The Customer Success Playbook

Unlocking AI-Powered Customer Support: Exclusive Insights from TheLoops CTO, Ravi Bulusu

July 08, 2024 Kevin Metzger Season 2 Episode 22
Unlocking AI-Powered Customer Support: Exclusive Insights from TheLoops CTO, Ravi Bulusu
The Customer Success Playbook
More Info
The Customer Success Playbook
Unlocking AI-Powered Customer Support: Exclusive Insights from TheLoops CTO, Ravi Bulusu
Jul 08, 2024 Season 2 Episode 22
Kevin Metzger

Send us a Text Message.

In this episode of the Customer Success Playbook Podcast, hosts Roman Trebon and Kevin Metzger sit down with Ravi Bulusu, the co-founder and CTO of TheLoops. Ravi shares insights on how TheLoops, an AI-driven intelligence support operations platform, is revolutionizing customer support by providing real-time data integration and analytics. He explains the platform's ability to act as an AI layer on top of existing support systems, enhancing both agent efficiency and managerial oversight. The discussion delves into the practical applications of AI in support operations, the integration process, and the future of AI in customer support.

Detailed Analysis and Insights

Key Themes:

  1. AI Integration in Customer Support:
    • TheLoops enhances existing ticketing systems with AI capabilities, offering real-time data integration and analytics.
    • Ravi emphasizes the importance of having an AI layer to aid agents and managers in resolving complex customer issues by aggregating data from various sources such as CRM and escalation data.
  2. Data Utilization and Real-Time Insights:
    • The platform creates a customer engagement graph that provides real-time insights to agents and managers.
    • By unifying disparate data sources, TheLoops enables support agents to access necessary information quickly, improving resolution times and customer satisfaction.
  3. Enhanced Agent and Manager Roles:
    • For agents, TheLoops aggregates historical case data and knowledge articles, providing actionable insights and resolution steps within a single panel.
    • Managers benefit from predictive analytics that highlight at-risk customers and identify trends, enabling proactive management of support operations.
  4. Automation and Efficiency:
    • TheLoops facilitates the automation of repetitive tasks, allowing agents to focus on more complex issues.
    • By predicting escalations and providing actionable recommendations, the platform helps in optimizing support workflows and reducing operational inefficiencies.
  5. Onboarding and Implementation:
    • TheLoops offers a structured onboarding process, integrating with existing support systems and ensuring data accuracy.
    • Ravi outlines the typical implementation timeline, highlighting the quick turnaround from integration to operational use.
  6. Future of AI in Customer Support:
    • Ravi predicts that AI will play an increasingly significant role in automating support functions, aiming for up to 80% automation in the next five years.
    • The discussion touches on the evolving nature of AI and its potential to handle more complex support tasks, thus augmenting human agents' roles.

Business-Relevant Insights:

  • ROI and Scalability: TheLoops provides significant value by reducing the time agents spend on data gathering and allowing managers to focus on strategic decision-making. The platform's ROI calculator can help businesses understand the potential financial benefits.
  • Customer Experience: By providing timely and accurate support, TheLoops enhanc

Please Like, Comment, Share and Subscribe.

You can also find the CS Playbook Podcast:
YouTube - @CustomerSuccessPlaybookPodcast
Twitter - @CS_Playbook

You can find Kevin at:
Metzgerbusiness.com - Kevin's person web site
Kevin Metzger on Linked In.

You can find Roman at:
Roman Trebon on Linked In.

Show Notes Transcript

Send us a Text Message.

In this episode of the Customer Success Playbook Podcast, hosts Roman Trebon and Kevin Metzger sit down with Ravi Bulusu, the co-founder and CTO of TheLoops. Ravi shares insights on how TheLoops, an AI-driven intelligence support operations platform, is revolutionizing customer support by providing real-time data integration and analytics. He explains the platform's ability to act as an AI layer on top of existing support systems, enhancing both agent efficiency and managerial oversight. The discussion delves into the practical applications of AI in support operations, the integration process, and the future of AI in customer support.

Detailed Analysis and Insights

Key Themes:

  1. AI Integration in Customer Support:
    • TheLoops enhances existing ticketing systems with AI capabilities, offering real-time data integration and analytics.
    • Ravi emphasizes the importance of having an AI layer to aid agents and managers in resolving complex customer issues by aggregating data from various sources such as CRM and escalation data.
  2. Data Utilization and Real-Time Insights:
    • The platform creates a customer engagement graph that provides real-time insights to agents and managers.
    • By unifying disparate data sources, TheLoops enables support agents to access necessary information quickly, improving resolution times and customer satisfaction.
  3. Enhanced Agent and Manager Roles:
    • For agents, TheLoops aggregates historical case data and knowledge articles, providing actionable insights and resolution steps within a single panel.
    • Managers benefit from predictive analytics that highlight at-risk customers and identify trends, enabling proactive management of support operations.
  4. Automation and Efficiency:
    • TheLoops facilitates the automation of repetitive tasks, allowing agents to focus on more complex issues.
    • By predicting escalations and providing actionable recommendations, the platform helps in optimizing support workflows and reducing operational inefficiencies.
  5. Onboarding and Implementation:
    • TheLoops offers a structured onboarding process, integrating with existing support systems and ensuring data accuracy.
    • Ravi outlines the typical implementation timeline, highlighting the quick turnaround from integration to operational use.
  6. Future of AI in Customer Support:
    • Ravi predicts that AI will play an increasingly significant role in automating support functions, aiming for up to 80% automation in the next five years.
    • The discussion touches on the evolving nature of AI and its potential to handle more complex support tasks, thus augmenting human agents' roles.

Business-Relevant Insights:

  • ROI and Scalability: TheLoops provides significant value by reducing the time agents spend on data gathering and allowing managers to focus on strategic decision-making. The platform's ROI calculator can help businesses understand the potential financial benefits.
  • Customer Experience: By providing timely and accurate support, TheLoops enhanc

Please Like, Comment, Share and Subscribe.

You can also find the CS Playbook Podcast:
YouTube - @CustomerSuccessPlaybookPodcast
Twitter - @CS_Playbook

You can find Kevin at:
Metzgerbusiness.com - Kevin's person web site
Kevin Metzger on Linked In.

You can find Roman at:
Roman Trebon on Linked In.

Roman Trebon:

Welcome to the Customer Success Playbook Podcast, where we bring you the latest insights and expertise from top leaders in the field of customer success. I'm your host, Roman Trebon, and with me as always, my co host, Kevin Metzger. Kev, we're getting to the end of June. It's getting hot here in Atlanta.

Kevin Metzger:

Yeah, first day of summer. Definitely a hot, hot day. I'm hearing, I heard somewhere on the news, we're expecting like six weeks of continuous heat.

Roman Trebon:

Yeah, the Groundhog's going to be its whole if it's six more weeks of 90, 95 degree weather here in Atlanta. So okay, we got it. We got a great episode of lined up. You were out in at the Paul's conference and you came across some software called the loops which is an AI driven intelligence support operations platform. And you said, Hey, we got to get, the guy behind this on the show. You want to kind of introduce the software and who we have on today?

Kevin Metzger:

Yeah. So we've got Ravi,

Built-in Microphone-1:

Balusu.

Kevin Metzger:

who's the co founder and CTO of The Loops. Ravi, welcome to the show. Can you tell us a little bit more about your background and how you got started with The Loops?

Ravi Bulusu:

Absolutely. Nice meeting all of you and very happy to be here. Yeah, my name is Robbie Bulusu. I'm the CTO co founder of the loops. Prior to this, I've been in this industry for about 25 years. I've done more than five startups. Three of them have successful exits. The last one's still going on. And this is something that we started four years ago. Fundamentally realizing that there's a lot of A. I. M. L. That's missing within the support stack, right? And we wanted to be the layer on top of your ticketing system that provides you with that A. I. Layer. Now, it doesn't matter whether it's for agents with the managers, whether it's for your analytics, we want to be that A. I. Layer on top of your support platforms. And that's the goal. And that's that's what we started the system for. I have about 27 patents in and around this technology. And that's Brief introduction of myself.

Kevin Metzger:

Thanks Ravi so let's go ahead and get into the loops a little bit. how do you get the data into the systems? And then how are you using the data? What kind of information am I able to get out of the platform?

Ravi Bulusu:

Yeah, absolutely. So if you look at any support system right now, they get you know, salesforce service cloud,

Roman Trebon:

you know,

Ravi Bulusu:

Zendesk, Freshdesk, most of these platforms, they're fundamentally designed to solve cases, right? And they're designed for agents to go through a list of cases, in fact, to the customer, respond back to cases and things like that, right? What they're not designed for is to be this data, the data layer that basically gets Your CRM data, your escalation data, your usage data, And all these different data are really important for agents to resolve the problem because it's not always going to be, how can I do something? And so I did something in your application that resulted in this particular problem. Immediately. What are you going to look for? You're going to look for, or is there an application crash in the system? Right? So. Obviously, every support agent has like 10 tabs open within the console to basically look for Splunk queries, look for, common cases that happened in the past, look for JITA escalations that are happening, right? That's the problem that we look for whenever we solve people or agents resolve problems. And that was the idea behind saying that in order to solve customer issues, you need data. And most of these support systems, the agent monitoring systems are not built up to be integration platforms. They're built up to be applications, right? We always believe that there are two fundamental stacks. One is a system of records and this sort of intelligence, right? There are two different architectures. One is an aggregate system. The other one is an you know, transactional system, right? So we are the aggregated, we are the intelligence layer on top of that's what we wanted to build. And that's where the loops came about to be.

Roman Trebon:

So, Robbie, with that analytics insight layer, is it just giving you information on what's happening and where there's areas of opportunity within the support structure process? Is it also sort of directing action? Like, is it going down to the agent and telling them to do something different, something So where does it kind of fit in and how does it play with the different rules that people would touch it?

Ravi Bulusu:

Absolutely. So if you look at support platforms right now, support organizations, there are two fundamental roles within it. One is an agent, agent solving the cases, right? Then you have managers who are looking at an overall picture of where are the cases coming from, right? Why is there a case uptake or this particular feature to give more cases than the other feature, right? Why some you know location having more cases than the other. These are the things that on an aggregate level, people managers observe, right? We cater for both these roles, right? From an agent perspective, if you look at every agent, what they do on a, you know, case by case basis, as soon as the case comes in, they're basically saying, Oh, is it a repeat issue? Is other customers facing this particular problem? Have somebody else resolve this particular problem for me before, right? All these things are something that's already there within your case system. What loops does, and again, most of the case intelligence systems do is to basically crunch all that information for you, provide the agent with necessary details about, okay, this is when it happened before this is how agents have resolved it in the past. And by the way, I always also found a knowledge article that you can share with the customer, all within. One panel. All the agent does is click, click. This one's back to the customer, right? What does it do? It helps the agent focus on the customer's problem, not gathering data, right? And the agent is focused on addressing the customer's problem and being that empathetic, human touch, right? Rather than being this data operator while actually trying to resolve the particular problem, right? That's what they do for a customer.

Roman Trebon:

Yeah,

Ravi Bulusu:

on a manager level, you're trying to look for that next escalation, right? Which is the next customer that's going to blow up, right? And this means a lot of data crunching, trying to look for hundreds of customers, all these cases, trying to find where sentiment is wrong, and things like that. And loops tells you that these are the top 10 customers at risk, right? And these are all data crunching operations, machine learning on top of it gives you that kind of an insight. And all these are actionable, right? You know this is an escalation. How can you resolve it? Right? Oh, by the way, assign it to a different agent or respond back to the customer. Get on a call with the customer, right? So that the managers can take it, to the next step as well.

Kevin Metzger:

So how are you being able to take all the data from various systems and basically bring it together to make sure you're looking at it in a way that makes sense. In other words, I've got CRM information from Salesforce, and then I've got issues coming in through service now, and I've got maybe some slack data coming in from my teams. How are you able to look at it and say, hey, I know these data is related to each other to be able to surface that to the agent.

Ravi Bulusu:

This is something that support organizations used to do before loops as well, right? And the way you should do this is because you have disparate systems, right? They used to put all this into data cloud. You know, data warehouse like snowflake or one of these things and then try to join these things and try to construct some kind of a report out of these things. The problem with this is most of these things are not that real time, right? And then the second problem was how does the agent query the system, right? When he's trying to look for a case or a customer, right? That's why we said that. Those systems are good for offline analytics, right? They're not good for real time insights, right? And that's a fundamental difference between our architecture versus what you know, Snowflake or these data clouds data warehouses can do, right? What we do is create a customer graph, right? So any case that comes in, right? We exactly know who the customer is. We exactly know what the customer did, how many open cases the customer has, How many escalations that the customer has. We know what are the related cases with this particular case, right? What are the, are there any of those escalations? So what we created is what we call it as a customer interaction graph, customer engagement graph, right? That graph lets us not only Join things or stitch things together very fast as the data is coming in, but also at a very real time, query the graph in order to provide those insights at real time, right, to agents as and when they need it. And it can happen in Slack, it can happen in Jira, it can happen in, Zendesk, right? Each of these three roles are very different, right? Zendesk is where agents are looking at, right? Jira is where engineering systems are looking at, and Slack is where experts are, right? But all three of them get a completely different view of the graph, right? Very contextual to their own insights that they really want, but it's using the same underlying customer engagement graph to provide all these insights.

Roman Trebon:

So, Ravi, I work with a lot of customers and customer support and they're all saying the same thing to us right now. Get, get the, get it out, right? Like I want to chat bot. I want to call the phone channel and I want the intelligent agent to help self service before it even gets to someone. How is the loops helping to sort of funnel those technologies and helping even avoid a case coming in or better educating the customer before they have to hit an actual agent?

Ravi Bulusu:

absolutely. So if you look at the journey of AI within customer support, right? So you have. Front end chatbots, which are particularly deflection agents, right? They help you not to create a ticket, right? They help you with deflection, right? Now, I've done this in the past. My previous company was something that we did chatbots for. But if you look at it, I mean, in most of the companies, we could not achieve more than 17 to 18 percent 17, 20 percent of deflections. Fundamentally, because if you don't have knowledge, right, you don't have any deflections, right? And knowledge is, again, you can do a complete podcast on that, like how, you know, inefficient knowledge systems are, right? So what, what happened is that all the simpler cases that used to come to agents are now handled by chatbots. Those are the 20 percent of the cases. Now, what agents are ending up with is very highly complicated cases, right? Which are not like a one click Transcribed thing that you can do. this is where the agent needs help right now, L1 agents who are initially equipped to solve the chatbot related problems are now solving higher complicated problems, Which means that they need to access data. And what we found out that they're not trained to access all these different systems. They're not trained to go into a Splunk query or a Datadog search or a look up on a CRM system. That's where loop helps you identify all that data and then bring you to the right panel saying, okay, you don't have to worry about all those things. All that information is right here, right? that's where it helps. And again, I think chatbots have a place in the organization, but they're not yet there to solve 100 percent of your problems. Yeah, that's the truth.

Kevin Metzger:

when you're looking at the AI agents here, and you've got your AI agent copilot, how are you ensuring that you're delivering accurate information? As with through the copilot to the agents and making sure that we're you're driving accurate information.

Ravi Bulusu:

Yeah, there is 2 aspects of it. If you look at copilot and fundamentally, when we say copilot is a very new concept, right? And even Microsoft, everybody trying to educate people of what a copilot really means and what it should be doing. I've talked to many of the industry. You know, people who are managing customer support at a very high level and they're saying, What is your. what a copilot should do, right? And fundamentally, they talk about three things, right? The whole copilot should help the agent respond to customers, right? It should help the agent understand the resolution journey, right? What is the next step that they need to do within the whole resolution journey of the process? The third thing it should basically do is to help you with The cleanup of your ticketing system. Like, okay, is there a feature request? How do you share information from your support system to other departments, right? There's a documentation request that's coming in. This could become a feature request within product, right? It should help them do those aspects of it, right? Those are the three aspects of copilot right now. Copilots are as effective as your knowledge within the system, right? And every system right now, including lubes, treats two things as knowledge. One is your KB articles, right? The second is your resolutions that your agents have provided in the past. And both of these two things, what we've figured out is, yes, you have a thousand knowledge articles, but are they even relevant to what people are asking for right now? Maybe you have to update your knowledge document. And how do you even know that, right? So the first thing, when we go to any customer to deploy, right, the first thing we tell you is like, let's look at your data, right? And let's, let's kind of figure out historically, right. from your past one year of cases, how many of your cases can even be resolved with your current knowledge that you have, right? Right. And trust me, I mean, we have not seen more than 30%. Right. 30 percent is the upper limit that we've seen. Which means that 70 percent of your cases cannot even be resolved with the kind of knowledge or information that you have, both in your cases, as well as KB articles. Many organizations don't have a way to capture how the agent resolved a particular problem. If you don't capture how the agent resolved the problem, Your cases are pretty much useless, right? And there's no information or intelligence that you're gathering with every ticket closure. How do you help that auto QA, right? People always think that auto QA is completely a different department and it should be basically something that worries about CSAT. Actually, no auto QA helps you build your intelligence within your support, support data, right? And the way we do that, we recommend the customer is saying that, okay, why don't you have an auto QA? Have a rubric within your system. rubric is something that we evaluate at the end of a ticket closure saying did the agent provide the resolution steps. Did the agent provide proper resolution to the customer, right? And these things could be something that an intelligent system can verify and then tell the agent. When you're doing the coaching with your agent saying it look at 30 percent of the cases, you've not even provided any resolution notes, right? But you're not helping the system be intelligent, right? But that training that coaching right will will help them put those resolution notes will help your system become more effective. and all this is something that an intelligent system to do. Right. And that's what loops us. First thing that we do is we basically tell you, yes, you have a hundred thousand cases, but only 10 percent of them have resolutions that we can use. Yes. You have a thousand knowledge articles, but only 30 percent of them are even useful. Right. But here's the gap and here's how you can fill them. Right. And that's the approach that you have to take with them. That's what makes your co pilot go from, Oh, I can help you with 20 percent of the cases too. I can help you with 80 percent of the cases. Right. And that's a journey, right? And all organizations are starting this journey very small right now. Fundamentally what you know, one of my you know, gurus that I believe he always said that your system, when it comes into the, into the organization can be 15 percent intelligent, right? But at the day, a hundred, if it is not 30 percent or 40 percent intelligent. And It's not going to improve. And that's the that's how we identify a rule based system versus an AI system. Right.

Roman Trebon:

Yeah, no, I love that. It's basically making the resolution repeatable and reproducible through having like you said like so many you're not even getting there's not even a support knowledge base or whatever it may be and so ronnie, what is it like from an onboarding experience? What does that journey look like? You know, I'm sure there's probably a lot of use cases. So you know, how do you get clients to start small, get value and then build? And then what does that kind of walk us through that?

Ravi Bulusu:

Absolutely. So loops has fundamentally three product lines, right? One is you have your co pilot, right? Then you have predictive operation center. Then you have your auto QA, right? We have different onboardings for each of them. You have a loops platform that I think any AI system should basically be integrating with your. Support systems, integrating your CRM systems, integrating with your escalation management systems. Once you do that, that usually takes us like two weeks, And then there's a data validation part of it, which is absolutely important for any AI system to make sure that the data that you're basing your intelligence on is clean. We spent two weeks on that. Once you do that, then we onboard all these different use cases. Copilot takes about, four weeks to onboard. Then you have auto create, which is a little bit faster. If you have a rubric that's already in place. That's another two weeks. You have backlog management, escalation management, CSAT prediction topic analysis, trend analysis, intelligent routing. automating that operating center, can take any time between two to four weeks to onboard most of these use cases. So we give customers anytime between three to six weeks is where you can onboard multiple use cases with them.

Roman Trebon:

Gotcha.

Kevin Metzger:

are you also doing anything within the loops to provide feedback to, the product teams and product management? So you've got all this great data that you've collected from a support perspective. It seems like there's a huge amount, a huge opportunity as well to kind of go towards bringing this information back to product management, bringing it into the CS organization to help understand where your customer is from, like you said, with the CSAT. But do you have paths for that as well?

Ravi Bulusu:

Support is a goldmine of data where customers come to you. This is the only organization where customers are more interested in reaching to you rather than other way around, right? Obviously, they leave a lot of information and unfortunately or fortunately, agents are very focused on resolving the problem and ignoring the noise, but the noise has goldmine of data in it, right? It has feature requests that can help the product team kind of build the next set of features. It might have, bugs or performance issues that engineering team should look for to prioritize these particular problems. how do you generate a document out of this thing so that somebody else can leverage that document? And all these things are lost, In traditional support environment, So what we have realized is that, you can, Offset this to the agent to basically say, okay, why don't you create a knowledge document out of it? I mean you resolve the problem, but they're not incentivized for that, They are incentivized to solve more cases, They're not incentivized to provide that shared data to other departments within your organization, but the vp of custom support is They are absolutely responsible for collecting this data, deriving insights out of this data, providing this nuggets of information to all these three departments. Right now, how do you make it easier? Right? So that's why we came up with this whole concept of topics, right? And topics for us are your product features, your product services, all the different areas of product where it. You have cases coming in, right? And then you have this intelligence that you can divide, right? With generative AI, it make it even more easier to basically look at all this data and then say that, okay, I need to generate a couple of documentation requests from the, authentication topic or reporting module, right? And once you do that, and then we give you a very easy way to go to navigate to this page, and then, click and generate this documents. But KCS itself is its own process engine, right? So when we create documents or when we generate documents, we always write it into cases system, but market as generated by the loops. And in a draft mode. So somebody can actually go and review them before publishing to an Excel, agency. And we integrate with most of the KCS processes that companies have right now, And even product like feature requests and stuff like that. We do flag whenever we see feature requests that customers are asking for, and we write that into product boards as well. So we have integrations into Asana. We have integrations into monday. com as well as You know, obviously, so we can automate this whole process for you,

Roman Trebon:

Yeah, which is great because right? Like if you're a support organization, you're not driving the issues like people are calling. They're calling because it's a product issue. Some issue that got driven that your way. And then so it sounds like the loops is not only helping you be more efficient when you have tickets in cases, right? But then hopefully driving some intel And driving some of those tickets out in the future with your product team and others, across the organization So I love Robbie on the site our audience if you haven't gone to the site you got to check out the loops. I love your roi calculator on there, which is awesome It has slide bars. It's super easy to use. So a good sense of how the value this can provide On that front, I'm sure this works amazingly. The more scale you have, the more tickets, the more, you know, but what was the, is there a minimum? What's like the sweet spot before you start to see some of the value from a loop solution? I think I saw was it. 25. I remember that right, Ravi? What's the minimum number typically to get this off the ground?

Ravi Bulusu:

So different use cases have different thresholds. Like we say that, you know, even if you have like five agents and a QA manager, loops can provide you value But what we say is that in order for the AI to learn, you need at least like 20, 000 cases, at least in the last three months, right? That gives you enough variability, enough repeatability that we can see that the models can learn. If you have less than that, then probably you're not ready for an AI solution. So that's something that you have to keep in mind

Kevin Metzger:

so Robbie, along those lines, you're talking about the amount of data you need what does the technology stack for loops look like? I realize it's a SAS kind of offering, but what's underneath the covers?

Ravi Bulusu:

So I think from there's two fundamental stacks here, right? One is the data processing stack itself, right? So we have our own connector framework, right? And a lot of companies actually don't build a connector framework right now. They outsource it or they OEM some of the other tools. We made a conscious decision to build one because we knew that it has to be bidirectional. It's support is not a universal thing. If you have insight and you can't write it back into a system record, if you can't have the you know, our manager take action on it, it's useless, right? So we went and built the connector framework. That's, that's completely proprietary. We have about 50 plus connectors right now spawning across 12 or 13 categories of data. And then, you know, everything that comes into, we have a SaaS platform that we host right now, both in EU and us on both GCP and AWS, because, you know, some customers are customers have their own preference and we want to make sure that we provide them with whatever they want. the stack is pretty straightforward. You have Kafka and then you have, you know, database and data. Other type of data stores that we have in the system. We have a graph database. You have a vector database. We have obviously SQL database in the system. All the tenants are partitioned. There's no commonality that there's a common database that we use across tenants and stuff like that. On the AI stack, we have three fundamental architectures within AI, One is what we call a generative or conversational reasoning framework. This is what we use our own custom LLM model, right? To do conversational reasoning. Then we have a predictive operations platform, which is our predictive operations library, doing escalation prediction and a lot of these Traditional ML models, That we use. And then we have what we call a conversational engine, where you don't have to create dashboards. You can just come and talk to the system and then we'll create those dashboards for you. Right now we're using GPT 4. It's a generative model that we work with GPT 4 at that site.

Kevin Metzger:

Cool. And you said you have your own AI, your own LLM for looking at support type tickets? Yes. Okay.

Ravi Bulusu:

Absolutely. So we are specifically trained to look for signals within support that could be, signs of frustration, signs of a feature request coming in, signs of a documentation request that that's not been able to answer, right? Again, can you use GPT 4 for it? Yes, you can write a prompt to do GPT 4, but the problem with GPT 4 is it takes a long time. And depending on the prompt and the amount of data there is no guarantee that it will finish in under a second, right? For us, it's very important to be real time, right? Because, you know, as soon as the case comes in, we take a lot of real time decisions, including who to route the case to based on language, based on availability, based on a lot of different factors around the topic So we have to be very, very fast. And that's one of the primary reasons why we've trained our own LLM. We host it within our environment, right? It's also secure that your data is not going to GPT 4 and things like that. So we give them all the security governance on top of it as well. But this model is not only secure and fast but it's also Traditionally for very support specific insights that we derive out of it, out of the

Roman Trebon:

So, so Ravi, you're obviously neck deep in AI, you know, I'm fascinated by this. So just from all the stuff you have in place with AI, Where do you see this in five years? Like, where do you see, especially in customer support and operation, like where, where is AI going to take us? Cause it seems like it's, it's going so fast. Just curious. You're on the front lines of it. Where do you kind of see the technology?

Ravi Bulusu:

Yeah. I mean, it's been a year since you know GPT came out and, most of us in this space kind of knew. And we're following this for a long time, and we knew that something like this was possible, but GPT kind of made it generally available for everybody to use, right? I'm more surprised by the amount of adoption that GPT had in the last one year and the amount of industries that are leveraging it. Some of them are leveraging it beyond its means, but, it doesn't matter. Once the pendulum sinks, it always sinks to the other end, right? It'll settle down at some point, right? There is a real chance that a lot of what support does can be automated, can be done by support agents in the future, right?

Roman Trebon:

Yeah.

Ravi Bulusu:

the piece of the puzzle that you really need to solve is the data, right? and people always ignore that fact, the more the case becomes complex, the more data gets involved. if you don't have the data, the right cleanliness, the right format, the right way to access it efficiently that creates a problem. I don't believe that a traditional data warehouse is the answer for this, Or a traditional graph database is the answer for this. It's a combination of these two, And that's a problem that needs to be solved and how AI can access data really efficiently in order to make those decisions faster. And I'm sure that people will come up with those technologies. And I think that automated support to almost like an 80 to 90 percent is really possible in the next five years. And I'm sure there's a lot of companies working on it at this point in time.

Roman Trebon:

Like you said, beginning of the show, 20 percent with chatbots, right? Imagine that getting up to 80%, right? When they can handle those complex issues that today they. You have to go to an agent, right?

Ravi Bulusu:

I think it's always going to be augmented, right? Right now, let's say an agent when they even look at complex cases, they're, they're doing 80 percent of the work and maybe intelligence or even co pilots are giving them that feedback about 20, 30 percent feedback on how to get there, right? That ratio is going to change. That ratio is going to be, the AI is going to be do 80 percent of the work, but it would still need that 20 percent of human input. And it's going to become very efficient to basically say, when do I reach to an agent or when do I reach to a human for help? Decision making. right. That is where AI should be. It should always interface with the human for real decision making, right? Yeah. But automate most of the things that are data related. Right? And that's where it's gonna happen in the next five years.

Roman Trebon:

It's awesome. All right. Can we ready to get to these? These hard questions here?

Kevin Metzger:

Yeah, I think it's time to hit the real hard stuff

Roman Trebon:

All right. We'll go back and forth here. Ravi Early bird or night owl,

Ravi Bulusu:

Book And then I would say early book. Is it's down so right now as age is catching up, right?

Roman Trebon:

I love it.

Kevin Metzger:

Yeah That's me too on As I got older i've gotten early. I Used to be an item. what's your favorite book?

Ravi Bulusu:

Favorite book was most of you know, I've not read books for a long time and it's been mostly knowledge articles from, from GPD. Yeah, I'll skip that question.

Kevin Metzger:

Let's go with the most influential knowledge article that you've recently read.

Ravi Bulusu:

The most influential one was, there was an article on how people created GPT the transformation from an AI model all the way to you know, the generative AI model and how the journey of, of that thing has been. And, and again, it's, it's been interesting for me because I've been observing this transition happen. I've been using technologies as it was evolving, Right. The tape from intent detection, 10 years back to you know File a CM models, you know, transformers and everything else. I've been using this and phases and when GPT came out you know, observing the whole transition and my journey along with it was really, really interesting.

Roman Trebon:

Place you'd want to travel that you've never been.

Ravi Bulusu:

Alaska.

Roman Trebon:

Oh,

Ravi Bulusu:

but yeah, I would really

Kevin Metzger:

like to,

Roman Trebon:

That'd be awesome.

Kevin Metzger:

All right. I think this is the last one. What's something on your bucket list that you'd like to achieve next

Ravi Bulusu:

I'm really afraid of heights, but I've promised that we exit and we'll do a skydiving. And that's a nightmare, but I would like to consult.

Roman Trebon:

Yeah, Robbie, you're a lunatic, man. I don't know how you do it. I'm super scared of heights. Like, man, let me know. I got to see a video or something. All right. Yeah. Robby, where can our audience find more about the loops? You find out more information?

Ravi Bulusu:

Well, the loops. io is where everything is. You know, reach out to us. There's obviously information about security. There's information about what's coming next. There's always, you know, emails and forms that you can fill us. There's a form that you can fill to watch the videos as well as, get a live demo from us. Very excited to show the product and what we've been doing so far.

Roman Trebon:

Yeah, it's awesome. Definitely check out the site, Ravi thanks so much for joining us. And everyone, thanks for listening. Please follow us on LinkedIn at Roman Trebon at Kevin Metzger. Check out the customer success playbook podcast page on LinkedIn. Like our show, subscribe, tell your friends about it, interact. Let us know what other guests and topics you'd like us to have on. And as always, Kevin,

Kevin Metzger:

keep on playing.