フィールドサービスの専門知識をAI時代にどう残すか:SightCall創業者 Thomas Cottereau 氏に聞く

Welcome, everyone, to the Emerge AI in Business podcast. Today's guest is Thomas Cottero, CEO and founder at Sitecore. Sitecore is a visual service intelligence platform that helps enterprise service organizations capture, structure, and scale the expertise behind real-world work. Siteful combines visual assistance, AI, and knowledge capture to transform service interactions, expert demonstrations, training, and existing content into trusted, reusable models, making expertise accessible to technicians, service teams, and enterprise AI systems. Thomas explains why service organizations need to capture the experience of their best technicians, not just more documentation, and why AI should structure what those experts already know rather than generate answers on its own. Today's episode is sponsored by Sidecall.

DSL AI products or services, Emerge gives you access through trusted content and real conversations. Learn how leading AI brands like NVIDIA and Google Cloud work with Emerge to reach Fortune 500 AI buyers. Download our media kit at emerge.com.ab1. That's emerj.com.ab1. Now the conversation with Tom. Thomas, welcome to the Emerge AI and Business Podcast Studio. Hi, very nice to meet you, Yolandi. Very interesting conversation today, and I know you have some interesting takes on it. So to give the audience an idea of where our minds are at today, we know that every service organization has that one technician that everyone calls when just nothing in the manual works out or it's just a very out of the ordinary case that they need to attend to.

And the problem with that is that most of what that person knows has never been written down properly. It's very much a let's call them, get the info, get the problem solved and continue with our lives. So I want to hear from you from where you said what's making it so hard for organizations to hold on to knowledge and pass it on. Very interesting question. I think that one of the big confusion today is confusing information with expertise. So most service organizations have, you know, manuals. They have a training deck. They have knowledge base. But there's a huge part of what makes a great technician great that is not in these systems. It's called tribal knowledge, right? It's all of these little things that have been learned after seeing the same failure 20 times, the sound that tells them that something is wrong, or even the diagnostic they have to run to define the very next step that they even learn from their experience.

So now, at this time in the enterprise world, and typically with manufacturing, services, there are kind of two forces colliding with each other. You have your most experienced people that are retiring or becoming harder to scale, while the new generation of technicians needs to become productive much faster, right? So the traditional answer has been, hey, let's try to document more. The problem is that asking an expert to stop working on seats with a technical writer, explain them what they know doesn't work and doesn't care. So for me, the challenge is not preserving documents, it's preserving experience. How do I capture what my best people know before it works out the door? And how do I make that expertise available to the next technician at the exact moment they need it? That's a very different knowledge problem that we have. It makes sense.

So I'm thinking we might have audience members sitting there, audience members in the service organization or industry, thinking that, okay, I work with technicians, we have technicians, we have some very valuable technicians retiring or maybe just finishing their project with our company. How do I know this problem even exists? What is the pain point that they feel before realizing, oh, it's too late? Let me ask you this question. When you have a problem with a physical device, what would you do if you don't know, if you don't find what the problem is? Probably Google. Yeah. Actually, interestingly enough, in the surveys today, when technicians are stuck in the job, there's more than 50% of the time where their escape route is to call another one that they know. Hey, I know that Bob might know the answer, so let's call Bob, right? That's the number one. Because Google might not have the answer, right? If you think about, you know, very specific machines, usually Google does not know. Then they implemented some AI systems, but the AI is also untrustable in most cases.

So they search for this very specific experience, right? And furthermore, if you add the fact that as you're working on physical equipment, physical device, you need to see because just explaining with words is hard, right? Is that this cable? Is that this pump? Or this very specific leak that you can see but that you can hardly describe. So again, that's where everything came back together with this need to capture the expertise from the field, even visually, on being able to restructure that to share it with the rest of the organization. That makes sense. So basically our pain point that will feel first is, oh, we don't know how to do this. Let's phone Bob. Oh, but wait, Bob retired last month, so what now? And then we'll realize, okay, we should have captured Bob's brain in some way. Now, I'm also thinking that, you know, a lot of the expertise within the organization is normally on show. Like you just explained, there will be technicians interacting with each other, sharing their experiences and saying, oh, I had this very out of the ordinary call today, and this is how I had to solve it.

Something that we won't find in the manual. And again, the problem comes to that is always in canteen-type conversations, lunchtime conversations, and never being written down. So let's talk about the soul. What can technology do to help us here? How can AI change the way that organizations capture that frontline expertise and turn everyday service work into something that people can actually use? I think the very important principle for me is pretty simple. You should not ask AI to create the expertise I think that a very very strong point because it could bring people to the wrong direction If you expect generative AI to generate you know this kind of fix, this kind of process without knowing, without having kind of the foundation to create that, you're going to fail. The starting point is that your team knows, right? So this expertise lives in your people, in your team. So what if we use AI to do the back office work that nobody wants to do?

So what I mean by that is using AI as a tool, not the other way, so that what your people knows becomes structured, trusted knowledge that human and AI, by the way, can also use in the future. So really think about it this way. Your expertise is here. Let's capture it, right? How do we capture it? We started as, you know, cycle where the leading the video assistant space, so helping people see on guide remotely. So that's where we started from. And then we really expanded out of this solution to generating knowledge because, you know, our customers told us, hey, when, you know, someone like in the remote fix team had explained the same fix 20 times, should he explain it another time to another person? So, yeah, I think that we can have the AI who learns from these 20 sessions and then regenerate the full step-by-step guided process that can be shared with the next technician facing the same issue or the next customer.

You know, so that's where we believe that AI can be transformative by being the tool that is doing this work that nobody wants to do. Nobody wants to watch these 20 recording videos on rewrite everything, on restrict your things, et cetera. So that's where AI is good at, right? And then you can show it, you can search for it to humans and, again, to other AI agents as well. So basically you are going to force the AI to teach itself, but from what it sees. Is that the right idea? Yes, that's exactly the idea. So that's why even in the AI transformation process, we have an analyzed video frame by frame to identify, you know, on this frame, what is this person doing? How did they remove this valve or whatsoever, right? So it's a combination of the image on the voice, the text, and all of the contextual information that we collect there, like a work order, all of these context signals that can help create the right tutorial for the next technician.

I can imagine that obviously we don't want to burden the technician with, okay, you now need to take a video of everything that you do. This obviously needs to happen while they're working. How does that work? Is that even possible for it to capture image or capture video while the work is being performed? Yes, you're absolutely right. That's a very key point. I've heard some crazy stories like, hey, you know, you should carve out some times from your best stack and give them the Friday afternoon to document. Guess what happens, right? Nothing. Nothing happens. Yeah, knowing that today everybody sees this gap, right, this knowledge gap, wants to capture it and restructure it and then share it to the new generation or the ones who didn't have the chance to get the full training program, all of this stuff. So to do that, you're right, we have to capture it on the job. So what I've seen today working well, one was recording all of these video assistance sessions because each time there's a problem or help that's provided recordly,

that's a perfect usage. Another one that's growing is even recording the trainer. So you have someone explaining how to fix a machine, how to do this maintenance, and then instead of having to repeat this session again and again and have even a very long two-hour video on how to perform this kind of maintenance, if you have the AI that can restructure that into step-by-step with short video snippets in there. We're more in the Instagram world now, right? Nobody wants to watch a two-hour video. 10 seconds is great. Show me exactly the fix here. And that's what we can do with AI by analyzing the image and taking just the small chunk that we want to keep in there. And then we've seen other very interesting also trends like ride-alongs. Get the genotech, Feel the senior tech. When they explain them something, don't just keep it for yourself. Record it so that it can be shared with everybody.

And even for you, if you face this same challenge, you know, in two years from now, you will not remember what you were saying this day, right? But if it's in the system, now it's becoming searchable just with a reference number or any other question in natural language. So you are literally kind of transparently building this knowledge. And that's what Gartner describes as the democratization of the knowledge creation, which has to go from a very centralized place where it has been over the past 50 or even 100 years to everybody has to be part of this. But it has also to be seamless. It has to be transparent. We cannot ask you to become a knowledge creator. So, yes, you are totally right. It has to be on the job and you have to capture what happens there and then use AI to transform it for you. I'm guessing that especially our very mature, for the lack of saying old, but for our mature service technicians, obviously content creation is also not their biggest thing.

So a lot of ideas that you shared there that make sense for every type of person that we have within the organization. Before we move on to the next question, I want to just also ask this, because I think this is something that our audience members might think about and wonder about, is who decides whether a piece of captured knowledge is actually correct? If we film a technician doing something, who decides whether they did it the right way? Is it just by default that we accept it, or is there some kind of intelligent layer that we will have? Can you take us through that idea? Yes, that's a very good point. So there are different trends in the market. One trend is capture everything, dump everything into a big AI, LLM, rag model, ask a question, and pray that the answer is good. Doesn't sound like the best idea. But that's what actually a lot of people are doing, right? And then there's deception, there's disillusion. So the other path is really to kind of create an ingestion pipeline with sort of a workflow that would analyze all of these knowledge capture videos recordings all of the materials that you have

and map them into what we call a knowledge graph. So meaning one graph that is designed for your organization with the right type of asset, of fix, of procedure that you want to have, and literally restructure everything, all of these signals, map them into this graph. And this graph will then lead to generating this how-to tutorial, right, where you extract the small video snippet, where you restructure the step-by-step, where we can also even generate image with self-annotation to highlight the very specific action that you have to take on this step number seven. And then, when all of that is restructured and rebuilt by the AI, have a final human review. That's how you have highly trusting knowledge because at the end of the day, that's going to be your foundation, the knowledge foundation for your team, for your AI agent that everybody will rely on. So it has to be very strong.

That's how all customers today have a human in the loop at the end of the process. AI today, in what we see on the feedback we have, is that these tutorials that are regenerated by AI, restricted by AI, right, where it's your knowledge restricted by the AI, 90 to 95% ready to go. But there's, again, maybe you have some very specific in your organization, some acronyms that you want to use a different way, or some process that you want everybody to respect, then you can do that in D&D at the end of the process. Yeah, I think for any listener that wondered about that, about how do we know if it captured the correct type of solution or was it executed right? That is an answer that will give them trust in the system and say, okay. Again, obviously, the term human in the loop helps a lot there, but to see that there's a process behind the scenes that basically a filter that will help us to know that, okay, this is the correct process, this is the right way to capture it, and the right information was captured as well.

And on that topic, and it fades in perfectly with what I wanted to ask you next, is from what I've seen, it looks like a lot of organizations start this transformation by loading their existing documentation into the enterprise AI platform. And I'm sure that approach will have some things that work very well there and some things that will create a big myth. Can you take us through your thoughts on that? Yes. This is why I think companies are at risk of taking the wrong shortcut because there are really two different problems to solve with enterprise AI. The first is access and intelligence, right? So take everything you already have, your manual, your service tickets, et cetera, put that into a RAG or your AI or even a giant platform. That's all good and you have to do that. That's a good start. But there's a second problem underneath it. Where does the trusting knowledge come from in the first place? So your service intelligence can only be as good as the knowledge foundation you give it, right, as we already surfaced.

So a work order might tell you that the technician replaced the component and closed the case. It probably doesn't capture what the technician saw, why he knew that component was the problem, the trick he used to replace it or the warning he would give another technician doing the same repair. That's the layer I think companies are missing most of the time. You need a knowledge creation engine that continuously captures what your expertise demonstrates and explains and reconstruct that expertise into this structured knowledge. So that's when it becomes very complementary. If you already have this type of AI platform, don't replace it. Just feed it now with better trusted knowledge. I think that's going to make a lot of leaders feel better and not think that, because I think a lot of them might think that, oh, we already have some training videos and some manuals, and we don't have to rebuild all of that.

We can repurpose it. Am I correct in saying it that way? Yes, you're absolutely right. There's something I've seen a lot is that definitely there's a lot of documentation. There's very few videos usually today because videos are very expensive to create, right? If you would take an agency, for example, on TeleMV, can you create a how-to video on this specific installation? The average cost for it today, if you take an agency, would be around $15,000. And it would take them one month. And furthermore, it does not even remove the burden of your team description because they don't have the expertise. So they're going to have to ask you. They're going to have to ask you to review the script one time, two times, to adapt it, all of this stuff. So that's why it's still pretty restricted today. Now, what we come up with is literally two things. One is everything that you already have, yes, can be absolutely repurposed, right? And I strongly believe in, you know, this AI system where everything should interact with each other. This type of AI agent with what we call MCP server where they can interact on shared knowledge.

They should not be disconnected. So if you already have all of these documents in an AI system, great. Just connect the two systems together. If you want to scale that and have videos and do more, don't need to go to a long, expensive, heavy process anymore. Capture them on the spot. The good news is that now the AI will transform it for you to make it available as new trusted knowledge. I think my favorite thing about that last sentence is the good news, And I feel like our solution part of this conversation, speaking on how AI can solve for this pain point, it really sounds like good news. It's a very exciting time for the ideas around capturing tribal knowledge and in an effective way, just as a side note. I mean, I think with the video capturing, we already live with a new generation that wants to use their phone for everything. So why not have that little training on there as well and all of their questions and gaining knowledge and harvesting the expertise from the older technicians makes very much sense.

So the last part of this conversation is where we get to speak to our leaders directly and give them some practical guidance on where to start, what to do, how to think about it. So for the leaders listening, what should they prioritize today to make frontline expertise easier to capture, to maintain, and to use for both their people, but also for their AI? I would start with one principle Don make knowledge capture another task that another silo that you put on your people Just put it inside the work people already do If contributing knowledge requires a technician to go back to a laptop at the end of the day and write an article, it won't happen. Let them recall the Reaper, explain something by voice, demonstrate a procedure, or capture a training session. Then second, I would build trust into the process. AI should help create and structure the knowledge, but you need governance. Who is the expert for this equipment? Who validates this procedure?

Is it generated, reviewed, or approved by who? In service, a plausible answer is not good enough if someone is standing in front of a $500,000 machine that needs a fix now. And then the third thing would be to think about knowledge as living knowledge. It's not a one-off. So don't try to run a six-month knowledge project and declare victory at the end. Product change, people change, procedure change. This technician discovers a better solution. AI has to continue to learn from their interaction and feedback every day. This knowledge foundation that we mentioned has to continue to absorb the expertise and continue to improve over time. And maybe the final one, if I may, is don't create another silo. You already have an AI initiative. This knowledge layer should connect to the system. So my goal is that the same trusted expertise can help, on one hand, a technician directly on the field watching this tutorial.

On the other hand, your AI agent to be much smarter and provide the best answer to the rest of the organization. Yeah, and that reminds me of, I heard you reference the living library before, and I know that we are just about out of time, so I'm going to work in one last question for our audience members because this is a very interesting concept for us to end the conversation on. You know, knowledge has been traditionally owned by central teams and maybe technical writers. So what does it take to make capturing and improving expertise that ongoing responsibility? How do we get to that point across the entire organization? I think it's a matter of engagement of the team. Here we're talking about more human behavior. So if you present this project as, hey, I'm going to get a camera on you, record everything so that tomorrow I can replace you by your bot, I think it won't work, right? Now, if you take another path where we say, hey, you've worked with us for 30 years. You are the most savvy person.

Can you help transfer this knowledge to the next generation? Can you bring that as a legacy to the people who won't have 30 years in the job? To this machine that's still around today that might be also 40 years old, that, you know, the product documentation would not help to service because we need your expertise. So can you be part of it? So it's really, you know, getting everybody in sort of this participating mode. That's where we discussed about the fact that this knowledge has to be decentralized. Everybody has to be part of it and even share feedback. So even if they try on the C1 tutorial that's provided here as a guideline, and they feel that there's a better way to do it, please share it. Because that's how, again, we can reprocess it and get the better solution behind. So it's really getting us most of the time. And maybe I would lead with this quote. to wrap up the discussion.

It's from John Carroll, who is the CEO of the Service Council. And he said at the board meeting during our last symposium in Chicago here in September, what he said is that essentially when every company is using AI and increasingly using the same underlying model, what will differentiate one company from another? So there's going to be two things, your people and your data. And I think that's exactly right. So don't just invest in AI, invest in your people and bring them to this transformation. And in parallel, invest in the knowledge foundation that captures what they know and make your AI yours. I love that that's the final thought that we leave our audience with. A great way to end the discussion. I think our big takeaways here would be that the most valuable service knowledge is not in your documents. It is in your best people, the people that you've invested in for years. And unfortunately, it will leave when they leave if you do not capture it.

So putting your old documents into your AI is a starting point, but it's not the only thing that we need to focus on. The trusted knowledge comes from learning from your experts as they work and having them confirm that it's right, keeping that human in the loop even after you've captured their expertise. And then the last part, the very exciting part of the conversation for me is treat knowledge as a loop that everyone feeds. Like you just said, it's not a single thing that one person will maintain. Everyone will share their little cheat sheets on, you know, what is the best way to fix this or what is the fastest way to get this machine up and running again so that our client does not lose out on a lot of money. And that is where we can see things like faster onboardings happening, fewer escalations, and just better AI all around. Thomas, thank you so much. It's been great to have you on the show. Excellent conversation. Thank you very much. Wrapping up today's episode, I think there are three key takeaways from our conversation with Thomas.

First, the expertise that makes a great technician is rarely in manuals or knowledge bases. So the priority is capturing experience before it leaves, not producing more documents. Second, AI should structure the knowledge your experts already have rather than create fixes on its own, with final human reviews so that the results can be trusted. And then finally, knowledge capture only works when it happens inside everyday work and keeps updating, so the same trusted foundation serves both technicians in the field and AI systems already in place. Do you sell AI products or services? Emerge gives you access through trusted content and real conversations. conversations. Learn how leading AI brands like NVIDIA and Google Cloud work with Emerge to reach Fortune 500 AI buyers. Download our media kit at Emerge.com.ad1 That's E-M-E-R-J dot com slash A-D-1 For further executive level analysis and to join our network of leaders delivering workflow impact with AI, visit Emerge.com. On behalf of the team at Emerge, we'll see you on the next episode.

Thank you.