Name: Document agents 101: A practical introduction
Uploaded: 2025-09-11T09:52:49.896Z
Duration: 35 min 8 s
Description: Document agents 101: A practical introduction

Transcript for "Document agents 101: A practical introduction": Hello, everyone. Welcome back to Templafy Talks. For those folks that are new to Templafy Talks, welcome. For those that have been here before, welcome back. I'll always start with a quick guide to our house rules. As you know, we're using, Goldcast for this, webinar. On your right hand side, you'll see a chat. Now you can comment in the chat. You can actually engage with other users. And if you want to, there is a q and a channel. For the first time in the back end, in the back end, on the back side of Templafy, we have our colleague, Casper, who'll be answering questions in the chat. Now we can test that now. We're coming to you live from Copenhagen. It'd be great for you to tell us where you're coming from live. K. I can see some, familiar locations coming in. Hello? So last time on Templafy Talks, we unpacked the hidden cost of document work. We discussed how to measure ROI and common pitfalls to avoid when you're measuring the success of your document workflows. Just wanna say thank you. It was a very well received session, and thank you for the feedback. If you missed it, don't worry. You can find a link to that in the chat, of and containing all of our previous Templafide talks. And so I think Casper's just gonna add that now. And, yeah, please take a look, and if you have any feedback, get in touch. For today's topic, we're discussing our newly launched document agent solute sorry, document automation solution, document agents. Now the agenda is gonna look something like this. We're gonna talk about AI enterprise organizations. We're gonna give you an understanding of what a, AI agents are, how they affect operational efficiency, how to maintain control, and understand the technical considerations. And finally, we'll tell you how to get your hands on it right now. This episode is part one of a two part series we're calling Document Agents one zero one. My guest today, returning to Templafy talks, needs no introduction, but it'd be very strange if I just started talking to him. Ultimately, we're all here because of him. It's our cofounder, Chris Inland. Welcome back. Thank you so much. Thanks for having me. You're very welcome. It's, always good to have you on Semify Talks. So first point, from your perspective, Christian, how has the, enterprise, organization's AI adoption evolved over the last two to three years? First of all, after I would say after a slow start, if we go, three to four years back, then with incredible speed over the over the past year at least. So, I I think any business that works around content creation, including, of course, what we do with documents, have needed to sort of really take into account what's happening around us and so have, every single business that we work with. We're having discussions every day and have been having that for a long time on how to get, first of all, their heads around, what AI can now what AI actually means, and what it can do. And then more recently to turn that into real value, which has been like a a big headache, actually. So tons of investments going into this, over the past few years and, more recently, strong focus on, okay, now it's here. We have these new brains. How do we make use of them and actually turn them into real value? But that's the interesting point, Roy. So the the gap between hype and actual impact in other organizations, where do you see the biggest gaps? Well, it it has actually been been there in terms of finding those specific use cases where you can actually use AI to get stuff done. Mhmm. And it's not new. I mean, it's, it's it's similar to what happens every time there's a big shift in technology. It starts, oftentimes with, with being with a lot of hype and being impressed with what new technologies can potentially do. So there's a lot of expectations what this might mean. Mhmm. And then after the first hype, it falls into a little bit of disappointment actually oftentimes before then starting to figure out where is it actually that we can find the use cases where this will make the biggest impacts first. And, that's also where we've definitely seen the gap, from having the way that we phrased it a few times is that there's been a lot of investments of consolidating all the knowledge that exists in large enterprises into these AI models and racks. And, but where a lot of companies have been to this, like, now having big brains with no bodies. Yeah. So they're really looking for the arms and legs that would put this to use and, and and, ultimately impact the businesses in a positive way. Yeah. And then, obviously, we're looking for arms and legs, but also controlling that brain. How are enterprise balance how are enterprises balancing speed, of innovation with governance, security, and any ethical considerations? Yeah. It's it's it's a good question. And another thing that we've we've been discussing with our customers a lot because, also, that's where we come from. With enterprises, typically, it comes with a layer of control and governance because one thing is to get stuff done quicker. Another thing is to get stuff done and making sure it's actually something you can use, that it's accurate and and trustworthy information that goes in to those areas. So, there has been tons of discussions with customers that were first impressed by what AI potentially could do, including building out things like PowerPoint decks very quickly based solely on a prompt. And then, sort of, graduating from that into thinking about, okay, it's possible. That's what we want from the future, but let's make sure it's actually also being done in a way that we can actually use. Otherwise, we're sort of left with nothing. Exactly. So what breakthroughs or what do you think the biggest shift is gonna be going forward for enterprises when it comes to AI? Yeah. I think I think what's been missing a lot, on the on the back of the first wave that with waves that we've seen with, with AI has been the governance layer. Yeah. The ability to actually put some control. So, the way that I see it is that, you have AI helping out a lot, but you still have you still have a lot of reliance on the individual Mhmm. The person. And, ultimately, what I think enterprises are trying to do is to get into the state where best practice is worst case, which is really what AI can help do. Because if you look across a large organization with, let's just say, a thousand people Yep. Not everyone is equally equipped to do something like business documents. Some are just better skilled than others. And it's been difficult for for companies to get to that point where they can actually, sort of scale that best practice because it sits with one person. But even if that person was willing to actually do more work about creating documents, for example, that would be a capacity limitation to how much that person can do. Yeah. So, of course, the big thing that that AI can do and specifically with agents is to actually have identify these best practices and continue to grow them. So every time something is being done, it's actually best practice. That's very strong. However, what that requires is that you take a lot of control about how these agents and how AI actually work. What it actually does is going to be incredibly important to get there. So you need to get a little bit away from relying on a person doing a prompt Yeah. And still having that capturing the intent from what a user is actually trying to do, but then have more control about what AI then does, have better instructions to how to actually perform that job in the best way that is replicable to actually get to those best practices. So that's been the big step and, what a lot of companies are investing in now is how to take more control about what AI does to help the users. So still requiring the intent from the user, but on the back of that, having AI actually do its work in a best practice state every time. And it's all about governance when we come to that point. So that's where we see not just in our business, but everyone we speak to in other industries as well doing AI. That's where the heavy investments are going to to be able to support enterprises. Fantastic. Well, eloquently answered as always. So moving on then to what Templafy document agents are. For those folks that missed our introduction, we actually did give, a short introduction to the document agents back in April. But what is a document agent, and how does it differ from our traditional template based automation? Yeah. Good question. So a simple way to look at a document agent is actually as a digital worker or digital colleague. That is just extremely capable of putting together documents, putting together business documents. So what it does is that it understands the intent of the user, what the user is actually trying to do. I need to create a pitch deck, for example, for this company. On the back of that, the the first agent is able to figure out what would actually be the best way for this user to get, that task done. And what it would do is to look for the best suited agent that would be able to support with that particular task. And that could be like a specialized agent that is really good at doing pitch decks, or it could be more a general agent that can do pretty much anything you throw at it. And on the back of that, it's then able to, to converse with the user if it requires a little bit more information like a normal person would do that is trying to help you do that. And it's also able to understand a lot of the stuff that exists within the organization to pull together the best possible output of that document that the user is trying to create. So it's a digital worker that supports the user, and as I said before, it it starts from the best practice Yeah. Level. So It's interesting. We, on the previous Tenpify talks, we had Paul Vitrano as a guest. And from his previous experience at BDO, we talked around how long it takes to generate a presentation or a proposal at BDO, and they needed to go to nine separate sources as a user. So they would first need to find a template they were looking for. That's that's part one. Right? Yes. Usually, the simple part. It's not always. Then they had to go and find, you know, as you say, the correct data. Was it a CRM system? Then if they wanted imagery, they need to go to a digital asset manager. Yep. From what you're describing, this would all be done in a conversational like, this would be done by agents now instead of this user having to go to different, sources or even people for, for that extra content. Exactly. So what we're trying to do is to replicate real world. We're actually not trying to reinvent the process as you're saying. We think the process is right. You know, you have an intent on the back of that. You need a template. You need to find all the data and content that should go onto the document, and you would need to compile the document at some point, and you need to verify that the quality is good enough. We don't argue against that process. We just think it shouldn't be, an individual that should be doing that. That should be technology doing that, actually. So, you asked before what the what is the difference between classic automation and then agents? Well, a big difference is that we've actually been able to output very, very accurate, documents for a long time with automation, with rules based automation, and we still do that because it's a great way to get to super high quality, very accurate business documents. Yeah. But the one downside is that it does require configuration because those are structures and rules that you need to set up and connectors to different sources. A big difference here is that with the agents, you have new brains that are able to understand what they need to do. So there are they oftentimes operate as project managers to your point, making sure that they can then reach out to the right resources or the right coworkers in terms of other digital agents that they can work with to support them of getting stuff done. Yeah. So it's very much emotion about replicating how things were even before technology. Mhmm. That's also what we're so excited about, that you just have these, super brains that are just really good at doing certain things and never run out of capacity so they can continue to do this work. And, it also allows, companies to be more in control or closer to pretty much every document ever created by anyone. Because with rules based automation, you would have to be specific on the type of document that you automate. Whereas with agents, if it doesn't have a specific agent for a specific purpose, it's still a general agent that can just do, anything you throw at it. Of course, to a limit in terms of what is the final how finalized is that document. You might still still need to do some work, but it can definitely get you started at least on a very good drafting state. Exactly. And I think the interesting part is, right, those documents that are extremely well tailored for a specific purpose, that's never gonna change. They're still gonna need you're still gonna need a lot and element of control on that part. Yeah. But from an operational perspective, it's very difficult to have a template for every single intent that the user has. So having that agent in the middle is extremely powerful. Do you have any examples that you could think where, a company would benefit from not having, you know, from having that sort of freedom in the middle? Yeah. You're you're definitely mentioning mentioning something that's completely right. Like, sometimes you have these very specific use cases, for example, of a proposal where you would wanna have very high level of accuracy of everything that goes on there. Because the there's it just have to be it has to be very structured. It's also a repetitive document, so it makes sense to do the investment to do that. However, that type of document might make up, like, less than 1% of all the documents that are being created in the organization. So you don't have, like, that specific document for every single case you would. And even if you did, you would have to put in a lot of work Yeah. To get the configuration going. So where agents come in very strong is that they're able to pick up, I would say, less defined types of use cases. Starting from the from the sort of the floor, that would be like one agent being able to to essentially create a document from anything you put at it to a certain, degree of finalization. Mhmm. So you might get to 50% of, a very good draft of a document just on the back of your intent and what that universal agent was able to support you with. Then as the next uptick from that, you would be able to identify actually from just how users work, what they actually ask, which agents it would be great to spin out. An example could be the pitch deck. So it looks like a lot of people are doing pitch decks. Maybe we want to do a specific agent that is really good at doing pitch decks. And then you would lift the, the level of finalization you can get to when you create your documents on the back of that. Because all of a sudden, you have one that is more specific that maybe connect to a specific model that is really good at doing that and is ground against its information in certain content that you control. So you sort of narrow in the field of where it should be good, which means that you, you get to higher quality levels. So these are two good examples of, of agents with very little configuration, allowing to get to closer to a lot of those use cases that otherwise you're just not able to get to. So, you still need the top the top level, I would say, for certain documents, but there's a ton of opportunity, for the other types of documents. Yeah. It makes sense. Very clear. I think the one question I think comes up quite often when we're discussing our agents with, with our customers and during our private preview, is how do we, leverage AI differently than Copilot, for example? Yeah. I mean, if you look at the I I mentioned the low tier before in terms of, the prompt to deck scenario, for example. I have an intent from a user. I put something in. I get a deck on the other hand. We're seeing quite a few actually in that field doing that, which is great. It's also supporting the education of what's even possible in the market. So we really welcome a lot of that competition for sure. Well, where we sort of stand out and what we're trying to do is that when you when you move from that level Mhmm. And you wanna get to higher level of accuracies for more important type of business documents, we just need more orchestration to get closer to something you can deliver to our customer. Mhmm. That's where we, come in really strong because we provide that. There is a full orchestration level, or admin interface where you can essentially control a full digital workforce of agents. You can connect to all sorts of different models. You can pull in data from your existing data sources into very specific areas. And even if you wanna move beyond agents, we still have all the rules based automation logic that we can still, sort of leverage on. So the big difference there is, when you run out of of steam, so to speak, on that prompt to deck logic and you wanna get further, you would need to look at something that has stronger levels of orchestration, which is where we've been putting a lot of focus because, the companies we work with are the ones that rely on business documents to deliver their value. And for that reason, they typically have a pretty high bar for for accuracy in the app. Mhmm. Awesome. So, essentially, then you go from something that can be sort of a one size fits all prompt to deck to something that you actually need real, like, concrete data that should not be you know, obviously can be pulled in by AI, but needs to come from a trusted source. Yeah. Pretty much. It's, it's like I know I simplified it a lot, but No. But we welcome, like, everything on on just, you know, throw throw a prom that mail. I'll give you I'll give you a a deck, you know, also on your own template. So it's following certain standards. 100% for sure we can do that, but we don't stop there because we know that the the next question that is going to be put us out at us would be things like, yeah. But we for this document, we require certain things that gets it to a to a different state. We would love to connect it to this model because that's been specifically trained for this, but we'd like it to ground its information these specific specific sources. Or we would like this agent to work here and this agent to work here. So the orchestration very quickly, becomes like a topic when you wanna get into those more business critical types of documents for sure. And we're no strangers at Semify to AI. Right? Well, AI assistant is utilized by some of our largest enterprise organizations across the, entire companies. And during that process, a question came out a lot. You know, are you a Copilot competitor? And we always we were very com we're very much a compliment to Copilot at the time considering we could in control the tone of voice. We could give, you know, control to the users on how they prompt. How does this new document agent workflow do we complement, or now is this something that you would say could potentially replace or compete with Copilot? I honestly think both because, there is a a ton of things that we really like about Copilot specifically. For example, the full, knowledge structure that sits underneath a Copilot, which is also now possible to connect to through APIs, which is something we do. So to leverage that knowledge pool is something we really, really like. Yeah. And it's also a way to put more value to an investment in Copilot by getting to higher specificity types of documents with Templafy. So there's a lot of combination in between. And then there are other areas where there's definitely overlap, where especially on the lower tier where, there's stuff that both of those tools can do with their own little differences. So where we sort of over index is always on the control piece and the accuracies and the quality piece because we know it's been our sort of business for forever. Yeah. And, that hasn't really changed. That's a really good answer. And I guess though I guess we did touch upon the admin side of things a little bit, earlier, but I think it's really interesting to dive into the configuration and understanding of how quickly organizations can get time to value from document agents compared to existing document automation solutions. Can you describe a little bit about how the, users go from, right, we have agents, and then how quickly users can start using the tool? Yes. I was mentioning before that, essentially, all we need is a template. Once we have a template, we love to have that because that's a good way to to get to the first level of company standards that actually follows your template. Then you're you're actually good to go because we also have out of the box universal, agents that can support just out of the box creating, creating documents. And then that middle ground that we've learned on the back of our ten plus years of experience of what types of documents are actually being produced. So we have some stuff that essentially works out of the box, with that. And even if you want to take it further and have, go in the direction of having your own specialized stations, It's also a very simple process because compared to what document automation typically has meant, which is binding structures and configuring templates to very specific purposes of each individual placeholder on the document. Agents just work differently. It's more like speaking to a very, talented colleague that essentially understands what you're trying to do. So so so you you it's it's more about giving instructions and then iterating over it. So you would iterate over to get figure out to get to the results, and you have ways to control the different sections that you're pulling out. Again, as I said, which models you wanna connect to or which data sources or knowledge sources that you want to connect to as part of that mix. And then very quickly, you can get to a point where you can iterate and improve it. I mentioned earlier that we we're using this notion on best practice is now worst case. So you very quickly get to best practice on what that agent is able to do compared to anyone in the company, and that's your that's your new base. On that, you just improve through the iterations to get to better and better results. As models get better, the agents get better too. So it's kind of a flywheel once you get started with that. So it's quick to get to value much quicker than it was with the older technology types. It's really funny. Like, when you first you mentioned that and we've heard it a few times, the the best case is the best practice is worst case. I think from an organizational perspective, that's music to their ears because they're they're usually the smaller parts that they, you know, battle with. That's speed versus productivity. But when there's speed versus productivity, there's always, you know, the risk. And I guess a question that comes up often, I remember when, you know, when ChatTBT first came out and there was all these news articles around, you know, people using it and getting caught and fined, for example. What steps are in place on the management side of things to make sure that we can that users sorry, the admins can really control the outputs from the, from the document agents. Yeah. First of all, I would say, we let our customers decide what they think is right for them. Mhmm. So it's not like you wouldn't be able to use, like, a a more general model and apply that to it to get started very quickly. So if you haven't gone through the the work and build out your own very specific stuff, you can definitely still get going. If also, that's what you what you what you think provides you with, with outputs that you're that you're comfortable with. But we also allow customers to make the decision of of going something that is more segregated and more specific to them. Yeah. That is not part of, anything outside. And that's where we're seeing a lot of uptake on, like, tons of models being being either built or specialized for specific purposes inside of companies. As I mentioned, all the all the racks and knowledge sources that we're seeing, coming in and the ability to ground information in existing content that are that comes from these system of records that are controlled by the company. Yeah. That has proven to be, like, an extremely important piece of taking that control. So it's really about again, it's like having the smartest person inside of the company just being available all the time to do to do all the work and also with some specialization. So, you know, it's it's one of the things that AI can really do, like being able to to, to understand and analyze enormous amounts of of content and data extremely quickly Yeah. And then apply it. And what the agents do is almost like the software layer that then allows you to turn that into something that is useful for a user because it's a it's a replicate of a person, the best person you have in the company to doing a certain thing a certain job. So Awesome. So before we go into the technical considerations, I'm interested like, again, we talked very, in great detail last time, October 5, talks around proving ROI and value. Now what how should companies measure success when it comes to document agents? Well, like they always did Yeah. I would say. Unless, there might be be be people in the audience here that have, like, different metrics to look for. But I would say 99% of the times I speak to customers, those are three things. Mhmm. Number one, how do I save money? And that's typically connected to saving time as well. Yeah. How how can I improve productivity, which is one big metric? Obviously, a lot of stuff underneath. Yeah. Number two, how do I make money, which is connected to revenue. So that is related to some of the quality, the improvements you can make to output, which I think is one area where where agents are really good because back to the notion on, on best practices, worst case, it's really an area where you can just get to better output quality. Mhmm. Also quicker, also better quality, but actually improves your ability to to turn that into revenue. And number three is risk. As I do this Yeah. How do I avoid ending up in scenarios where I just put risk on my business because I'm using information that I'm not supposed to or I just have inaccurate information coming onto my presentations. So, and the third one is really, that shoe it's the one that is oftentimes forgotten Yeah. And also where enterprises oftentimes are like, they cannot not have it. Yeah. And, and it's an important one to look at. And it depends how they define risk as well. But, something for customers, and prospects is that the customer success team at Templafy are working on, some extremely, valuable metrics. So when, you do get your hands on agents, we can show you every step of the way how it's, performing. So from the technical side, I imagine the guests here are wondering about deployment and management. Mhmm. They're probably thinking it's just another tool to manage, you know, and it's is it how complex is it? So how are documents documentations deployed? Well, I I would say for the ones that are already existing customers, it's already there. So it's a it's a a click of a button, I would almost say, and then you have the whole infrastructure both on the on the front end, how it's met how it meets the users, but also on on the back end, how it actually how to actually, maintain, the agents on on the back end for sure. But one thing that we've always been, fans of and have optimized our technologies for is to be where users are. Yeah. Where work gets done, you need to be there. So you don't wanna put an extra barrier. You know, you need to go somewhere else to do something. So, you know, obviously, there is, sort of a fully fleshed website where you can where you can trigger the agents. You can do it from your phone. You can do it anywhere, essentially, where you wanna where you wanna start a process of generating a document, which is actually strong now because it's, since it's it's, in a conversational flow, you can actually get to pretty finalized documents alone with that, which was that building something from your phone actually ends up with pretty good outputs just on that. But on top of that, we also like to be very, ingrained with those technologies and applications that people are working with. So, of course, that's PowerPoint, it's Word, it's Outlook. It's also the equivalent from Google. Mhmm. It's your CRM system. It's your document management system. It's where work gets done. We like to be there so you don't have to go somewhere else to find it. And even also having the agents triggered automatically by some of those work some of those workflow processes where you don't even need to have a user involved. Yeah. As as simple as me changing stage on my opportunity in in a CRM system, triggering an event that actually generates documents where where agents are triggered to support that process. So we just like to be very ingrained in the workflows and applications that that users are already using. Yeah. And, so they can just consider this as another really brilliant colleague that supports them getting back to their real job, which is typically not creating business documents. Well, that's the exciting part for us, especially, from my perspective as in the customer success team. You know, when a new solution or a new workflow is then optimized for, there is configuration. There is time that goes into it and effort, but the result is always, on point. I'm really excited to actually have these agents become part of everyday life and just being able to sort of see value even quicker than before. But, but so far, we've discussed around enterprise organizations and how they've measured their sort of viewed AI in the last few years. I'd like to think now everyone, that's viewed this understands what document agents are and their effect on operational efficiency in an organization. We've talked about measuring we touched upon measuring ROI and just discussed some technical considerations. I guess almost that time, but, Christian, if the guests wanna get their hands on document agents now, how do they do that? Well, I would say it's very simple. You get in touch, and we get you up and running very quickly. That's the simple answer. As I said before, at least to get started, all we would require is just a template. And, and we're pretty much ready to get you up and running. And, of course, on the back of that, we would be in conversations with you on what you're actually trying to achieve Yeah. And how agents can be helpful. And if agents is not, the exact or is not the technology to use, we definitely have other stuff, to support on that area still. So it's basically just about reaching out. I think there was a call to action here. Yep. You can access the link in in the chat, and we'll be in touch. Yeah. We'll be in touch and get you up and running quickly. Well, thank you, Christian. It's been a pleasure having you on Templafy Talks. As, I mentioned at the start, this is part one of two. The next session is gonna be around use cases. We're gonna show you how the agent works, from the back end. Also, obviously, on the front end, what the user experience is. And, by then, we'd like to figure that some of you are gonna have your hands on the tool. As Christian said, fill in the, the form that we've just posted in the chat. Get in touch. Your account manager, customer success manager, maybe even Christian will get in touch. Not only can we set up the agent for you, very, very quickly, we can also, dive in and understand exactly how else to build upon those agents. So once you've got a strong agentic foundation, we can add more and more agents to your workflows. So very, very exciting times ahead, from myself, from Christian, from Isabella behind the scenes, and Casper, in the chat. Thank you, Casper. That's all from Tenify Talks, and we'll see you on the next one. Thanks for joining.