Explore the complete conversation below

Doug Cullen (01:09.20)

Just wanted to get some perspective from you on what do you think AI means for dealmakers, and how are we thinking about it collectively in terms of the impact that we can have on the market?

Raj Bakhru (01:22.17)

It's been transformative. The last few months, we've seen step changes in what's possible. We've seen models generated. We've seen entire IC memos generated. We've seen work against a data room that's never been possible before. AI is in a new place today. Over the course of the last year, it's just completely transformed the way that we're getting work done. And that's for both the buy side and the sell side. We're seeing folks prepare and create data rooms much, much faster, find issues, and reconcile data. On the buy side, we're seeing folks work through diligence a lot faster. So people are empowered. AI is actually having a real, material value that they're seeing ROI out of it, which is awesome.

Doug Cullen (02:00.28)

Yeah, I remember, you know, you and I, not too long ago, thinking about content creation as an example, that sort of last mile, I feel like maybe even ten weeks ago, we were sitting down and saying, oh, this is probably a year or two out. You know, we didn't think the sort of LLMs would be able to get there in terms of the high-quality content. And yet, here we are. What are some of the things that we're doing, you know, with Blueflame AI, in order to solve that problem.

Raj Bakhru (02:28.21)

We brought the best of all the LLM worlds in for content creation. Now, it's been super exciting to see people leverage this tooling. We've had over 3,000 Excel and PowerPoint files created in the last month alone. So that's a huge volume of content creation, and that's across LBO models, operating models. It's across IC memos. It's across customer cohort and retention analysis. Just a huge amount of really valuable, time-consuming work that's now getting handled largely by the LLMs. Not 100% of the way there yet, but we're seeing it make material gains and get 70% or 80% of the way there for folks.

Doug Cullen (03:06.24)

So, you know, as you know, when we met about a year ago, we sort of had our own organic vision for what we could do within the data room. And we've done some amazing things. But one of the things we focused on very quickly was really the opportunity that we felt was created by Blueflame broadly, because you were sort of building this, a platform focused on the buy side. And we literally dreamed about an opportunity to potentially, you know, make some of those Blueflame capabilities available in the native data room experience, specifically for Datasite Diligence and Acquire, and maybe even Prepare. Can you tell me a little bit about how that journey went and then kind of help us understand kind of where we are with that?

Raj Bakhru (03:53.15)

Yeah, we were equally excited about that opportunity, obviously. And that's why we came together. We knew that there was just so much anxiety, time-consuming nonsense that goes into syncing content, right, downloading it out of the data room, trying to interact with it manually. We wanted to take all that away. We knew it was going to create a world where you'd just be able to directly interact with this content. It would be able to come to you and say, these are the issues that you should be focused on in this data room. And the data room is just where all of the content and all of the real material lives, and deals need to move faster. AI is obviously great at working through unstructured data and making deals move faster, so we knew it was the home, the right home for all of the AI tooling that we were building. Over the past year, we've endeavored on a number of things together.

We wanted to make sure that we AI-enabled all of the content, and what that means is you can't just pass the content directly into an LLM. There's security concerns or privacy concerns. There's also just fidelity concerns. You want to make sure that you're reading through charts and graphs and logos with really high fidelity so that you don't make a mistake in your write-up. If you're trying to, frankly, be cheap or use the wrong models or things like that in processing this content, you can make a really bad mistake. And the stakes, as we know, in these deals are so high that you need to get it right. So we spent a lot of time working together to make sure that we knew how to work through watermarking, and redacted content, and finding all that really high-fidelity content, processing the tables, and bringing all of that into Blueflame, exposing that to Blueflame users.

And now, more recently, what we've said is Blueflame's got all this great capability for largely our client base. But we recognize not everyone's going to be a Blueflame subscriber. We have folks who have corporate edicts who say we need to use Claude, or ChatGPT, or Copilot, and we want Datasite in all of that. The value of the AI that we've been building to be accessible to every dealmaker out there. So we've been embarking on a journey to integrate MCP and bring Datasite's AI tooling through MCP into Claude, and ChatGPT, and Copilot just the same way that you have it through Blueflame today. So the first step, right, was really connecting Datasite into Blueflame so that Blueflame subscribers would get, and, by the way, this has really never happened, right?

The opportunity to get Datasite data for all their deals that they're looking at, and really be able to utilize the Blueflame toolset in order to interrogate that, which I think has brought tremendous value to some of the private equity firms that we have.

Doug Cullen (06:25.27)

It's been huge, and people have loved using it and interacting with that VDR content, working through diligence on the buy side, working through their IC memos off that content, finding and reconciling data and building data packs and all that type of stuff. Now, through MCP, we're exposing an additional layer of tooling that's largely sell-side focused as well. Now, you'll have the ability to create data rooms directly and populate those data rooms, and leverage AI to auto-sort and index that data room, to add people to the data room. You'll have to interact with much more of the structure of the data room for sell side, just the way that Blueflame's been bringing huge amounts of tooling to the buy side.

So, building on the initial set of capabilities, we'll bring Blueflame into the data room experience, and now you're kind of even putting another wrapper on top of it, or another layer, which is MCP. So I think, confession time, I'm not even sure I knew what MCP stood for about three or four months ago. Now it seems like we're talking about it all the time. Can you give me a little bit of a sense of like, what is MCP?

Raj Bakhru (07:33.02)

Yeah, it's a fancy term: Model Context Protocol, that's a standard that was built for AI agents to communicate with systems. It's a fancy API, at the end of the day, designed around AI agents. And what AI agents can do with MCP is interrogate a third-party system and interact with that third-party system. So, the natural language that you have of, 'tell me what projects I have,' tell me what the files are in that project,' tell me what text files we have for this deal.' All of that can be translated directly into queries against the APIs. The MCP server that Datasite's now hosting, and that means you can really do anything that a system would want to be able to do against Datasite content and index the members, as well as read through the materials in a data room. So, you know, and this has not been available historically.

Doug Cullen (08:22.11)

Right. So you had to log in to Datasite as either an administrator or a reviewer. You'd have to sort of use our interface. And now what MCP is allowing us to do is sort of make that experience prospectively available in another environment altogether. So something like Claude or ChatGPT Enterprise, are you saying that I could be in that type of operating environment and effectively connect into a project or multiple projects?

Raj Bakhru (08:53.17)

So we were groundbreaking with having Blueflame's access into managing your data rooms. And now we're going to be groundbreaking with giving that access to Claude Enterprise, ChatGPT Enterprise, and Copilot to be able to interface with your data rooms through MCP as well.

Doug Cullen (09:09.09)

So, you know, these platforms have become a bit prolific with the growth of Claude, and certainly ChatGPT and Copilot. What's your perspective? What do you sort of think about people maybe wanting to do within that operating environment? So I'm sitting on my desktop, I have this tool which I'm probably utilizing for a bunch of other use cases. But as you started to think about with the product team, what were some of the things that we dreamed about prospectively an administrator on a project wanting to do?

Raj Bakhru (09:42.21)

Yeah, you know, we had our own ideas, and what really surprised us and part of what we've loved about this journey of building it is we have pilot teams using this today, and they've had their minds blown and come up with a ton of their own ideas in leveraging it. So really, the sky's the limit in how you interact with it. We, you know, contemplated the obvious things. I want to be able to set up a data room. I want to be able to permission it. I want to be able to upload my content. I want to be able to index my content with AI intelligently sorting it. I want to be able to find certain content in the data room. What we've seen folks using it for is a level beyond that. It's leveraging reasoning and deep search capabilities. It's reconciling data points; it's building customer cohort and retention packs. It's building a CIM.

I think we built the first-ever prompt-generated CIM off a data room, and that was by leveraging the interaction through MCP, of being able to query the data room, assemble the slides directly. Leveraging the LLMs literally would have saved weeks during our last sale process at my last private equity firm. So you can, you can really do anything with this content now. And if you think about all of the workflows that happen during the deal process, they've all just gotten infinitely faster.

Doug Cullen (10:54.01)

And talk to me a little bit about context and the context window, because one of the things that I think we realized very quickly when we started to take Datasite content and make it available for the Blueflame subscribers, was we have a lot of content in some of these data rooms, I think probably exceeding some of what you initially may have believed. So talk to us a little bit about that, because one of the things that I think the models, although they're getting better and better and they're increasing their context window, you know, they maybe have, you know, 1,500, 2,000, maybe pages worth of content. We have hundreds of thousands of pages in a data room. So what are some of the ways that we've made that context a little bit more accessible to people, either in Blueflame, in the data room itself that is Blueflame-empowered, and then ultimately through MCP.

Raj Bakhru (11:46.08)

Yeah. So you're absolutely right. You can't fit an entire data room into a context window. It's just not big enough. Even with the largest context windows in the millions today, it can't fit an entire data room, and it never will be able to effectively. Even at those larger context window sizes, you see recall rates drop materially. So the LLM, while it can fit the content in, technically, can't actually understand and refer back to all of that content. So there's a lot of techniques that we have to leverage in terms of AI enabling the content. We have to push through all the content. We have to chunk and vectorize it, all of these technical things to allow for deep search through all the content. We leverage something called GraphRAG. So we're actually finding named entities— people, companies—building edges and connections between these things.

So we know Company X is referenced in this doc. And this is the board for Company X and the management team for Company X. And these are the vendors and their suppliers. And we build this entire graph to understand the relationships out of all the content that's being built. Then we leverage that graph. We leverage all that information that's within all of the documents in the data room to execute searches and answer questions that people have off all of that content. It's a very similar technology, groundbreaking but similar technology to what Google's been using for years in terms of how they index the entire internet. This is just a lot harder to do, because we have to build all of those nodes and connections, and we have to do it in a permissioned way. We have to understand who's able to see what, right, through that entire graph. So we're maintaining security throughout.

The process of searching is non-trivial. The process of finding the right answers is non-trivial. So because you can't just stuff this whole thing into an LLM and say, you know, here's my entire data room. We work through a lot of pretty advanced techniques to enable search, and that's a lot of what Blueflame is built behind the scenes to allow ChatGPT, Copilot, and Claude to ask questions of the data room through MCP

Doug Cullen (13:39.24)

And what types of questions do you see people asking? And then just trying to think about maybe a traditional way that would have been done, you know, a couple years ago with different search techniques and really before RAG is what it is. I mean, what types of questions do you think are really now at people's fingertips that maybe never were before?

Raj Bakhru (14:02.12)

Yeah, I mean, it's all the way from the easy ones that you still have to flip through a lot of materials to just find the answer. Things like: who's the chief product officer? Who's on the management team? What is churn? What is customer retention? What's their gross retention and net retention, all the way to the much deeper analyses of: If you were to look at all of the customers in Q1 of 2025 and look at the cross-sell that's been executed on those, you know, what's been the take rate on that cross-sell? It can actually do that deeper analysis through a multitude of queries that it'll run together and stitch together for that analysis. So it can be anything from the single one-off. And where you see a lot of that is actually the MDs and seniors, where they feel like they should know something, maybe forgot it, have a lot of things going on.

They don't want to pester a junior analyst with a simple question. Easy to just go fire off that question, get an answer off all of the content and I'll find the right answer. Cite it back to you, tell you exactly where it got it from. And give that to the analyst. If you're doing much deeper and more intense work, you're able to execute that entire workflow, often through single-shot prompting, and a single sentence is able to run that entire analysis.

Doug Cullen (15:13.13)

Yeah. I mean, I remember you describing to me because, you know, single shot was sort of a new language to me, right? But ultimately you're sort of asking a question. And then behind the scenes, Blueflame is effectively triggering a bunch of other inquiries by the inference of what we think the user's looking for. Have I got that right?

Raj Bakhru (15:35.02)

Yeah, exactly. And what we've actually seen over time is the complexity of the planning and the number of tasks that get executed to complete that piece of work has gone up dramatically. So we used to have task-duration horizons in the single-digit minutes at most. Now you'll sometimes see an agent run for 45 minutes, an hour, and it's really crunching through hundreds of tasks. We actually had something recently in our user interface where we show all of the tasks that it's running on the right side, and we realize, actually, that list is getting too long for the screen, and now we have to collapse that because it can run so many. So the agents are getting really smart. The reasoning—they used to lose the story, if you will. They used to lose track of, 'What am I trying to get to?' Context windows would blow out. They weren't able to follow the trail.

Now it can go for pretty extended periods of time on much more complex work.

Doug Cullen (16:25.05)

Yeah, because I don't think users really want a black box, right? They don't really want to put in this inquiry and just have it spit out an answer. You know, you talked about sort of streaming the reasoning behind the inquiry, like how do you how important is that to kind of show your work, if you will, particularly for dealmakers, where, you know, the stakes are pretty high.

Raj Bakhru (16:45.23)

Transparency is huge across the board, not just in terms of the way that it's executing the plan of action, but all the way down to referencing sources. It's kind of a trivial thing that people don't think a lot about. But if we're citing a statistic back to you, gross retention is 92%. You need to know where that came from, because if your MD asks you and says, you know, actually, I thought it was 91%, you need to be able to say, 'This is exactly where it says it's 92%, right?' Justify and validate all of the material that's coming back. So citations are actually something that we put a lot of focus on: How do we make sure that we can tell you exactly where we got that data point from? We're actually also seeing a lot of really interesting work around reconciliation of data.

It doesn't sound very sexy, but what you'll find is that sometimes a CIM says something, sometimes a data pack says another. Sometimes there's a business update that changes that number, and you want to really see the time series of of all of these numbers and find cases where things just don't jibe. Right? Like this number looks really funny or it doesn't make sense in context of that number. And now we're seeing things like teardowns and validations that are doing a much better job at spotting those things. So if you're on the sell side, what this does is it drops the number of questions that are coming in, and that speeds up your process. It gives greater confidence to the buy side when they're putting in bids. On the buy side, what it does is it gives you obviously more confidence and gives you a leg up, makes you find things that you might not have otherwise thought about.

Doug Cullen (18:09.14)

And I'm going to date you a little bit. We were sort of talking about you selling your company in 2015 to to ACA at the time. So you were a dealmaker. You ran strategy. I mean, just give a little bit of a compare and contrast. Like how did this used to get done? I mean, we've now lived in this world of AI, certainly within the Blueflame context for a little over two years. I mean, how did this work this used to get done on the sell side, or even your experience on the buy side?

Raj Bakhru (18:37.19)

Yeah, I mean, the amount of desk work that was happening by the analysts, it was just, you know, truly painful. I felt so bad for the analysts at our bank, or at our investment bank, when we went through our last sale process, and they sat there ticking and tying and working through the operating model. And our CEO was on the call and saying, 'I want that number changed, and that formula looks wrong, and this doesn't tie out. And why, when I change that thing, didn't it update that thing?' Now, all of that is just so much faster and easier because the models are being built. You can fact check them, you can audit them. Different LLMs are really good at actually validating models versus creating them. And we can get the best of both worlds on Blueflame. So the world has really changed in the way that you get this done.

Just the manual work that was happening in Word, Excel, and PowerPoint in particular, and in building all these documents is just dramatically easier today. It was easy to make mistakes. It was hard to find cases where data didn't match up. It was really hard to massage the data that was coming from the selling company, and getting that into the right form and trying to make sense of the way they book things and where they were finding adjustments and things like that. So all of this is just much, much easier. You don't need to go back and ask the question. You can directly find the answer when you need it, just, you know, far faster, less friction, less pain. And frankly, people are enjoying their jobs more because they can do this through AI.

Doug Cullen (20:01.08)

We think a lot about how do we how do we turn people into prompt-first dealmakers, right? In the coding world, you've had a lot of people who have gone prompt-first with coding. You know, you go to Cursor, you go to Claude Code and build out an entire application with a prompt. We're seeing now that you can actually build entire data rooms with this MCP connector. You can now conduct diligence just through prompting. And what we'll see transpire is that that's going to become the preferred way of getting deals done. And Raj, you sort of live on the edge. I think you are in front of almost everyone I've ever listened to on the topic or think about it, but not everyone's there necessarily. You know, we do live in a compliance-driven world. Bankers, lawyers, accountants: precision is paramount.

And I think for a while, this idea that I would be taking sensitive content and exposing it to an LLM seemed like it was crazy-town. Talk me through that a little bit. Maybe for someone that's not as comfortable with this content and is very worried about exposure or breach, what are some of the guardrails that we have in place? How do we approach security? I'm just looking for a little bit of a reflection, knowing that everyone's not Raj in terms of the knowledge about these tools. And I think we want to give dealmakers a level of confidence that this is safe.

Raj Bakhru (21:34.02)

Yeah. So we were obviously super thoughtful about all of this. Datasite's brand is built on compliance, security, and privacy. Blueflame's brand is built on security, privacy, and compliance. And that's part of the reason that we wanted to take our time to design this right, architect it correctly. Make sure that we got permissions correct, and redaction and watermarking and disclaimers and all the things that you would expect us to get right. We also wanted to make sure that any interactions with the LLMs were held to the same standards that we would hold them to. So that means things like zero data retention with the LLMs; they don't have the ability to even log what we're doing with them. So if they have a breach and someone sees their logs, it doesn't matter. Our content is not there with the data room specifically, we didn't want to hand over the entire data room to the LLMs.

We just didn't. We didn't love the idea of the data sitting with them. Part of the benefit of Blueflame being behind this is actually all of the data room parsing and vectorization, that all of that data that has to be leveraged for search still sits within the Datasite ecosystem. It doesn't actually go over to the LLMs until you query something and say, I want to ask a question of this. It'll send back the minimal amount of information that it has to send back to the LLM. So we're being very, very thoughtful about security, privacy, and compliance, as you would expect Datasite to be.

Doug Cullen (22:52.15)

Yeah. I mean, it was one of the reasons why I think we got on so well, right, is the background that you had previously. And just generally the understanding of the Datasite brand, what Blueflame was building. It was really quite vertically specific. I think that was super, super relevant. And I guess just talking a little bit about maybe I would say not the current, but the future, because, you know, one of the things that we've launched recently with Blueflame are these add-ins within probably the most powerful application that any dealmaker uses, arguably Excel. Can you talk a little bit about the add-ins, their maturity, the ability to really solve big problems within either Excel or PowerPoint, which is probably where a ton of dealmakers spend a lot of time, particularly around that, that last-mile content generation and/or just digesting of the information.

What's been our experience with some of the Blueflame customers around the Blueflame add-ins?

Raj Bakhru (23:57.18)

Yeah, the add-ins have been super valuable. Obviously they're relatively new, so everyone's still learning the best ways to use them. We announced our Excel add-in about two months ago to beta users, and we've been rolling it out gradually. And we're going to do the same thing with PowerPoint. What you can do with the Excel add-in is just tremendously time-saving. It's all the similar things that you could do through Blueflame directly a while ago, creating Excel documents, but once you have that Excel document, being able to manipulate it directly in Excel with the add-in is just so much faster, so much easier to iterate with it as well as to check content. So if you want to say things like, you know, 'Just find any formula errors,' it does an exceptionally good job at saying, you know, this one cell doesn't match the others. Or I think you've miscalculated EBITDA.

This is the way you should think about it. So it's doing a tremendously good job at helping you get confidence in the work that you're doing, as well as doing that work itself and making changes to the document itself. We expect to see the same thing in PowerPoint, being able to create the deck. We're already creating tons and tons of PowerPoint decks at this point, but being able to iterate with those is super valuable. Oftentimes, you know, as I mentioned, it comes out 70% or 80%. Getting it to 100% might require a bunch of prompting back and forth and saying, actually, I want this there, or I want you to cite this stat, I want you to, you know, focus less on this content, focus a little bit more on this. Providing that guidance that steering directly in the add-in is just a much better experience. So super exciting.

We've got Datasite feeding into Blueflame for people that are enterprise customers of Blueflame, huge amounts of value being created, then Blueflame is coming into the Diligence and Acquire experience, and Prepare as well. MCP can kind of meet them where they are, right? If our dealmakers are in love with Claude or ChatGPT Enterprise, they'll be able to operate there. I mean, it's a fast-moving space.

Doug Cullen (25:59.02)

So I guess, could you give us a little bit of a sense of where you think things are going? You know, some of the things maybe a bit more topical now around the prospective next releases of the frontier models, Mythos or Spud, you know, some noise, you know, out there in the market around reactions to it. As I said, you are really sitting on that bleeding edge of the technology. I mean, how do you see this panning out and kind of what would be your advice to some of the people that are newer to the adoption? Or how do you think the world's going to evolve from here?

Raj Bakhru (26:36.19)

Yeah, I know the pace of change can be really scary. It's actually obviously also very exciting. And I think what's actually happening is we're seeing a lot of step changes, one right after another, such that we're getting to a point where AI is actually becoming more accessible, not less accessible. A lot of people get scared by the number of changes and say, I don't, you know, head in the sand. I don't want to play with this. The reality is, if you were to go to AI today versus if you were to go to an AI solution two years ago, you're going to have a dramatically different experience, right? It can do so much more today. And those step changes are continuing. So the new Spud and Mythos models— we're quite excited about them. Right. They're going to bring another layer.

The task durations of being able to run agents that last for 45 minutes to an hour really well are going to go way longer. It's going to be five, six, maybe even a full day soon that you could run an agent and have it, you know, execute a really long-running task for you. So just the capability set is dramatically improving. That's going to flow straight through. So everything that we're doing with MCP connections of Datasite and Blueflame into the LLM providers means that you can have it operate against that data room with longer context and smarter reasoning and deeper reasoning. Those step changes that we've seen over the last few months of being able to generate content really well, you can now leverage your data room content and build those Excel files and PowerPoints directly off the data room or into the data room and populate the data room with that content.

And that's all facilitated through the MCP connectivity and facilitated through these agent model step changes that we've seen.

Doug Cullen (28:10.05)

And also, I think one of the things that was very early for you in terms of a value-realization opportunity, was the notion of a multi-model environment, which Blueflame is natively, of course. And now Datasite will be by extension. Why multi-model? I mean, what was the initial thinking, and why do you think it matters and why? Why do I care as a client ultimately that you're multi-model versus maybe picking one of the frontier models and running with that?

Raj Bakhru (28:41.11)

Yeah. You know, the leapfrogging effect is still very real today. You know, different models are better at different things. That changes over time and actually changes surprisingly fast over time. If you think about it today, Claude is still the best, or is the best, at code execution and leverages that code execution to create PowerPoint decks and Excel models. GPT-5.4, which was recently announced, actually is better at checking Excel files than Claude is, so you can actually pair them against each other and vet them against each other in a really nice way. And Gemini is actually the best at multimodal and content creation, particularly infographics and charts.

So if you're thinking about, 'I need to build a CIM,' or I need to build a teaser or I want to build an IC memo, obviously there's a lot of infographics, technical architecture diagrams, things like that you probably want to put in there. You'll actually be best off pairing Gemini's content creation of those charts, graphs, images or diagrams, and pulling that into Claude's content generation. So pairing them together can have these really nice synergistic effects. Even xAI's Grok, which a lot of people write off, but it's actually the best at real-time data if you were to ask for, as an example, stock quotes from all the major LLMs, Grok is going to get that the best in news in particular because of its integration with Twitter/X, and it actually is the best news feed out of all of the LLMs. And then that's not it. That's all the frontier models.

I wouldn't write off all of the open source models. There's a long tail of open-source models that are being built in the US and China that are extremely capable. And actually some of those models have even better PowerPoint and Excel generation capabilities and code generation capabilities. than Claude today. So we're actually seeing a lot of leapfrogging. We're seeing this change quite dramatically. We'll see how Spud and Mythos change the equation again, updating sort of where you want to go for each of these different tasks is our job. We don't want to make that our clients' job. So we build all of this logic into Blueflame to say, for this, you want to go there, and for that, you want to go there and just try to leverage the best of all of them.

Doug Cullen (30:50.26)

And that's fully abstracted. The user doesn't have to think about that at all. Right? There's a Blueflame agent when I'm utilizing it. I don't need to parse, you know, which model is good at or bad at. I'm just asking it. And we're responding effectively with the most effective model and the most kind of the highest quality response.

Raj Bakhru (31:12.05)

Exactly. Our data science team has taken on the work of doing all the benchmarking and deciding where to go for what and who's best at the current time. So we bake all of that in and it's behind the scenes.

Doug Cullen (31:23.02)

And how frequently does that change? I mean, is it a matter of like, hey, these models are always good at these things, or does that actually evolve a lot?

Raj Bakhru (31:33.04)

Oh, it evolves quite a lot. And actually, surprisingly, not even just when there are new model releases. We've actually seen recently, For example, you know, one of the major model providers has been overwhelmed with load. And there's a lot of talk out there. They've actually taken the computing complexity and thinking time down on the model. Behind the scenes, it's transparent to everyone. But it's noticeable. And that's actually changed some of the benchmarks that are out there around who you might want to use for different things.

Doug Cullen (32:00.06)

And you talked a little bit about open source. I mean, this could have been an alternative approach. A lot of our clients have been building their AI capabilities. Many of them have sort of decided to run their own infrastructure with maybe these open source models, I guess. How have you advised clients around, okay, well, that is an approach? Why do you think this multi-model and kind of highly responsive model is probably going to yield a better result?

Raj Bakhru (32:30.04)

There's been no question that open source has made a lot of strong strides. It's not up to the same level as the frontier models and not up to the same level as frontier models in terms of broad usability. They'll be the first to admit they're particularly good at different things. If you look at it, most people don't know what Mercury Coder is, but it's actually a closed LLM that's a competitor to some of the frontier models on coding, much faster, doesn't use a transformer framework that people are seeing out there. It's a diffusion-based model, which is relatively new for coding, and you only want to use that for code generation. You don't want to use that for anything that's multimodal. You definitely don't want to use it for anything that requires real-time data.

There's a lot of specialized LLMs that are coming out, and pairing them together can just provide much better outcomes than going to a single frontier model or going to a single open-source model.

Doug Cullen (33:22.10)

And so, what are you most excited about? I mean, we're sort of on the verge of a pretty incredible moment for both our partnership as well as just dealmakers as a whole, you know, bringing Blueflame into the most prolific data room that exists out there in Datasite Diligence, Acquire, and Prepare. We, of course, have this incredibly enriched experience for Blueflame enterprise customers and subscribers. And now we are laying in MCP, I mean, what's next? Like, what are you so excited about? Like, what can't you wait to see happen for someone to try within one of these frameworks?

Raj Bakhru (34:05.02)

I, you know, honestly, I'm astounded by what people do every day. I get surprised by some of the things that our teams are doing and users are doing on the platform. So they've gone a level beyond even the types of things that I've ever dreamed of. But we're seeing that prompt-first dealmaker really come to the fore, and that's someone who says, 'I have a deal task,' and they go first to an LLM to get that done. That's really exciting to me to see this transformation because that unlocks so many things. It unlocks more deals. It unlocks faster deals. It unlocks better deal certainty, and everyone wins. If we can get this whole system moving faster and better and much less grunt work and grind in it, people enjoy their jobs more. So much benefit that can come from this. So that excites me with the MCP connections that Datasite is doing.

Now, I know that we're opening this up to a much bigger world, and that's really exciting to just see this larger population get the benefit of all of the tooling that we built out. And, you know, there have been there are plenty of firms out there who say we're a Copilot-only shop, and we've said to you, 'Sorry, you don't have Blueflame. We can't help you very much, even though we know this would be super valuable to you.' Now, we can actually solve that problem. We can say, actually, we can plug Datasite and Blueflame into your Copilot, and you can get the best of being able to work with Datasite directly in your Copilot and interact with querying Blueflame directly out of your Copilot. That's going to be transformative to so many people who just didn't have this access into our tooling before.

Doug Cullen (35:33.15)

And then I guess the last thing I'd ask, and maybe just as a wrap up, like, you know, it's always hard to anticipate where these things go. And I, like you, am absolutely fascinated by some of the things people do within the platform. I feel like our clients are really pushing us continuously to make it better and better. Can you just talk a little bit about maybe the Blueflame-Datasite ethos of responding to customers? How quickly do we try to sort of exceed their expectations? I mean, we really take the feedback super seriously, and I think it would be just great to hear from you in terms of how maybe the customers interactions have really shaped what Blueflame is and likely what Blueflame and Datasite will be.

Raj Bakhru (36:15.04)

Yeah, it's a huge part of the ethos and the way that we approach product building, and that's true on both the Datasite teams and the Blueflame teams. One of the things that drew us together was that we both really believe in white-glove service with our clients and making sure that we were hand-holding them where they needed to be handheld, and we were taking all of the feedback from those hand-holding sessions and trying to make product better and improve it in different ways so that people can get things done faster. That's always been a part of our ethos. It jibed too when we first met, we saw the same thing, and that's a lot of what's been happening here. We've been building. Part of the reason we're releasing these MCP connectors, for example, is we heard our clients say, 'I have Claude or ChatGPT or Copilot, and I can't have another tool.

IT is not going to allow that,' or whatever it may be. And now we're responding to that feedback. We're saying we can support that as people experience the new tooling and they start to work against the data rooms, and they start to query them and build really rich content and leverage the LLMs. With all this content, we're going to get a ton of feedback, and we're going to respond to that really quickly. And we're excited about that. That's what we get up to do every day.

Doug Cullen (37:26.18)

Yeah. And I think it's like we're dealmakers, right? I mean, that's really how we focus our time and our energy. And, you know, the MCP story is a fascinating thing. I mean, you literally came up with that idea about three weeks ago in terms of really building this type of rich capability around MCP. Our teams collaborated, kind of launched it out there, and it's fully functional right now in production and soon available via the different publishing environments of the frontier models. So Raj, I got to ask you. I mean, there's been so much going on. Can you give me a little bit of a snippet? What makes you really excited these days?

Raj Bakhru (38:05.01)

Oh, there's so much. The world is so fast in the AI world. The space is changing so fast. So many new and fun things that we can take advantage of. I'm excited for these MCP connectors to get out there. I know that there are so many users who are going to get value out of this. Being able to connect directly to the data rooms and leverage Blueflame AI Search and be able to automate so many more of their tasks because they're confined to Claude or ChatGPT Enterprise or Copilot. Now they have access to this rich set of capabilities that we've had and they've not had access to before. I'm excited by what they're going to surprise me with. Every time we put some of these tools out there, people come to us and say, I was able to do this, and it was really cool, and we hadn't even thought of that. So we will be pleasantly surprised.

I know that there are going to be people who save hours, and they tell us about how they saved hours and got to go play tennis with their kid because they saved hours. I'm excited for all of that, and I'm excited for how this space is going to change really fast. We have new models coming out pretty much every other week. at this point. The capabilities are stepping up so fast. These are step changes. They are not incremental. So where we're going to be in six months, we're definitely going to be at a place where we have prompt-first dealmakers and folks who say, 'I have a task to do on a deal; I'm going to go prompt that and get that done.' And that, to me, is just so exciting to see this whole space and process and workflows change in front of our eyes.