# Use cases for deploying voice agents in SMB and enterprise [06# Jannis Moore]

## Метаданные

- **Канал:** n8n
- **YouTube:** https://www.youtube.com/watch?v=5nHyNpa-k7I
- **Дата:** 07.11.2025
- **Длительность:** 41:12
- **Просмотры:** 1,886

## Описание

In this conversation, Jannis Moore discusses the integration of voice AI with n8n, highlighting its benefits for businesses. He explores various use cases, including lead qualification, customer support, and appointment setting, emphasizing the importance of prompt engineering and measuring ROI. The discussion also touches on the impact of voice AI on employee happiness and the challenges of navigating complex tech stacks. Jannis shares insights on human-AI collaboration, particularly in customer service, and the value of warm transfers in enhancing customer experience.

00:00 - Introduction
01:10 - The power of voice AI for businesses
03:05 - Four core Voice Agent solutions 
05:10 - Welcome & guest introduction
06:08 - Connecting voice AI to backend systems
09:20 - Use cases: Voice agents in different industries
11:27 - Lead qualification and reactivation explained
14:00 - Adding value with lead reactivation
16:06 - Customer support automation with voice AI
18:00 - The art of prompt engineering
21:00 - Markdown and structuring prompts for AI
24:00 - JSON vs Markdown in AI prompts
26:45 - Voice AI for small businesses and after-hours calls
29:00 - ROI and value of voice AI solutions
32:00 - Employee happiness and human error reduction
35:00 - AI evaluation and ongoing improvement
38:00 - Reporting and dashboards for voice AI performance
41:00 - Sales training and AI-driven analysis

#n8n #n8nMasterclassPodcast #podcast #masterclass 
The n8n Masterclass Podcast

## Содержание

### [0:00](https://www.youtube.com/watch?v=5nHyNpa-k7I) Introduction

You don't need to know how to code to create voice agents. We're not in that age anymore where you actually need to know how to code. I'd say even for like a one oneman shop, a oneman army, it would be worth it implementing a voice AI solution that automates things for you. Simply imagine that you have a business that collected, let's say, thousands and thousands of lead over the last couple of years. It's humanly just not feasible to have a team literally just calling through them and following up with every single one. We had campaigns for Fortune 200 companies where we called 20,000 people in like a month. That would have taken a human multiple years to fulfill, right? And these things have not been possible before. But generally the AI has no connections to anything. Now you got to connect it to something, right? And that usually happens through workflow automation tools. And NAD is just happen to has a lot of more really cool functionality, especially if it comes to custom code that we really, really like. It's more like a set of solutions that work extremely well. And that's four in total. That's lead reactivation, appointment setting, customer support, and lead qualification. These four solutions work mostly throughout any kind of industry. And we really have proven that to ourselves because we had companies from like small businesses to Fortune 200 companies that we serve with these solutions. Simply imagine you you own an ecom store. You sell a thousand different kind of products. Having an employee trained on knowing all of those products and what to do with them is incredibly hard.

### [1:10](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=70s) The power of voice AI for businesses

— Hey Yiannis, welcome to the podcast, brother. So glad that you're here today. — Yeah, thanks for having me. I appreciate it. — 100% man. Uh love your channel. Love what you're doing. uh you have a really technical back end and you really focus in on voice AI and integrating into init. Can you talk to me just a little bit about like the benefits of connecting a voice AI platform to the back end of init? — Yeah, I mean there's a gazillion things that pop in my head and I'd say probably from like more sophisticated solutions towards implementing anything that makes the voice agent talk to external services is usually orchestrated through it end right. So if we talk about voice AI itself, voice AI is basically just that orchestration layer. You have platforms like BPY, Lemon Labs, Retail, they allow you to have these voice conversations with AI, but generally the AI has no connections to anything. Now you got to connect it to something, right? And that usually happens through workflow automation tools. And NAD is just happen to has a lot of more really cool functionality, especially if it comes to custom code that we really like. And that allows us just to build really sophisticated solutions towards helping N8N connect our voice agent to external services stuff like QuickBooks, uh, Facebook, whatever that might be, right? And that is one of the major use cases. But recently since AI obviously we going deeper into MCP which works incredibly well with Nadu and I think that is something that we are trying to leverage even more especially now since you have that S um SSC integration so you can stream data back and forth which was the only bottleneck that we had to make the data interaction quick enough so that you can actually talk with MCP server in voice AI and that has been I would say the biggest USP for now at least if it comes to MCP. Yeah, because with voice AI, you don't want that big lag effect and you don't want to have to wait until you get the entire data set of voice being sent over. So, being able to stream that, you can kind of get more real-time conversations back and forth if it makes sounds more human. Is that accurate? — Yeah. The thing is basically when you

### [3:05](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=185s) Four core Voice Agent solutions

spin up an MCP server that usually you had to have this middle solution, this middleman and whenever you send a request there, it would need to spin up this MCP server. It would initialize all the services, then provide the tools, then make this one tool call and shut it down. This whole process takes anywhere from 5 seconds up to like a minute depending on how quick the system is and you can imagine if you're on the phone and you need some quick data waiting a minute for an answer is just not a thing right so that is where this thing changes and now we get responses in like less than a minute less than a second which is really good — amazing can you talk to me about I want to first talk about like the like use cases so you you've been really deep in the voice agent section of things what are some of like the the big industries or the big use cases of voice agents for businesses — is a very interesting one. I wouldn't even say there is a specific industry that stands out. It's more like a set of solutions that work extremely well and that's four in total. That's lead reactivation, appointment setting, customer support and lead qualification. These four solutions work mostly throughout any kind of industry and we really have proven that to ourselves because we had companies from like small businesses to Fortune 200 companies that we served with these solutions. So they have been incredibly powerful. When you say that, could you say that again? Lead force solutions. Is that what you're saying — for lead qualifi? You mean the type of solutions? — Oh. Oh, you said lead qualifications. — Lead qualifications. Yeah. — Got it. Okay. So, so lead qualifications. So, they're calling it up. They make So, real estate, you want to make sure that they can they apply for things and so you're looking to get all the data extracted for uh real estate is a popular one for lead qualification to know should I pass this should I even pass this on to a human or is this someone that we don't even want to direct towards this? We may be sending them some educational content. Is that correct? — Yeah, exactly. You have Facebook lead forms, right? They fill out a form. You just call them with voice I read after. Speed to lead is super important. That's why it works so well. You just qualify them. You ask them a couple of questions. They can either book an appointment straight on the call or you just drop them or send them an email, whatever it is. And then just have them book there. But yeah, that's one of the big things that works super well. — What were the other ones? And I just want to go down them one by one. You said there's four of them. — Uh lead reactivation. — One of our favorite ones. Yeah. Simply imagine that you have a business that

### [5:10](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=310s) Welcome & guest introduction

collected let's say thousands and thousands of lead over the last couple of years. It's humanly just not feasible to have a team literally just calling through them and following up with every single one. So we have voice a we can literally just call them and qualify them because in some way they like obviously they need to be validated and you need to have permission to call them. So there's a lot of legal work around that. But when you have that it's incredibly beneficial. — We have campaigns for Fortune 200 companies where we call 20,000 people in like a month. that would have taken a human multiple years to fulfill, right? And these things have not been possible before. — Yeah. So, lead qualification uh reactivation I should say. I'm curious on that one because there's a piece of this. Yes, you can do volume, but I know as someone who's hyperfocused on value added activities besides just calling them up and getting the data, is there any way with that lead reactivation that you've seen of use cases to besides just extracting uh to reactivate them? Is there any value added activities you're doing with that voice call? You mean specifically regarding lead

### [6:08](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=368s) Connecting voice AI to backend systems

reactivation? — Yeah, I I know for the company itself, say the Fortune 2000 companies, right? It reactivating is highly valuable for that Fortune uh 200 company, right? So, they're getting they're reactivating leads, but for the lead itself, is there anything like, hey, man, uh I don't know, it's your birthday or we found these really cool things or is there something that you're doing in order to reactivate that lead to add value to that lead's life? — Oh, yeah, 100%. I mean, there's always a reason why we call him, right? So we did that for example in the recruitment space uh recruitment you can imagine people are often looking for new jobs if you can call them and you can match that position of what they previously tried to apply for you have that information right we can enrich the voice they are literally just focusing around that and can request more information about their current status how much they want to earn uh etc so this is how we can really tailor those calls specifically towards people and that's obviously just one call right then you have things like what happens if the person doesn't pick up and calls back what happens If you implement like a call cadence, so you call them multiple times over a set of days. There's tons of extras and extra utility you can build on top of that. So it's usually not just a one-time call, it's usually like a whole system around that just to make sure you get a higher output and success rate. — Yeah. So what you're saying there is like what for obviously if someone's looking to get a job then you're okay um you have these certain set of skill sets. Let's say um product designer by the way if you're looking for a job we have uh three positions that are that maybe pay this range or something be very valuable. Is that something you'd be interested in? And then you then it's a double-sided value market because you know the recruitment space gets a high quality lead for them. They can get employment and then you're pairing them up with something that they want to have. — Yep. — Correct. — Love it. Okay. What was the third thing? — Uh customer support. — Customer support. Yeah. Yeah. — Simply imagine you you own an ecom store and you sell a thousand different kind of products. Having an employee trained on knowing all of those products and what to do with them is incredibly hard. So with voice AI, you can nowadays just take all of those products. You basically clean, that's how we call it. You clean that information. You feed it into the voice AI agent and it knows everything about every single product. Then you have a person just calling in and saying, "Hey, look, my item is broken. It's XYZ. U can I please have a replacement? " And the voice AI can say, "Oh yeah, no worries at all. Uh what is your order ID? " For example, you say the order ID, it can literally schedule you the placement for you or offer you a refund. So that's stuff that works incredibly well and throughout all kinds of industries. — Yeah. And I can see that being valuable especially for if you have a large uh data set of knowledge because a lot of these customer service agencies they're mostly following it's usually some sort of uh possibly offshore customer service support person and they're just running through an SOP. Tell me what your issue is. They look it up in the SOP. They say okay this widget broke here then try X Y and Z steps. Did you try XEP? Move on to the next phase of solutions. And so it sounds like you're probably feeding in the manual of whatever that SOP is for customer support and then training the agent on it so that it can follow the same type of steps. Correct? — Yeah. Yeah, pretty much. It's like I [snorts] I can go pretty deep in this but to make it on a basic level. Simply imagine like we I would not consider myself like a technical engineer if it comes to building voice agents. We are more of conversational artists which kind of goes together with

### [9:20](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=560s) Use cases: Voice agents in different industries

you what you just said. So we are actually prompt engineering more uh more of the time than we do actually the tech stuff behind it. So if it comes to SOPs, yes, in some way we feed in the SOPs. On the other hand, it's making sure that it has some sort of guideline that would kind of match the pattern of the SOPs that they have for certain products. That makes sense. So you basically try to merge everything into one conversation that kind of like helps the AI to understand what to do in certain scenarios. — Yeah. And it's really interesting because I think part of these things is when people with AI agencies, right, they think they need to know um technical skills which is it's an important piece of it, but there's an artistry to this. There's a certain art behind say prompt engineering. There's a art behind say uh lead getting um if you're writing cold email copies. There's an art behind the technical science of things. Can you talk to me about a bit of the art of prompt engineering? like what are some of the elements or structures or styles or patterns that you try to do to to make a really good prompt for these types of voice AIs? — Yeah, I'm happy to answer that. But before that, there's one more thing I want to mention. — You don't need to know how to code to create voice agents, right? We're not in that age anymore where you actually need to know how to code. If you use NA and you know how to use chat GBT, you can most likely build whatever you want literally just without any kind of technical knowledge. It's just such a powerful weapon. So I just see that all the time. People are afraid of going into building these agent because they want they think they need to code. They don't. Like not at all. — Yeah. I know like back in the day when you'd go to like a like a coding boot camp, the first thing they do is they teach you how to Google. — All right. What's your Google Fu? Can you Google the problem first instead of trying to solve it? And I feel like we've shifted away from Google fu to, you know, chat GPT fu, right? Are you your ability to be able to interact with the AI and know the different methodologies to be able to work with an AI to solve coding problems? — Yeah, exactly. It's it all comes again down to prompt engineering, right? How to actually say something proper so AI gets the best value out of it. — Yeah. — And yeah, there's tons of techniques around that. I mean, one of the biggest ones that we've seen is least voice

### [11:27](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=687s) Lead qualification and reactivation explained

which has been around with OpenAI since the beginning. It's called markdown. Does that does it ring a bell? Do you know about that? — Mhm. — Okay, cool. Let me just break it down a bit easier cuz initially it was — some sort of formatting a textbased formatting language that helps us to format text visually so that it's easier to read for humans. So imagine if you add two hashtags in front of a single line. It would be considered an H2 title, right? So if you're in a Google doc and you try to select a title, you can select from H1, H2, H3. Two hashtags would basically be the equivalent to these H2 titles. And it's just written in text. So it would be literally just hashtag space and then the wording which would then be rendered by some external service wherever you upload those markdown files so that it's visual and you can see it with more patterns and that is amazing because uh it not only allows us to actually see things with certain priority but also AI obviously if you go on a website the first thing you read is usually the title that's the thing that is most relevant so you know it has a higher meaning now if you imagine that you have this H1 title so one hashtag in the word you see this thing first. So it has a higher meaning and the AI can interpret the same way even though it's not visual but AI just literally sees the hashtag but it knows that this thing that comes after that until the next line break is a lot more valuable. So when prompting things you can literally use this markdown to incentivize the AI to value certain things better or more to get a better outcome. So that is one of the good ways on how you can structure things in blocks. So basically like paragraphs and titles. Make sure you give it like guidelines and instructions and constraints all properly formatted in blocks. That AI has a better understanding of what they actually want from it and a better guideline with less hallucinations cuz hallucinations are always the worst and probably one of the biggest things that we see people struggling with when building any kind of AI solution. two points of that but one thing is actually I had this at I was running in a meetup in LA uh two days ago and they're asking about the value of sending in using JSON to send that in as the system prompt. Um and they're asking is that more valuable than sending in text. Um, but it seems like my positional belief around that is if you want to have an output that's structured, you know, using a JSON like a structured output parser that to have structured output is good. But do you think there's any benefits of sending in JSON as a system prompt in order to understand that the to prompt the system better or is it or simply will the markdown be sufficient enough? I — I think there's some edge cases where a JSON would potentially be better. Um, to be fair, it's probably that edge case that I can't even think of one like out of the box. I would probably say

### [14:00](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=840s) Adding value with lead reactivation

just markdown is better. If you want responses in JSON, that's a different thing. You can have a markdown section that contains a JSON example, like a response example on how you'd like to have the answer back. That makes sense. Or where it also makes sense to have JSON, but again, it's just part of this whole markdown prompt is when you have certain structured data that you want the AI to understand better. So if you for example have a tool call, so you have functionality within this AI agent that retrieves information from somewhere else. Let's say your AI agent can retrieve information from QuickBooks about an order and when it retrieves that information, you can put that back in as JSON because it just helps AI to understand, aha, okay, this is actually information that comes from somewhere else and it knows better how to leverage that. But again, it would just be like a subpart of this whole markdown prompt. — Got it. Yeah. So it's essentially saying hey if you need to call a tool uh in your tool belt for the AI agent this is you can feed in a certain data set and say this is kind of how you can call that tool and this is the to understand better how to call the tool to get the data back say from QuickBooks or from any of the other tools that you're calling. — Yeah. So I would see JSON only as a part of a markdown prompt rather than as a prompt itself. — Got it. Yeah, that makes sense. Um and so question about the customer service bit. So high level, massive volume, being able to train the AI to answer SOPs or maybe there's some sort of training materials. That makes a ton of sense. I see another use case. And tell me at what point is this valuable? There's many businesses out there, small businesses, uh, mechanic shops or other places that can't always answer the phone calls, but they might be missing out on responding to leads that maybe look them up in Google Maps and give them a call and they don't get them there, right? Uh, have you seen value for those sizes of businesses? Maybe it's a one to five person shop or something. At what point does it make sense to bring in an AI agent to help answer customer service calls when the attendance or the person isn't there to answer the call? A — very good question and I would probably see this from two different angles. The first angle is obviously the

### [16:06](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=966s) Customer support automation with voice AI

lucrativity of voice itself and the way of tracking the data and making sure you spend less time on it. Uh I'd say even for like a one oneman shop, a oneman army, it would be worth it implementing a voice AI solution that automates things for you. So and even we had this many times that we had some afterhour calls, right? Or after call solutions where the shop for example implemented the voice caller afterwards. So whenever the shop was closed, they just route that call over to the voice AI agent and they just qualify it, see if it's something important and then either route them somewhere or just note it in a database. Right? again simple connection with Nand into an air table and that's done. That is pretty valuable. The question and that is what usually holds people back is the other side which is the price. Usually it costs quite a bit of money for someone to implement that and that might be the bottleneck that keeps people from really going for it especially because that whole technology is still so new that most people don't really understand how it works and the value of it. So that's kind of like where we come in. I mean this whole market is out there since maybe a bit more than one and a half years that it's like an actual feasible thing that makes a lot of sense to implement and — it's still very new. So people they are not even they're often not even solution aware. So getting over this gap and then charging them anything from like 500 to 10,000 for a basic solution is very hard to reason if they don't really understand the concept. So that's probably one of the biggest bottlenecks of making them aware and uh showing them that it actually brings that value, right? And this is kind of the other side where I think we have a lot of people that don't really go for it at least right now. — Yeah. And so diving into that because I think that is a key piece is there's I think there's two probably blanks or two bottlenecks that people might face. you know, one, will this have a negative impact on the view of my business um as if someone's using voice

### [18:00](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=1080s) The art of prompt engineering

AI? And the other one is, will I actually generate revenue from it? I think speed to lead is really important. Um, but I want to touch on this uh you call it lucrativity or ROI with these companies. It how do you show with data that this thing is delivering an ROI? What does that look like for you? — There's a couple of different angles to it. And interesting enough, it's not always money. There's a lot of other aspects involved that make a lot of sense. Um, an interesting one is for example the health and the recreational space of or happiness of employees. Let's for example say and this might be a bit further out but we for example had some let's say dentist offices right they have always a front desk. The front desk takes all of the calls and that front desk can be fairly easily replaced. However, replacing it is usually not a way how we market that because even though a lot of people do that, they don't understand why because they don't have the data. The fact is that most people, especially those local companies that have like one dentist office, they love their employees. They don't want to get rid of them. They have like a personal connection. They feel like family. So rather than talking about replacing them, it's more like enhancing their functionality and possibilities of just leveraging their time better and helping work on more high leverage tasks, right? And where this comes very interesting is because if you're on the front desk and you need to take calls, you never know who you deal with. Someone might be in a bad mood. They get screamed at. They have really bad interactions and that really pulls them down. So telling a business owner that you can actually help their employees be a happy version of themselves, which not just gives them time back, but probably also better resources and more uh more flexibility on working on higher leverage tasks. This is a massive game changer. It helps them in so many ways. And a the business owner is happy because they know they also don't need to replace them. They know they can leverage them in a better way. They solve phone problems. So all of that is a is one of the unique selling points. And these specific connections, they're often even priceless. What we've seen, right? So they don't even put a price to that. If they can make someone happier, especially like in a very tight environment, great. So that is one of the aspects. Then obviously we have the time saving and we have the human error reduction which is another one. All of them kind of contribute towards the same thing. And these are like the basic values, right? You can even calculate that up if you have enough data. How much money do you save them? How much is a reactivated lead worth it? How much more money or how easier does the company scale now with having more resources available rather than having people on the phone? All of those things will fall into that. So it it's surprising, but we had a lot of cost companies where we brought them at least like half a million worth of more value, right, from like a simple phone interaction. — Yeah. And I didn't think about the mental space of health and happiness for the customer services people because I noticed like you know long time ago I was in the customer service space and I recognized that some of the most unhappiest people were customer service people because they constantly had their cup be drained of happiness because they're constantly facilitating these really unhappy

### [21:00](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=1260s) Markdown and structuring prompts for AI

people. Uh, and so I could see how if there was a chance to basically offload some of that capacity, which says, okay, you don't have to focus on these people out the gate, you can maybe focus on the higher leverage activities, whatever those things might be. You can then say, okay, you're not being spread too thin, which then, you know, lowers the overall, you know, happiness meter of the company. That's a really cool one for like a I would say kind of like a second tiered value. Like there's primary value and there's second tier value. Primary is profitability and all that stuff. And the other one is kind of like the psychological safety in an organization. Uh overall health, wealth, happiness, you know, feeling like you're heard, you're understood, you know, and then, you know, nice little value bombs of, you know, wishing someone a happy birthday or whatever it might be. So I like that as a secondary. The thing I — actually sometimes even a primary one by the way, like depending on the company size, the smaller companies for them, the secondary part, like you mentioned, is usually even the primary one cuz they don't often want to grow. They just want to have more time. They freedom. And that's something you can enable that. So it really depends on the kind of target group which is I just want to emphasize that because this whole thing is so new no one really understands it. So getting the data out there is one of the most relevant things at all. — Yeah. With businesses I've noticed this. There's like there's two sides to this. Right. Like so the first half it comes down I wish I had more leads. And then the second half is I wish I had more time. Right. And it's usually one or the other. And then if you get all these inbound leads then you have to filter through the unqualified leads or you got to re you got to reactivate the leads or whatever it might be. And then you translate that to great, we got all this inflow. Now, how do we make it free up more time with these people coming in? So, if you can parse them out using a voice agent, um, I could see how that could add to more time, more happiness, and then ultimately more money because then you can focus on those higher leverage activities versus answering phones all day or doing redundant activities like going through, — I don't know, tech less human error. It's a very big one, too, right? — Less human error as well. Yeah. And in terms of training the AI, um, one of the things that I don't know if you have you used the AI evaluation system for init. — Yeah, we have different methods for that, but yeah, — cool. Well, because that's one of the questions is a lot of people want to use um AI agents to do things and since it generates sometimes when it generates it doesn't always do the output that you want and so there's that when you talked about the confidence of is this thing answering in the way I want it to be able to answer you know we have the AI evaluation system where you can kind of give it the data set of the evaluations and then the back half you can then measure that in terms of an output with the evaluate evaluation testing and the metrics right um whether you use that system or another system how do you how do you have confidence that you're training the AI to have the output that you want with these voice calls? — It's usually always a work in progress. Like even if we talk about like what we usually serve is handover solutions. So we basically build out the whole system on the client's text and they kind of own everything after. And they usually expect this to be still a onetime thing which it kind of is. The thing is the reality is that you need to always work on voice

### [24:00](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=1440s) JSON vs Markdown in AI prompts

agents on an ongoing basis because there's always a way on improving quality and getting a better throughput. So for us this whole testing thing is often something that is ongoing and we usually even splat that in uh split it into two things. We have on the one hand the maintenance and on the other hand the monitoring both are automated and both are using some sort of voice AI testing tools and obviously like in your in your sense with NAD one really cool way that we've been using is hooking up NAD with u let's say superbase or air table we use both for that to basically just visualize the data that we get out of those phone calls. So there's usually after a phone call you get an end of call report that contains a lot of information with relevant details about the call how it went the success valuation maybe some extracted values and in air table we can just bristed fairly easy with superbase just feed in all of the values visualize it in an interface and then we can see aha okay in month uh January we got x amount of people and there so many people booked an appointment so you get actually insights and these insights again you just leverage to understand how well it currently works and usually there's some rule of thumb. If you build a voice agent and you release it, it's usually around 80 to 85% accuracy. And the more you progress, going getting to the last 10% is incredibly hard. And the last 5% is super duper hard. And even then, you know, you never get to 100%. It's just impossible the way how AI is designed and that you can always just extend it. But the higher you go, the more sophisticated it becomes and the harder it becomes as well to adjust things. So these kind of tracking solutions they are really good of getting these first 10 to 15% really well. — Yeah. I'm curious with that too because you're talking about you know taking these it's optics really it's taking these unseen conversations that most people say higher level people in the organization don't have access to pulling that out in this conversation and then dropping that into a database so that you can review that to see okay what are the edge cases what are the questions that the AI wasn't ready for or what are most of the times with SOPs and things like that they're not fully fleshed out. they're good, but then you have the human intuition that is usually bridging that last, you know, 20% that are these edge cases that just humans might know how to handle that aren't in the SOP. I'm curious with getting that reporting from the phone call into the database and then delivered to leadership, whatever that leadership might be, do how do you deliver that? Do you live that do you deliver like a report as like an email or do you have like a front end for uh air tableable or do you call them with a voice agent and give them a daily report? What does that reporting to leadership look like? — Yeah, let's go with the air table example because it's easy to understand. Air tableable has something called interfaces which allows you to just create a visual dashboard by hooking up the data of whatever is in the air table database. Right? So we can actually just give the client access to the air table

### [26:45](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=1605s) Voice AI for small businesses and after-hours calls

database or either they have their own account and they just see those visual interfaces. they can interact with it and can maybe leave comments and or comments and other things. Right? So that is one of the things or you can even just externally generate a report within an end flow for example and just send that out via an email. So these are the two ways. However, the interfaces are usually the more relevant ways for people because they like to see things visually. — Yeah. It's about how what's a good way to process the data and so if it's and so especially because any business owner anywhere they don't have a lot of time. So, can I glance at something and see what it is? And so, like the Air Table visual interface is nice because then you can probably graph things, put things inside of there. You create a dashboard and they can just at a glance see who's performing. Uh when I say who, it's like the AI agent, which one of these AI agents are performing well, what's the data coming back? What are we getting? Have you seen anything with I don't know if you do this at all. I was just thinking about it with um reporting on like sales phone call conversations. Um is there anything that you do with that? I know you do a lot of voice agents, but I know voice agents for lead qualifications, then that usually gets passed off to some sort of either setter or closer to then close the deals. Is there anything that you've been able to use like in terms of like humanto human voice conversations or anything around sales that you're pulling out the data from those conversations to be able to convert those sales to higher performance? — Yeah, there's we did I think like one solution where we specifically trained AI on the existing transcripts. So we basically analyzed the transcripts and created a solution around that just helps their own staff to be trained on. Right? So you would basically simulate a customer with AI and that customer is just talking to the actual sales rep and you just see how well they actually perform based on certain questions. So you can have certain scenarios built into those prompts to make them dynamic. And let's for example say one is a little bit more picky, the other one doesn't really want to buy or doesn't have the funds. you can just see how the sales rep interacts with it and at the end you do an analysis just like you do for a customer to push that again into a database and then analyze the performance of those sales agents. So that's something we did before. — That's awesome. Yeah, because now you're using basically two sides of the AI agent. One is customerf facing, one is employees try facing. The customer facing answers the questions the employee facing trains the customer on best practices that you can then do a report whether it's going up to a sales

### [29:00](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=1740s) ROI and value of voice AI solutions

director or the owner of the company to go how is this person overcoming the objections? How are they answering needs and concerns? How are they, you know, adding value to the conversation or booking the next meeting? I like that. Um, and so then that would also be visualized in some sort of dashboard like Air Table. — Yeah. — Cool. — It depends usually on how big and sophisticated it is. Air Table has a few limitations if it comes to massive amounts of data. So, anything where we have more than 50,000 calls, it's usually something that we don't put into Air Table. — Got it. at or at 50,000 uh calls where would you put it superbase or — yeah superbase or sometimes even just a my SQL on their own AWS infrastructure so it's some more custom stuff because especially with the bigger ones you have a lot of more regulatory compliance to follow so they usually have all of that in their own little infrastructure environment — yeah AWS is a very big um place for enterprise users to dump data into great to put data in a little expensive to pull data out um so as data if anybody doesn't know that AWS is really low cost and but they and it's really easy to get up and in there but they kind of have a walled garden that if you want to extract and like get all your data out of it they charge a lot of money uh to do that. So it's it it's very convenient to get in but if you want to switch it's can be a little painful. — Um love it. Now we talked about three of the four use cases. Can we go back to the fourth use case? Right. We talked about lead qualification, lead reactivation, customer service and then what was the fourth one? — Appointment setting. So that kind of goes together with lead qualification but can be a separate solution too. How do you handle uh appointment settings especially across different time zones? I know sometimes that can be a little tricky. — Yeah, that's only one of those 500,000 issues that can appear with appointment setting. Um — give me the topic. Give me the top three. — That's okay. One of my favorite examples was one of the I'd say bigger dental companies in the US. — They had around they had more than a thousand offices. And it's not just that you have an office with doctors. You have multiple offices that have doctors that rotate in between them, right? You have a doctor, Dr. A in office B on Tuesday and then the next week in office C on for like only half day and all of that is basically just managed in a very random and not even defined way in their CRM system that they are using and connecting all of that and this even like without any kind of time zone switches can be very difficult and it's always there there's not like a one solution that suits everything. It's always a trial and error that needs or that meets the exact criteria of what that specific client needs. So time zones to be honest is often something that we've discussed and some people want it but the people that really want it are either really big companies or they just want it because they have no idea why they want it in the first place. This literally just these two cases. Um but yeah it's very individual and the outcomes vary tremendously. If you have something with

### [32:00](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=1920s) Employee happiness and human error reduction

rotations like in that case, you can't for example just book slots during a call because it would take way too long because you have such a line of things that you need to check and make sure it works. So with these CRM systems, you can freeze slots, which means you just block it and then at the end of the call, you do extra analysis to go through and see, aha, this person actually wanted to book here. Let's just really book them in. And then you walk through this whole process, extract all the things you need and then book them in. So there's tons of logic that goes into that. And yeah, as much as I'd like to have like a predefined solution, there isn't one. — I'm very familiar. I've had to build these systems out. So I I know in terms of the booking can be complicated, especially if you have like a non-local company, right? And you can and they're they do work all around the world and they want you to do a roundroin with the different, you know, closers and each person's calling in from different places around the world and you need to do a roundroin system. You know the part of the thing I think the challenges I see is around you need to be able to one say okay are you what kind of tech stack are you are you using uh Google calendar are you using go high level are you using calendarly or and then can you that's the one piece the tech stack and then can you get people to actually have their availability and use the calendar consistently so that if you're actually booking the right people you actually are pulling the right data sets and so you have a thing in place where you're setting the meeting but you're not confirming the meeting unless everything crosses and checks after the call. Is that correct? — Yeah, it's a lot of money work and especially with the calendars, it can be very tricky through. We had people with Microsoft calendars. You can imagine you got to authenticate a lot of stuff just to make this work. — And uh same with bookings, Calendarly, for example, for a long time. They had they have an API, but they didn't really have an API for booking appointments for whatever reason. And we had it often that customer came in with a roundrobin structure of different uh sales agents on Calendarly. there's just no way for us to use it because we couldn't book appointments, right? We can get the availability setter, but it doesn't help us if you can't book it. So, you got to find really creative workarounds about solutions that seem to have an API, but then you can't integrate them. It it's really weird. And all of these kind of little nuances, they make a project go from two weeks to two months pretty much. — Yeah, it's those edge cases. If you know, if you can come in and say, "Okay, we're going to use this tech stack with these things and these this is what we're doing," then it's great. It's just that friction between trying to integrate what the company's already doing and then uh what the actual employee like what the employees are actually doing and trying to get them to integrate with uh stable structures I would say and technologies. — Okay. — Yeah. It's a hard one. It sounds it's high value uh but also uh high complexity. So if you looked at those things versus like customer service is probably high value less complexity which is probably an easier task to take on. — Yeah. True. It just depends then again on how easy it is for other people to jump on as well, right? You have more competition in one that is easier to fulfill, but it usually yields a lot better results to go for something more complex. — Do you do anything with human loops like

### [35:00](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=2100s) AI evaluation and ongoing improvement

where uh the voice agent can't handle and then a human jumps in and helps take things over — all the time? — Yeah, — it's some very common thing. I mean obviously it's people might lose business if some stuff doesn't work. So in the majority of cases, especially with bigger companies, we always have fallbacks where we route certain cases or edge cases or emergencies to a purified hotline. — Yeah. So is that I was going to say how do you then connect people from I'm on the phone with the voice agent things aren't working well for whatever reason you hit the panic button and for a human in the loop. How do you end up routing them to do you tell the agent to tell them to call this number? Do you is there some way to have them hop in on that phone call? What how does that work? depends a bit on the infrastructure that you use and what kind of service you build the voice agents on. Um but generally it's a prompt engineering. You just try to define rules for how to handle edge cases or let's for example say the customer repeats or the AI doesn't really understand their insurance ID or whatever it is and it can't figure it out for three times then you basically just build in a mechanism that routes them to a certain place when they for example don't hit it in three times and that's usually a call transfer. So again depending on where you have your phone number hosted there are different methods for it but generally uh what we do nowadays a lot is warm transfers. Initially we did call transfers basically meaning that whoever we call with or whoever we transfer it to that person doesn't really know what happened on the other side. Now we have one transfer. So now while the user is on the phone call we actually transfer it. The other person picks up and the EI summarizes what happened in the current conversation. So the person already knows something about it and then we transfer. Right? That's kind of the difference and that's what we use most of the time. — Love it. That's a great use case because one of the things if someone picks up the phone they have no context around the conversation. Much like an unprompted AI uh agent, you have what's going on. But say, "Hey, this is Bob. Here's what he's trying to do. Here's what he's struggling with. Um and if you could help him do this, this would be great. " And then you bring them in and they go, "Hey, Bob, nice to meet you. I see you're struggling with this. Let's help you get across the finish line. " And then that way it it feels like there's been a you know it's not like the most frustrating thing in any business with these things is when you get transferred like you're calling a credit card company or something. You get passed from one company one service to another service and you got to keep giving them your card number. your pen. You got to keep giving the same like man I already gave you all this stuff and I have to do this one more time. I'm going to be so upset. So the warm transfer I think is a really understated value ad that you can do to to keep the customer happy and engaged as you after you shift them from AI to human. I love it. — Yeah, exactly. It's those small little things that you add on top of it that people usually don't think of when they build a voice agent. But this is exactly these few percent that make a difference. — 100%. Now uh shifting out into uh basically I'll say text stacks but overall structure right what are the common so we talked about this like say vappy nad in air table right what are the different elements that you need in order to set up these voice agent infrastructures — it can it can get really complex but

### [38:00](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=2280s) Reporting and dashboards for voice AI performance

I'd say especially if we talk about beginner things the best possible setup you can have as a beginner for building voice AI agents is NADEN, Vapi and Air Table. These three literally can get you mostly anywhere, at least in the beginning, to build our voice agents. If you have more sophisticated needs for custom infrastructures, we for example, we build a whole community around that where we just educate those people in the voice space and we basically hand them out to other people that actually need those uh kind of this kind of support. So, this is I would say probably one of the easiest things. Yeah, go with Vap and Air Table. — This is this has been super helpful and a and a great communication. Um, is there anything else you'd like to let people know about um before you tell them how to get hold of you? — Biggest one is probably if you're interested in learning about voice AI or knowing why voice AI or what voice actually does because that's usually not even clear. I highly suggest just looking into any kind of either your own company or into friends company and see if they have phone interactions and just ask them how happy they are with those phone interactions or if there's anything that can be optimized. Majority of times there is plenty of need and I think this is literally one of the initial things that you should do because that helps you to see how much demand there's actually out there. Everyone, everyone whenever they join a space, they think it's completely saturated because they get content all around this thing. It's kind of the same. You try thinking about a blue elephant or not to think about of a blue elephant and you elephant. It's kind of like the same principle. You just get bombarded with the stuff that you're already interested in or that you think about. So you think everything is saturated and my biggest thing is just making yourself aware that you can leverage or you voice AI is at a level where it can bring a lot of value to businesses and a making it sure for yourself to understand so that you can tell them that they understand because this business opportunity is massive and you can make quite a lot of money for both yourself and the business which is in my opinion one of the best things out there especially given that voice AI is the quickest medium we have nowadays to communicate with technology. That's a fact. — Speaking uh the way that humans normally do, using their voice to communicate is one of the fastest things and it's a growing marketplace. Love it. Uh so with all that being said, uh could you please let people know how they could find you? — Yeah, sure. Uh easiest is probably on YouTube. Uh Jiannis Moore, J A N I S M O R E. I push a lot of Voice AI content there. Or if you are a school member or would like to learn more about Voice AI itself, you should look into Voice AI boot camp on school. That's probably the best places to find me. — Awesome. Yiannis, thank you so much for your time. I appreciate you being on the show. Have a blessed and beautiful day, my friend. And I'll see you on the other side. — Thanks. I appreciate it. Wish you the same, man. — All right. Bye now. — Bye-bye.

### [41:00](https://www.youtube.com/watch?v=5nHyNpa-k7I&t=2460s) Sales training and AI-driven analysis

—

---
*Источник: https://ekstraktznaniy.ru/video/15201*