Skip to main content
April 29, 2024

Self-service content in the age of AI with Patrick Bosek

In episode 165 of The Content Strategy Experts Podcast, Sarah O’Keefe and guest Patrick Bosek of Heretto discuss how the role of customer self service is evolving in the age of AI.

I think that this comes back to the same thing that it came back to at every technological shift, which is more about being ready with your content than it is about having your content in the perfect format, system, set of technologies, or whatever it may be. The first thing that I think either of us will say, and a lot of people in the industry will tell you, is that you need to structure your content.

— Patrick Bosek

Related links:



Sarah O’Keefe: Welcome to the Content Strategy Experts Podcast brought to you by Scriptorium. Since 1997, Scriptorium has helped companies manage, structure, organize, and distribute content in an efficient way. In this episode, we talk with Patrick Bosek about the changing role of content in self service and whatever the opposite of self service is, maybe just service. Hi, everyone. I’m Sarah O ‘Keefe, and I’ve got Patrick Bosek, the CEO of Heretto with me today. Hey, Patrick!

Patrick Bosek: Hey Sarah, it’s good to be here. I guess to be back, technically.

SO: Yeah, you’ve been here one or two times before, so I’m going to cut to the chase here. And our topic today is self-service content and how things are changing in self-service content. So talk a little bit about that. What’s going on?

PB: Well, I think to talk about self-service content, we have to talk about what’s changing in, I think, self service more generally, which you kind of alluded to in the idea of, you what is the opposite of self service, right? So the landscape is, as I see it, is very interesting today, because historically we had what was very obviously self service, and then what we had was very obviously not self service. So I guess just people or service or something—people service maybe. And for the most part, self service was content, right? So if you went someplace and you read something and you figured it out on your own, over the last decade or so, self services started to involve a little more action. You can go to a McDonald’s and self-service order a coffee today. We can talk about it in a minute whether or not we think that’s a good idea or a bad idea. But now as we’re getting into the age of true…intelligent, if you want to call them, virtual assistants, AI, those types of things. Now we’re in a place where the things that were very traditionally handled by humans. So helping you figure out what you really mean, helping you dig through something when you’re not exactly sure where you’re finding, finding it, or if it is possible, those types of questions are actually performing actions. Some of those are going to continue to bleed over into systems. So now self service isn’t just content anymore that you go and look up and then you yourself go and figure it out and do something. There’s going to be this mixing between these two things where the service that’s provided by automated systems is going to perform some of the things that humans were performing. They’re going to need a bunch of content in order to be able to do this properly. And probably also as instructions, you’ve got to teach these things what to do somehow. And as we all see with the way that we interact with them, you use words, you don’t use programming languages as much. So content plays a role in its traditional form. It continues to play a role in training people. It also plays a role in establishing what this new generation of systems that are going to help us perform actions and learn things and answer questions, what those things are going to do.

SO: So it feels like we’re adding another dimension to this, because when mobile apps first came out, the big development there was that they were contextually aware. So you can get your app to tell you what the weather is at your location because it knows where you are or it can know where you are. And it feels like this is a similar kind of thing that some of this is more a matter of not just, “Hey, here’s a page with some instructions,” but rather, you know, let me, let the system do some work around what your context is and what some of your knowledge is and adapt accordingly.

PB: I think there’s absolutely an aspect. So I would actually put that in the category of just like maybe even traditional personalization. You know, feed metadata in those are things about yourself and then you perform some type of a matching or a computation. And then you feed content back. So that’s, that’s effectively personalization out of its core. Like here’s some things about me. Okay. Then those match to some things about the content. Give me just the content that matters to me. I think where this starts to become new and really interesting is where you start to have systems, so probably, you know, AI-based systems that are actually not just like filtering or personalizing content, they’re actually manipulated in content or they’re manipulating your journey with the content. And one of the places that, you know, we are seeing more of that I think is an interesting place for this is in learning. So you start to think about how learning systems and self service have worked for a long time. And as much as there’s been like micro learning content, and then there’s been more organic things like that kind of stuff, by and large learning has been linear when it comes to self service. Here’s a guide, here’s a course, whatever it may be. And it didn’t matter if you didn’t need to know a good chunk of it or if you already knew a good chunk of it or whatever it may be. Like you went through that, right? That’s how that works. So that, I mean, it’s how colleges, college courses work. It’s how, it’s how learning works in general. Like unless you’re sitting across from a tutor, like learning is linear by and large, when it’s being taught. Well, with AI, all of a sudden we can deploy systems at scale potentially where you can always be sitting across from a tutor. The entire paradigm around how it is that we, you know, air quotes here, “self service,” if we still want to call it that, the things that we need to learn can fundamentally change.

SO: So what does it look like for us sitting in the content universe when customer experience is moving in this direction towards this, I guess, more sophisticated self service? Not just here’s what we have, deal with it, but rather here is information or learning or a chunk of content or whatever that is adapting to that person’s requirements. I think it is, right?

PB: In certain circumstances, it certainly has the ability to be adapting to people’s requirements. I think that’s the thing that we will absolutely be seeing. But more generally, what does content need to look like to give you the range of things you might want to do as an organization in this new paradigm? And I think that this comes back to the same thing that it came back to at every technological shift, which is more about being ready with your content than it is about having your content in the perfect format or the perfect system or the perfect set of technologies or whatever it may be. So the first thing that I think either of us will say, a lot of people in the industry will tell you is like, you need to structure your content. And I do think that the story for this on the learning side, on the traditional self service side, on the AI-agent side, and I think even on the people service side of things still does start there. But I don’t think it’s because self service is intrinsically something which is powered by structured content. What I think is that structured content really just gives you, the organization, you, the content creator, a lot more control over what goes into these systems no matter the range of intelligence that they have. And that means that you have input control on the experiences. And as we all know, LLMs, even when they’re backed by like RAG, retrieval augmented generation, or other systems that are meant to kind of keep these things fenced in, they’re black boxes. And the bigger the box, the blacker the box. So, the strategy, if you look out over the industry, there’s a lot of very sophisticated stuff, but some of the stuff that works the best is input control. And that’s where I think that structured content is really gonna be a key element of this, no matter how far in the future you look.

SO: Yeah, and I mean, my explanation of this to people, which I’m sure makes the actual AI experts cry, is AI likes patterns. And so if you feed it content that follows consistently the same pattern, you greatly improve your odds of getting good output from what you’re putting into the system. When you have stuff that’s not well organized or structured or anything else, you know, garbage in, garbage out—you’re gonna get a mess.

PB: So I think there’s that is true. There are caveats, but the thing that remains true at the center of that is that if you don’t have very precise control over what goes in, you lose an enormous amount of control over what comes out. So even there’s, there’s such things like overtraining, right? Where you can actually get AI that will produce, less high-quality results with certain quantities or certain types of training. And what you end up with in those circumstances is that like, okay, well, if your stuff is just a bunch of stuff and you stuff it all into an AI system. 

SO: That was excellent.

PB: It’s good, right? I gotta have some fun. So you don’t have any ability to say, okay, well, these pieces really, these are the ones that are impacting our outputs. Let’s pull this out, or even just iterate in an intelligent way. So, which part of the corpus, which part of the things that we put into this system are the ones that are having the negative impact? Retraining becomes a much more complicated process. And at the same time, when we’re looking out over deploying across multiple experiences, right? So, you know, let’s take the learning and reference. So, you know, probably speaking documentation portals, whatever they may be. Some people call them knowledge bases in certain circumstances. Those are the two obvious things, because in the past they’ve been highly bifurcated, even though they use a lot of the similar information underneath the hood. Well, if you’re trying to build AI-backed much more like personalized learning systems. Well, you can’t have the content in those systems being different than the reference content, because when you go and you look at the stretch of things that go into that stuff, well, if you have the AI system telling your, user something which is wholly inaccurate. And you can’t pull it back to the rest of the stuff that’s published on the internet, you can get highly divergent results, and you could end up in a circumstance where you have no ability to actually deploy these things properly. So you can’t have one set, which is very cottage based, like, you know, we will go in and we craft these things and one set, which is highly structured. And then you power all of the learning, the intelligent learning systems of the structured stuff, because it’s going to be easier. So you have to find a way to pull these things together, and then use the mechanisms underneath the content to put the right inputs into the right places.

SO: And we’ve been talking for years and years and years about problems with silos and how they’re an outgrowth of the organization itself, right? You’ve got a learning organization and a documentation organization and a tech support organization. They’re all producing content into their respective silos. And the question becomes, if organizationally that’s what the company looks like, then it is almost impossible to rip those silos apart or put them together, destroy them to collaborate across them because the org chart doesn’t encourage it or even makes it impossible potentially. And so then you’ve got different terminology being used by the same company but in different departments, which is really common. And then what? So we’re back to, you know, the fundamental truth that when you have a website as a company, even if you segment that website into like, oh, and and KB or, your customers don’t care, right? I mean, they’re not interested in the fact that you have three separate organizations that all hate each other. That’s just not on their list of things they care about.

PB: So I mean, this goes back to the classic, like don’t ship your org chart, which is, yeah, obviously, right. So obviously we’ve been doing this in content for ever, basically. I do think that it’s gonna be, we’re gonna be forced to change because, you know, again, you go back to the idea of like, if you contradict yourself in your learning content and your docs content and it’s being read as written or as built and presented to a human being. Human beings are incredibly flexible creatures. We can go in and be like, oh, well, okay, fine. So like they didn’t update that piece, those dummies, they should have done this, but I understand what’s going on. But when you put an abstraction layer over that, now you have a system that just basically does what it’s told, you know, by and large, or understands what it’s, what it’s educated in, and what it’s told, it’s not going to have that same intuition. That’s a very human thing, even at this point in time. I don’t, I don’t really see the current generation of LLMs getting to a point of having that style of intuition. I mean, the thing that just happened with the ASCII art is a great example, right? And it’s a little bit divergent here, but bear with me. So people figured out how to hack these AI systems by going and asking them questions with ASCII art, which in one sense shows their brilliance because they’re able to understand it. But in the other sense, it shows their lack of intuition because they were like, oh, well, this doesn’t apply to my rules. Who cares? Right? And whereas a human being would have been like, oh, I’m still not allowed to talk about bombs. Right? It doesn’t matter if you’re in a theater, if you write, don’t yell bomb in ASCII or in Sans Serif, people understand that it’s still talking about the same general concept. So this is the same thing that you run into where you can’t have these discrepancies and stretch a single system over the top of these. And you can’t have a really strong customer experience that properly educates people, properly answers questions, and all those types of things, unless they’re joined together.

SO: Yeah, apparently in addition to ASCII art, if you use Morse code, that will also work around all the guardrails, which sounds fun. So, okay, so in our couple of minutes that are left here, how does Heretto and CCMSs in general, but Heretto specifically, how do they play into this?

PB: Yeah, sure. So that’s a great question and one I appreciate you asking it for obvious reasons. I’m going to answer the CCMS part first, then I’m going to answer the Heretto part. So CCMSs as platforms, and I think this is probably true for pretty much every major CCMS in the industry or in the space. They’re going to give you the ability to manage more structured content at a higher velocity and a higher level of governance. Now they’re all going to be able to do this, you know, to different efficacies, right? So some are going to do it better or worse for your particular circumstance. But broadly speaking, that’s why you buy a CCMS. A bunch of content, you want to be able to have your per author or per information developer content be higher than it is, and you wanna make sure that you have the proper amount of governance so that what you deploy is what should be deployed, right? It has to be good enough, it has to meet the criteria, especially today. So those are the things that you build into the process in the CCMS. And then, obviously, they help you track things like localizations as well, but I would broadly put that in the same bucket. So as we’re looking at, how does this relate to the future of the customer experience, you know, be it directly with the content or be it derived from the content through some intermediary system like AI. It’s the governance piece, and it’s also the quantity piece. You have to have enough to be able to answer all the cases to be able to touch all the learning points to be able to educate and guide these systems in all the proper ways. And now, because you don’t have that human intuition as your last fail save, the level of governance has gone up a bar. You have to be able to have much better governance on your content to be able to control inputs. So I think that this is a new age of CCMS. And it’s funny because we have seen an acceleration in interest from people and an acceleration in interest that’s educated where people are coming. And they’re like, we have to get this in order, because we realize that if we don’t have our hands around this, we’re gonna have a huge mess at the end of the toolchain. So I do think that people are more aware today, you know, it’s probably still relatively niche in the grand scheme of things, but there’s a growing awareness that the having the right systems in your content operations ecosystem to produce the right outcomes down the chain is gonna be critical. And CCMS for this style of content is 90% of the time gonna be the best place to start. Not to say there’s not other ways to get there. On the Heretto question, Heretto does all that stuff like the other CCMSs. Obviously there’s some aspects of collaboration and things like that that we think we do better. But I think the key thing as it relates to the future of these technologies that Heretto provides that you don’t really get in other CCMS technologies is our ability to efficiently and agilely deploy content into specific pods. So we have static publishing, we have the ability to generate HTML, PDF, all that kind of stuff, like all the other CCMSs. But we also have the ability to dynamically deploy content into an API layer where the content is in its own little pod. So you can kind of deploy as many little content APIs as you want. Most organizations have one big API that they deploy that powers an entire doc site. It can be tens of thousands or more, you know, topics. But you also have the ability to say, all right, I want to just deploy this here, and this API is only going to have this content in it. And that comes back to the critical aspect of the bigger the block box, the blacker the box as it relates to AI systems. So you don’t go hook your AI system up to the totality of, you know, your large API that serves your web experience or general web experience. You hook your AI up to this specific pod that only has these specific things in it. And therefore, you know exactly what’s going into that AI system, you know, be it for something that’s more like RAG-based, which is really just like search and summarize, I think it’s a much better name for it. Or BSL which is more training-based. So I think that’s kind of the critical piece that Heretto offers right now that is not present in other systems that, you know, relates to what we’re talking to today anyways.

SO: Alright, I mean, that seems like a good place to leave it, because basically you’re saying, hey, this stuff is coming. All these things are changing, and here’s a helpful roadmap for how to get there. Any closing words before I wrap this up?

PB: No, other than this is a really exciting time to be part of content. You know, we’ve both been here for a little while, and we’ve seen a lot of changes in this industry. But this is certainly unique. There is no doubt that the acceleration understanding and the change in the landscape over the last, you know, year, 18 months, I would say has been unprecedented. You know, I think you see that all the way from, you know, the types of experiences that we want to start deploying that we believe are possible, but we’re not totally sure based on new technologies. And then also the change in approach to things like learning content, seeing organizations suddenly starting to say, okay, so like PowerPoint is not our primary method of training people. Like our primary method of training people is going to be dynamic digital experiences. And we need to be prepared for that. And then whatever comes after that, you know, I think that this is a shift that, I’ve been waiting for, you’ve been waiting for, for a long time. I mean, like, for gosh sake, like we implemented the experience for DITA learning and training. What is it? One dot one or something like that, like 10 years ago. And we’ve had a couple of customers that have used it across that time, but it’s just been recently that we’ve had more and more people coming in and starting to use it. And there has, there seemed to be a bit of a Renaissance in the understanding around these things. So, I don’t know, this is just, it feels very new. It feels very fresh again, which I think is part of the structured content cycle. And this is the fun part of the cycle. So I’m enjoying it.

SO: Yeah, I would I would agree with that. So I’ll leave it there. Patrick, thanks for being here. Always good to see you. And with that, thank you for listening to the content strategy experts podcast brought to you by Scriptorium. For more information, visit or check the show notes.