Good afternoon. I'm Cynthia Hiponia, Vice President of Box Investor Relations, and welcome to BoxWorks 2025 Investor Product Briefing. Since our comprehensive product strategy briefing that we provided at our Financial Analyst Day in March, our product leadership and pace of innovation has continued. I'm sure. Uncertainties and assumptions. Further information on these risk factors that could affect our forward-looking statements we make in this presentation can be found in the documents that we file with the Securities and Exchange Commission. With that, I'll turn the session over to Aaron Levie, Box Co-founder and CEO. Aaron?
Hello. I'm Aaron Levie, CEO and co-founder of Box, and welcome to our BoxWorks I nvestor Product Briefing. Just this morning, we announced at our BoxWorks 2025 conference some of the biggest product announcements we've ever had as a company, and this afternoon, we wanted to share some of those key highlights with all of you as investors and analysts. Legal, finance, or sales that could work a thousand times faster than any other person. Think about all the ways that we would be able to use these agents to drive more productivity and accelerate our team's or individual productivity or businesses. Now, imagine if those agents could run in the background. They could be involved in any kind of background tasks. They could be run in parallel. You could kick them off and wake up in the morning and see what they've done.
That's a future where we're going to see vastly more AI agents than even people inside of organizations, and this is going to change every single aspect of work. Think about the individual productivity that we've already seen in our business, broad organizational efficiency. This means you can onboard clients faster. We can get personalized marketing in any segment that we're going after. We can accelerate product development to ship more product to customers, and we can reduce business risks and bring more automation to our supply chains. This means we could get more workflows from end systems that go into an ERP system or an HR system. This is the data that goes into a structured database, but what's amazing, that's only about 10% of the data that we work with as an organization, and it only represents a small fraction of our overall work.
Think about all of the business processes that we have in a company, and then think about how many of those are much more dynamic and involve people sharing data with one another or onboarding data into a process or reviewing information to figure out what next step it goes to, and the vast majority, for the first time ever, lets us begin to tap into the value of all of that unstructured data, where we can now automate any workflow and get insights into that information at scale. We could never have done this before we had AI, so think about the kinds of workflows and processes that we can now transform. We can accelerate product development processes with the power. Challenge is that most enterprises can't yet tap into the full value of their unstructured data to actually get the benefits of this automation.
If you think about it, where is most data today across a lot of organizations? It's in all of these data repositories on-prem, sometimes in the cloud, legacy systems or point tools, document management systems, collaboration technology, storage infrastructure. And these technologies have long been a problem for most organizations, and it's generally always been a major pain because it creates security risks. It means you're duplicating data, so you're spending more on technology. And you have workflows that really get broken because people have to hop between different systems. So it's always been a painful challenge for organizations, but now it's actually existential.
Think about a world where you have 100 times more AI agents that are roaming around and need access to information to make decisions or to automate workflows or to enable an employee to get access to the right answer or the right piece of data to do their job or playing with the wrong information or an answer. You'll have agents that leak information. So this is a scenario where you don't have permissions that are in sync between multiple systems, or somebody has overexposure or access to data, and now an agent's actually exposing answers. So, as your architecture, and in particular, unstructured data architecture, represents an existential challenge if they don't get this right. So enterprises need a platform that can connect content to AI securely and then integrate across all of their applications, and this is the intelligent content management platform from Box.
At the center of our platform is content. This is all the user will be able to tap into, and then we integrate all of this technology and the customer's data across their entire tech stack. This could be with products like IBM's watsonx Orchestrate. It could be inside of Salesforce with Agentforce. It could be in Slack or ServiceNow in their Agent Fabric, Microsoft Teams, or any other product that a customer is working within. And more and more with our no-code app builder, customers can also build important information from that content. Now, the companies that thrive in the AI-first era will be those that are able to take full advantage of their unstructured data and information at scale, and today, we're incredibly excited to have announced some of the biggest breakthroughs in our platform's history.
Today, we're going to revolutionize how we work with our unstructured data. To share a bit more about our overall platform strategy and where we're going, I'm going to bring up Diego Dugatkin, Box's Chief Product Officer.
Thanks, Aaron. Hello all. I'm Diego Dugatkin, Chief Product Officer at Box. As Aaron shared, the potential to give our customers the best AI agents for content. Since our Financial Analyst Day in March, we have delivered on our roadmap with significant releases across our portfolio. We have added the latest AI models and empowered multi-doc query and new formats. We launched Box Archive and new security and governance enhancements. Box AI for Hubs has been enhanced as well as in April, we also released Box Archive. Effective archives are a key tool in regulated industries. Now with AI, good content hygiene is even more important, where we need to ensure that only relevant data is accessible to improve accuracy and to avoid leaks. Finally, mission-critical business processes currently run across multiple systems, and so as these platforms develop agents with deep specialization, we foresee multi-agent workflows.
This is why we have been enhancing our AI APIs with the integration of Box AI and key partners like.
I'm Ben Kus. I'm Chief Technology Officer at Box, and I'm here today to talk about our AI platform. Now, as Aaron mentioned, the potential for AI in an enterprise is huge, but AI needs enterprise context to be successful. Many of our customers tell us they don't actually have an AI problem; they have a data problem. And for cost, unlimited storage, internal and external collaboration, integration with workflow, metadata, all done very securely, and more. And on top of that, we layer in the ability to do enterprise-grade AI, where our customers are in control of the AI. It uses trusted models, making sure that everything that you do is safe and permissions-aware. And then on top of that, we then layer in our AI content foundation.
This is where we provide the idea of model flexibility, including using our OpenAI GPT models, our Gemini models, our Anthropic model, or Mistral, and all underlying powered by Box context, and this context is critical because this is what then enables your agents to do the kind of specific work in your organization to accomplish the tasks that you need. At Box, we specialize in unstructured data agents so that you can get the most out of all of your unstructured data. These include things like our... and all of this is customizable inside of our Box AI Studio. You can build custom agents. You can now add knowledge so that the agents can reference material, in addition to selecting your model, testing it.
That's their content. Now everyone in the org can set up and manage extraction processes through an enterprise-specific content processes and deploy them at scale on a wide variety of content. And finally, Box Extract provides customers the AI tools to get reliable and consistent results. At its core, Box Extract uses agentic reasoning to understand the document and extract information. These agents are powered by our standard and enhanced agents that many of our customers already use. We then use advanced techniques to enable higher accuracy and reliability in the process. AI needs to be accurate to be helpful. This enables users to handle...
For a large retailer, our team receives hundreds of thousands of vendor contracts for all the different brands that sell merchandise at our stores. We needed a smarter way to extract data from these contracts so that we can manage them effectively. That's where Box comes in. To get started, we'll create a new custom extract agent to automatically capture key data. The catch is that since they are created by other companies, each one can be quite different in formatting and content. Take contract end date, for example. Some of our vendors clearly state an end date in their contracts, while others do not. Fortunately, I can solve this by giving Box Extract custom instructions. Box Apps makes that so easy.
My vendor app has a chart that categorizes contracts by status, so I click on this bar chart to filter to active contracts and sort from there. However, Box AI allows me to take it a step and still allow me to cancel one and read all the policy terms. I'll just ask my contract analysis AI agent of these contracts, which allow for cancellation or reduction in quantity. Our company created this custom agent to read and interpret vendor contract terms. The agent analyzes the contracts in this view and reports back that three out of the 15 have favorable terms, which will allow me to reduce the order. Right from here, I can securely share these contracts with the vendor manager who handles contract negotiation. As you've seen, Box Apps and Box Extract accelerate your business.
Thanks, Julia. We are already seeing incredible outcomes at customers using our data extraction. For example, Valmark Financial is now able to extract over 250,000 data points from complex insurance policies for downstream processing. In comparison, their previous solution was able to pull out about 4,000 data points. That is a 60 times increase in insights from their content. A leading investment firm was preparing for an impending audit and faced a daunting challenge of categorizing and identifying key information from the client-related documents. With Box Extract, they were able to process over 3.8 million pages in one weekend. And it's not only... chose the redesign of workflows as the number one contributor to their ability to see revenue impact from the use of AI. These leaders and enterprises as a whole are fundamentally rethinking how work gets done.
When it comes to case management process, it's content, AI, and people. And that is exactly why I'm incredibly excited to announce Box Automate. Box Automate is an agentic workflow automation built natively in Box. It automates content-based... These assets are curated and tagged with AI. They go through a publishing workflow and are available to everyone in the company with powerful solutions. Here's a property management app that a real estate company built to manage all their content silos.
My team spends most of their time reviewing loan applications and all the supporting documentation to determine which ones to approve. But I didn't hire them to read through mountains of paperwork. I hired them to make smart decisions. I know AI agents can help. Let's see how I can use Box Automate to optimize this entire process from application to approval. I'll start from a blank slate and drag in the first step, a form trigger. Next up, I'll add in some AI agents to automate work previously done entirely by my loan officers. We've got the extract agent that pulls data from the loan application and verifies that all the required information is there. There's a risk assessment AI agent that helps loan officers determine if the application meets our company's risk thresholds.
Finally, I added some Box workflow actions, like assigning tasks to my loan officers to complete the credit and finance reviews, along with steps for doc generation and e-signature. This is how easy it is to create workflows with Box Automate, and these workflows combine custom AI agents with Box's powerful content collaboration features without requiring any coding or help from IT to set up. Now that I've built the workflow, let's dive into the risk review and approval step you saw before. Earlier, I created this custom risk assessment agent to read through the loan application and submit documents like bank statements, pay stubs, and use it to calculate key metrics like debt-to-income ratio. Also, I supplied the agent with our company's risk evaluation. A new loan application is ready for review.
He clicks through the task and sees the agent's risk evaluation as low, along with the key data points it calculated. He also has the complete application package and extracted data, all in one view, so he can verify it as needed. The ratios look good, and so does the credit score. Having reviewed the application, Ray agrees with the agent's evaluation and approves it. As you can see, Box AI agents and Automate have revolutionized my team's operations by handling the document processing, analysis, and assessment in accordance with our company's guidelines. And this doesn't just happen once. Every time a loan application is submitted, my team of AI agents is there to lend a helping hand.
Thank you. Next, I'd like to welcome Manoj.
Thank you, Kailash. As you've all been hearing, Box is integrating AI throughout the entire content lifecycle, and as many of you could guess, this also includes security and governance. At Box, we work hard to remain the most trusted platform for creating and collaborating on your most critical content. With our products such as Box Shield and Box Governance, we've delivered security solutions across the content lifecycle, from our powerful malware and anomaly detection tools to classification labels, driving access controls and retention policies, to the new features like content recovery and Box Archive, but as we know, security is a moving target. The scale and volume of content continues to grow at a rapid pace. This is further complicated by the threats getting more complex and sophisticated.
AI agents offer organizations a chance to scale their efforts beyond anything we've ever seen before, better securing content security operations. It significantly increases the security team's reach and coverage for classifying content with a classification agent, helps them with analyzing and responding to alerts in an efficient manner with a threat analysis agent, and detects and protects organizations from ransomware activity. Box Shield Pro is in active beta right now and will be available as an easy upgrade to Box Shield. Let's first talk about the classification agent and how it's delivering step function improvement and increasing the reach and understanding and securing of the content. To have the best content protection capabilities, you want to be able to understand what the content is about and its sensitivity to the organization. Let's take an example of a technical architecture document.
It contains deep technical IP for the company, but not necessarily anything that is obviously sensitive, such as PII. Using traditional solutions, distinguish starting from customized prompts and applying security across the content of the entire organization. Now, I've told you a lot about what this agent can do, but let's take a look at it in action. Over to you, Julia.
Today, we'll take a look at how a manufacturing company called InnoTech leverages the classification agent to enhance their security posture. InnoTech designs tech products and uses their Box account to store sensitive information about their widgets. A product manager is about to upload documents into Box related to their product, Atlas. Some of these documents are technical papers explaining the inner workings of the Atlas cutting-edge design, and some of these documents are access. However, they can't always rely on end users to add the proper security classifications to content, and since technical papers don't always contain the same keywords, they're difficult to automatically find and classify. This is a challenging nuance for most classification tools to understand, but with the classification agent, the security admin can write classification policies in natural language, describing exactly what types of documents the tool needs to classify.
AI classification is perfect for this sort of challenge because it relies on more than keywords and specific identifiers. It inspects the context of the document based on custom criteria to determine the correct classification. The security admin at InnoTech wants technical papers to be classified with the confidential label. So they type out that documents classified as confidential should include engineering designs and technical documents. But these product designs aren't the only documents that InnoTech considers confidential. They also don't want to leak documents with PII, financial data, or information about mergers and acquisitions. The security admin is able to define all of these prompts and test the behavior before enabling them. Simply by defining what confidential means to them, InnoTech is able to monitor a large surface area of their organization's unstructured data, more effectively labeling and protecting it from security threats.
With the new policy in place, now when the end user uploads these documents into Box, the technical papers are automatically detected and labeled as confidential. Importantly, Box captures audit details, including the rationale for the classification decision and when the classification action happened. Inspect and classify confidential information across your Box account. Instead of solely relying on the presence of keywords or identifiers, the classification agent uses custom instructions to protect sensitive data based on real context, allowing companies to protect what matters most without slowing the business down.
Thanks, Julia. That was amazing. Our customers can't wait. Customers such as the Santa Barbara County Public Defender see significant value in deploying these agents. Our early beta customers agree as well. They have seen a 10x in addressing this challenge. Threat analysis agent takes large, detail-filled threat alerts that can be difficult to parse and refines them down into simple, straightforward summaries using Box AI. By distilling these alerts to the crucial information and context, using easy visibility into a new type of user that is agents, Box is committed to helping organizations keep their most critical content secure. That's it for me. Let me hand it back to Aaron now, who's going to wrap it up.
Innovation. Hopefully, you have a clear ability to come up with company's information. To deliver this innovation to customers, we are continuing to double down on our Enterprise Advanced plan. The all-new AI agent-powered Box Extract will be in Enterprise Advanced. Our agentic workflow automation capabilities will show up in Enterprise Advanced. Our new enhancements to our custom AI agent builder or new advancements to our no-code application development will empower our customers' work with their enterprise content. And with that, looking forward to opening up for questions to all investors and analysts. Thank you.
Great. Thank you. Welcome, everyone. And as a reminder, if you'd like to ask a question, the virtual audience, please, you can enter it into the chat online or email at ir@box.com. Our first question comes from John Messina with Raymond James. As we think about the pace of adoption for AI solutions and AI agents on the Box platform, can you talk about the delineation?
Particular knowledge. We're seeing great traction across really most industries at this point, with, I think, extra emphasis on things like public sector, financial services, life sciences, areas where we already have natural differentiation because we can offer customers a more compliant, secure AI experience on their unstructured data.
Only bold trend. A lot of applause for the two.
Hi, George Kershaw from Citi. Thanks for putting this event together. I wanted to ask about the AI world that you see coming here, and particularly where you see Box's role in terms of what types of use cases do you feel it makes sense for Box agents to handle within the Box ecosystem versus third-party agents doing the work outside of Box, but maybe leveraging Box data?
Yeah. Maybe I'll frame it up, and then I think maybe Ben from a platform standpoint could build on this. But one of the things that we've seen is how important context is for AI agents. And this was obviously a core theme of our keynote today, is to get set of, for instance, if you had your sales team inside of an organization want to instantly get feedback on a deal that they're doing based on all of the prior sales materials and training information that you had as a company. Well, Box is a very natural place where that data might already be living, and then you would build a custom agent in the Box AI Studio to let you go do that. Similarly, if you had an AI agent that you wanted to create to review contracts coming in for all of your Box Extract.
Yeah, I think we need to do specialized work for you. It turns out that there's a lot of work for every platform of different types: HR systems, CRM systems, structured data systems, unstructured data systems. And you need to get your AI agents to specialize in doing that and doing it really well. And then so we start to see that all of our partners start to develop their own agents. And then so as part of an AI ecosystem, we work with them. And so we'd have our agents talking to the Salesforce agents or talking to agents from these other systems. And that is, we believe, the emerging way by which these enterprise platforms are talking to each other.
So then our role is to provide the AI capabilities on unstructured data, but also just provide specialized AI agents that can then be called the MCP via Ada via API for our customers and our partners to call.
That's great. Maybe one more, if I may. You showed right at the end of your prepared materials that it seems like most of the new innovations you're announcing are included in Enterprise Advanced. From a pricing and monetization perspective, does that imply maybe the realized pricing on Enterprise Advanced adoption may be a little higher than what you'd previously talked about? Or is this more of a carrot-to-speed-up upgrades?
Yeah. I think we would still stick to that kind of 20%-40% uplift framing from Enterprise Advanced. And we've continued to see that from the first two quarters where the plan has been available. We want to just keep doubling down on the momentum of Enterprise Advanced. We're starting to build up more of the kind of steam for our Salesforce, really in every customer conversation, highlighting the capabilities of Enterprise Advanced. So we want to follow up with Box AI products. Is there more people looking for growth or efficiency? Thank you. Yeah. So I think we're certainly focused on both, actually, really kind of three dimensions.
So, price per seat increase because of Enterprise Advanced, seat expansion because of the use cases that we can now kind of increase the value proposition of Box for, and then a consumption dimension, which is this consumption of AI units for any high-volume AI over time that I think is maybe out there in the conversation simply because we are in the tens of millions of seats scale as a platform, and we're really the potential market for Box is hundreds of millions of knowledge worker, billion knowledge worker kind of scale demographic. So, I don't think that a seat compression in one area of a business or other would really kind of impact the overall seat dynamics in our business.
In fact, if anything, what's happening is by virtue of AI agents, we're actually seeing the expansion in seat areas that we wouldn't have sold into before because now all of a sudden in a business, again, the legal team might now have a use for Box because metadata from contracts, which more acceleration of their business and companies that want to drive more efficiency. We'll have scenarios where customers say, "Hey, I used to spend $2 million on this type of business process." With Box AI, I could imagine saving a meaningful chunk of that. So we're seeing kind of every variant right now of customers finding ROI potential within Box AI.
Thank you.
Thanks.
Thank you. The next question is from Lucky Schreiner at D.A. Davidson. Do you think adoption of more AI agents will be a catalyst for adoption of Box Shield Pro, or is the reverse true? And, when customer adoption of Box Shield with security functionality built in, are they more willing to adopt and trust AI agents?
Yeah. Maybe just the meta answer would be we continue to have this kind of core brand value proposition, which is we should be the most secure and governed place where you can manage your unstructured data. And so agents just give us another way that we can add to the data security of our customers and organizations.
Shield Pro would continue to increase the content gravity that Aaron is referring to, and we believe in it so more of the latter in the way the question was phrased, I think is going to help bring more traction to the rest of the platform, to everything else we offer.
Great. As a follow-up on.
Through all as they start to adopt more.
I think what we offer by virtue of us being kind of an applied use case on AI is very easy out-of-the-box use cases. This is why you're seeing so much adoption with things like metadata extraction, is just every company on the planet has unstructured data that they've always wanted to be able to pull out the structured data from that content. They've never been able to do that at scale. It's an instant out-of-the-box use case that then lets them go and automate downstream processes. We offer customers a lot of use cases with AI, maybe the AI sphere or all of the different agentic kind of capabilities inside of workflows. They don't have to redo the certification cycles because they've already approved the fact that Box does secure and compliant AI on your data, which is one of the.
Follow-up on functionality.
I mean, usually it's capabilities within Box by default lets you talk to any of your data within Box content or a marketing asset and just interact with it and bring the expertise from the model to bear with that content. With Box Hubs, you can take a collection of data, hundreds or thousands of marketing files or sales files or HR materials, and then let anyone interact with that data as a knowledge base that's intelligent by design. So these are, and that new Box Extract lets you do data extraction at scale. Then I think extremely importantly, and I want to underscore this capability, Box Automate then lets you start to deploy these AI agents in workflows in a repeatable fashion.
And so Box Automate will provide the underlying guardrails and the underlying kind of pathway for agents to then go get involved in more and more mission-critical work that you want to deploy at scale, and you want that work to happen as efficiently and in as repeatable of a manner as possible. And so we had to build a next-generation workflow automation system to go and be able to power that, and it's agentic by design.
Shayne, my mind goes to Relay. I mean, Relay.
was obviously our first foray into workflows. Relay is a little bit like an if-this-then-that type structure. So you say, "A file comes into this folder. I want to move it to this person, add a task to it." So it's simple by design, highly powerful because you could do that tens of thousands or millions of times. As we saw agents enter the scene, more advanced, some kind of workflow automation. Automate is a full business process builder of your workflow. It can be as complex or as simple of a process as any customer would like. And then you can choose whether either there's system events that occur, people are involved in those system events, or if agents then get dropped into parts of that workflow. So again, we got to build this with agents right front and center on day one of our workflow system.
Basic set of.
Thank you. Just a follow-up for me, Morgan Stanley. So more and more we're hearing customers, or sorry, enterprises talk about AI lowering the barriers to service different types of customers, and would be curious to know how your different innovations are allowing you to service new customers that you wouldn't have been able to service before.
Yeah, so I think maybe at a high level, I'll share a quick thought, and Diego, if you want to add to this. But dramatically we are dependent on how many people do they have in those functions to go and review documents, to review data, to move documents through that workflow. And so by definition, there's a kind of a TAM constraint to some extent as to how many customers have that sophistication, how much have the teams to go and do that, do all the lines of business, have the talent for those types of workflows. AI agents effectively are bringing that talent in. Now we can go to a business lead wasn't a particular problem.
And with the power of AI agents, we can actually go in and you saw an example of this on stage today with Box Apps of the kind of full breadth of apps that we can begin to enable for our customers. You saw an example of contract lifecycle management. You saw an example of sort of a wealth onboarding type lifecycle. We talked about things like insurance processing. So these are all new applications and categories that we can be entering because agents traditional applications capabilities reintegrate with. And smaller companies have small departments that actually cannot handle too many of them. So the smaller companies want to work with a vendor that actually solves the problem for them. The bigger companies want to remove 200 of the 300 applications they have.
So we have actually an amazing opportunity in front of us because across the whole market, small or big, companies need a way we apply to all of them. So applicable to is yeah. Actually, and if I can build on that because that's a very key point, a way for your resources to solve them. So to exactly Diego's point, we can now bring more use cases. So I answered the question more from the lens of the labor side. But if you just think about even the software use cases, small companies typically weren't in the market for having a contract system or a full digital asset management system because they just didn't have the scale where their processes kind of require that, and they didn't have the people to go implement them even if they did.
So we can make those types of use cases incredibly simple out of the box and then expand the number of workflows that we're powering in those organizations and then bring it all into one single platform. So that's where we can get a lot of leverage. I called this out, I think, in the last earnings call where we had deals in Q2 that automation.
I'll just give it back to you, Aaron, for any closing remarks.
Sure. Yeah. Thanks, everybody, for attending in person or virtually.