Welcome to our echowin getting started guide. echowin is a versatile platform built for beginners and advanced users alike. As a platform, echowin is capable of automating almost all types of inbound calls. In this guide, we will look into how we can get an agent trained for general business use case, using a fictional Gym called Boulder Fitness as an example.

Introduction

Before we start, lets look into the fundamentals of how the AI agent works. Your AI agent is able to understand natural language and can understand, reason and respond to callers just like you would. There are three core parts to this:

Knowledge: Your AI agent can learn about things like your business, your products, pricing and everything else using either References or Web pages.

Workflows: Your agent can perform actions using workflows that we call Scenarios. They are composed of multiple tools like "Ask Question", "Send Email" etc. which can do powerful things combined together. Your agent knows when to start a workflow based on the description of the workflow and what the caller is asking for.

Coordination: You can coordinate or script the call flow using instructions. You can instruct the AI agent to handle the calls a certain way, or tell it to call workflows in certain cases.

For the purposes of this guide, we will train the AI agent to be able to answer any questions the callers may have about the gym such as business hour information, available amenities and more. We will also setup some scenarios to automate a couple of workflows:

1. When the caller asks for the address of the gym, we will tell them the address and offer to have the gym address texted to them.
2. When the caller asks to talk to someone, we will collect their name and callback number and have that emailed to our gym manager. Then we will tell the caller that someone will call them back soon.

We will also instruct the agent to not answer any questions related to health or fitness advice. Let's get started.

Teaching Using Website

The easiest way to teach the agent about your business is through the website, if the business already has one.

1. First, lets go to the training page of your agent by clicking on "Training" on the left sidebar of your client portal. Then we will go to the "Web Pages" tab on top to get to the web pages section.
2. Click on the green "Add" button on the top right.
3. Let's give the webpage a name, we will just call this one "Homepage".
4. Since this is a fictional business, I am just going to use another gym's website I found Google. Fill out the address and click on the button at the bottom to finish.

After a few moments, you will see the information the AI agent gathered from the website. At this point, your AI agent is ready to start taking phone calls. But lets keep going and add the other things we talked about earlier.

Adding A Reference

Our website did not have information about our newly installed sauna. Let's add that information using References. Think of references as your knowledge-base items.

1. Click on the "References" tab on the top.
2. Click the green add button on top right.
3. Fill out the question and answer as shown on the screenshot and hit save.

That's it, our agent now knows about the sauna and can answer the callers with information about it when asked. Feel free to add other references for things that your callers may ask about.

Add Instructions

Now, let's instruct the agent to not answer anything related to health or fitness advice.

1. Click on the "Instructions" tab on top.
2. Add instructions as shown on the screenshot below and hit the save button.

Now your agent knows not to provide health or fitness advice.

Using Cubie

Let's keep going and create a scenario to handle the address of the gym. For this, we are going to use Cubie, our AI solution developer friend that can build this for us.

1. Click on the "Scenarios" tab on the top.
2. Click on Cubie button on top of the scenarios page.
3. Describe the scenario to Cubie just like a person. It will ask follow up questions for the details, answer them.
4. Once Cubie has all the information needed, it will queue the scenario creation. After a few minutes it will be ready. Head on to the "Scenarios" tab on top to see your scenario.

That's it, your agent can now automatically offer to text your callers with the address when needed. Cubie even used an advanced tool called Evaluate to evaluate the caller's response to decide if the user wants the address texted or not.

Creating Scenario

Let's go ahead and create the other scenario that handles the case where the user wants to talk to someone. We could definitely use Cubie, but let's make this one manually.

1. Click on "Scenarios" tab, then "Add" on top right of the scenarios page.
2. Enter "When the caller asks to talk to someone" as the scenario description. This description is what helps the AI agent decide when to enter the scenario. You can also make the scenario run during specific hours so that your agent can handle things differently during office hours and off hours, but we will not do that for now.
3. You will now be in the scenario view. Let's add some tools.
4. Click on add on top right of the scenario builder and add the "Ask Question" tool. This will let us ask a question to the caller and store the answer in an identifier. Identifiers let you use the answer from one tool in another tool within the scenario.
5. Check the box that says "Auto rephrase based on context". This will let the AI rephrase the question to make it more personalized. You do not want to enable this option if you want the agent to ask the question exactly as asked.
6. Also check the box that says "Use AI to extract response". This will let the AI normalize the response to match the question. For example, when we ask "What is your name?", the caller may say something like "My name is Rob". Checking this box will only extract "Rob" and not the "My name is.." part. This also normalizes dates, emails and other information.
7. Add the question to ask the name as shown on screenshot, and use "name" as identifier so we can use it later to send an email.
8. Click "Save" to save the scenario.
9. Connect the "Scenario Starts" block with the newly added tool.
10. Do not forget to make the scenario "Active" using the switch on top right.
11. Using the same exact way as before, add a "Send Email" tool.
12. Fill out the details like shown on this screenshot:
13. As you can see, we used the name identifier from the "Ask Question" step using $name. When the email goes out, the $name gets replaced with the caller's name they provided during the call. We did not even have to ask for their phone number as there is a built in identifier $call_from that provides this information. Feel free to add a new "Ask Question" tool if you would like to explicitly ask for a number. Click save and connect the tool with the rest.
14. Last but not the least, let's tell the caller that the email has been sent. In the same way, add a "Speak" tool.
15. Make sure everything is connected. Feel free to check the option to rephrase like before.

That's it, our new scenario is ready! Once the scenario is finished, the AI agent will ask the caller how else it can help and continue the conversation, no manual steps are necessary.

Configuring Greetings and Voice

Now as a final step, lets change the initial greeting message that plays when the call is picked up by your agent and pick a voice. 

1. Click on "Configure" tab.
2. Change your greeting message to whatever you would like it to be.
3. Pick a voice that matches your brand the most. The "Recommend" section is full of good choices. Be careful with the experimental voices as they may cause the agent to respond a bit slow.
4. Feel free to check the other settings on this page and change them to your liking.
5. Hit save at the bottom.

That's it, your agent will be pretty hands off at this point. Feel free to go back anytime and update your agent as your business needs change. We have not even scratched the surface in terms of what the platform is capable of, as you can integrate with over 6000 apps with Zapier, get information from third party sources and use it during the call using webhook calls and much more. But I hope this was a good starting point for your journey ahead. Feel free to reach us anytime at [email protected] if you need any help with setting up your agent. We wish you the best!