Building a No-Code AI Customer Support Agent

I'm super excited to be here and to show you something today. My name is Dennis Otto. I'm a senior product manager and AI builder. So for the last 10 years, I was working in different industries from SaaS, e -commerce to fintech.

Start Small and Focus on Business Problems

And let's face it, the AI world is changing every day. So there are like a lot of passwords out there, a lot of tools, new tools, every day something new that we actually should learn but at some point I always sometimes I believe we forget a bit about actually what we should do and it's about solving business problems and to solve business problems it's not always you don't have to have the most complicated technology for that so that

reminds me also about the book that I'm currently reading so it's called Ikigai I hope I pronounced it in the correct way from Ken Mogi. It's a Japanese philosophy and it's basically about finding purpose in joy in life and one of the key principles is to start small and that's what we're going to do today.

So we build something simple, something small and hopefully something that will inspire everyone to try it out themselves and just to start the AI AI journey or enhance the AI journey.

Tools and Data for the Demo

All right so to do so we are going to use three tools today so one of them is N8n, 11 labs and you need an API key.

Anyone has some knowledge how to get an API key from an LLM? All right I'm going to show that then anyway but before we start we actually need also some kind of data so

I prepared here already some kind of dummy data and what we're going to build today is a customer support agent so first of all it will be a chat version and in the second iteration I'm going to show you how to build like an voice

agent I prepared here the file so you can see we have some kind of dummy data so for the purpose of presenting and showcasing it so we don't stick to a database so our database will be the google sheet for today and here we have like different orders with email addresses and the status and so on and we're going to do that in in

n8n we're going to build the workflow and the workflow we're going to build looks like this one so basically we we trigger it through a a chat message.

So everyone who's new to NLM, you can sign up for NLM. There's a 14 days free trial and also 11 laps. You can do that at zero cost.

And also if you want to use an LLM, so you can go to OpenRouter. So it's like where you can get like some kind of models for free.

Creating an OpenRouter API Key

So you're just going to sign up for OpenRouter.

And then basically if you logged in or you have an account, you go to dashboard basically then you find on the right left hand side you find api keys and then here there's like the cta to create a new api key and then you can call it whatever you want

you can set the expiration date since there's no kind of money set or some credit card connected you can stick to no expiration date at the moment and no limits and then basically you can create one and basically then that's the key that you need to copy once and you have

to save it in a in a safe place or in a place where you can find it again because after you close the window you won't see that API key once again and

Building the Chat Support Workflow in n8n

then in NNATN if you're the first time in NNATN it's super easy to start actually to create a new workflow you click on create new workflow and you You have a blank canvas here.

And in the first step, we stick to, basically you set here the trigger, what should trigger the workflow. And we going to choose on a chat message, so we can do that for now.

And then we going to add an agent. And to that agent, we can create at the chat model. And as I showed you, like you can choose whatever you want. you can use open ai you can use gemini whatever in that part in that example we were choosing open router so you click you choose open router and for the people who do that the first time

you can actually see here i have already some credential set but if you're the first time user you create a new credential basically and that's where you have to copy paste basically the api key that we just saw on the other screen and then you just save it you will get the success message and it's connected since I have already the account we don't need

to do that and then we just add a simple memory because memory is needed so the LLM has some kind of memory about the conversation so there's already some solution for that in -house for NNN and then as I said we're going to use tools

we can connect tools what I said like there are many tools out there but for that purpose we stick to Google sheets and basically you just look it up Google sheets and edit and then you have basically the bare minimum of the skeleton that we need for the for the agent and so um not to go too much into the details how to set it up and so on so there are a few things to do to connect your google account but basically

Testing Order Lookup with Google Sheets

that's the agent that we have right now so going back to our sheet let's give it a try and ask about our order so let's do that hello I have a question where my order is and then you can see actually how the workflow is running so obviously here in in the customer support agent,

I gave also some instructions. So just like for the background. So I said like it should ask some kind of clarification questions, ask for the email address for the order ID. And I think that's basically what we got here.

So hello, I'd be happy to help you check the order status and so on. So we need to provide the order ID and the email address. And we can do that. We go back.

So I'm just going to copy paste that that and bring it to the chat and now you can see that's actually the cool thing about a night in because you see exactly how the workflow is executed and now you can see the agent the agent hasn't has a model connected that which decides how to solve the problem and you can see it's going to call our tool for the lookup orders and it's giving us probably in some seconds like an answer

where our where our order is and you can see it was a success so it gives us the estimated delivery time it will be October 23rd and with that we have already our first agent support agent that can fully cover like all the requests there's also like one thing I implemented here but maybe in the matter

of time we're going to skip that but basically it can also create the ticket so if you say like your item is damaged you can do that because we also added like a policy here as you can see in the file we i added just like some kind of company policies so it can also check like if it's like eligible for some refund or not and that's the and that's the first thing

Turning the Chatbot into a Voice Agent

and now as i said like i wanted to show you how to build the voice agent so to bring it to the next level we have to connect it to eleven labs and just to highlight you that so we know now the chat chat workflow is working and so I did a side -by -side comparison because we need to adjust a bit the workflow that we had before so that's the one we just we just executed and it basically had the tooling and it was triggered by a chat message but now one important thing to know so if you

Connecting ElevenLabs with an n8n Webhook

want to connect two tools you have to have an option where the first tool like 11 labs can call the NNN and tell like hey I have some information for you you need to execute the workflow and how to do that it's basically you do that through a webhook and there's also like an option in NNN where you can basically say the trigger of the workflow should be the webhook and when When you do that, you get like multiple options.

In that case, we need to have like the post request because we want to ping the tool and say, hey, listen, I have some information for you. You need to execute something. And then you get here already the URL for that. So that's the URL that we're going to provide 11 labs with

so that it can basically call and then tell like, hey, I have some user on the phone, whatever here, the information, execute the workflow and look up like where's the order where's I know where's the parcel and so now if you if

Configuring the ElevenLabs Agent

you have a 11 labs account you go basically to agents on the left hand side I hope it works here and it's basically the agent I already set up here but in case you want to create the new one there's also the option to do

that but here basically you have the probably some similar or familiar things that you know already there's a system prompt and also like some kind of settings that you can have for for setting the language and so on and actually the thing that i told you now we have to make sure that nn and 11 labs they can communicate to each other and that's where we need to add the tool to the agent and And here, I have added this already,

but if you're new, you can go on, add the tool, add the new tool, and then it's the webhook tool. And here, you can see you have multiple options. You have the get option, that's basically when you receive some information or you want to fetch some information,

but we want to send our information to NNN and tell them, hey, NNN, you need to do something. So we just go on post, and then we basically would copy paste this URL here into the into that form and there are a few things you need to fill out as well because here it's also the body parameters and that's important because

as I told you because 11Labs will now take care of all of the conversation so that's why we also don't need like this kind of simple memory but basically here we need to instruct 11Labs what kind of information it it should collect from the conversation. So I did that already, as I told you, so I can quickly show that with you, share that with you.

So basically here we are going to ask in the conversation about the order ID, about the customer email, and then in case there's like some problem with the parcel or it's damaged, we also have the message to identify what's wrong with the message.

Testing the Voice Support Experience

and yeah basically that's it what we need to to set up so i mean we can give it a try now and try to call the the agent and see if it can basically answer a question so just give me a sec to find some good um good example here so maybe we can also call it now and ask just for our order and see if it works.

Hello, I would like to know where my order is.

I'd be happy to help you track your order. Could you please provide me with your order ID and email address associated with your order?

Sure, the order ID is 1024 and the email is alex at example .com.

Let me look that up for you right away that's actually a feature you can have that you can add that i found your order it's been shipped via dhl and the estimated delivery date is the 23rd of october which is tomorrow your tracking number is dhl1024 is there anything else i can help you with today no that's

it thanks thanks a lot you're very welcome alex i'm glad i could help have a wonderful day and and ensure your order. Thank you very much.

All right, you can see, like, that's it. How can I help you today? Today, that's it.

You know, that's the fun part of it. It's still listening, Quade. It is? Yes. All right, that's the ultimate AI experience.

Key Takeaways

But you can see, so I hope I was not rushing too much. but yeah what i wanted to show you today so that's super easy there was no code involved and it was super easy to solve a real business problem like to have to create a customer agent or customer support agent to answer simple questions like we had it first in the in the

chat version and now as an um voice agent and it was really helping us to um to solve a business problem and in an easy way I hope this was fun for you guys I hope you learned a bit about it and you're going to experiment on your own but yes thank you very much for your attention