David Grant 0:08
OK, awesome. Thank you everybody for joining Ahmed and myself here today for AMC Technology’s Agentforce Voice Unboxed session. We’re really excited to walk you through what we’ve prepared for you. I know Ahmed especially has been putting a lot of work into making sure we can walk you through the process of enabling Agentforce Voice and show a few use cases at the end.
So, a couple of introductions. My name is David Grant. I’m an Enterprise Solutions Manager here at AMC Technology. I work with current and prospective customers to look at how we can use our DaVinci Toolkit to solve their unique business use cases. I’ve been here about seven years, worn a few different hats, and I’ve had a lot of fun meeting a ton of different people and helping solve some pretty cool problems. Ahmed?
Ahmed Mahmoud 1:15
Hello everyone. My name is Ahmed. I’m a Solutions Engineer here at AMC Technology. I’ve been with AMC Technology for about a year and a couple of months, and I’ve been having a lot of fun working with everyone here.
David Grant 1:28
Awesome. A few housekeeping notes: We are recording this session. If you have any coworkers or anyone you know who wanted to attend but wasn’t able to, we’ll be sharing the recording, slides, and some follow-up resources after the session.
Please post any and all questions in the chat. I’ll keep an eye on it while we’re going, and we’ll be sure to answer anything that goes unanswered at the very end of the webinar and demo.
Also, this is live. We’re configuring this live. It’s incredibly possible—and dare I say likely—that we’ll hit a little hiccup, because we all know how live demos go. If we do, that’s totally OK. That’s part of the process. It’s brand new to a lot of us, and anyone who’s worked with these tools knows there are things you can’t always plan for. We’ll work through any hiccups live, and that’ll be part of the fun.
Anything from you, Ahmed, before I keep going?
Ahmed Mahmoud 2:58
No, sounds great to me. You can keep going.
David Grant 3:00
All right, awesome. While I work through these few slides, if you’re just joining, please take a moment to answer the poll in the chat. We’re curious: are you using AI tools more at work or more at home? I voted “home” because I am not a culinary mastermind—however, I can follow a recipe, and ChatGPT and other tools have given me some great dairy-free waffle and pancake recipes. So I voted home. What did you vote?
Ahmed Mahmoud 3:41
I voted “home” as well. It helps me plan a lot of things I want to do, and it helps me figure out the best ways to do them—so I use it more at home.
David Grant 3:54
Awesome. I’m not solving many problems except my appetite, I guess.
All right—moving on. Here’s what we’re going to cover today. We’ll show you where you go in Salesforce to configure Agentforce Voice. It’s possible you already know this, but it doesn’t hurt to start at the ground level.
We’ll show you how knowledge sources are connected—what your agent will be pulling from to provide the answers you’re looking for. We’ll also touch on how CRM data grounds and informs the conversations that Agentforce Voice (or any Agentforce tool you’re building) is going to use. We all know this, but the data out is only as good as the data in—so we want to make sure CRM data is clean and in a good spot.
We’ll also cover how routing and escalation work—if the agent is unable to solve the problem, or if the caller just wants to talk to a person right away. What does that look like? How does escalation work?
And then Ahmed will show a real live test using a business scenario we’re all too familiar with this time of year. We’ll be focusing on filing taxes in this demonstration.
Now, what steps will we go through?
First, we’ll enable Agentforce Chat and Agentforce Voice. We’ll assign permissions to ensure they can access everything needed, and we’ll configure our Salesforce user.
Next, we’ll configure the voice stream we’re using, assign phone numbers, and set up routing.
Then, we’ll create the agent. Ahmed will show topics and actions, how to configure them, and where to connect data sources and knowledge bases.
And finally, we’ll run the live test.
Anything to add here, Ahmed?
Ahmed Mahmoud 6:41
No, we’re good to go.
David Grant 6:43
Awesome.
Now, the situation we find ourselves in—the problem—is that it’s tax season. In our scenario, the people we want to call and talk to are busy.
Ahmed works for Tax Filing Experts LLC. He’s underwater with calls—people want to set appointments for every question they have, and the questions are repetitive and can be answered quickly.
So what Ahmed wants to do, as CIO at Tax Filing Experts LLC, is build an autonomous voice agent that will answer frequently asked questions and automate scheduling, confirming, updating, and canceling appointments—so the tax preparers and CPAs can spend their time on higher-value work, like deductions and maximizing refunds.
So, without further ado, I’m going to hand it over to Ahmed, who will walk us through creating this autonomous voice agent using Agentforce Voice, and we’ll do a live demo at the end. Ahmed, take it away.
Ahmed Mahmoud 8:38
Thank you, David, for walking us through the scenario. Now we’re going to dive right into it and start building the AI agent.
To start, we need a Salesforce org with the appropriate licensing. We’ll need the licenses required for Agentforce Voice, and we’ll need user access so the agent can interact with the objects it needs to use to create appointments.
Here I’m in my Salesforce org. I’m going to navigate to Agentforce Setup.
Here we can see the Agentforce Voice setup steps we need in order to configure and build the agent:
Create a voice-enabled agent
Create an Omni-Channel flow to route work items to the agent
Create a phone number and add it to the channel
Create the messaging channel
Route incoming calls and configure escalation in case the agent can’t solve the issue and the caller wants a human
Next, we’ll start with Step 1 and click Agent Builder…
David Grant 10:40
Now, Ahmed—be honest with us. How many times have you tried to create the agent without having that toggle turned on?
Ahmed Mahmoud 10:47
It may have been once…
Ahmed Mahmoud 10:51
Let’s go ahead and start building the agent in Agent Builder. Here we can either create an agent from a template or create one with GenAI. In this scenario, we’re going to create it from a template.
David Grant 10:51
OK.
Ahmed Mahmoud 11:06
From a template, we want a service agent because we need an agent that can make changes to records and perform tasks in Salesforce.
David Grant 11:21
Ahmed, is there a reason creating from a template is more advantageous than using GenAI in this scenario?
Ahmed Mahmoud 11:31
Creating with GenAI may be faster. However, when creating from a template, we can tailor it more precisely to our specific use case. I’ve tested this previously and found it gives more consistent results for what we want to do.
David Grant 11:46
Awesome.
Ahmed Mahmoud 11:47
So I’m going to click Next.
David Grant 11:50
Better results using the template. Awesome—thank you.
Ahmed Mahmoud 11:57
Here we have pre-built topics and we can add or remove them. For our scenario, we don’t need these topics. We’re only going to leave one: General FAQ.
Just as a heads-up, a topic is a defined category with a defined scope. When the user asks about something, the agent identifies the topic and follows the steps and instructions relevant to that topic.
So I’m going to uncheck all of these and leave General FAQ, then click Next.
Now we configure the agent: name, description, role, company info, and assign a user. To save time, I’ll be copying and pasting some prompts.
David Grant 13:10
Have you found through your testing that certain prompt wording works better than others?
Ahmed Mahmoud 13:22
Yes—more context is better. It reduces ambiguity and helps the agent understand exactly what you want. There’s also a feature in Salesforce that uses AI to review your wording and suggest improvements.
David Grant 13:32
Cool.
Ahmed Mahmoud 13:41
We’ll look at that soon.
David Grant 13:46
Awesome.
Ahmed Mahmoud 13:51
Here I’ve provided information about the company, what it does, what the agent’s role is, and I’m assigning a user with access to the objects we need—so the agent can run flows and perform tasks. I’m going to click Next.
Now we select a data library containing the knowledge the agent needs to answer FAQs.
David Grant 14:29
And that data library is loaded, in this scenario, with information people would normally ask when trying to schedule an appointment for taxes or get advice on filing.
Ahmed Mahmoud 14:50
Yes. The data library contains a PDF with the knowledge we need for Tax Solutions Experts—appointment information and tax filing information.
It’s straightforward: you create a data library, upload the PDF, and then the agent can pull information from it to answer questions.
David Grant 15:31
Cool.
Ahmed Mahmoud 15:33
The data library is ready. Next, we’ll test the General FAQ topic.
I’ll ask: “When is the tax filing deadline?”
It says the standard federal deadline is April 15th, with exceptions for weekends/holidays.
This answer is coming from the PDF. You can also see the reasoning panel showing what the agent is doing—topics selected, what it’s executing, and where it retrieved information. It selected General FAQ and pulled the answer from the knowledge base.
Next we’ll create a new topic for managing appointments. I’ll call it Tax Service Appointment, with the description: “Handle requests related to scheduling and maintaining tax service appointments.”
When you provide a description and context, Salesforce can generate scope and instructions automatically. Here, it’s generated instructions like: if a customer requests an appointment, provide dates and times, etc. This looks good.
If I click the prompt quality checker, it says it looks good—nothing to improve. I’ll click Next.
Now we’ll add actions. These actions will run flows that create, update, retrieve, or cancel appointments, and verify customer identity.
There are prebuilt actions, but for our use case we’ll use custom flows, so I’ll click Finish without selecting the prebuilt actions.
Now I’ll select the Tax Service Appointment topic, go to Actions, and create new actions from the flows I built.
An action can be Apex, an API, a flow, a predictive model, or a prompt template. In our case, we’re using flows that create records for the appointment object.
First action: Customer Authentication (verifies identity).
The action instructions are already populated from the flow description.
Loading text: “Verifying your identity.”
Inputs: email, full name, phone number (required).
We’ll instruct the agent to prompt for each of these.
Outputs: contact ID (we won’t display this to the user).
Click Next.
Now we’ve created the first action (identity verification). We need four more: schedule, update, retrieve, and cancel appointments.
Next action: Schedule Appointment
Inputs: date/time (required).
Contact ID is stored but not entered by the user.
Output: confirmation number.
Loading text: “Scheduling your appointment.”
David Grant 22:04
So when someone schedules an appointment, Agentforce will respond with something like, “Great—we’re scheduling your appointment,” while it works.
Ahmed Mahmoud 22:16
Yes, exactly.
David Grant 22:17
OK, cool.
Ahmed Mahmoud 22:26
Next action: Update Appointment
Loading text: “Updating your appointment.”
Inputs: confirmation number (required), new date/time (required).
We will show the new date/time to the user afterward.
Click Finish.
Next action: Retrieve Appointment
Loading text: “Finding your appointment.”
Input: confirmation number (required).
Appointment ID output stays hidden.
Finish.
Last action: Cancel Appointment
Loading text: “Canceling your appointment.”
Input: confirmation number (required).
Contact ID is stored but not asked for.
We won’t display internal output values.
Click Finish.
Now we have all the flows needed for the appointment management topic. I’ll refresh and test.
Question: “What documents are needed to file taxes?”
The agent lists typical documents: W-2 forms, 1099 forms, photo ID, Social Security card, prior returns, deduction records, etc., pulled from the knowledge library.
Next: “Can you help me file taxes?”
It responds that it can provide general guidance and help schedule an appointment.
Now: “I would like to schedule an appointment with a professional.”
It asks for full name, email, and phone number to get started—running the customer authentication flow.
I’ll provide:
Name: Demo User
Phone: 804-123-4567
Email: demoemail@gmail.com
Now it asks for the date and time. I’ll say: “February 26 at 2:30 PM Eastern Time.”
It confirms the appointment is scheduled.
Now I’ll navigate to the appointment object records. We can see the new appointment, linked contact, confirmation number, and the provided date/time.
So we’ve completed step one: build and test the agent in Agent Builder.
Next steps: attach it to a voice channel by procuring a phone number, creating a messaging channel, and creating a flow to route work items to the agent.
To get started, I’m going to activate the agent. While that’s loading, I’ll navigate to Flow Setup.
We’ll create an Omni-Channel flow to route work items to the voice agent.
Click New Flow → Omni-Channel Flow.
Add an element: Route Work.
Label: “Tax Service Voice Agent Route.”
We’ll create a variable for record ID (Text, available for input), and use it as the record ID.
Select the Phone service channel, route to Agentforce Service Agent, and select the agent.
We don’t see it yet—likely because it’s not activated. I’ll choose Ignore and Activate.
Quick note: the warnings are about steps we’ll configure later. We can proceed for now.
I’ll save the flow as “Tax Service Voice Agent Route,” then refresh.
Now the agent appears in the list. I’ll select it, add the fallback queue, and save and activate.
Next, we set up a phone number.
Go to Unified Number Management → New Number.
Country: US
Type: 10-digit long code
Voice + SMS
Next → enter area code → select a number → review and save.
After refreshing, the phone number is added. Now we’ll add it to a voice channel.
Click New Channel → Name: “Tax Service Voice Agent.”
Developer name: “Tax_Service_Voice_Agent.”
Select the phone number and select the routing flow we created, then save.
(We hit an error due to a naming conflict.)
Let’s give it a different name.
Now the channel is created with the phone number assigned.
Next, we add the agent to the channel.
Go to Connections → New → Telephony.
Inbound routing is already configured. For escalation routing, you can configure escalation to a human representative if needed. For the sake of time, we’ll focus on inbound routing.
Now I’ll test from my phone. I’ll activate the channel and dial the number.
(Warning appears about missing escalation flow—again, we’re skipping that for time. We’ll ignore and activate.)
I dial the number…
Voice Agent (call test)
“Hello… When is the deadline to file taxes?”
Agent: “The standard federal tax filing deadline is April 15th…”
“What documents are needed to file taxes?”
Agent: “Common documents include W-2s, 1099s, photo ID, Social Security card, prior returns, and deduction records…”
“Can you help me file taxes?”
Agent: “I can’t file taxes for you, but I can help schedule an appointment with a tax professional…”
“Yes, please.”
Agent asks for: full name, email, phone number.
Confirms the email and phone number.
“Verifying your identity…”
Identity verified.
“What date and time would you prefer?”
“February 27th at 3:30 PM Eastern Time.”
“Scheduling your appointment…”
Appointment scheduled; confirmation number provided.
“I’d like to update the time.”
Agent asks for new date/time; updates appointment.
“I have an appointment I’d like to cancel.”
Agent requests confirmation number.
First attempt fails due to mismatch; asks again.
Second attempt succeeds; appointment is canceled.
End call.
So as you can see, through the voice channel, the agent was able to schedule appointments, update them, and cancel them. That removes repetitive administrative work from the human team so they can focus on more important tasks.
Now we’ll open it up for Q&A for a few minutes. If you have questions, please drop them in the chat.
David Grant 41:10
Awesome—thank you, Ahmed. That was really cool.
