Introduction

Phonely lets you use voice AI to answer your phones.

But we do so much more than that. We give you the no code platform that lets you build AI voice agents with automatic websites imports into knowledge base (RAG vector DB), prebuilt integrations with your software and SOTA call monitoring and testing and reporting. In short, we give you the tools so that you can rest assured that you customers will get the support they need, when they need it.

Want to jump right in? Try out Phonely for completely free here.

Building an AI agent in Phonely


Step 1: Learn the concepts

Phonely combines a few different technologies together to create aI that accurately answers your phone. We have a few key components that make this happen.

1. Phonely transcribes your customers calls

  1. We use SOTA speech transcription models to transcribe what the users are saying. Sometimes, these models aren't perfect and can occasionally mis-transcribe information. Humans also do this over the phone, and just like humans, you can adjust what Phonely says when it doesn't understand the user in settings.

2. We respond using a Large Language Models (LLM)

  1. The LLM's are what actually react to your users message to generate responses to their queries. These LLM's get their answers from several areas, 1) the Phonely knowledge base 2) the agent guidelines you set in Phonely 3) the workflow stages you've built

  2. All of "how" the Ai responds can be configured in the agent design page in Phonely's platform

3. We generate a humanlike voice

  1. We rely on a a number of different providers that produce realtime speech from text in milliseconds (i.e. elevenlabs, cartesia, play.ht and others). You can change a voice, accent, language in the agent design settings

4. We analyze and display your calls

  1. Every call gets an AI summary, list of action items and an automatically generated transcript. These are stored in the call history page, and can be viewed or filtered by the call performance, number, duration or more.

  2. We also have other handy dandy features like the dashboard and call paths to help you view how your calls perform in realtime.


Step 2: Plan your agent

For businesses who answer over 2000 calls per month, we do white glove onboarding and agent build out! Talk to sales here

Just like a person, your agent needs to be trained. You can do this by configuring the agent design page in Phonely's dashboard to include all of the information about your business, software etc to build your agent. This knowledge base contains all of the information on how to do this, but first we still need a blueprint.

Fortunately, with Phonely's templates, we have taken a lot of the legwork out, and have probably built an AI agent for your use case that should need just a few minutes of tweaking! Check out the templates here

Even if you're using a template, the following is a helpful exercise to go through to help you build your agent:

  1. Define your 80% use case

    1. To do this, think about what 80% of your people are calling about. This is usually 1-3 things (i.e. scheduling/ rescheduling appointments). Jot down the questions that need to be asked, and the steps you would do to perform this task over the phone. These will be each structured as workflow later.

  2. Think about your software

    1. Oftentimes, having the AI integrate with your software is a crucial step of the process. When you're doing this, write down which software you need it to integrate live (during the call) and which software you need after. This includes SMS messages, emails, etc that need to be sent.

  3. Gather your knowledge base

    1. Phonely will integrate with a knowledge base to answer FAQ's via an inbuilt knowledge base. To do this you need to have a knowledge base handy. If you don't have one premade, you can type your own or connect a website

    2. Note: if you connect a website, it is handy to remove the unnecessary pages to prevent the AI from grabbing information that isn't relevant

  4. Think about your call forwarding

    1. To get your AI actually answering your calls you have to either forward your existing phone line to a Phonely number, port your number to Phonely, or give your customers Phonelys number. In virtually all carriers there is settings for this, and the easiest way to find out is to google "how to forward numbers on xyz provider"

  5. Start simple, nail one use case and scale from there.

    1. Oftentimes we see customers come with us wanting everything to be handled right away. Barring enterprise customers, this almost always ends up being a ton of work. To prevent this, we recommend trying to nail one use case, and set up call transferring or, a generic message taking for other use cases. This lets you get comfortable building out one workflow and you can scale incrementally over time


Step 2: Build your agent

Now that you understand how Phonely works, and have a plan, lets dive into how to actually implement it in the platform

1. Chose a voice and greeting message

Head to the agent design page to configure the voice and greeting message. The greeting message will be set as soon as someone answers the phone and can be completely customized to your business use cases.

Phonely has hundreds of prebuilt voices to choose from providers such as Elevenlabs, Cartesia, Openai and more. However, if you don't like any of the prebuilt voices, you can also clone your own to really be specific. Additionally, you can select and filter voices based on languages, genders, accents and more!

2. Set your agent guidelines

Your agent comes pre-formatted with agent guidelines. Agent guidelines will be something that the agent always has context to, across all workflows and when retrieving information from the knowledge base etc.

Agent guidelines should be behavioral related, meaning if you want the agent to act a certain way, in a specific manner, or just want it to understand what your business does from a high level. You can write in or change the agent guidelines in any way you please, and if you do it always a good idea to keep in mind good prompt engineering context.

If the AI was a human, the Agent guidelines would be the information general manners, business background and behavioral traits that the person had.

In more technical terms, agent guidelines are where you set the "base prompt" of the agent.

3. Import your knowledge base

Now click on the knowledge base tab on the part of your screen. The knowledge base can be imported from either

  1. PDF documents, word documents or other files

  2. typed manually

  3. Imported from your website

Phonelys knowledge base works using a vector database retrieval. Vector database retrieval works in a similar way to how a search engine works - meaning that all information you put in here is not immediately available to the agent. However, if you as a question that is similar to something in the knowledge base, it will retrieve this information and use it in the AI's response to the next question.

Knowledgebases can be tweaked and improved to optimize performance, and this is a service we provide to our enterprise customers.

3. Build your workflows

Workflows are a series of questions, steps, or interactions with software that an agent should follow. You can think of workflows as specific tasks that would need to be done over the phone, for example, every time you schedule an appointment you have to gather the same information - workflows allow you to create, and define this script while keeping the agent completely focused on the task at hand.

There are two types of workflows, inbound and outbound workflows.

Inbound workflows have a few key concepts:

  1. Start Flow block

    1. The start flow block lets the AI know when it should perform this workflow. When it meets these conditions, the AI goes directly into this flow and it will proceed through the questions.

    2. The most common use case is to set one flow, per distinct action. I.e. appointment scheduling, callback information, and call routing would be individual separate flows.

  2. Ability to transfer in between workflows.

    1. With workflows each are available at every point in the conversation, so if someone wants to trigger a different workflow later, or during the flow, then the AI can switch between those flows. This lets the AI act more like a person because it can change dynamically based on request.

Outbound workflow key concepts:

  1. Trigger condition

    1. Trigger conditions are the actions that are used to place a call. When these trigger conditions occur, they will immediately trigger an outbound call to be placed by Phonely

  2. Trigger variables

    1. Each trigger condition comes with its set of 'variables' this would be information gathered from the source. I.e. if you are using the Google Sheets trigger condition, the variables would be the headers of each column in your table (first_name, phone_number etc.)

    2. For every trigger condition we will use the dynamic information we receive from the trigger condition to place the call. For example this lets us customize the initial greeting voicemail message and other so that it actually captures the correct information

4. Test, deploy and monitor

After you've got y