汇波动使第二季度营业利润率减少了约100个基点，每股收益减少了0.04美元。在第三季度和2027财年展望中，外汇仍是逆风因素。
• 管理层将季度账单开立额描述为本质上存在不均匀性，原因在于签约时间点、对比基数以及汇率变动。
• AI Agent 处理数据的规模比人类用户更大，这带来了额外的访问控制、异常检测和监控需求。
• 管理层强调，生产环境中的 AI 模型必须满足零数据保留、合规性、基础设施托管和区域可用性等方面的要求。

分析师问答环节要点

AI经济学与模型中立性：管理层预计，运行时间较长的 Agent 和大规模文档处理仍将属于 Token 密集型任务。Box 的模型中立架构旨在将工作负载路由至满足客户所需准确度与成本平衡的模型，适当时包括开放权重模型。

传统内容迁移：管理层表示，AI 的采用正在促使企业将合同、理赔单据、研究文件和其他非结构化数据从本地或传统内容系统中迁移出来。Box 计划推出更多产品更新以推进这些迁移。

Agent 安全：Box 预计 AI Agent 比人类用户需要更严格的控制，因为它们可以访问和处理大得多的数据量。计划推出的功能包括额外的护栏、使用量阈值、实时警报、审计日志以及针对 Agent 的专属身份控制。

全公司范围部署：管理层认为，将非结构化数据整合在受控的单一事实来源中，降低了将多个 AI Agent 连接到多个内容系统的复杂性。这一动态正推动更广泛的席位扩展和全公司范围协议的签署。

前向部署工程：Box 正在扩大对实施 AI 工作流客户的技术支持。合作内容包括模型选择、基于客户文档的评估，以及在新模型可用时进行反复的工作流微调。

账单前景展望：管理层表示，第三季度账单指引所暗示的环比放缓反映的是正常的季度波动和审慎的预测，而非基础需求发生了既定改变。据公司称，销售管线、Enterprise Advanced 的采用、RPO 和净留存率依然保持健康。

业绩电话会议完整文字记录

---

完整财报电话会议逐字稿

管理层陈述

Cynthia Hiponia

Good afternoon, and welcome to Box's Second Quarter Fiscal 2027 Earnings Conference Call. I'm Cynthia Hiponia, Vice President, Investor Relations. On the call today, we have Aaron Levie, Box Co-Founder and CEO; and Dylan Smith, Box Co-Founder and CFO. Following our prepared remarks, we will take your questions. Today's call is being webcast and will also be available for replay on our IR website. Supplemental slides are now available on our website.

On this call, we will be making forward-looking statements, including our third quarter and full fiscal year 2027 financial guidance and our expectations regarding our financial performance for fiscal 2027 in future periods, including gross margins, operating margin, operating leverage, future profitability, net retention rates, remaining performance obligations, revenue and billings and the impact of foreign currency exchange rates and our expectations regarding the size of our market opportunity, including the growing opportunity driven by the increasing role of unstructured data in AI agents in the enterprise, our planned investments, future product offerings, go-to-market initiatives and growth strategies, the timing and market adoption of and benefits from our new products, solutions and pricing models, our ability to address enterprise challenges, including enabling organizations to automate critical workflows and deliver value for our customers. The benefits from our deepening partnerships with leading AI labs, hyperscalers and systems integrators and our capital allocation strategies, including potential repurchase of our common stock and future share count reductions.

These statements reflect our best judgment based on factors currently known to us, and actual events or results may differ materially. Please refer to our earnings press release filed today and the risk factors and documents that we file with the SEC including our most recent quarterly report on Form 10-Q for information on risks and uncertainties that may cause actual results to differ materially from statements made on this earnings call. These forward-looking statements are being made as of today, August 25, 2026, and we disclaim any obligation to update or revise them should they change or cease to be up to date.

In addition, during today's call, we will discuss non-GAAP financial measures. These non-GAAP financial measures should be considered in addition to and not as a substitute for or in isolation from our GAAP results. You will find additional disclosures regarding these non-GAAP measures, including reconciliations with comparable GAAP results in our earnings press release and in the supplemental slides, which can be found on the Investor Relations page of our website. Unless otherwise indicated, all references to financial measures are on a non-GAAP basis. Finally, please see our earnings deck posted on our IR website for a more detailed look at our Q3 and full year '27 guidance. Thank you.

With that, let me turn the call over to Aaron.

Aaron Levie

Thanks, Cynthia, and thank you all for joining the call today. Box delivered exceptional second quarter results, continuing the strong momentum we saw in Q1 and led by the rapid customer adoption of Enterprise Advanced. Second quarter revenue exceeded our guidance, growing 9% year-over-year or 11% in constant currency and produced operating margins of 29%. We drove a net retention rate of 106% ahead of our expectations of 105%, driven by both price per seat increases and seat expansion. Our Q2 billings growth of 17% year-over-year and RPO growth of 15% year-over-year reflect the success of our strategic investments in both go-to-market and product road map in delivering solutions to customers that address their most critical challenges in AI.

Some examples of our Enterprise Advanced wins in the quarter included a leading multinational investment bank that upgraded from Enterprise Plus to Enterprise Advanced transitioning its legacy file servers to Box Platform. This deployment will expand its license to a wall-to-wall agreement to deliver unstructured data insights across its global banking teams.

Next, a major federal agency upgraded from Enterprise Plus to Enterprise Advanced with a 4x seat expansion to replace its legacy contract life cycle management and collaboration platforms. In partnership with Salesforce, Box will power secure out-based CLM and document management across key legal and research divisions, replacing multiple SaaS vendors. This agency-wide modernization is enabled by Box's FedRAMP high compliance, our secure identity verified e-signatures and Enterprise Advanced capabilities. With record Q2 bookings these wins and many others make it clear that our role in enabling enterprises to get the most out of their enterprise content and transform an era of AI is becoming increasingly significant.

During the second quarter, I spoke with many enterprise technology leaders who highlighted their primary goals and challenges in implementing AI. One of the most common topics is how enterprises can get the right context to AI agents in a secure and governed way as well as tap into the full value of their unstructured data. To do this, enterprises need a secure platform that can connect all the intelligence and capabilities of AI models to enterprise content and workflows. The world's most advanced super intelligence is only as useful as the underlying enterprise knowledge and corporate information that it is access to. Instead of companies sitting on millions or hundreds of millions of files that they know very little about with AI agents, they can now ask questions about this data, mine at all for intelligence and automate nearly any workflow that involves this enterprise content. This is the intelligent content management platform that we are building.

These technology leaders that I'm speaking with are also recognizing that as AI model capabilities advance rapidly across an expanding set of vendors like OpenAI, Google, Anthropic, Meta, XAI, NVIDIA and more, that enterprises will need a model neutral platform that connects their content and workflows through these models and agents securely. With AI costs continuing to rise, the ability to draw the right cost performance mix with any vendor becomes essential rather than migrating content and workflows into separate systems to unlock AI's benefits, our intelligent content management platform gives enterprises a single platform where they can swap models or agents on their content at any time securely.

Now Box is at the center of the greatest transformation in how enterprises work, and we are continuing to drive our product and go-to-market strategies to take full advantage of this massive opportunity. Building on our product leadership. In the second quarter, we announced a range of new capabilities that help customers transform the value of their content with AI. We introduced new security capabilities designed to give organizations greater control over AI agents working with their enterprise content with new agent guardrails, third-party agent activity oversight, prompt injection detection agent classification-based access policies and more customers will be able to extend Box's enterprise-grade security controls to both Box Agents and third-party agents such as Claude, ChatGPT, Gemini and more.

To support our headless initiatives, Box announced new MCP integrations with Anthropic's Quad for legal Databricks, Harvey, [ IBM's Watson X Orchestrate ] agent catalog, Notion Custom agents, [ Slackbot and SpaceX's rock ]. Box partnered with Anthropic as a launch partner for Quad's new legal industry solutions using the Box MCP server as the secured governance layer for agentic legal work. New MCP tools now let Quad execute multistep matter operations directly in box, copying and uploading files, taking metadata, managing collaborator access. Turning Claude from a QA chatbot into an active practice agent.

All actions stay governed by the firm's existing Box permissions and ethical walls, avoiding the governance gap of moving sensitive client data into unsanctioned tools. Also, earlier this month, we announced the release of the Box MCP server for Databricks now available in the Databricks marketplace. This integration lets data analysts, scientists and engineers, combine, connect and query unstructured content from Box, including their contracts, clinical records, financial assets and specifications alongside structured sources like CRM and ERP, all without duplicating data or moving it outside of Box's secured governance boundary. This unlocks the use cases across industries from health care teams spotting care apps by combining clinical records with referral and billing data to financial services firms accessing borrow and covenant risk by joining loan documents with banking data.

Now as we look further into the second half of FY '27, we're continuing to drive significant innovation across our platform to help enterprises maximize the value of their content in the era of AI. Building on the momentum of Box Automate, Box Extract and Box Apps, our platform is evolving into a premier agentic workflow automation system designed to streamline critical content processes like client onboarding contract reviews, brand asset verification, supply chain automation and thousands of other workflows in an enterprise. Additionally, we're advancing Box extract to help power complex document extraction needs across a range of industries from financial services to life sciences. Our model neutral agentic carnass ensures that customers can both improve the accuracy of this extraction and lower their costs by choosing exactly the right model they need for any document type.

Box is also modernizing its core content management infrastructure with improvements in metadata management, large file support and file system capabilities, we are paving the way for enterprises to retire legacy on-premises ECM systems and migrate their unstructured data to a secure cloud native platform where it can be easily accessed by AI.

In Q2, we've continued to see more and more enterprises look to migrate off these legacy systems in favor of a much more modern AI-driven approach. At Box, we're also optimizing our developer ecosystem to support AI agents working with enterprise content at scale and introducing new tools and improvements such as Enhanced MCP support, deeper integrations with leading agents like Claude, ChatGPT, CoPilot and Salesforce agent force and improved context retrieval APIs, which will allow developers to securely connect enterprise content to AI agents. We're focused on delivering the world's best headless experiences for working with enterprise content securely across any AI agent and monetizing this usage through our AI units and API volume.

Finally, all of these innovations are anchored by Box's industry-leading security and compliance foundation. As we recently saw with the OpenAI hugging face incident, enterprise will increasingly need platforms that can securely protect their corporate data and ensure that neither humans nor agents can get access to information they shouldn't have access to. As external AI agents interact with enterprise data, Box is implementing robust guardrails, comprehensive audit logs and real-time security alerts to ensure that content remains protected, governed and visible at all times. We'll continue to deliver industry-leading data protection and governance capabilities to ensure the security of unstructured data in an enterprise.

Now we will be sharing much more about our product road map at this year's Box Works in San Francisco in early November, where we'll be making major product announcements, we'll hear directly from customers that are taking advantage of the Box platform and hear directly from our partners, including the CEO of NVIDIA Jensen Huang; [ Lip Guan ], the CEO of Intel; and [ Michael Trell ], the CEO and Founder of Cursor.

Next, for our go-to-market strategy, we remain focused on accelerating the adoption of Enterprise Advanced, enabling customers to power their intelligent workflows with content while driving the growth of platform revenue. To win in key industries such as financial services, life sciences, government, education, media and entertainment, legal and other key verticals, we will continue to deepen our vertical-specific marketing sales motions, collateral solutions and ecosystem partnerships. We are also expanding our FDE or Forward Deployed Engineering efforts to ensure that customers can successfully implement and tune AI agents on their enterprise content for everything from document processing to Agentic content workflows. Additionally, we are expanding our system integrator ecosystem, collaborating with vertical regional and global system integrators to embed our platform deeper into enterprises critical content workflows.

Finally, our partnerships with major hyperscalers like Amazon and Google will be central to expanding our enterprise distribution and enablement.

In the second quarter, we continued to see strong momentum in customer wins enabled by these partners, a critical part of our go-to-market strategy. For instance, in partnership with DataBank, a leading insurance provider has adopted Box Enterprise Advanced with [ Shield Pro ] and purchased additional AI units to drive a comprehensive platform modernization. These deployments leverage Box's platform APIs, the Box [indiscernible] APIs and Box AI to connect box directly into the firm's custom middleware for core systems, including Guidewire. This positions the insurance provider to modernize more than 100 terabytes of content, retire multiple legacy platforms and integrate Box AI across high-volume workflows like mail room and policy processing.

Working with [ Slalom ], a large U.S. state DMV upgraded from Enterprise Plus to Enterprise Advanced and purchased additional AI units as the foundation for a new intelligent document processing initiative. This agency is replacing a costly legacy document processing system with classification and metadata extraction powered by Box AI. This solution will extract key information from identity documents at scale and automatically populate sales force records associated with each driver profile, streamlining licensing applications and renewals across the state.

At Box, we have an extraordinary opportunity to serve as the defining platform for securing, managing, governing and applying intelligence to unstructured enterprise data at scale. Nearly all mission-critical workflows, such as processing regulatory data, automating insurance claims with AI, reviewing legal contracts, managing aviation research or facilitating collaboration in pharma are all fundamentally powered by enterprise content. Our intelligent content management platform sits squarely at the center of these vital business processes. We're incredibly excited at Box about the market transformation happening right now due to AI and we have the team, the technology and the customer base to fundamentally take advantage of this massive opportunity.

Now let me turn the call over to Dylan.

Dylan Smith

Thanks, Aaron, and good afternoon, everyone. We had another very strong quarter in Q2, driven by record Q2 bookings and increasing Box AI adoption. As a result, we exceeded guidance across all top and bottom line results, delivering our fifth consecutive quarter of accelerating revenue growth in constant currency. As Aaron Levie discussed, we advanced our leading intelligent content management platform by deepening our AI and agentic capabilities while investing in key go-to-market initiatives to drive continued enterprise advanced momentum.

Q2 revenue of $321 million was up 9% year-over-year and up 11% in constant currency, exceeding our guidance. Customers paying us at least $100,000 annually grew by 10% year-over-year. Suites customers now account for 69% of revenue, up from 63% a year ago. We ended Q2 with remaining performance obligations, or RPO, of $1.7 billion, a 15% year-over-year increase or 17% in constant currency. Both short-term and long-term RPO accelerated sequentially with short-term RPO up 11% year-over-year and up 14% in constant currency. We expect to recognize roughly 55% of our RPO over the next 12 months.

Q2 billings of $310 million were very strong growing by 17% year-over-year or 16% in constant currency. This result exceeded our expectations for low double-digit growth with the outperformance driven primarily by Q2 booking strength. In Q2, our net retention rate improved to 106%, above our guidance of 105% and up from 103% in the year ago period. Our annualized full churn rate remained at 3%. This outperformance was driven by continued improvement in our seat expansion rate as well as the impact of very strong net retention results within our Enterprise Advanced customer base, which exceeded our overall net retention rate. We now expect our net retention rate to be 106% exiting FY '27.

We delivered Q2 gross margin of 81.2% in line with our expectations. Operating income of $95 million resulted in operating margin expansion of 90 basis points from the year ago period to 29.4% which reflects a 100 basis point headwind from FX. This was above our guidance of 28.5%. In Q2, we delivered EPS of $0.40 which was above our guidance of $0.39. This includes an FX headwind of $0.04, $0.01 higher than our prior expectations.

Turning to our cash flow and balance sheet. In Q2, we generated free cash flow of $60 million in cash flow from operations of $71 million, up 67% and 54% year-over-year, respectively. These results were driven by strong linearity, allowing us to collect a healthy portion of our Q2 bookings within the quarter. We ended Q2 with $446 million in cash, cash equivalents, restricted cash and short-term investments. In Q2, we repurchased 2.6 million shares for approximately $66 million. As of July 31, 2026, we had approximately $378 million of remaining buyback capacity under our current share repurchase plan.

With that, let me now turn to our Q3 and updated FY 2027 guidance. Note that our second half expenses will be more weighted towards Q4 and versus our typical seasonality due to the expected impacts from [ Box Works ] occurring in Q4 this year as well as the recent extension of our Redwood City headquarters lease.

For the third quarter of fiscal 2027, we expect Q3 revenue to be approximately $329 million, representing approximately 9% year-over-year growth or 11% in constant currency. We anticipate our Q3 billings growth rate to be roughly in line with revenue growth of 9% which includes an expected tailwind from FX of approximately 70 basis points. We expect Q3 gross margin to be approximately 80.5%. We anticipate Q3 operating margin to be approximately 28%, which includes an expected headwind from FX of approximately 80 basis points. We expect Q3 EPS to be approximately $0.39, which includes an expected headwind from FX of approximately $0.02. Weighted average diluted shares are expected to be approximately $142 million.

For the full fiscal year ending January 31, 2027, we are raising our revenue expectations for the full year by $10 million to approximately $1.29 billion, representing 10% year-over-year growth or 11% in constant currency. We expect our FY '27 billings growth to be roughly in line with revenue growth. This includes an expected headwind of approximately 150 basis points from FX. We expect FY '27 gross margin to be approximately 80.5% with Q4 gross margin expected to be roughly 80%. This reflects the strong and growing adoption of Box's platform and Box AI as well as the capacity dynamics of our public cloud providers.

We continue to expect FY '27 operating margin to be approximately 28%, which includes an expected headwind from FX of 80 basis points. This reflects our ongoing focus on delivering operational efficiencies even as we continue to invest in driving durable revenue growth. We now expect FY '27 EPS of approximately $1.54, which includes an expected headwind from FX of approximately $0.09. Adjusting for the impact of the currency and share count movements versus our previous expectations, this represents an increase of $0.01 versus our prior guidance. Weighted average diluted shares are expected to be approximately $141 million. This represents a significant reduction from 149 million shares in the prior year as we continue to execute our disciplined capital allocation strategy. The $10 million raise to our revenue expectations this year reflects continued momentum across the business with demand for Box AI and the growing adoption of Enterprise Advanced, driving continued acceleration in our revenue growth rate and continued improvements in our net retention rate.

As Box's intelligent content management platform is increasingly becoming the foundation enterprises rely on to securely unlock AI's value across their content, Box is well positioned to drive durable long-term growth.

With that, Aaron and I will be happy to take your questions. Operator?

Operator

[Operator Instructions] Your first question comes from the line of Lucky Schreiner with D.A. Davidson.

分析师问答

Lucky Schreiner

Great. Aaron, I thought it was really interesting to hear about your expectations for innovation in the back half of the year. And I wanted to follow up on that. The longer Agentic workflows tend to be more token intensive. So can you give us an update on how you view token cost evolving here and how OpenSource model adoption factors into customers implementing some of those longer form workflows and maybe sneak in, like any difference in unit economics between frontier model versus OpenSource for you guys internally?

Aaron Levie

Yes. So you're exactly right. The kind of momentum we're seeing generally correlates to interestingly, 2 dimensions, either one longer-running agents that do more processing work in a single session or the ability to run agents off of large amounts of data, which can be broken up kind of discretely on a per item or a per document basis, both of which have the exact same type of tendency to be very token intensive, very consumption heavy which is both great for us.

I mean, basically on every dimension because it means that customers will increasingly want those types of workflows to happen inside of platforms that are model neutral because the more tokens your use case requires, the more, obviously, over time, you're going to be price sensitive because you want to make sure that you're optimizing that cost structure for the use case. So by having a model neutral layer, which is our agentic harness, we can then make sure that we are directing the workload to whatever is the effectively cheapest model at the accuracy level that the customer is looking for.

In some cases, that can be an open weight model, and we have some -- there's models on the horizon that we're quite excited about, that we'll be opening up. We assume in the second half based on some of the visibility we have from partners. In other cases, it can be just the sheer competition that's happening between the labs. You've seen things like OpenAI bringing down their prices or Gemini bring down its prices. That actually also flows into our product as more either margin or relief or or more consumption from customers.

But in general, the again, the trend that we're going to continue to see is customers are going to say, "I have millions, tens of millions, hundreds of millions of documents. I want to be able to run agentic workflows on these documents to automate processes or extract intelligence from data or be able to use all this information as kind of critical knowledge from my organization." And then our layer is really in the best position to both, again, deliver the highest level of accuracy at the cost profile that our customers are looking for. So you're going to see AI unit growth continue to go upward. You're going to see more upgrades into Enterprise Advanced and even Box as a headless platform performs well in that environment as well. So these are all great trends for us.

Lucky Schreiner

Awesome. Last one for me. It was interesting to hear about the legacy migrations. Can you give us a sense like those tend to be long and painful processes. Obviously, your partnerships improve that. But maybe like how have you been able to speed up that process, I imagine Box Shuttle and AI capabilities in general help you out there? And what's the customer demand to go through that kind of painful modernization process today?

Aaron Levie

Yes. Yes. It's interesting. So we've obviously talked about this a bit over the past couple of years. I think it started out as something that we assumed and kind of could feel would happen but it was still very early in the trajectory. Now we're actually seeing example after example on the rise, which is if you're an enterprise and you still have a large amount of your unstructured data in legacy systems or on-premises environments. Your contracts, your research files, your insurance claim data, your loan processing documents, your KYC documents, your agency documents in a government agency, all of that data is often sort of effectively trapped and closed off from AI agents.

So as you have an AI strategy that assumes that an agent is going to read a document or process a claim or look through a contract or be able to process an image for brand guideline failures, all of that data needs to be available and accessible to AI agents in secure ways in these workflows. And so many of the legacy approaches to doing document management or enterprise content management just simply don't work as companies are modernizing how they're going to work with and automate their enterprise content workflows. So that's leading to this catalyst of more and more customers reaching out to us calling us and then obviously going out into their environments for either large-scale migrations or really kind of replatforming their next generation of these workflows. We had a number of deals in Q2 that kind of represent this combination of very much agentic workflow-driven use cases that have a data migration or a legacy ECM system migration as a part of the where the dollars are going to come from.

So on the horizon, we actually see quite a bit of this opportunity continuing to grow. And so in the second half and certainly as we go into next year, you'll see a continued amount of product launches and updates that help facilitate and accelerate that migration, all of which are these next-generation features that help customers manage their content in the cloud at scale in these business processes and be able to bring AI agents to that content very seamlessly and securely.

Lucky Schreiner

Congrats on a phenomenal quarter.

Operator

Your next question comes from the line of George Kurosawa with Citi.

George Michael Kurosawa

I'm on for Steve Enders. Maybe just a question on the security role that Box plays within these agentic workflows. If you could talk about the governance and security side, what is maybe new or different from -- in terms of enterprises needs in an agentic world versus dealing with human users? How does Box role there evolved?

Aaron Levie

Yes. This is an area of an incredible amount of surface area. So which is very exciting for us. Obviously, it's an incredibly dynamic and sometimes kind of stress space for the customer ecosystem, but we have -- there's a lot of innovation really available here. If you think about what agents tend to need, not only do they need basically all of the same level of security and controls that humans need. So they need access levels, they need -- they need to be only able to view or edit the documents that you want them to, you need to be able to obviously be [indiscernible] if they're either going rogue or you're seeing kind of too much usage happen. So that's kind of the basic foundation that we're drafting off of in the core of Box's security and access control capabilities, which is why we're in a very strong position to take a lot of these workloads.

But what's interesting is, over time, they're actually going to need additional protections that we didn't commonly think that end users needed. Things like an end user could only -- a human user could really only process a certain amount of data at scale and so you could quickly kind of detect if maybe a user was doing something that they shouldn't be or there be very limited kind of blast radius or damage that they could do if they kind of went rogue or there's some malicious actor in the system. Whereas AI agents have the ability effectively with just kind of -- it's kind of just correlated to their compute level, to be able to either work with large amounts of data, access the wrong information.

In the case of the hugging phase or OpenAI very much go out and execute on a goal to the kind of ultimate level that its compute is allowing. And so what that means is that enterprises are going to need an all-new set of guardrails, alerting mechanisms, anomaly detection capabilities, ways of really having a better a better set of controls on what agents can do with their data. And that could be things like, okay, agents inside of these folders or at these usage levels shouldn't be able to write data or shouldn't be able to download data, they can only kind of view it or we need to be able to monitor their activity or we need to be monitored -- or we need to be alerted after a certain threshold of activity happens under certain kinds of agents. And maybe we don't want to let these kinds of agents into our systems, but we're fine to let those other agents in the system.

So all of that functionality is effectively kind of net new for the agentic era. But what it's all built on is the same foundation and capabilities that we've been working on for a number of years. So Shields, for instance, which is our advanced threat detection and security products from Box, we're going to continue to build out more and more features that help our customers protect this data that both again, let them protect human users, the antigenic users in the system. There's new forms of identity controls that we need to be building out that we're excited to share more about over the coming quarters. We obviously are going to continue to integrate with the broader security ecosystem to help protect enterprise data.

But these are all things that continue to reinforce our value proposition and reinforce the need for secure very, very robust systems of record on data. This is kind of why we -- in the -- maybe the first part of this year when there was a lot of commentary on maybe people just vibe code different kinds of applications, to us, it was a little bit kind of humorous just because we know the level of security and protection that is necessary for these enterprise systems. And I think as people saw the OpenAI hugging face incident has kind of made it more resonate what's really possible out there from a security risk standpoint.

So we take our position very seriously. We're going to be doubling down in all of our security investments and make sure that we are the best platform for helping customers protect all of the unstructured data.

George Michael Kurosawa

Okay. That's great color. And then maybe one for Dylan. You referenced a couple of different dynamics on the gross margin side, AI usage and then capacity in the public cloud providers. If you could just double-click what are some of the moving pieces there? How should we think about that going forward?

Dylan Smith

Yes. So as noted, really, the 2 biggest drivers are the kind of strong and growing adoption of Box's platform, Box AI specifically as well as the capacity dynamics with our public cloud providers. To double-click a bit, really, number one, we're really pleased with, especially the new features a lot of the kind of heavier workloads and kind of agentic processes that Aaron mentioned are certainly kind of a quickly growing and net new type of use case that customers are using the platform for.

And then on the capacity dynamics, that's really related to kind of limited access to certain components of the infrastructure just given what's going on in the environment, and that really an impact on the kind of level and impact of the infrastructure efficiencies that we expect to deliver this year. So certainly, as those constraints ease, we'll continue to unlock those efficiency projects. But those are some of the things that we're seeing that are impacting gross margin versus our initial expectations during the year.

Operator

Your next question comes from the line of Matt Bullock with Bank of America.

Unknown Analyst

This is [ Jake ] on for Matt Bullock. I think we're hearing kind of more and more about kind of enterprises going wall to wall. You mentioned a few on the call today. Could you talk specifically about how kind of the AI governance or AI agent governance opportunity is like driving this? Why is there a benefit to going wall-to-wall rather than just having Box maybe within a certain department when it comes to like embedding AI workflows.

Aaron Levie

Yes. Definitely, we've had some great wall-to-wall wins. I would say we're in a position where we can both capture the individual line of business use case that a customer is trying to automate or be able to bring AI to as well as the wall-to-wall notion. So I think we're actually both motions are humming at the moment.

But to your point, the wall-to-wall benefits for a customer are really imagine a scenario where you have rolled out a variety of AI agents to your enterprise, maybe some people are using ChatGPT, others are using Claude Opus co-work. Maybe your developers are using Cursor, your sales force has Salesforce agent force. Now all of those agents are running around, they need access to corporate information to be able to make decisions or be able to work with your information. So that could be your research materials, your marketing assets, your HR documents, your contracts.

The challenge is if you have 3 or 5 or 10 different systems where all of those agents all have to be able to work with and equally use successfully, and you have to be able to secure the access controls to all of that data, it's just a very, very difficult problem. It's a many-to-many relationship problem, and that's generally not a clean architecture for most enterprises. So what you're going to see is a consistent pattern across various data planes where how do we move our canonical data in different topics into sources of truth. So you've seen that in the structured data world with things like Databricks or Snowflake, we're seeing these kind of mass migration projects. You've obviously, over the years, seen that in ERP systems or CRM systems. But we also think there's quite a bit of momentum on the same for unstructured data.

So you'll go to an enterprise and they'll say, "Hey, I want to be able to have all these agents have access to corporate knowledge." That, again, is we need to make sure that those agents are working off of the real source of truth the authoritative copy of data, which means that you can't have a fragmented landscape that everybody is working from. So that's one of the core wall-to-wall benefits. That's the kind of like knowledge worker end-user productivity benefit. There's also security benefits, there's governance benefits. We've been building a lot of features in Enterprise Advanced that help you with things like data retention, data archival. We have new capabilities around records management that we're excited to share in the next couple of quarters. But all of that is really around the need for robust infrastructure that helps you manage all that enterprise content, which is really the context for those agents.

Dylan Smith

Yes. And the only thing I'd add, this is Dylan, is, as we talked about our confidence as we launched it in Enterprise Advanced, also being a potential catalyst for seat expansion because of the capabilities that it enables and the types of workflows that customers can now do using those capabilities A lot of those are really cross-departmental, right? So if you think about whether it's a sales enablement workflow or a contract life cycle management workflow or something like that, yes, there might be a primary business unit who's really driving it, championing it, but that might go across 4, 5, 6 different departments. And if they don't have access to Box, they're going to run into all the challenges that Aaron was mentioning. And that's why we've seen -- I mean you look at the increase in our net retention rate and those trends, the biggest driver of that has been higher seat expansion for exactly that reason.

Operator

Your next question comes from the line of Chris Quintero with Morgan Stanley.

Christopher Quintero

Congrats on the nice acceleration here across the board. Aaron, maybe for you kind of high level, we've been hearing a lot about zero data retention as a key enabler for AI growth and compliance. And we've seen how some of the recent models have influenced some potential changes around that. So curious how you're thinking about that from the Box perspective and what's the opportunity there for you all?

Aaron Levie

Yes. So this is obviously a very hot topic at the moment, mostly driven by the release of [indiscernible]. In general, maybe one of the underappreciated reasons for just the rapid rise of AI growth is the fact that very quickly, most of the leading labs coalesce around the idea of zero data retention, which basically in simplest terms means that when I kind of am interacting with an AI model and I have information in the context window, that data doesn't sort of get stored and sit around for a week or 30 days in the servers of those AI labs, it's sort of just a kind of a femoral pass-through. And that's what led to enterprises being very comfortable with the adoption of AI in their organizations, whether that was direct adoption or through more of these kind of applied AI layers like a Box or a [ Harvey or Sierra ] decagon, et cetera.

And so the challenge, obviously, was unlike [ Fable ] was they launched without zero data retention, which obviously means that there can be less adoption, you have to have kind of a separate exception handling that customers have to go through. And we've made it very clear in our platform that the sort of in production GA generally available models we'll have a set of criteria that are met around zero data retention, certain compliance requirements, ways that the infrastructure is hosted, being able to have certain regions that it all operates in. And that's the enterprise trust that we've been able to establish with organizations as we deliver AI to them. And that's again where I think you're going to see a huge benefit to this applied AI layer is being able to really sort of bridge the breakthroughs of AI models with the actual real enterprise workflows that need compliance, they need guardrails, they need security, they need governance. They need these kind of regulatory controls.

So that's kind of the state of the industry. My guess realistically is actually just Anthropic will evolve their stance on this because they'll see it in the revenue. and they'll have to change course. And I think we'll just see more and more modern approaches to how these companies will both meet their safety -- internal kind of AI safety requirements with offering things that kind of technically resembles ZDR for their users.

Christopher Quintero

Got it. Very helpful. And then great to hear about the FDE motion, the expansion you're putting through there. So curious what are some of the key learnings you've had on that whole motion as you've now kind of ramped it up and grown over the past few months?

Aaron Levie

Yes. So this is a pretty exciting area because if you kind of think about the types of challenges, workflows, goals that customers have when working with enterprise content and unstructured data and agents, let's say, you're a bank, let's say, you're a pharma company, let's say, you're an insurance provider. You're seeing all of this potential with AI. And certainly, in your internal use cases, things like coding agents, et cetera, are showing incredible breakthroughs and productivity gains.

But the rest of your organization, the back-office processes, the customer-facing workflows, you're trying to figure out how do I actually get the same AI gains in those parts of the business? And some of that's a data challenge, some of that's an architecture challenge, some of that's an information governance challenge, all of which obviously Box is very, very expert in. But a lot of times, it's also how do I tune these agents, how do I pick the model, how do I run evals against my data and compare what model is the best to use for certain types of workflows or certain kinds of document processing tasks.

And so Box from just a talent standpoint, a brand standpoint and a technology standpoint, I think best represents being able to help customers go through that journey. And so our the FDE motion really kind of originated by us effectively doing that with a number of our individuals across our consulting team and our solutions engineering team. So we've been increasingly more programmatizing that and scaling that across customers. I think this is going to continue to be an investment area for us, both for the rest of this year and certainly next year. And again, in essence, it's really about helping our customers transform with AI, making sure that they can actually get the -- again, the data environment set up in the way that they need, making sure that they understand how to do the right types of evaluations of AI models and agents on their documents, continue to kind of tune these workflows.

There's been some engagements we've worked on with customers where we've gone back maybe nearly a half a dozen times because there's been a new model breakthrough, and we want to introduce that into the customers' environment and make sure that they can get those either productivity gains or cost savings as a result of that. And that really, again, requires a high degree of technical expertise on the side of both the vendor and the customer, which is why you're seeing this big push on FDE's right now.

Operator

Your next question comes from the line of Brian Peterson with Raymond James Company.

Unknown Analyst

This is [ Jonathan Carry ] on for Brian. I'll ask one just multi-partner. Aaron, I realize it's early days, but I wanted to ask on the consumption trends. It sounds like that's ramping nicely. How much of the traction there is customers that are seeing enough success to actually come back and refill the tank, so to speak, on credits prior to renewal.

And then relatedly, is the ramping feature adoption there on the AI platform predominantly from existing customers? Or are we at a stage where that's actually driving material competitive wins on the net new side as well?

Aaron Levie

Yes. Great question. So consumption is obviously off of a lower base than many other parts of the business, but has been growing, I think, quite rapidly, especially kind of on a year-over-year standpoint in Q2, we saw really nice wins on the growth side. A lot of it is the customer still doing their initial big kind of growth workloads just because the -- we're still relatively early in the AI unit monetization. And then by virtue of just the scale of our existing customer base, a lot of it is from existing customers. But what's interesting is that is often leading to materially bigger upsells on those existing customers. So it's a little bit might be even incidental in some cases, that it's an existing account because the new transaction is so different or bigger because of the new use case that they have with us.

So we're seeing really great opportunities across the board where customers are saying, "Hey, I thought of you as a place where I put my documents or I put my content. Now it's a place where actually I can run the workflow and intelligent workflow on you, which is really changing the calculus of the conversation." And I was in a conversation just literally 3 days ago with a customer that has been at Box probably nearly a decade. And the moment we showed them our Agentic workflow automation tool it like completely changed the type of conversation we are having. And this is a customer that theoretically could have already known about all of those use cases and capabilities, but we caught them in when they were at a juncture of a new set of use cases that they need to run their documents and document processing and that will absolutely kind of turn into now another sales motion where we go and work with that customer to expand what they're doing with Box.

So that kind of how it's working every single day right now across our customer base. And what that's leading to is almost back to the last question, more on the FDE front, more on the verticalization piece, more with system integrators all of which we're seeing great success in from a go-to-market investment standpoint.

Operator

Our next question comes from [ Joshua Trautman ] with RBC Capital Markets.

Unknown Analyst

This is [ Josh Trautman ] on for Rishi Jaluria. Congrats on the quarter. You guys have mentioned strong product adoption and some different tailwinds to retention. And I was just curious around how international deals have helped participate in that and how those conversations with those clients have been developing.

Aaron Levie

Yes. Great. For us, it was a broad-based quarter. So we're quite happy with the results. We had great wins in Japan, some of the leading enterprises out there that we're hope we'll be able to announce relatively soon some of those big wins, some great wins in EMEA. So we had great traction and participation from EMEA. And then obviously, U.S. business across everything from public sector, U.S. enterprise, commercial, great momentum. All of which very similar contours of conversation around how do you transform with AI agents.

I was out in Japan in the beginning of June and just overwhelming excitement around how do you bring more automation to your enterprise content with the power of AI agents. But again, those are very similar to the conversations that we're having here in the U.S. and those that we're having in Europe.

Operator

Your next question comes from the line of Jason Ader with William Blair.

Jason Ader

I wanted to ask about billings, very strong in Q2. Billing for Q3, I think you guided to 8%. Just can you help us understand if there's some seasonality there, why the step down from Q2 to Q3?

Dylan Smith

Sure. So a lot of moving pieces in there and billings is inherently lumpy. And if you look back to even and the year that's kind of the dynamic that we had expected, which is a combination of very different -- everything from very different kind of FX impact quarter-to-quarter to kind of ease or difficulty of some of the comps.

So I would say if you think about the full year billings outcome, even since our initial guidance 6 months ago, we've raised those expectations pretty significantly more than $20 million and showing a strong acceleration from where we were last year. And so we really think about it as that dynamic as not at all indicative of the change that we're seeing in the demand and momentum in the business as the underlying kind of leading indicators of that growth and bookings remain very healthy from pipeline to Enterprise Advanced adoption to RPO and NRR, our net retention rate, both moving in the right direction.

So really a function of we say just some of the variability quarter-to-quarter as well as the fact that, as always, we want to be thoughtful and prudent about how we guide. I mean, even looking back to this year so far, we've had pretty significant outperformance against the guidance that we set up. And hopefully, that trend continues, but just really wanted to be kind of thoughtful about that.

Jason Ader

Okay. And for the year, I think you said billings growth and revenue growth about the same. Wouldn't -- I guess, it just -- it would seem to me, given the momentum you have, the billings growth would be ahead of revenue growth just because it's more of a leading indicator. Can you just help reconcile that?

Dylan Smith

Yes. I mean I would say -- I mean I do expect it to be ahead on a constant currency basis as there is more of an FX impact to billings than revenue, and we did give the revenue as we're talking about growing at roughly the same rate, that's on an as-reported basis. But again, some of it is just kind of the way that we think about setting expectations and there's as you'd imagine, more variability in billings plus just some of the dynamics around compares that didn't show up in the same way as revenue.

So I would say kind of other factors outside of the underlying momentum are impacting each of those metrics a little bit differently. But to your point, we do expect billings growth to be a leading indicator of revenue to be ahead of it if we keep up the momentum that we've been on. And I think you see that type of dynamic for example, in our short-term RPO growth, which is both a few points ahead of either of those metrics and you're kind of moving in the right direction and accelerating.

Operator

We have reached the end of the Q&A. I will now pass the call off to Cynthia Hiponia for closing remarks.

Cynthia Hiponia

Great. Thank you, everyone, for joining us again this afternoon. As Aaron mentioned, we're hosting [ Box Works ] in San Francisco on November 5, and we'll be once again doing another Investor Relations product briefing at the event. So we look forward to giving you more details, and we'll talk to you on our next earnings call. Thank you.