Unlocking Collective Intelligence with AI Search and Knowledge Management | Hillary Thompson of MBH

My name is Christopher Parsons. I'm the CEO and founder of Knowledge Architecture. Welcome to Unlocking Collective Intelligence with AI Search and Knowledge Management at MBH Architects. Our guest is Hillary Thompson, who is a principal at MBH and leads up all things marketing and brand as well as knowledge management, learning and development, and AI. Just that. Hillary is fantastic. We've had the pleasure of working with her for— and her team for a long time, and she's one of those people Early in my career, you know, somebody told me you can't innovate without innovative clients. And I have found that to be true more and more the deeper I get into my work. And Hillary is one of those clients that makes it possible for us to go forward because they're willing to go out on the edge with you and push. And so I'm so grateful that Hillary has decided and, and agreed to share her journey in terms of knowledge management and AI and what's happening at MBH. And we're really excited to have you. So welcome, Hillary.

Thank you so much, Chris, for that nice introduction. Sure.

I want to say one thing before we get started, a quick disclaimer. It's a good disclaimer. Um, Hillary's going to show some new stuff that people haven't seen. So Hillary's in the Knowledge Agent private beta. Um, if you see— if you like what you're seeing, you can email us at support@knowledge-architecture.com if you wanted to be added to the waitlist for Knowledge Agents. Um, And with that, I'll turn it over to you, Hilary.

I am incredibly excited to show knowledge agents. So thanks for following along. Thank you so much, everyone, for joining today. I want to start today with something that I think, especially based on your responses, might feel pretty familiar to a lot of you. Most firms right now are not dealing with a lack of knowledge. If anything, it's the exact opposite, right? We have more projects, more data, more documentation, more tools, more messages, more communication channels, more lessons learned. And yet when we actually need something, when you're in the middle of solving a design problem or a, you're in the middle of a proposal, you're trying to answer a client question, it can feel surprisingly hard to get the right information. It's not because it doesn't exist, but because it's not usable in the moment that you need it. So today I wanna walk through how we've been working on that problem at MBH Architects, not in theory, but very practically, and what we've tried, what worked, what hasn't worked, and how AI search and now knowledge agents fit into all of that. So this is Unlocking Collective Intelligence with AI Search and Knowledge Management at MBH. Just a little context on MBH, um, because it matters for how our knowledge behaves. We've been around since 1989, so that's 4 decades, almost 4 decades to accumulate knowledge. We are about 240 people spread across 5 offices, and we have a really wide time zone spread, 12.5 to 13.5 hours depending on which side of daylight savings time we're on. And we work across a wide range of sectors, and we also operate across multiple service lines. So the type of work that our teams do and also how they execute it can be pretty diverse. And then lastly, our synthesis intranet is called Archie. You'll hear me mention Archie throughout the presentation. And that's what we've called our platform. So all of this has added up to a pretty complex knowledge environment. So anything that we build has to work across different teams, different project types, different levels of experience, and different ways of working. Throughout today's webinar, I'm going to talk about how we're working on 4 layers of our knowledge structure. They're not the only layers, but these are the 4 that we're gonna concentrate on. And in, in They're layers not because they're completely separate. They're definitely not. But they do behave differently. The first is content and storytelling. This is how we describe what we do. Proposals, case studies, how we position ourselves, how we talk about our work. The second is practice knowledge. This is how we actually do the work, the standards, the workflows, the lessons learned that teams rely on every day to get the job done. And then the third is project data. This is the structured record of what we've done, the information that helps us make better decisions and iterate as we go. And the fourth is learning. This is how we take everything that we know and make it usable for other people, how we teach it, how we scale it, and how we build on it. One of the things that's become pretty quickly realized here is that you can't solve all 4 of these the same way. They require different levels of structure, different ownership models, different incentives. But they do need to connect because when they're working really well, they reinforce each other, and when they're not, you start to feel the gaps really quickly. So let's start by looking at where this knowledge lives besides our intranet. It is everywhere. It is in drives and folders. It's in project teams and studio silos. It's in people's heads. It's in emails. It's in PDFs. It's in meeting notes. It's in some recordings. And sometimes it is even literally in places that no one thinks to look anymore. Sometimes it is in really thoughtful, well-organized places, and sometimes it is buried 13 or so folders deep in something that has not been opened in 5 years. So the problem becomes not that we don't know enough, it's that we can't find it easily, so we can't connect it easily, and then we can't reuse it. Easily. And that's where the framing has really shifted for us. We do not have a knowledge shortage here. We just have a usability problem. Around the time that AI tools became more accessible and more mainstream, the immediate question that everyone seemed to be asking was, what tool should we use? And that makes sense, right? It is a totally fair question. It is an important question. But when it comes to leveraging AI to activate our firm's knowledge, the better question is really, is our firm's knowledge actually usable by AI? Because if it's not, the tool doesn't solve the problem. It just exposes the problem. I remember uploading a bunch of spreadsheets to Claude a few years ago when it was first released, thinking that I was about to create this magic fee calculator. And instead it didn't work. Surprise, surprise. That was because the data wasn't consistent. The naming conventions were all over the place. There was no structure for the model to really recognize patterns. And so our output wasn't useful. And that was actually kind of a turning point. So we set out to incrementally make our knowledge more usable. And along the way, we've discovered that the integrated AI tools within Synthesis are actually exactly what we needed. I hope this is a relatable statement to everybody. Knowledge is essentially everything, everywhere, ever. And because of that, it feels like this monumental thing to tackle. And so we identified a framework to try to simplify that feeling. Pick one use case, capture knowledge as part of the work, not after the fact, structure it so it can be reused, prove the value, then expand. That first step, picking one use case, was super important because otherwise that overwhelming feeling comes in very quickly. And that fourth step, proving value, there's an important aspect to that that I haven't explicitly called out here, and that's internal marketing. I find that I have to socialize value quite a bit to implement a change. And if you don't internally market your measurable results, it's kind of like the tree falls in the forest and people don't see the value, so they don't contribute to it. So let's look at this framework in place with our first case study. A marketing proposal language library. Around 2023 into 2024, we started feeling real pressure in brand communications. Proposals were getting longer, more detailed, more technical. Clients were asking really great questions. Procurement processes were more structured. And they weren't just looking for expertise and relationships. They were really looking for demonstrated thinking. And at the same time, proposals are how we sustain growth, and we were growing really rapidly. So the expectation was do more, do it faster, and make it better. But our team size wasn't increasing. At the same rate as the firm. So something had to change. We had to find a way to improve both speed and quality without simply adding more people to the team. The way that we historically have composed proposals, um, was driven heavily by memory and involved a lot of manual recall. We had tried to really organize a lot of our information into documents, but a lot of that was boilerplate. So you had to do quite a bit of iteration for it to feel specific and relevant to the pursuit. You searched through folders, you went back into old proposals, you copied and pasted, you edited, um, and you often relied on the people that just knew where things were. I cannot tell you how many times I have probably turned around to my teammates and said something like, hey, do you remember that thing that we did that showed the process for X, Y, and Z? Do you remember what that was for? And then we all try to remember what receipt that was for several months ago. And sometimes that works, right? But it is slow, it's inconsistent, and it is heavily dependent on individual memory. And that is not a system that scales well when volume and complexity are both increasing. I cannot say that we didn't fall into that tool trap that I mentioned a few minutes ago. It is very tempting to look for a tool as the solution. And we did start by looking at what tools could help us solve the problem. We did some exploration. We did some vetting. And we couldn't find anything that did exactly what we wanted. And then AI Search Beta was released and a light bulb went off for us. We could build this thing within our existing knowledge base and freely input all of the confidential information that felt just absolutely crazy to release to a startup tool. So the solution that we ultimately arrived at was to create a community on Archie— again, that's our name for our synthesis intranet— and devote it to Proposal Language Library. Initially, we created one giant guide within that community, and then we very quickly realized that we should create different guides for different work types. And the reason that we did that was because we wanted to use search filters for AI Search and eventually use knowledge agents to recall that typology-specific content. And then we started filling in the guides with curated structured information. And within our separate task management platform, we made updating this language an automated task that appears as part of our proposal closeout process. So we made it part of our workflow, and therefore, this database of information increases value incrementally, and we don't have to do this once in a blue moon massive effort to go in there and add information. So now with AI Search, instead of digging through documents, you can ask a question, get back a response. That is already assembled from the structured content. And as a marketer, authenticity is incredibly important to us. And in this era of just endless streams of AI-generated content, this answer is authentically ours. It is grounded in our knowledge and our knowledge alone. It is grounded in our very real project experience. It reflects how we currently talk about our work, and it's something that you can use as a starting point, or in some cases even as the end product. So instead of spending time searching, copying, editing, you're spending time refining. You're working from something that already has the context to it, and that changes both the speed and the quality of the output. It also reduces variability among team members. Different people are no longer starting from completely different places. As I started sharing this with people, one of the most common questions that I got was, why not just put all of your past proposals into a library and use that knowledge? And one of the clearest lessons here was that more data does not automatically produce better results. Um, and in many cases it does the opposite of that, because our knowledge of evolves. What we said about a topic 2 or 3 years ago may not reflect how we think about it today. We find out new information, best practices change, processes improve, language gets refined. And so if we simply include everything, we are introducing a certain level of noise and inconsistency, um, and it doesn't make the product better. What mattered more has been creating content that has better structure and is more curated. And once we started using AI Search really regularly, and I do mean this is part of our daily workflow, to help us draft content, something kind of fun and unexpected happened. Um, we were doing a proposal for an affordable housing project that was along a major freeway, and the proposal asked if we had experience with acoustical attenu— attenuation for this particular circumstance. And normally the process would've been to, uh, email our very busy housing team right away. But on a whim, we decided, hey, let's put this question into synthesis and see if by any chance there's any result. And immediately AI Search surfaced information from a firm-wide staff meeting video transcript where the housing team had presented about an affordable project they were working on that happened to be along a busy freeway. And they included in that presentation how they had dealt with the noise. So instead of sending the team a vague question, we sent them a draft response for them to review. And that's a totally different starting point. This kind of knowledge in particular can be really easy to overlook. But when we have a tool that can automate how it is captured and how it's surfaced, it becomes something valuable. And instead of just disappearing when the meeting ends. It becomes part of our firm's collective knowledge. So in this way, we see the AI flywheel at work, right? Knowledge gets captured. In this case, it was the video transcript from our all-staff meeting. Then AI helps surface and connect it to real work in real time. We formalize it into something reusable. In this instance, it was an update to our case study about that particular project, which also lives on our intranet. And so that improves the outputs for the next cycle. So the system is getting better over time because the quality of the information is improving. We saw meaningful results very quickly. And the same month that we started building that initial library, we received an RFI from a potential client, and it was an Excel document with multiple tabs and well over 100 questions. We were still building the initial library, so we answered this the old way, and all in all, it took the team about 15 hours to complete. We decided that this was a great test for our new library. So after it was completed, we created a column for the AI search results, and we recompleted the RFI using AI search, and it took about 3 hours. And what was really surprising, we went through and graded each answer. And for the most part, the answers were as good, if not better, than what we had already inputted. And in the cases where they weren't, it was usually just because we hadn't created the information within our library yet. So we are now spending a lot less time on recall and compiling and more time on improving and refining. So we're producing better responses. And more consistent language. Not only that, but as we saw from that housing example, the creativity of the response improved as we had more sources of information. And now as we add new team members, which we did recently, they can add this— access this information instead of relying solely on individual guidance from other team members. At the same time, this also highlighted a gap for us. We needed more input from our architectural studios, which I'm going to talk a little bit about in the next case study. But before I get there, I am very excited to give you a special preview of Synthesis Knowledge Agents in action. If you hadn't joined when Chris explained, the K team has allowed me to start a private beta test of these agents. And wow, do they unlock the next level for us. In this first example, I'm asking my new project experience matcher agent to match project experience to the criteria for a pursuit. In this case, I typed it in. I also could have uploaded the RFP to give it more context. So let's see how it works. Not bad. So this is the kind of task that historically was incredibly manual, right? If you're lucky, you're simply filtering a project directory on Synthesis and extracting the information that you need. If you're like us, you are also pulling information from Deltek, searching folders and digging through case studies, asking around for more context and details, cross-referencing typologies and dates, etc. And it's not that any one of those steps is particularly difficult, right? It's just that there are a lot of them and they do add up, especially when we're talking about like really specific experience types. I could tell you all of the life science labs that we've completed, but I definitely couldn't tell you every elevator that we've modernized. The other day, we used this agent to find all relevant experience within a 50-mile radius of a specific site, which was a question asked in a recent pursuit. And that sounds simple, but if you've tried to do that manually, you know it's not. You would probably use AI to make a list of places within that radius and then run a report matching those locations to a field. And then you're exporting data and resorting it for your output. And here it's just answered. And it's not only answered, it's explained. And I think that is a really important distinction. You're not just getting a list, you're getting reasoning and information that you can actually work with. In a later case study, I'm also gonna talk about project data and enriching our project information database will absolutely supercharge this resource for us. In this second example, we're going to draft a scope based on an RFP that we uploaded to the system. Historically, this is frequently a start from something and rewrite it process. You find a similar scope, you copy it over, you customize it, you adjust it, you fill in the gaps. So let's see what the agent can do. Here I'm uploading the RFP. And just a quick note here, as you can tell, I have sped up the response. And voilà. So what the agent is doing here is starting from a grounded base of knowledge, the scope library that we built out within our proposal language library. And then it is tailoring it to this specific pursuit. So instead of starting from a blank page or a scope from a previous project, we are starting from something that is informed based on that criteria. And so that's really changing the way that you spend your time, right? You're not asking, what should this include? You're asking, how can I make this stronger? And that is a much more high-value question. I don't know who drafts scopes typically within your firm. For us, it is typically done at the principal and director level. And so this is also a great way for us to communicate value and buy-in to leadership for some of these initiatives. This is how we will save you time and add value. So the takeaway here is pretty simple: recall is not scalable. At some point, you do hit a ceiling where relying on memory just doesn't work anymore. Structured knowledge, on the other hand, scales. It gets better as more people contribute to it. And it becomes more valuable as it gets used. So this is the content and storytelling layer at work. It's not just about writing proposals. It's about taking what we've learned from designing and building projects and turning it into something that can be reused. Something that is reflective of how we think today, not just what we've done in the past. The next case study moves into practice knowledge with what we call Studio Workspaces. We introduced Studio Workspaces as a way to organize knowledge closer to where the work is actually happening based on some challenges that we were repeatedly experiencing. One of the things that we kept running into was that even when good information existed, people either didn't know it was there, or it wasn't worth the effort to go and find it. And at the same time, firm-wide standards only get you so far. As I mentioned earlier, we are a pretty diverse firm in terms of the project types and workflows that we have. And what works for one of our studios doesn't always translate cleanly into another. So instead of trying to force everything into one centralized model, we created these spaces where studios could build knowledge and standards that's specific to how they work. Our first studio workspace actually dates back to the last version of Synthesis that was still tied to SharePoint. But we didn't really formalize the concept until launching the Synthesis native platform, which introduced things like content outliner, communities, the freedom to easily move things within the system, versioning, and, um, building standardized pages across different parts of the intranet, um, in a way that felt really cohesive to people when they visited those parts. So for us, that was a tipping point because it removed so many of the barriers to entry. But adoption was uneven. Some studios really leaned in in the beginning. They built out their content, they organized their workflows, they started capturing those lessons learned in a structured way. Uh, other studios moved much slower. Not because they disagreed with the idea, at least I hope not. I don't think so. But because, and I think this is important, the value was not immediately obvious to their day-to-day work. And this is something I've kind of come to expect with knowledge initiatives. Even when the concept makes sense, it does not always translate into immediate behavior change. Contributing knowledge takes time, and if that return on energy isn't totally clear, it can be hard to prioritize. So we ended up with a mix. Some are really strong, mature workspaces, some that were just getting started. But it has reinforced something really important for us. When people can see how knowledge is used, they will engage with it differently. Along the way, we have started documenting what works, what doesn't, what's sustainable, what tends to fall apart. And we have built that into essentially a playbook for how to build and maintain studio knowledge in a way that actually lasts. And that lives within what we call our Intranet 101 guide. I think one of the big risks with any knowledge system is that it starts really strong, right? Like you have a couple people that are super excited about it. But if you don't spend the time to build something really sustainable, it can degrade over time. So we've really focused on making sure that these things are maintainable and sustainable. Now, I wish that I could show you more detailed examples, um, of what are in Studio Workspaces, but fortunately our teams have now put in so much information into Studio Workspaces that there is a ton of confidential information that I can't share out today. But what I can show is how the experience changes. Before AI search, I sometimes heard from people that they felt like they needed to understand the navigation to find things, and because they didn't understand the navigation, they couldn't find things. And sometimes you did know, have to know where to look, or you had to know what things were called or how they were categorized. But now you don't. You ask a question, you get an answer, and not just an answer, you get the source, you get the context, and you get who to talk to if you need more detail. I was not expecting that when we first started using AI Search, and it's such a cool feature because that, that's really important, right? It's not about replacing people or replacing human-to-human interaction. It's just about connecting people to the right knowledge faster and in a way connecting them to each other faster. The second example shows something else. Unfortunately, I had to redact pretty much every ounce of information out of the response, but I couldn't resist including this query because I found it illustrates something just so cool. The number of sources that it's pulling from. In this case, what used to require clicking through 20 different pages on the intranet is now condensed into one response. That is a really big shift. When the effort required to access knowledge goes down, usage goes up. And when usage goes up, contribution tends to follow. So not only that, it also shows how these responses can help teams effortlessly create new training materials, right? Like you could create this one-pager now that explains all the different deliverables to a new hire, and you're doing that not by doing more work, but by leveraging the work that you've already done by creating this documentation in the first place. A good example of this in practice is the PlanCheck comments database that we're building. PlanCheck comments are incredibly valuable. They tell you where things went wrong, what things were unclear, what standards weren't aligned. And traditionally, they have lived in our project folders on our network drives. And unless you were on that project, you probably won't ever see them. That means that we are constantly relearning the same lessons. So we created this centralized library on Archie, and then we layered AI search on top of that. And instead of digging through files, you can start to see patterns, you can start to ask questions like what comments come up repeatedly, where are we seeing friction in this jurisdiction. And now something that used to be isolated is available to everybody and allows us to operationalize these lessons learned from past projects and offer key insights to clients. This knowledge becomes accessible beyond the original project team, and it also allows our different teams to connect and share insights based on the search result when otherwise they may not realize that they're working on a project in the same location. I mentioned earlier that we created Studio workspaces in part to respond to the diverse ways our teams work. That is absolutely true, but when we originally formulated the idea, we also hoped that it would help surface some areas where we could standardize firm-wide. So this firm-wide best practice codifier agent looks across how different studios are working with the grounding knowledge being our studio communities and enforce— instead of forcing a top-down standard, it shows what's happening, where there's alignment and where there isn't. And then it recommends best practices based on that. And what I liked about this is that it grounds standardization in reality. It's not something that's theoretical. It's based on how people are actually working. And going back to that idea of starting from scratch versus starting from a draft, it is so much easier to get everyone on the, on the same page same page when you have something like this drafted that says, you know, hey, we took into account how each of you are doing this. How about we all do it this way? Rather than starting from a theoretical conversation that, you know, so often leads to a lot of circling around progress. So the takeaway here is knowledge contribution only scales when retrieval is easy and valuable. If it's hard to access, people won't use it. And if people don't use it, they probably are not gonna contribute to it. These Studio workspaces are just one example of how we're building out the practice knowledge layer. It's still evolving and hopefully will continue forever evolving along with our teams. Some parts are mature. Some parts are still taking shape. And what's starting to happen is that knowledge is getting closer to the work. It's becoming something that teams can actually use in real time, not something they have to go searching for after the fact. And that shift from stored knowledge to usable knowledge is really the goal across all of these different layers. Deep breath for project data. This is really a different kind of challenge. If anybody has really just nailed the solution here to this problem, I'd love to hear from you. Because this is where structure really matters, and for us at least, this is where friction shows up the most. For us, the issue has always been kind of friction in, friction out. Some data gets entered at the beginning of a project, maybe again at the end, hopefully. If it's a big or complex enough project, the brand communications team will collect more information from the team. Typically a small group is responsible for it. And then except for marketing materials, it kind of disappears. And the problem is that that return on effort isn't obvious. So from the team's perspective, it feels like I'm putting this in, I don't really see where it's going. And then when that happens, participation drops, not because people don't care. But because the value just isn't visible. This was a challenge that I had grappled with for years and years. Frankly, it just felt like a problem too big to fully solve, like a someday problem. But project data is incredibly important. It is absolutely vital to our work. It informs strategy decision-making. It informs the projects that we create next. It informs proposals, certainly. But the way that we were collecting it just didn't support how people actually work. Everything changed for me when I heard Ron Himes from RDG Planning and Design's 2025 KA Connect talk. And it was the first time that I saw a model that made sense, more specifically, a model that would allow us to collect the data as part of the workflow, not before, not after, but during. So here is what we've been designing. I will say the first iterations of this were about 1,000 times more complex with all sorts of crazy branching for all sorts of crazy situations, and we've really simplified the workflow to try to make it successful. We are designing a system where data is captured in small moments throughout the life of the project, not all at once, not as a big task, and rather as really small targeted inputs tied to specific phases of the project. We are using phase-based microforms, automatic triggers, and much broader participation, kind of democratizing the whole experience, and integrating with Deltek VantagePoint so that it all connects. The goal here is absolutely to reduce friction, because if we do reduce friction, participation will improve. At least that's the working hypothesis here. So when people start to see the value of that data, something definitely shifts. People are more willing to contribute, which improves the quality, which improves the outputs, and that creates this self-reinforcing loop. And this data is what enables something like this design precedent explorer agent to dive deeper. This is a tool that allows teams to explore past work in a much more fluid way. And the possibilities here feel kind of endless. You can look at really broad things like I've done here, facade strategies, or very specifically, we could dive into the minute details, something like a flooring material specification. And this is where internal marketing comes back in, because when people can see this kind of thing in action, when they can see how they can benefit from it day to day, it clicks. They understand why the data matters. They understand how it benefits them specifically. And so they're much more likely to contribute. The takeaway here is data quality improves when data is actively used. So if we take a step back for a second, you can start to see how these layers connect. We've talked about content, how we describe what we do. We've talked about practice knowledge, how we actually do the work. Project data is what ties those things to reality. It is the record of what has actually happened. And without it, the other layers start to drift a little bit. They become less grounded, less reliable, and harder to build on. So this layer is really about making sure that we've anchored everything that we know in what we've actually done. And the last layer we'll touch on today is learning. This is where everything really connects. Historically, learning can feel separate. It's something you step away from work to do. But the more we look at it, the more we realize that the most valuable learning is coming directly from the work itself and is embedded directly back into the work itself. I cannot say that we have this one packed. It's very much a work in progress, but we aspire to build this highly tuned version of this model, right, where we capture insights from real work from our projects, from QA/QC, from plan check comments, from day-to-day experiences, turn those into structured insights, and then turn them into something that's reusable, like courses that reflect both our current and our evolving needs. Then we can deliver them through the LMS and make them accessible through AI search. And people apply that knowledge in their work, which then generates new insights and feeds back into the system. So learning becomes part of the system, not something outside of it. This is a course in our LMS on egress. While the insights that drove the creation of this course were surfaced pretty manually. Um, the same principles apply here. Our technical director recorded and categorized the most common issues found when reviewing drawings. He identified egress as a really important concept to reinforce among emerging professionals. And pre-LMS, he structured those insights into a post series with related quizzes. Now with LMS, we've converted them into a course. We will now assign the course to new design professionals. And also those insights that are captured can be surfaced by AI search by anyone in the firm asking a question. And even from the post and quiz model, a measurable improvement has been seen. Which makes better projects, and then that feeds back into the system. So the takeaway is, if we can surface and codify it, we can teach it. At this point, the layers start to feel less like separate categories and more like parts of the same system. Content draws from practice. Practice improves through data. Data really feeds everything. And learning feeds back into all of it. And when that learning loop is working, knowledge doesn't just sit somewhere. It's moving through the firm. It is improving things. And it's becoming something that the firm can build on over time. As we wrap things up, here is a snapshot of some of the other knowledge agents I have created in a very short amount of time. I can't tell you how excited I have, I felt about these things, and I have plans to build many more. I also can't wait to see how these grow and change our knowledge infrastructure. And I'll close by sharing just a little story about the NDA and contract reviewer agents. Our CFO retired last Friday after 26 years with the firm, and for much of his tenure, he reviewed our contracts and filtered kind of that pass-through to legal. And there was this moment of realizing just how much knowledge lives with people over the course of their career and how easy it is for that knowledge to walk out the door. So yes, he created reference guides for contract language, um, and we created a skill, um, based on that knowledge, but can that really replace the wisdom that someone has accumulated over that many years. So with that in mind, in his final week, our absolute rockstar studio manager sat with him and recorded an hour-long deep dive specifically about contracts and NDAs. So now we will use that video transcript as part of the grounding knowledge for our NDA and contract reviewer agents. Years ago, I remember creating a slide for a KM presentation. I can't remember exactly what it was for, but I remember putting this picture of a brain in a wheelbarrow to represent knowledge walking out the door with someone. And at the time, that felt really accurate. But what's different now is that we have these tools and systems that allow us to capture that knowledge in a way that is usable beyond the individual. And that is really the goal, not to replace people, um, but to make what we know more accessible. So thank you all so much for joining, and thank you to everyone in this community for sharing your ideas along the way. A lot of what I've talked about today has been shaped by conversations and ideas within this KA community of firms. So thank you.

Thank you, Hillary. This is fantastic. Um, I, I want to let everybody know, we, um, Hillary and I can both stay after to keep answering questions. So I'm going to try and ask the most common ones up front, but there's— I'm sure there's a lot here that people want to know about. The thing I saw the most, Hillary, was— and this is going to be a little bit of a long setup, but I think It's some version of, like, all the way at the beginning, you talked about, you know, rather than just getting a bucket of proposals, your team took the time to build, like, kind of distill the best of our knowledge and make something curated. But there are some other examples you showed, like the plan check, where it's actually just a bucket of plan check letters, or even the recording in the NDA example. So there's this kind of— it feels like there's sometimes when you want to take that, the raw materials, and make something more polished out of them, and there's sometimes when you're thinking the raw materials are good enough. Um, and is that true, or is that just we haven't gotten to the taking the raw materials and turning them into something else part yet?

I think it matters what the goal is, right? Like with the PlanCheck library, um, the, the goal is not to source the the language, the goal is to source the pattern. So the language doesn't really matter. We're trying to find, we're trying to surface what the commonalities. And with something like the Proposal Language Library, the knowledge, the voice and the language really matters. The specificity which, with which we talk about our approach. Matter so much. So yeah, I think, I think it is a mix. You have to look at what the goal is. But I think knowledge agents are gonna unlock a lot of that. Currently our proposal library is built in our brand voice already. So what's coming out is in our brand voice. But what we'll be able to do with knowledge agents is layer that onto pieces of information that aren't already built in the voice and tone and language that we want. So I think that will unlock a lot of new possibilities for what we do.

Great. And, and to kind of talk about the, the kind of knowledge distilling process, like let's stay with the proposal library if we can. Someone asked a question, right? Like, what was the actual process though? Like, did you, like you said, this is once in a blue moon, we're gonna do this massive lift. To try and make our best stuff. Um, was that a team project? Was that you by yourself? Did you even use AI to help you do it? Like, how did you actually get to the gold standard?

Um, it was absolutely a team project. Um, it was a monumental amount of information. Uh, we started with, um, a document that we had already created. I mentioned it earlier, kind of just like a boilerplate, like, this is language that we repeat a lot. These are different versions of things that we're using commonly. And then a lot of the iteration actually came from as we were doing different pursuits, like surfacing the things that were coming up a lot and that we wanted to make sure were stored in it. And then after we did the initial page kind of for each topic, we had our PR specialist go through and like really make it better, really polish it up, put it solidly in our brand voice and make the whole narrative more cohesive. So it was a multi-layered process. We tracked it in our workflow software. And yeah, it, it is something that is tricky to iterate and evolve. And it does involve, you know, the, a little bit of kind of a manual process of going through when you do finish a proposal and figuring out that like, oh, these are, these are the chunks of things that I wanna resave. Mm-hmm. Um, but the more that you build this thing and the more that you know what's in it, the easier that becomes to do.

You know, there's a gap that you'd be filling if you captured this piece.

Yeah. And I know when I find out new information about something. Mm-hmm.

Let's talk about the, um, so one thing that people should know is that the knowledge agents can be personal. Just for you and you can build them for your own purposes. You can build them shared for teams and departments, and then you can build— administrators in the product can build global agents that are for everybody. You've started off by building some personal agents, and I'm kind of curious, is it just you building? Are there other people in your team you're opening this up to, to start experimenting? Like, what is your thought process of what the first month, 2 months, 3 months of using this within the company looks like?

Kind of with everything, I like to figure it out a little bit before I introduce it to anybody else. That way I can kind of guide and ask questions. With this, I was like really so excited about it that I think I made it like 36 hours before I had to let the team in and have them, have them start looking at things. For me, having the team in there provides a lot of new perspectives. I mentioned our PR specialist. He really brings in an emphasis on like trying to write for a specific purpose, whether that's a blog post on our website, kind of fine-tuning it to get to that brand voice. I mentioned that we have someone that joined us 3, 2 and a half months ago, and having the perspective of someone that's new to the team is incredibly valuable. And then We have some specialized functions within the team that also offer a different perspective. And I intend to have a couple of more people join us in there from different studios, but they're people that are already really heavily kind of devoted to content in Archie because I want people that will talk about these as much as possible. And people that will give us feedback as much as possible. And in the beginning, people that do have a certain understanding of kind of the inner workings of the system so that they can build high-quality agents.

Yeah. One thing you shared with me is that that new hire also has built an agent for herself to help, like grounding it in your proposal language to try and get a first draft of answers before she comes to you. Like, so it's been kind of a buddy, right?

It was so, so cool to see this week. Um, she built an agent and, um, was working on a pursuit and was asking questions and then verifying the results with us, um, in the office this week. And, um, it's, it's so cool, um, to see someone else's excitement, um, and see them find a way to like kind of take initiative in their own role, in their new role, and be able to act more autonomously.

Amazing. We got a question about the team. So kind of like, how is the knowledge leadership structured on the ops side? Like, what is technology's role, marketing, business development, knowledge management? Like, what does that look like?

What is our, what is our knowledge team spread across all of the different operational departments? We have people from each studio that are devoted to our intranet content and responsible for those things. Within the architectural studios, we have people that are really spearheading those efforts. They lead smaller teams within their studios to accomplish the work. And then I communicate directly with kind of the leaders from each of those studios to see you know, how well are they doing? We issue monthly analytics to every studio. And then we also work with our operational departments quite a bit, looking at our search data to see how well people are finding that sort of content on the intranet and what can be improved. That's great.

Thank you. Rachel asked to go to one of the, one of your, to staffing's recommendation. Agent. Rachel wanted to know, where do the experience and skills live and how does that get updated? How often? Like, how does the agent know who to recommend?

Oh, I'm so glad that someone asked this question. There was so much personal information in there that I couldn't pull it up, but we grounded it in the employee directory, which also includes people's profiles and also the proposal language library. Because we have a lot of information drafted there about how we structure teams for different project types and programs. And so something just super cool happened, which was that, like, I started immediately discovering people's skills that I hadn't realized. I entered in a small and very specific project type and asked me to— asked it to build out a full team for me. And it started surfacing skills from people's bios that they had written. And then also we have a field with kind of like standardized responses that you can punch in so that all the naming convention is the same. It started surfacing those, and then it also pulls from those people's project experiences, which live on their profile. And then pulling in that team structure information from our proposal language library and building a team based on that information, like just absolutely blew my mind.

One of the things that when you showed me this, was it Tuesday? It feels like a long time ago when you showed me this on Tuesday is that some of that, so to, to kind of Rachel's question. It's a mix of like who owns that information and keeps it up to date. But some of it is actually the employee, right? Populating their skills and their bios and which kind of could influence what opportunities you might get or how you get brought back into these recommendations.

Yeah, absolutely. I did notice in a lot of the queries that I was doing that I was getting a few employees that had like really put a lot of time into updating their profile. And their skills. So there's a good incentive there to update your program.

Return on effort goes up and more work.

Yes. Great.

Bill asked about the HR piece of this with employee manuals and policies and if you've experimented at all in building on it in HRE agents. And if so, if this was different than BD or practice stuff that you've been showing.

I was, I was kind of wondering if somebody was going to ask about that sort of content. The reason that I haven't is because we've traditionally done that really well on the intranet. People have long seen our intranet as a great resource for that sort of operational content, and people don't seem to have too hard of a time finding it. And so in the beginning, in, in this like kind of internal marketing generating value time, Mm-hmm. That hasn't been like totally my focus, but absolutely, like, will we build an open enrollment agent this year? Yes. Mm-hmm. Um, will there be some great specific use cases that we'll build soon? Yes. But in this initial push, yep. Um, like we already have value there and so it just hasn't been my focus yet.

One thing, and, and I, I do not wanna turn this into a knowledge agent briefing, but one thing that's important, just Bill, cuz you brought up HR, Hillary, I know you experimented with this. When you configure an agent, you can configure how it's fidelity to the source material. It can be creative and inferential all the way to verbatim. A lot of the HRE stuff is going to be more verbatim, but then some of the marketing stuff you showed will be able to be a little bit looser. On that front, how are you identifying use cases and priorities for both foundational knowledge you're capturing, agents? Is that just you've worked there for a long time and you understand where the pain points are? Are you working with a team to surface this knowledge capture priority? How does that work?

I think twofold. There, there are the, there's the reactive ways, right? Like responding to pain points. Those, you know, might include like some of the examples from today. Indicators from our work, QA/QC, RFIs, pain points that teams are feeling. I do a lot of talking to a lot of people too within the firm. So surfacing pain points that way. Pursuits are also a really great way to learn what our clients', potential clients' pain points are. And we see a lot of patterns there. But those are kind of the like reactive ways. More proactively looking at where we want to go as a firm, what new project types are emerging that we need to become skilled in, and following the pipeline, or it might be a new geography rather than a new project type. So looking at the future and where we want to get to. is equally as important as, you know, the, the more reactive kind of responding to those. Where are we currently feeling?

And then I'm guessing kind of upcoming projected retirements of kind of one and only, one and onlys, right? You know, some things like, like, is it Ollie? I think your CFO. Um, so I'm guessing there are more of those coming. And so like, how does your, how are you thinking about this kind of critical knowledge? Capture and transfer and how it fits into this whole project we talked about today?

I don't have it fully unlocked. I, we, we actually just went through a pretty big series of really big retirements. And I wish that we had done a little bit better job at that. I think this most recent one where it was imparting a lot of technical knowledge in a really condensed way in a really short amount of time definitely will inform our strategy going forward.

Mm-hmm.

Um, but yeah, I don't know. I'm gonna have to get back to you on that one. Sure, sure.

Um, that's a good model. I mean, you, you kind of— I wrote a piece on LinkedIn today that was because somebody had asked me recently, how is AI going to impact succession planning? And to me, it's like, I think it will impact it in terms of, you know, it's easier to transcribe things and capture them and make it available.

Yeah.

But it can't do that without the knowledge foundation. Like the two are interconnected. And I think, so if you want this, like we think about succession planning a lot of times as the people, but actually a lot of what we need to think about is the capability also. So MBH needs to preserve the capability to review NDAs or design X type of project and the values need to proceed. So like, how does, how, what's really kind of like a business continuity question in some ways.

Yeah. And You know, you said video. There's such a reluctancy, I think, in some cases to make things a video because people think that it's going to just like destroy human interaction. But it's amazing what the video transcription has done for that model and the fact that like we are able to, you know, surface the insights from that without having to watch an hour-long piece on it. Yeah.

I think I got one more question from Wendy and she's got a question around kind of, I'll just read it. I don't recall the term you used, maybe plan review, but it sounded like quality assurance reviews of CDs. Are these shared as documents stored in Synthesis or is the review team posting them and @mentioning the project? Project. I think she's talking about the plan check comments piece. Um, and those are documents that you've uploaded into a private document library that you grounded the knowledge in, correct?

Yeah. Um, currently they are located in a private community. Very soon they'll be in a public community. Um, the reason that we have not moved them into a public community just yet, um, is because we're still trying to get— figure out how we're going to both amass the critical volume where you can search anywhere, but also how are we going to let people know, like, which places you can search and which places you can't so that people don't learn that this is a non-usable resource. But the document that we're using is our— actually our plan check response letter. So it's a very standardized format for every single project. It's not trying to, like, read comments on on CDs. That's great.

Awesome. Well, I think, I think we got through everybody's questions. So thanks everyone for hanging around. Um, and, but especially thank you, Hilary, for putting together such an awesome— I, you didn't see the chat. We'll send you the responses afterwards, but people said very nice things about you.

So, oh, well, thank you so much. I, I really appreciate it to be here. And again, just really thankful for everybody in this community for always being so generous with your ideas.

Yes, and speaking of that, Susan's gonna post our upcoming events link in the chat. The big one coming up is K Connect, obviously August 11th through 14th. Hillary will be there. You can corner her and ask her all kinds of questions. That's excellent, and we look forward to seeing you all at our next webinar or around the neighborhood. So thanks, everybody.

Thank you.