What if you could double your customer support team's effectiveness without hiring a single new employee? Sounds too good to be true, right? But that's exactly what businesses are achieving with Salesforce's Agentforce platform. Every day, support teams are drowning in tickets, burning out their best agents, and watching their customer satisfaction scores take a nosedive. The cost of scaling support teams traditionally is skyrocketing, while customer expectations for instant, 24/7 support keep climbing.

But there's a solution that's transforming support teams across industries. Today, we're diving deep into real-world success stories and practical implementations of AI-powered support that's cutting response times in half and revolutionizing how teams work.

I'm thrilled to welcome Aaron Godby, Founder and CEO of Green Irony, who's been at the forefront of this transformation. Aaron and his team have implemented some of the first Agentforce solutions with Salesforce, including a groundbreaking project at UNC Charlotte that's completely revolutionized their student support system. As a seasoned technology executive and AI implementation specialist, Aaron has helped countless organizations turn the promise of AI into actual, measurable results. He's the perfect guide to help us understand how to achieve these incredible efficiency gains.

The AI Hat Podcast host Mike Allton asked Aaron Godby about:

✨ Strategic Implementation Matters → Success with AI support tools requires careful planning and clear success metrics.

✨ Human-AI Collaboration → AI doesn't replace support teams; it empowers them to focus on high-value interactions.

✨ Measurable ROI → Companies can achieve verified 50% time savings while improving customer satisfaction.

Learn more about Aaron Godby

• Connect with Aaron Godby on LinkedIn

Resources & Brands mentioned in this episode

• Green Irony
• Salesforce Agentforce
• AI as Your Co-Host: Unleashing NotebookLM’s Potential in Content Marketing with Mike Allton
• AI Primer: A Comprehensive Guide
• Explore past episodes of the The AI Hat Podcast podcast

Full Transcript

(lightly edited)

Cut Your Support Time in Half: Real Results with Salesforce's Agentforce with Aaron Godby

Cut Your Support Time in Half: Real Results with Salesforce's Agentforce with Aaron Godby

[00:00:00] Aaron Godby: So for me, AI could completely eliminate that thing from the equation. And all of a sudden, support teams are not dealing with these angry customers that have had to spend 15 to 20 minutes navigating the phone system to get to them and then repeating everything they just did. The customer's a lot happier when they get to them because maybe they've spent less than a minute on the phone system because it knows exactly what they're after.

[00:00:24] It's grounded in generative AI can interpret why they're calling to begin with. And then it's handing off to the human agent in a very seamless fashion so that they can go address the problem. So I I, I think it's gonna evolve to the point where the employee experience is also much, much, much better, and they're getting fulfillment out of solving real problems instead of telling somebody information that's available on the website.

[00:00:50] That's, that's the vast majority of tickets. It's just somebody could not find information that's already available. So they're calling and then they're, because of the IVR system, I just, I just mentioned in that paradigm, they're getting more and more and more frustrated, and by the time they're on the phone with the human, they're a nine outta 10 on the annoyance scale.

[00:01:09] So I, I think all of that can go away and make the customer experience better, along with the support agent experience in tandem. So everybody wins.

[00:01:18] Mike Allton: Mike's got The Hat, the AI kind. Unlocking power in business, you'll find. Cut through the hype, explore tech. Every week, new ideas to check.

[00:01:42] Allton leads us on a digital quest. AI secrets, he'll always invest. Tune in tight, don't miss a beat. Knowledge, groove, keep you on your feet. Listen up! To the AI flow. Business growth, here we go. Cuttin' edge insights, that's the plan. With The AI Hat you'll understand.

[00:02:27] Greetings program. Welcome back to The AI Hat Podcast, where I selfishly use this time to pick the brains of experts at harnessing the power of AI across all areas of business. From streamlining operations and boosting productivity to gaining a competitive edge and driving explosive growth. Oh, you get to learn too.

[00:02:43] Subscribe to be shown how to prepare yourself and your brand for this AI revolution and come out ahead. So what if you could double your customer support team's effectiveness without hiring a single new employee? Sounds too good to be true, right? But that's exactly what businesses are achieving. With Salesforce's agent force platform, every day support teams are drowning in tickets, burning out their best agents, and watching their customer satisfaction scores Take a nose dive.

[00:03:11] The cost of scaling support teams traditionally is skyrocketing, while customer expectations for instant 24 7 support keep climbing. But there's a solution that's transforming support teams across industries, and today we're diving deep into real world success stories and practical implementations of AI powered support that's cutting response times in half. And revolutionizing how teams work. I am thrilled to welcome Aaron Godby, founder and CEO of Green Irony, who's been at the forefront of this transformation. Aaron and his team have implemented some of the first agent force solutions with Salesforce, including a groundbreaking project at UNC Charlotte that's completely revolutionized their student support system.

[00:03:51] As a seasoned technology executive and AI implementation specialist, Aaron has helped countless organizations turn the promise of AI into actionable, measurable results. He's the perfect guide to help us understand how to achieve these incredibly efficient gains. Hey, Aaron, welcome to the show.

[00:04:07] Aaron Godby: Hey, Mike.

[00:04:07] Thanks for having me.

[00:04:09] Mike Allton: My pleasure. I, I can't wait to dive into this with you. This is gonna be so much fun. But before we dive into those solutions, can help us understand the current state of customer support. What are some of the challenges that companies are facing that make some of those traditional scaling methods that we might be familiar with?

[00:04:24] Unsustainable.

[00:04:27] Aaron Godby: Sure. Yeah. So let me walk you through it a little bit and what we've seen, and I'm thinking of a, a few of our customers in particular but this is just how it works broadly with customer support right now. You'll have a call center let's call it, 500 to 2000 agents or so in this call center.

[00:04:45] And they're hiring like crazy and training like crazy because the turnover of these resources is, is significantly high. If these operators are dealing with technology systems that, in most cases are not very easy to use and their tasks are pretty complex because people can call these call centers needing an answer for just about anything.

[00:05:05] So think of an airline where there's a, maybe a weather event in Atlanta, a weather event in Atlanta is gonna impact all sorts of things because that's Delta's hub. That's a big time international airport. A big hub on the East coast of the United States. You name it. So. It's gonna result in a lot of spikes for support needs for an airline.

[00:05:25] So they're gonna be overwhelmed with all sorts of different requests and these, let's call it 2000 agents, are gonna have the phones ringing nonstop with all these different types of requests, and they're gonna be interfacing with these support systems that can be fairly cumbersome to use. So if I, if I take a step back and I look at this as a business leader and I do some quick napkin math.

[00:05:50] The cost on that is pretty significant. It's, if you look at 500, a thousand agents, it's, it's easily over a hundred million in labor costs. When you start considering management, you start considering, and then you bring in other operational costs, like, like office space, like equipment training, costs management.

[00:06:09] It, it starts to really add up. And what you're left with is, pretty significant investment with not great results. A couple of 'em that I'm thinking of actually have net promoter scores below negative 50. And it's not because they're not trying, it's not because they're not investing in this.

[00:06:27] It's just a really challenging problem to solve because of the, the resources, because of the technology, the high training costs, you name it, it's just a difficult, difficult business problem to solve. But with AI, we now have a tool in our tool belt that makes things a lot easier.

[00:06:47] Mike Allton: Got it. My first job outta college was working for a Dana Corporation, 80,000 plus company at the time.

[00:06:54] I think they've shrunk a little bit since then, but they were a tier one supplier to automotive companies, and I was in their global IT group in the internal IT support system. So I was handling those internal calls. Now it was only 80,000 employees, which sounds like a lot. It's not the kind of volume that like a, a United or, or some of their airline might get when it comes to customer service.

[00:07:13] But still it was a lot of volume. We had to have different areas. We probably had 30 people on the team. Would you say a lot of the same problems and struggles that you're talking about just now would apply just the same to a team that size, not 500 to 2000 like you mentioned before?

[00:07:28] Aaron Godby: Absolutely. So if I look at University of North Carolina, Charlotte, their IT support teams, roughly, let's call it 30 people or so.

[00:07:36] Yeah, even then they're it's, it's difficult to support their student body of 30,000 people even on the IT side because they're calling all hours of the day. So you're not necessarily restricted to eight to five business hours. Somebody could be locked outta their system at 11 o'clock at night and they need to get in because they need to access and submit a homework assignment, for example.

[00:07:58] And in their case, when we, when we started working with them, we realized that 20% of their tickets. And 20% of their time was just going to system access issues. So I, I've forgotten my password. Nope. Can't use my two-factor authentication, what have you. And that resulted in, a lot of inefficiency there because that's something that could be fairly easily solved with ai.

[00:08:20] Mike Allton: Yeah, just yesterday I was delivering a keynote on AI for business, and that was one of the questions that came up. How did I personally feel about AI agents answering the calls? And I said, I love it because of this exact reason. When I was at Dana Corporation, I was supporting all of our ERP systems, Baan, Oracle, SAP, and so on.

[00:08:38] I'm sure 80% of the calls that I got were, Hey Mike, I forgot my password. Can you get me back into Baan? Which is fine, but that wasn't like fulfilling work for me as an individual. That wasn't the best use of my time when there's so many other creative things I I could have been doing. So it's a fantastic point and I love that you brought up UNC 'cause I definitely want to dig into this.

[00:09:00] Walk us through what that implementation looked like. What were some of the specific problems that they were facing and, and what results have they seen since working with you?

[00:09:09] Aaron Godby: Sure. And Mike, do you mind if I take a quick step back just into the higher ed landscape as it exists today? Yeah, there's, okay.

[00:09:17] There's some fairly significant challenges just with budgeting. The policies that have come out in the last couple of months have significantly restricted the amount of overhead that higher ed institutions can have. So there's, so in addition to, what we were seeing last year when we first started working with Charlotte, the problems amplified because of the new policies in place, restricting the amount of overhead that can be used in a grant, and also restricting the, the number of grants that are given out.

[00:09:45] So there's a bit of an operational expense crunch happening right now in higher ed across the United States. And this is an area I'm really passionate about. My sister's actually a professor over at University of North Carolina Chapel Hill. She does some great work in infectious diseases. So my thought is how, how do we help them be success as successful as they are today with a more limited budget or perhaps even more successful at driving student outcomes, at driving research outcomes?

[00:10:11] Everything that's important to the university. Which has nothing to do with helping people, unlock systems when they forgot their passwords, right? That's, that's just something that comes along with it. That is a necessary evil in supporting everything. So how do we take those things like resetting passwords and make it a lot easier so that they can focus on the big problems?

[00:10:32] Because as a higher ed leader, you've got all these really smart people in these positions. And if they had more time, they could solve bigger problems and more of them for the institution. But in a lot of cases, they're stuck doing these mundane tasks with really high percentages of their time.

[00:10:49] Kind of like you mentioned in, in your role. So our thought is we can use agents to free up their time and then this is gonna allow them to focus on these really big problems. So at at Charlotte specifically it really started with that reset password scenario, like. How do we, how do we pick this low hanging fruit and show some immediate value by eliminating 20% of these tickets, and also prove things out with ai because their executive group there is very forward thinking, very into ai.

[00:11:20] They were looking for what is the first way that we can really, start to build the AI muscle while also realizing value And this password reset scenario came up. So what we did was we, we stood up the agent force agent, and the way that works at a high level, and we can talk more about this later, you essentially ground the agent in knowledge.

[00:11:43] So everybody's got, well, not everybody, most big organizations, especially if they're larger support teams, have a knowledge base. And these human support teams leverage the knowledge base while they're talking on the phone, while they're operating a chat bot while they're answering an email, what have you?

[00:11:59] And they use the knowledge and the knowledge base to answer questions. Same exact paradigm with Agent Force. So you ground it within this knowledge base. Now it is capable of answering ques every, every question that it possibly can based on the knowledge that it has available to it, which in theory is the same knowledge that you're feeding to human agents.

[00:12:20] So that's the approach we took at Charlotte. It's a relatively simple configuration as far as AI goes, pretty straightforward. Nowadays we could probably get it done in eight weeks or so, at the end you're left with an agent that can all of a sudden has access to every piece of knowledge that your human agents have access to in a blink.

[00:12:41] So that's really powerful and that's really valuable. Because now these humans can go and do you know, more valuable things with that 50% of their time. That's all of a sudden being deflected by the agent, prior to them receiving a conversation.

[00:12:58] Mike Allton: Oh, fascinating. Now, some of my listeners who, first of all, if you haven't been following the news in the United States, or maybe you're not, not from this country, I know a lot of you listening are from outside the us.

[00:13:08] What Aaron's been talking about are changes in decisions at the executive level. That are reversing or withholding billions of dollars that was previously earmarked to go to a very specific higher ed institution. So as you can imagine, like any organization or businesses, suddenly you drop a billion dollars off of your budget.

[00:13:26] You've gotta scale back in a lot of ways and, and remove a lot of services and functionalities that, that previously you were planning on doing. So that's, that's the context for this conversation. Now you mentioned. Cost savings and time savings, that sort of thing. One of the things that I'd seen from Goldman Sachs, and this was, this was two years ago now in a study that they did, their state of AI was that support centers were seeing a 20 to 25% reduction in cost and a 20% improvement in operational efficiency.

[00:13:56] Now I hear you and others talking about a 50% time savings. That sounds like a bold claim. And, and an improvement in just a couple of years. Can you break down exactly how Agent Force helps them see these kinds of efficiencies?

[00:14:11] Aaron Godby: Sure. Yeah. Let me, let me start with kind of the, so chatbots have been around for a while.

[00:14:17] They're, they're nothing new. And I think in a way folks are thinking about this and they're thinking about chatbots and maybe it's just a more advanced chat bot. So how could we possibly get, 50% right out of the gate? But the paradigm's actually much different between an agent and a chat bot.

[00:14:33] So let me, let me dive into that real quick. 'cause I think that will add some context that illustrates why you could, why you can get these, these levels of gains with agent force. So with the chat bot, you're essentially programming every piece of behavior in that chat bot. So it's, think of it like a decision tree, either suppose you're pulling up Delta's chat bot.

[00:14:55] I need to rebook, rebook my flight. I need to know where my gate is. I need to understand where my checked luggage is, those sorts of things. So you'd have to program every single one of those and then behind the scenes make it actionable in order for that chat bot bots work. So you start to get into a situation where you're only able to pick and choose certain scenarios to address.

[00:15:20] Now, let me contrast that with what we do with Agent Force that I described earlier. You were able to ground it on whatever data you want. We started with the knowledge base at Charlotte, and when you start with the knowledge base, what you're left with is an agent that can answer questions as good or better, then your best human agent.

[00:15:40] So now all of a sudden you have that, you know that, let's call it defense in front of your support team. That's able to answer, basic questions as well, or better than your best agent because it's consumed all of this information, just like chat, GPT, for example, is grounded on the history of the internet.

[00:15:57] This would be grounded on your knowledge base. Then you can take it to the next step and say, all right, I also want to ground it on customer specific information or in the case of a, a university student specific information. And that could be things like. What financial aid packages are that student qualifying for?

[00:16:16] What courses are they enrolled in, what activities they do outside of class? So now all of a sudden, in addition to being able, able to answer any general q and a that's available in the knowledge base, it could also answer things that are very specific to that, to that human that it's interacting with.

[00:16:34] Then you can think about how do we take it to the next level and let the agent actually perform actions. So in the past with those chatbots, again, you're limited to whatever, whatever you've programmed into it. In the case of agents, you have an action framework. So you're essentially giving that the new term for it as digital labor.

[00:16:56] So think of it as a digital support agent. You're giving it additional capabilities that it can perform. You're expanding the scope of its job role. So if you have a robust set of APIs, and in case anybody listening is not familiar with those, that's basically how computers communicate with one another.

[00:17:15] So one system will have a contract of how do I want to be communicated with to do a certain thing? That's an API. So if you have a good set of APIs, like one of our clients spirit Airlines, and this is publicly available information, they've done a lot of success stories on it. They've got a great set of APIs.

[00:17:32] It allows them to do basically anything within the airline that's passenger facing, pilot facing, flight attendant, you name it. They can perform that within their systems, through these APIs. So you can imagine that if you were to configure an agent like Agent Force and you had a set of APIs like that, all of a sudden the possibilities are fairly limitless.

[00:17:51] And to me, and this is, this is speculative at this point because the technology's so new that nobody's done this yet. But what I'm talking about right now is technologically possible today in, on April 10th, 2025. This is possible you could build the abilities for agent Force to interact with every one of those APIs and essentially perform any action that a human agent could by leveraging those, plus being grounded in the right data and grounded in the right knowledge.

[00:18:23] So, think of it like a digital representation of your best support agent. That's, that's essentially what you can build today with Agent Force, and that's why you can get those sorts of results, like a 50% plus right out of the gate because it's able to do anything that these humans can, and it's not incurring the tax of having to work with a, a pretty ineffective user experience in order to do it.

[00:18:49] Mike Allton: Yeah. Well first thanks for dropping the timestamp because this will come out a little bit after what we've discussed and when we've recorded it. So capabilities will be even better than what we've talked about and longtime listeners of the show, this might sound a little bit familiar 'cause we've talked about RAG on the show Retrieval Augmentation Generation Systems, which Agent Force in some respects is we talked about it on an episode all about Notebook LM, which I'll link to.

[00:19:12] In the show and Notebook LM is a fairly simplistic version where you simply load up sources, documents, files, whatever, and any conversation that you have then with the LLM is restricted to those sources. And that's something that I wanted to underscore for the listeners. When we're talking about setting up systems like this.

[00:19:31] We're not talking about giving them access to the worldwide web. And the ability to generate answers based on whatever the AI happens to find out there. 'cause we all know that's where hallucinations and incorrect information are, or even irrelevant information. If I want to quote from one of my guests, I want to go to Notebook lm, not Chet gt.

[00:19:49] 'cause Chet is not gonna limit itself to my actual podcast episodes. If I say I want a quote from Chris Penn, it's gonna find the first quote from Chris Penn. Not necessarily something he said to me. So think about that. But the other thing that I'm wondering though is tools like Notebook L Lab are fairly easy to get started on.

[00:20:08] You open it up, you drop in some sources and then you can talk to it and it's fairly straightforward. What's the learning curve like then for support teams implementing something like Agent Force and how do you manage that transition?

[00:20:21] Aaron Godby: Yeah, good, good question. So I, I think two things there. There's, there's the implementation side of actually standing it up for the first time.

[00:20:28] That's something that to me, it, it's wise to find somebody that's done it before that understands agent ai, that understands large language model concepts and prompt engineering, and also understands data integration, data lakes and everything that it takes to actually make this thing be effective.

[00:20:46] That's where they should probably go for some level of outside help, but that outside help, if they're skilled in this, they can do this relatively quickly because Salesforce has built a great platform for spinning up agents. As a quick tangent, we've done this the hard way in the past.

[00:21:02] So as soon as open AI came out with an API that we could interact with, we started building agents custom. And that takes a lot of work behind the scenes for all the reasons you just mentioned, Mike. Preventing hallucinations is not easy grounding in the right stuff with RAG that takes a lot of engineering infrastructure.

[00:21:20] So we, we had dabbled in that in the past. And our thought always was, somebody's gonna build a platform around this and make it a lot easier to focus on adding business value versus building infrastructure. And that's what Salesforce did starting last September with their Dreamforce announcement.

[00:21:36] So I think to me the more important part of that question there is, what's the change management to a, a support team if they were to add an agent force? And I think short term, it's minimal. It actually makes their lives a lot easier because if there is an escalation from this chat bot they'll receive a summary.

[00:21:55] So it's, it's smart enough to know what is it that the customer is actually asking for support on, and it'll summarize that as soon as it hands it off to the human. So they don't have to go through this chat transcript or come in cold with no knowledge. They know exactly what the customer's after, as soon as they start fielding that support ticket.

[00:22:14] Longer term. Now, anyone, it's anyone's guess as to where this goes. To me it seems like the entire support paradigm will, will start to evolve. Because all of a sudden, if you're deflecting 50%, 60%, even 90% of the work all of a sudden these human agents can now focus on providing more of that white glove type of service.

[00:22:37] Because the things getting to them are no longer these trivial requests that are chewing up their time. Maybe it's a more complex request and they can go and solve that on behalf of the customer and start to, and then you start to experience a lot higher CSAT scores, a lot happier customers. And you start to build a better business that way and happier support teams for that matter.

[00:22:59] Mike Allton: I think that's a huge point and I'd love it if you could expand on that a little bit more. 'cause I think a lot of the folks listening. If you're the business owner, the idea of increasing response time rates and, and decreasing the number of tickets that go to your human reps sounds amazing. Mm-hmm. But if you're the human rep, you're starting to wonder, I think, where does that leave me?

[00:23:21] Am I gonna be out of a job? What am I gonna do? And you, you kinda touched on that a moment ago, but I'm wondering if you can expand on that a little bit more.

[00:23:27] Aaron Godby: Sure. Yeah. And I, I, I could see where they're coming from and it's evolving so quickly that it's. It's, it's difficult to keep up. Even for me, that's in AI every day.

[00:23:37] There's, there's something new every day. So I can understand the fear behind that. But I think where it's gonna move is more of the, the humans providing white glove service and the customer service getting much better not eliminating human resources. So as an example, like there's these IVR systems that have been around for a while that folks are probably overly familiar with, and for me it's.

[00:24:01] It's one of the most frustrating experiences you can have, and I try to avoid it at all costs, but sometimes you just have to call the 800 number and go through the menu where it says, press one for this, press two for this. And then all of a sudden you're like eight layers deep into this thing and you're in the wrong spot.

[00:24:16] And either you have to start over or go back somehow. And I find myself getting annoyed more and more and more annoyed as these things go along. And all of a sudden I've spent 15 minutes trying to get through this phone menu to get to a human. And then you get to the human and they don't have any context of what you're after to begin with.

[00:24:35] So you're starting over again. So for me, AI could completely eliminate that thing from the equation. And all of a sudden, support teams are not dealing with these angry customers that have had to spend 15 to 20 minutes navigating the phone system to get to them and then repeating everything they just did.

[00:24:54] The customer's a lot happier when they get to them because maybe they've spent. Less than a minute on the phone system because it knows exactly what they're after. It's grounded in generative AI and can interpret why they're calling to begin with, and then it's handing off to the human agent in a very seamless fashion so that they can go address the problem.

[00:25:13] So I I, I think it's gonna evolve to the point where the employee experience is also much, much, much better, and they're getting fulfillment out of solving real problems instead of, telling somebody information that's available on the website. That's, that's the vast majority of tickets. It's just somebody could not find information that's already available.

[00:25:34] So they're calling and then they're, because of the IVR system, I just, I just mentioned in that paradigm, they're getting more and more and more frustrated, and by the time they're on the phone with the human, they're a nine out of 10 on the annoyance scale. So I, I think all of that can go away and make the customer experience better, along with the support agent experience in tandem.

[00:25:54] So everybody wins. That's where I would like to see it go.

[00:25:58] Mike Allton: Those are such huge points. Routine inquiries are gonna get taken care of automatically and and a much more rapid clip. When something does need to be escalated, the context comes with it, which is better for both parties on that call and the support center rep, as well as the customer on the other end of the call, and by not having to deal with those routine tasks.

[00:26:20] Your customer support agents now can devote even more time to researching, talking to customers, talking to second level engineers, quality assurance folks, whoever they need to do to help them solve the bigger issues. So instead of issues lasting four to eight hours, or. A day or more, they're free to just focus on it until they get that thing solved as as tenacious as they need to be.

[00:26:41] Folks, we're talking with Aaron Godby about how organizations are successfully implementing and benefiting from Agent Force. I've got several more questions you aren't gonna miss, but first, a quick word from our sponsor.

[00:26:51] This episode of AI for Business Leaders, The AI Hat Podcast is brought to you by Magai. Tired of juggling multiple AI subscriptions and trying to figure out which tool to use and when? Magai simplifies AI for businesses by offering access to all the leading language models and image and video generators for one low price.

[00:27:12] With Magai, you can create custom instructions to personalize AI for every business need, and use them across all LLMs, including Chat GPT, Gemini, Claude, and even Deep Seek. No more building custom GPTs or gems that are limited to just one platform. Magai also lets you create customized workspaces for your entire team, giving everyone in your organization access to the same AI assistance and tools and prompts ready to streamline your AI workflow and empower your team.

[00:27:41] Tap the link in the show notes to visit Magai today to learn more, and now back to the rest of today's episode.

[00:27:49] So, Aaron, how do we know if this is working out? What are some of the metrics that businesses should actually be tracking when they're implementing these kinds of AI powered solutions?

[00:28:00] Aaron Godby: Great question.

[00:28:00] So I'll caveat this and say that it's evolving really every couple of weeks. Salesforce is updating agent force and dropping releases every two weeks. So that's rapidly evolving. The first thing to look at is. What we have been looking at in support for a long time, which is case deflection.

[00:28:17] So how many different cases can be deflected by this chat bot where somebody's happy with the response that they've gotten and the service they've gotten, and they don't need to talk to a human and then consume their time. That's obviously the best one. And then you start to dive deeper. What types of cases are being deflected?

[00:28:38] What types of cases are not being deflected and what are they about? So that lets you start to improve the agent over time. To me, those are, those are the two main things you should be looking at because we, we like to look at the, the agent as if it's digital labor and it has a job description of its own.

[00:28:56] And that job description can be ever expanding as you add new capabilities to it. So you configure new behaviors, you ground it on new data, you give it new actions to perform. You need to know like what are the big ticket support requests that are slipping through your initial line of defense with the current agent so that you can level it up over time.

[00:29:17] Because this is really an ongoing thing. You're investing in your digital labor the same way that you would your human labor by giving it additional capabilities and additional things that it can perform on behalf of the customer.

[00:29:30] Mike Allton: Love that. We've been doing something similar internally at Agorapulse, we use a system called Slite for all the Documentation, and Slite has its own AI system so that it can treat all of your Slite documentation like a rag we talked about a moment ago.

[00:29:44] And so via Slack or in Slite, we can ping our knowledge base and ask whatever question we want and whatever language we want and get an answer with citations. But then we're also able to add our blog, our support center articles, and all these other external things and bring in those sources. And just last week, we learned some of additional capabilities, it's adding. So it, it's great to see that you're absolutely right. It's changing all the time. We've been talking a lot about routine tasks. I wonder if, are there any examples that you have in mind of more, a little more complex support scenarios that AI's been able to solve?

[00:30:18] Aaron Godby: Sure.

[00:30:19] Yeah. So that's, that's really that's really gonna be limited by your data and your actions that are available. So if I go back to what I mentioned earlier about Spirit, who we've worked with since 2020 to build an API framework and connect all their systems together surface actions. If you're a company that already has something like that at your disposal, the possibilities are fairly limitless because it's just a matter of calling the right API to facilitate a request.

[00:30:48] So an example I'll give you is a prototype that we built around a year and a half ago on a, with open ai. Where we mocked up a data model that could be used by an airline because we were familiar with that. And we took a scenario that I was actually living. I was scheduling a golf trip with some buddies.

[00:31:06] We were all coming from different cities. We wanted to get there at roughly the same time to the airport so that we could drive together to the house and then make our tee time. That's not that easy in airline scheduling systems today. To take different cities and different requirements and all get there at the same time.

[00:31:25] So what we did at Green Irony is we built a prototype based on the way that these airline APIs work, and we knew them to work. That would allow us to essentially just describe that scenario and that goal to the large language model in real time. So I've got this person coming from Tampa Bay, I've got this person coming from Atlanta, this person coming from Denver.

[00:31:48] And this person coming from New York City, we all want to land in Bend, Oregon between this time window. How do we do it? What are the best flight options for us? And it chewed through all the data through rag techniques like you mentioned. That was, that was how, how we implemented that and then would serve up the best flight options for each person.

[00:32:11] So that's a relatively complicated scenario, and to my knowledge, it's. There's no way to do it today other than brute force, searching every airline or using one of the crawlers that aggregates flights, but everybody's gotta do it themselves. In this case, we're giving it a goal, we're giving it different cities, and it's giving us all the flight itineraries that get us there at the, at the same time, or we're at, within a time window as best as possible.

[00:32:38] Mike Allton: I love that example because as a business, I actually see that kind of thing. All the time. Maybe we are exhibiting at an event and we're gonna have four to six team members go into that event. Well, the event's at a specific time, but everyone's gonna end up arriving at random times 'cause they're gonna make their own travel arrangements.

[00:32:54] One guy's gonna, show up two days in advance. The next side is gonna show up at day in advance. The next side is gonna show up the morning of the event, and none of that works. You want everybody showing up at approximately the same time. 'cause then they could ride share from the airport and they could check in at the same time and be on the same page as a team.

[00:33:10] Or maybe you're having a retreat or something like that. I love that use case for businesses and that's something that is going to be deployed more and more and it's gonna become easier and easier. As businesses to do those kinds of what used to be, to your point, very manual, brute force, cumbersome, often confusing tasks.

[00:33:29] What are some of the mistakes that you're seeing though when organizers make organizations make, when they're trying to implement these kinds of AI solutions into their support operations in particular?

[00:33:40] Aaron Godby: Yeah, I think there, there's a few the first one that comes to mind is around the, the mindset. I, I think.

[00:33:47] There's, people have just been inundated with so much information about ai that the expectations have have changed quite a bit, and a lot of times somebody will think it should just plug and play it's magic. It does everything. You don't need anybody to configure it. That's, that's just not the case though because of, everything that we've talked about on this call regarding the need to ground the agent to make sure that you're not hallucinating to get the right data in there.

[00:34:12] So. With that in mind some of the mistakes that I've seen as far as implementation goes, are under emphasis on data and integration. So the agent's only gonna be as good as the data that you ground it with. So if that's a knowledge base, you need to make sure that knowledge base is really, really good because the agent is gonna take that as gospel.

[00:34:36] It's gonna assume that that information is correct and it's gonna give it to the customer. So the first thing is the quality of your knowledge base. The second would be the quality of any other data that goes in there. So enterprises have, hundreds of different systems with data in there. And in order for the agent to know about that, you have to get that data into a data lake.

[00:34:54] Salesforce has a data lake house called Data Cloud, and that's what's used to ground agent force. So you need to get that data in there in a consumable fashion and it needs to be accurate in order to ground the agent. That's the second thing. The third would be surfacing actions that the agent can perform.

[00:35:12] So if I go back to the airline example, maybe that's rebook a flight or if I'm a pilot and I'm sick, maybe it's call in sick. And then that triggers a bunch of different actions in different systems that have to take place. But if you don't have the systems integration in place, you don't have the APIs to drive that, that agent can't perform that action.

[00:35:32] The last one that I'll touch on. Is just the human agent interaction. So these, these agents are not replacing people. They're not not at scale, right? They're, they're replacing some of the work that people are doing so that they can do it more effectively. So what we found in our exercises really important to nail down.

[00:35:50] How does that handoff work? So when the agent escalates to a human, what does that handoff look like? And how do we make the human's job as easy as possible? Once that handoff occurs so that you've got a happy customer, you've got a happy customer support agent, and and everybody wins.

[00:36:10] Mike Allton: Y your point about data, I think is huge.

[00:36:13] That is, that is a massive concern that people need to be thinking about. I mentioned our system at Agorapulse, we're slowly implementing it out to customers because people have questions all the time about a social media management tool. Do you support this or that platform? Do you support videos to Instagram?

[00:36:27] Those kinds of things. And we discovered this when someone had asked, do we support Google my business? And the initial answer was, no, we don't. Which was an incorrect answer. And when we looked. At the data and we looked at the citation. It was referencing an old release note where it had mentioned that we don't support Google My Business from like several years ago.

[00:36:50] And then fast forward, that was something that, that we added support for later on. And the AI system often can't know that or can't tell that. So that seems to be almost a daunting challenge. Any advice for businesses or organizations listening who might be thinking, wow, I have so much. A documentation I wouldn't even know where to begin.

[00:37:12] Aaron Godby: Yeah. I think that's, that's gonna be the bulk of the work to enable these agents to work effectively. It's, it's integrating your systems. And that's, so I, before this, I was in mobile applications. I was in digital transformation. To me that was always limited by systems integration. That's, that's the brute force dirty work that has to happen to enable these things.

[00:37:33] And it's gonna be the same with ai. It's just maybe even more of an emphasis on it and even more of a priority to get it right. You have to make sure that your, your API strategy's in place so that you can interact with these systems and that your data's clean so that it goes into the system and you can, it goes into the data lake and you can trust it with ai.

[00:37:53] To me it's there. There's not really a shortcut around it. It's, it's been an issue since I've been in consulting. Starting in mobile consulting in 2012, it's always been the issue. It's always been what's dictates success or failure. People see the fancy user interface, or in this case the ai agent, but what's driving behind that is always the data.

[00:38:14] And if the data and the systems interactions are good, it's gonna provide that good experience. But if they're not, and you're grounded on the wrong data, there's gonna be a lot of problems. So. I, I think it's just, it's a matter of having a strong engineering team that understands systems integration, that understands APIs and that understands data hygiene and, it's, it's a, it's an engineering undertaking that needs to happen.

[00:38:37] Mike Allton: If they're looking at multiple sources of data, multiple integrations, maybe a knowledge base, maybe a blog. Maybe they've got a cloud drive like Google Drive or something along that. Is it smart to just do one at a time? Integrate one and then test rigorously before, like, and rather than just plugging them all in at once.

[00:38:57] Aaron Godby: Absolutely. Yeah. We, we will typically start with the knowledge base because that's, if somebody has one that's the most bang for their buck smaller institutions may not have that. And in that case, we, we've had a lot of success with using the website, using FAQs and PDFs that they, that they know have the right information to start to ground the, the agent force agent.

[00:39:19] But yeah, I, I typically would, would recommend that crawl, walk, run type of approach. Don't try to just do everything at once and eat the entire elephant, like start small. You're gonna get a ton of value just starting small. Even the, the Charlotte case that I mentioned that's just currently grounded on a knowledge base.

[00:39:37] They've got a vision to drastically expand that because of the success they've seen. They started out with just the knowledge base and they got that level of value just from that. So that's always where I'd recommend somebody start.

[00:39:50] Mike Allton: Fantastic advice, Aaron. This has been so insightful and hopefully very eye-opening to everyone listening For folks who've got more questions, maybe they wanna reach out and work with you, where should they go?

[00:40:02] Aaron Godby: They can find us on on Green Irony.com. There's a lot of content around Ai. We'd be happy to talk to you about it as well. We'll, we'll, we'd be happy to help your entire audience to sign their first agent. So reach out to us and let's talk.

[00:40:16] Mike Allton: Fantastic. Thank you, Aaron. Thank you all of you for listening.

[00:40:19] We will, as always have all the links in the show notes below. But remember, if this is your first foray into this world of AI and business, I've got the perfect resource for you in the show notes below the AI primer designed to get you quickly up to speed and all the terms and the tactics and the underlying technology that's powering this AI Revolution.

[00:40:35] Check it out and I'll see you in our next episode. Welcome to the grid.

[00:40:40] That's a wrap on another episode of The AI Hat Podcast. I hope you've gained some valuable insights and actionable strategies to boost your business with Ai. If you enjoyed this show, please subscribe, leave a review, and share it with your network. And don't forget to visit the ai hat.com for more resources and to connect with me.

[00:40:58] Until next time, keep innovating and remember, the future of business is human and it's powered by Ai.