Beyond Forums: Building the Agentic Front Door to Customer Success

All right, thank you, Natasha. And thank you so much for joining us for our session. As we said, I'm Brian from SearchUnify, which is a Graziidi interactive company. And with me is Shweta, who is our head of our community practice for Graziidi Interactive.

And so Shweta has actually lost her voice pretty much completely. So I've been told I've actually volunteered to try and actually help her with her talk track. So I can confirm. She wasn't at the sphere.

She wasn't at some after party last night. So we're just going to have to craft a really funny Vegas story to actually justify why she can't even speak. But again, we really appreciate you guys joining, especially after lunch. The last thing we want to do is put anybody to sleep, especially after a meal's so filling.

So I think we'll just go right into-- we'll go right into a couple things here. Before we get to the slide, one thing Shweta wanted to say is that when we look around this room, everyone has built something. A forum, a knowledge base, a peer network that they genuinely believe in. And that belief matters, and it's what keeps those spaces alive.

But if we ask ourselves one question before we go any further, does the community that we've built actually resolve problems? Not service information, not connect people, but resolve end to end without a human needing to step in? And I think for most of us, the answer is, well, not yet. We're on our way.

And so that gap between our communities, what they do, and what our members expect them to do is exactly how we'd like to spend these next 20 minutes with you. So let's go into this next slide, which is what we're calling the 2 AM problem. We're seeing here that you recently onboarded a customer. It's 2 AM, and they've got a problem.

And so what are they going to do to try and solve it? So if they go to the community, these are some of the things that they might run into or they might get from that experience. Number one, a form thread from three years ago. Number two, keywords, dead ends, and broken links.

And at the end of that frustrating self-service journey, they then have to submit a support ticket and hurry up and wait until somebody can actually come to them and help provide that resolution. Now, we have to think about the disconnect in terms of expectations, because those three bullet points aren't really what a customer is looking for when they come to the community. They're looking for accurate up-to-date information. They're looking for context-aware conversational answers.

And while we're not saying that every question can be automatically resolved, but they're hoping for more of an immediate resolution without having to get into a queue. And I think the key takeaway we have here is that behind every modern community is a connected system, a system that understands intent and resolves queries faster. And we're looking at across three key pillars that will unpack with you today. So if we think about knowledge, specifically trusted knowledge, knowledge is basically the memory of your organization.

Search is a way to get the information Search is the nervous system. How do we find that knowledge? How do we deliver it to our customers just in time when they need it most more than anybody else? And the agentic AI layer, that's really the intelligence that makes both of these make sense and take action for your customers.

So these three pillars form what we call an agentic success center. And this is the front door to customer success. And again, because it's asynchronous, because your customers are reaching out and trying to engage with your community, we shouldn't just be thinking about it as a ticket queue or a support portal. The community is going to be an engagement hub if you build it right.

And so let's go ahead and walk through how we did this for one of our clients. [AUDIO PLAYBACK] Let's see here. Oops, sorry. [INAUDIBLE] Sorry.

[INAUDIBLE] OK, cool. So first, let's start with a pillar that is on everybody's mind and probably on most vendors' boots, agentic AI. Not a chatbot, not a search, not an autocomplete, but something fundamentally different. So I know there's only a few of us in this session.

But one thing I'd like to ask is that if we could just raise our hands, have you deployed a chatbot or been a part of a decision to deploy one over the last couple of years, whether it's, hey, we went to a vendor and we decided that we wanted to deploy their chatbot? Or somebody in our organization said, hey, why don't we just throw chat GPT on top of our community? Show of hands, how many of us have gone through that experience? At least more than once for many.

Now, let's keep our hand up or raise our hand again if you describe that outcome as genuinely successful. Yeah, that's what we've seen. And the good news is we're in the trust tree. It's a very normal occurrence right now.

If we actually resolve the majority of what the customers needed, we wouldn't be in this loop where we'd be frustrating them just by deploying these experiences that confidently deliver answers but do so in a very incorrect manner. So traditional chatbots were essentially pattern matching engines. And a user typed something. The bot looked for pre-matching keywords and it returned a pre-written response.

That works fine until a user asks a follow-up question. The context changes. And they ask the same question in slightly different words. And so what's changed is the shift from pattern matching to action or reasoning.

So modern AI agents don't match keywords. They understand intent. They retrieve information from multiple sources simultaneously, either through MCPs, through looking up unified indices, or even thinking about using solutions like a data cloud, for example. So by retrieving this information from multiple sources simultaneously, we're actually able to not just focus on what the user needs but far surpass their traditional experience by just connecting them with answers based on what they typed.

OK, so the outcome of that shift in practice is something quite amazing. OK, and the cost per interaction that we actually saw with one of our Search Unified customers, when they first started with an existing chatbot canned experience, every single time that an issue wasn't resolved, it was passed over to customer support. And the customer actually going through that was basically being towed behind a chariot and hitting every bump of the road before they were able to actually get help from support. And so our customer said, there has to be a better way.

We're already providing good insights and good engagement and good behavior for the customer. And so we started to see how that might be when they're able to have the Assistant support level. We need to change our delivery method in self-service, and that's what they did. And so they deployed our AI support agent, which is called Suva, that retrieves reasons and responds in context, handles multi-turn conversations, routes intelligently when the assistance is needed, and isn't just focused on providing effective self-service hey, I've reached the end of my chatbot session with this virtual assistant.

Why don't I have the chatbot post a question in the community? So it's changing the dynamics of how our customers are engaging in self-service completely. And again, I think the bottom line shift that we've seen is that for every successful resolution that's happening in the AI agent, it's pennies on the dollar compared to the cost of a fully burdened support agent solving the same question. And as a support practitioner, oftentimes when we see a support ticket, the first thing we ask ourselves is, well, have we seen this before?

And so the more known issues that we can help shift out to self-service so our customers can effectively engage and receive the answers there, that's what they want. Invariably, that's what we want. And the economics dictate that that is the right decision. Cool.

So we'll go ahead and move on to trusted knowledge. And I think the key thing when we talk about generative AI, especially early on, is that there are all these pipe dreams that we're going to take our content, we're going to throw it through generative AI, and we're going to get these robust, really great answers that we're going to satisfy our customers. And what a lot of clients, a lot of practitioners out there discovered is that, yes, you are getting very well-written answers, but they are confidently incorrect. And I'm sure we've all faced that, whether it's our own company and the answers that are being given to our customers in self-service, or when we actually go use the self-service products of a company that we actually have a relationship with.

So I think the key thing to bear in mind when we think about generative AI is that all outputs are derivative in nature. They have to come from somewhere. And so, you know, Shweta and I talk a lot about garbage in and garbage out, especially-- but except now, these agentic solutions are confidently delivering garbage to you and telling you that it's great. And that's going to cause some cognitive dissonance for you as a user when you know that what they're throwing you is actually not what you need at all.

So in terms of building trusted knowledge, what we're looking at is that having trusted knowledge takes care of some of the traditional KB flaws-- wrong or outdated answers, content duplication, lack of governance structure, KB governance overflow. And I think the key thing here, Shweta, if you want to-- I don't want to have you chime in because I want to save your voice. But when your AI draws from trusted, governed, reviewed knowledge, users start getting the consistent answers. And it's the same answer consistently delivered across customers and across channels.

And we want to make sure we have that consistency, because if we're getting a different answer every single time that we ask that chatbot solution, how do we know what's correct? We're just going to be reaching out to our customer success manager and say, hey, I tried to get an answer in self-service. This sucked. Now I need you to spend your valuable time helping me solve this issue or helping me learn how to use this product in a better way.

So we're on the next one. OK, so how does unified search come into play here? So I think one of the key things to think about is that when you have trusted knowledge, you still have to pull it into the RAG pipeline. Relevance is the currency of an efficient digital self-service engagement experience.

And I think the key thing to think about is that if your search mechanism is not delivering the right knowledge to your customer before generative AI, how do we expect it to all of a sudden figure itself out and start delivering relevant knowledge now? OK, and I think the key thing here is that you need one search layer that reaches across every knowledge source. So it'd be a single centralized cognitive search experience across all knowledge sources. And just on that point, what we're saying is that if you're able to bring all of your content sources together using MCPs and using calls to an index, you can take that knowledge wherever you need to deliver it to your customers.

So instead of relying on individual systems or saying, look, we've loaded up chat GPT with our documentation. Let's go ahead and try and create a window to that in our community site. This actually gives you granular control over these different knowledge repositories. And then you also have these GenAI-fueled analytics to surface content gaps and underperforming articles and emerging user needs.

And I think that's the main thing, is that traditionally search was seen as a black box. I mean, how many of us 10 years ago said, hey, we want the search on our website, the search in our community to work just like Google? I'm sure we all probably felt that, right? Now, the thing is, if you talk to folks at Google, they squabble internally about search results.

So it's not just working automatically, even internally at Google. So I think the key thing is that you want to be able to have a search solution that you can actually own, and you can bend it to your will for the needs of your organization and for your customers. And when we do that, there's going to be better content visibility. Content gaps get identified immediately.

And I think it's this self-monitoring ecosystem where we know that if we fail at connecting one customer with the right knowledge because it doesn't exist, how do we pick up on that content gap? How did we then curate knowledge or documentation or community assets where the next time a similar customer comes in with the same issue, we're meeting them exactly where they want to be instead of going into a long tail experience that's not going to be valuable for them? So I think this brings us directly to the Agentic Success Center. So again, thank you for your patience on me reading through this.

So when we have three pillars that work together, this is where some magical things happen. So we can create a space where members get real help through simple conversation. The system actually understands what they need, and we're understanding intent, not just keywords. So if we're asking a multipart question and get a response, we can have our Agentic solution or our Agentic Success Center.

We can have that surface the responses in each part of context. Now, I think this is where it also comes into play. Your LLM or your Agentic approach has to be honest. And by having a search index for your lookup, you can automatically understand that when a question is asked by a customer that you don't have content for, the last thing you want to do is have your solution deliver a confident hallucination.

That's the last thing that we want to do. So the good news is that search is basically the guardrails for that. And if we think about some of the improvements that we're seeing, some of the business outcomes that are listed on the slide here. So I think that some of the key things we've seen from our customers is a total night and day difference in response speed and accuracy.

The reduction costs, if we're able to help our customers solve their issues or answer their questions out where they are actually seeking help, then we're going to reduce our cost-per-support interaction. I think one thing that's not mentioned here as well is that when you have robust analytics, think about the value of the community. When I think about a community, I'm thinking of a place where there's peer-to-peer engagement, but we should be able to capture the voice of the customer at scale. What is happening?

How do we inform our content to be improved so we can eliminate friction in that customer journey? But then also, how do we give that feedback to our product and engineering teams? I've spent enough time in sales where I view my community as a sales differentiator. Because if we're having engagement in our community and our customers are happy, they're more likely to refer us.

They're more likely to be a reference for us. But also, too, if they go to a competitor and they see that that community is dead by comparison, that's going to give a window into what the post-sales experience looks like. So I think it's really, really critical for you to have these components in place. And then you can see when they're leveraged properly some of the key benefits at scale.

So we're going to close out with three questions that you can take back to your team. I know this week was all about getting your questions answered, coming up with new ones. So here's a few that you can take in your back pocket. So number one, where does your community lose customers before they get a resolution?

And what does that cost you per month? And we're not just thinking about support operations and cost per case. We're also thinking about where does trust get lost? Where is the friction where a customer says, you know what?

I might not want to try that again. So it's not just a cost in terms of monetary value, but it's also the currency of trust. Number two, is the knowledge that your AI draws from accurate, governed, and regularly reviewed? Or are you simply slapping or bolting a GPT solution on top of unmanaged content and hoping for the best?

I think this one is key because there's been many organizations that I've done analysis on and I figured out that when you don't have the insights in terms of how that knowledge is performing or when it starts sliding off in value for your customer, again, like Schweta said, the last thing-- or what you were going to say earlier, think about the 2 AM problem. What's worse, reaching a dead end or getting a chatbot that is consistently telling you the wrong answer but acting like it's correct? It's just going to create more friction in your experience and you're going to tell somebody about it. And then number three, now I know not everybody may be hooked in with their customer support colleagues, but think about every single time that a case comes in and your support team says, yeah, we already knew the answer.

We know the answer to that. It's a known issue. What would a reduction in support cost per interaction unlock for your business? And what is standing between you and that?

So I think the reduction in support cost is very interesting, but I think there's also a hidden benefit is that when we're helping our customers solve more issues on their own, that unlocks more time for us to do things that we've never been able to spend time on before. So every time when I talk with support leaders, they're saying it's not just a productivity gain, but there are things on my list that I've dreamed of being able to spend my time doing that I never get to because of my current environment. Sorry for my voice. We're going to go to mics on this.

Yeah, so what Shweta is trying to say is that communities were built on a fundamentally human idea. The people who've already solved the problem are willing to help people who have it yet. And today we want to ensure that when an answer exists, the system delivers it reliably, even at 2 AM, and handles follow up questions just as easily. And I think that's the key thing that we're seeing with our customers and the success that they're reporting is that, again, the capability to help customers be more awesome using your products and services without you having to hand hold them every step of the way, that is where the magic is happening.

And I think with communities, what I've always loved about communities is that it's a place to learn. And I always go back to the Winston Churchill quote who said, I love to learn. I hate being taught. So when we can allow our customers to engage and learn instead of having to reach out to somebody and be taught, it's just going to help them feel that they can do even more with your product and they can go deeper.

So that is our talk track for today. We are the Grazidi interactive booth that's past the coffee station if you'd like to stop by. Shweta and I will stay here for a few minutes. I might have to do some vocal interpretation for Shweta.

But any questions that you have, please feel free to let us know. And thank you so much for spending time with us. Hope you have a good rest of your day.