Today on the Salesforce Admins Podcast, we talk to Elizabeth Awad, Associate Product Manager for Prompt Builder at Salesforce. Join us as we chat about how you can use Prompt Builder to simplify users’ daily tasks by integrating generative AI moments powered by prompt templates into their workflows.
You should subscribe for the full episode, but here are a few takeaways from our conversation with Elizabeth Awad.
There’s been a lot of buzz around Prompt Builder these days, so I was excited to get Liz Awad on the pod to pick her brain about it. She’s a product manager on the Prompt Builder team, so you could say she knows some things.
The first question I had for her is one I hear frequently from the community: What’s the difference between Einstein Copilot and Prompt Builder? “They go hand in hand,” Liz says. “The prompts that you create in Prompt Builder can be connected to custom actions in Copilot.”
In other words, Prompt Builder is where you as the admin create custom prompts to write a sales email or summarize a case, and Einstein Copilot is what allows your end users to invoke those prompts.
We’ve had more than a few guests suggest that you jump on an LLM to practice prompting but I wanted to know from Liz, specifically, what she’s found helpful. She had a really interesting answer, which was to try to get ChatGPT to write a birthday text to her mom.
If you think about it, that’s a perfect use case for practicing writing better prompts. It’s pretty simple in terms of information but it’s tricky to get the right tone. More importantly, you probably have a good sense of what sounds right and what doesn’t. And so when you give adjustments, you instantly know how well they worked.
Liz points out that this process is what people mean when they talk about “grounding” a prompt. It’s the extra bits of information you give an AI to adjust the response, like “that sounds too formal” or, “here are three other texts I’ve sent my mom.”
As a reminder, there are four prompt template types that you can use in Salesforce right now. They are:
Josh Birk and Raveesh Raina went over all of this in detail in their episode last month. However, Liz wanted to draw special attention to a new addition to the flex prompt template, which is the ability to use free text as an input. That means you can use something like a chat transcript or a case summary to ground your prompt. It’s super flexible, and the sky’s the limit.
There’s so much more great stuff from Liz about what you can do with Prompt Builder and how she approaches her role as a PM, so be sure to listen to the full episode. And don’t forget to subscribe to hear more from the Salesforce Admins Podcast.
Mike:
Earlier this year, we talked with Melissa Scalercio about Prompt Builder, because she was on the customer side and she had some pretty neat things to say about how they were using Prompt Builder. You can go back into the Salesforce archives there to listen to that episode.
In case you’re not familiar, Prompt Builder allows you to simplify users’ daily tasks by integrating generative AI moments powered by prompt templates into their workflow. It’s really cool. You get to bind a field and put sparkles on it. That’s what it’s called literally, sparkles. I wanted to talk to one of the product managers that is working on Prompt Builder. That is Liz Awad, who is based in New York. She is helping build the future of Prompt Builder. She not only gives us some really useful insights into Prompt Builder and how she’s seeing other customers use it, but also into what’s coming with Prompt Builder, and you may have already seen it. Then, fun fact, we get to learn a hobby. I bet it’s one maybe that most of you do. I’ve seen it on a ton of TV shows. But I’m not going to tell you because you have to listen to the episode. With that, let’s get Liz on the podcast.
Liz, welcome to the podcast.
Liz Awad:
Thank you. I’m excited to be here.
Mike:
Well, I’m excited to talk about Prompt Builder. We’ve talked about Prompt Builder in the past, with Melissa, earlier. I wanted to hear from somebody that’s actually working on the product and doing some really fun stuff, because not only is Prompt Builder a really cool tool for admins to use. We’re seeing that in all of our AI Now tours. But also, there’s a lot of really cool features coming as well.
Look, I’m getting the cart in front of the horse, like I always do. Let’s start off with you. Liz, how did you get to Salesforce and start owning Prompt Builder?
Liz Awad:
Sure. I’ve been at Salesforce a little over a year now. I’m actually part of a rotational product management program here at Salesforce. Previously, I was on Sales Cloud, and I was working on adoption. Which is actually admin relevant, as a new feature on Sales Cloud is launching soon called Sales Cloud Go. I was part of that development. Then I knew my manager, Avantika. She’s the one that pulled me into Prompt Builder, where I am now.
Mike:
Wow.
Liz Awad:
I’ve been working here for around five months now. I’m very excited. It’s been a great journey. Things have changed really fast, as you know, in the large language model space. There’s new models all of the time. We are, like you said, launching some amazing new features.
Mike:
Yeah. Just trying to keep up with … I remember when Einstein came out and we suddenly had to learn large language models as part of our nomenclature. Then now, we’re having to learn generative AI, and prompts, and grounding. My most fun word to say is hallucinations. Because you know, hallucinations, what could go wrong there? But that’s a different podcast that you can listen about.
Prompt Builder, what are exciting things that you can dug into and are happy to work on out of the gate? We’ll preface that by saying tell me how Prompt Builder is different than Copilot.
Liz Awad:
Yes, great question. Prompt Builder is a low code prompt management tool. This basically allows you to create reusable prompts that are being sent to these large language models. Prompts are just instructions that you send so that the model outputs the correct text.
The way that it’s different than Copilot, the way that we view it actually is they go hand-in-hand. The prompts that you create using Prompt Builder can actually be connected to custom actions in Copilot. Then within Copilot, when you ask Copilot a question, it can invoke the prompt that you have created. They’re different in that Prompt Builder is this tool that allows you to create the custom prompts. Copilot is maybe one area when your end users actually invoke the prompts.
Mike:
Yeah. Copilot’s a little more conversational.
Liz Awad:
Exactly.
Mike:
Yeah. That’s also one of the things that, as a former Salesforce administrator, I love about Prompt Builder is it’s just a little sparkle button that my users have to press and they get magic.
Liz Awad:
Exactly.
Mike:
It doesn’t depend on their ability to write a good prompt.
Liz Awad:
No. It’s the admins, now you all become prompt engineers. I will say you do need to spend some time reworking your prompts, testing your prompts. In Prompt Builder, you can even change the models. Certain models are better at writing sales emails than others. Depending on your use case, that Prompt Builder workspace is really where you can test. Just so, like you said, your end users just get to see magic.
Mike:
Yeah. Let me ask. What do you do to get better at writing prompts?
Liz Awad:
I was introduced to the concept of prompt engineering, like the rest of the world, when ChatGPT came out in November 2022. The first thing I did was just go to the website and start asking it questions that I knew the answers to already. Or I would be able to tell that’s a good response, and that’s a bad response. It was as simple as, “Write a text to my mom, wishing her a happy birthday.” When the LLM responds, it used maybe a formal tone, or something that I personally would never send to my mom. Then I’m like, “Make it sound more casual. I’m very close to my mom, so make sure you include that.” That idea of editing a prompt, and trying to have it fit your tone and your style can be used in an enterprise context as well.
The other thing to think about is, when you’re telling ChatGPT or your LLM more information, that’s called grounding. That’s giving more data to the LLM about the type of response that you want back. All of those things are really important in prompt engineering.
Mike:
Yeah. Well, I’ll be honest with you. The first time I ever heard the term grounding was when I was doing the Data Now workshop. We talked about building a prompt in Prompt Builder. I think the very first line was, “Acting as,” and then you could fill in $user.organization, however that worked. It was literally meant to pull the running user, which is you. I thought, “Oh, that totally makes sense.” Then you could pull in, “Here’s different information.” Because I think back to the way you brought in writing prompts, the rest of the world was just like, “Oh, neat. With ChatGPT I just ask it questions.” But here, we can actually pull in and say, “No, I want you to go into my Salesforce org and really pull that data out,” as opposed to just asking it for a text for your mom.
Liz Awad:
Exactly. That’s what really makes it powerful, is that it’s dynamic. What you said here is you can think of it as a variable, or a field, that’s user.name. When that prompt is invoked at runtime, that is replaced with the actual end user’s running the name in Salesforce. That prevents you from having to create multiple iterations of prompts.
When we say grounding, all we’re really saying is data. It’s just a fancy way of saying injecting more data into your prompts.
Mike:
Right.
Liz Awad:
With Prompt Builder, that data can come from a wide variety of places, not just your CRM. It can come from Data Cloud. You could use a flow to inject data. It can get really powerful, which is really exciting.
Mike:
If you think about it, if you really wanted to go back in time, this is where a time machine would help to tell our parents, “If I’m grounded, it just means I’ve got the data. Thanks.”
Liz Awad:
Yeah, exactly.
Mike:
“You’re grounded for a month.” No, I’m grounded for life because I’ve got data. That’s how you get more grounded. Just realized it, that dawned on me.
One thing that I did for the longest time. I love using the little sticky notes feature on my Mac. When I would write a good prompt for ChatGPT and it would turn something out, it was like, “Oh, awesome.” I would copy that and I’d hang onto it. Then I remember sharing stuff with my team early on. They’re like, “Oh, how did you do that?” I would share the prompt with them, the text.
Liz Awad:
Yes.
Mike:
Then they would plug that in, and they would get something different. I realized early on, with AI, the learning curve for AI is everybody’s different, as opposed to when we were trying to learn Flow Builder. When Flow came out, the admin created the flow, and then the record, or whatever the action was, worked the same for everybody. Then as I started looking at AI I was like, “Boy, this is going to be tough for users,”
because it really depended more on what is that end user’s ability to write a good prompt. Or how do I disseminate good prompts to my users?
That all changed for me when I saw Prompt Builder because we can put all that in there, which means the end user gets a solid experience every single time without having to be good at writing prompts. Right?
Liz Awad:
Exactly. The end user doesn’t need to be the prompt engineer because, as an admin, you are created those instructions every time and it’s following that template.
Now I will say you can have different outputs because it’s generative. Even if you’re invoking the same instructions, the response might be a little different every time you call the large language model. But it should be somewhat the same every time you call.
Mike:
Right. Obviously, if it was the same, then it’d just be like email templates.
Liz Awad:
Exactly. I do want to say, now that you’ve brought up email templates, just a small use case or a small story.
Mike:
Yeah.
Liz Awad:
One of our customers was telling us that they used to use the email templates. Now they’re using the sales email template in Prompt Builder. Because it’s generative, they’re delivery rate was much higher. As in Gmail, Outlook, and these email clients wouldn’t mark it as scam. It would actually deliver in whoever they were contacting, that inbox.
That was really interesting for us to hear. That was the first time we had thought about that metric of tracking delivery rate. And how, when you’re using generative AI, it’s not a pattern, it’s not a template. Those words can change every time you’re calling the large language model, with the same theme throughout.
Mike:
Right.
Liz Awad:
That was really interesting for us to hear on the product management team.
Mike:
No, I like that. That actually leads into the next question I had. Which was where are you seeing admins deploy Prompt Builder the most?
Liz Awad:
It’s a great question. It really depends on industry and use case. But we’ve seen a lot of field generation templates, which is what you described, where your end users see that magic button. You use a prompt template to fill in a field on a record.
Mike:
Okay.
Liz Awad:
We’ve also seen a lot of sales emails use cases. Case summarization use cases. It’s really across the board. The best part is, you as an admin get to be creative with it. Because you can really think of 100 plus use cases, that’s why we have flex template to allow you to customize and put in as many objects, or even now strings or free text, as input to a prompt.
Mike:
Tell me more about the templates. I forget, there’s four types of templates?
Liz Awad:
Yes, there are four types of templates. Sales email.
Mike:
Okay.
Liz Awad:
Field generation. Record summary. And flex templates.
Mike:
Flex sounds the coolest.
Liz Awad:
Yes. Flex, it’s in the namesake. It’s allowing you to create those custom templates where you can call an invokable action, you can use flows.
One feature that we just released is actually called string inputs or free text. Where, using a flex template, you can have text as your only input. Previously, we required templates to be associated with objects. But we heard from customers, “Hey, what if I have a transcript of a chat and I only want to use that as my grounding data?” Well, that’s why we just released this, actually this month. It’s very exciting.
Mike:
That’s interesting. Tell me more about that. Because I think I saw that in one of the release readiness’ that you did.
Liz Awad:
Yes. The use case here is that, if you want to invoke a template from a flow, or maybe using Apex, and all you want as the input is just text itself … Another example could be an agent gathers a question from a user in a chat, and you want to use that question as input into the prompt. Previously, you would still have to associate that template with a record, which would just be wasted time. You’re not actually using that record as grounding data. But now, you can just have that template with the input as a user question, for example.
Mike:
Interesting. Yeah. No, I like that. I think, in the earlier podcast we talked to Melissa, she was using the sales email stuff a lot. I’m curious, do you have any insight … Maybe this is just me, and I’ve worked in these overly micromanaged organizations where marketing needs to see every period, and comma, and I that’s dotted, which is why they loved email templates so much. You mentioned sales templates. Sales email templates. How are marketers adjusting to AI, where it’s not word-for-word every single time?
Liz Awad:
That’s a really good question. I just want to point out that we still recommend that a human’s eyes are on the responses from the large language models before they’re sent out. We don’t recommend automated generation and sending just yet. Primarily for that reason, just to make sure that either a marketer or a sales rep understands exactly what is being sent to their customers. It’s really just used as a tool to help, like an assistant, as opposed to replacing that job of writing the email. It gives you a baseline for you to go in and potentially edit or add certain things that you want, depending on the customer.
Now, like you said, some organizations are stricter than others. Maybe they’ll, in their company, have a process which is you still use the sales email template to start, but then someone either needs to check it before it’s sent. Or the admin can actually go into the audit trail and see all the generated emails, and someone can review those and make sure that that pass is their standard of quality.
Mike:
Sure, sure. Yeah. No, I like that. Plus, you could probably include some stylistic cues in the prompt when you’re building that as an admin as well.
Liz Awad:
Definitely. That is recommended.
Mike:
Write in your voice and tone.
Liz Awad:
Yes, voice and tone is definitely recommended. You really want to be as specific as possible. You can even give an example of a great email in your instructions.
Mike:
Oh! Well, that works out nice.
Liz Awad:
“This is an example of a great email we would want to send to a customer.”
Mike:
Interesting. I like that. There’s so many areas now, I think of Salesforce, to talk about with stakeholders as an admin. Not just features that are coming out, but it’s also … I know working with sales teams, thinking of I’ve always had that sales manager that would come up to me and say, “So-and-so writes great emails. How do we make everybody a great email writer?” Prompt Builder now has that ability. You can work with that great salesperson to get some of their great email ideas, and use those as some initial early prompts, that you can then disseminate to all of your sales team. Which, to me, I’m sure that salesperson doesn’t like that. Salespeople are a little competitive that way.
Liz Awad:
Yes, that’s interesting. That’s true, I never thought about that.
Mike:
As a product manager, I guess, admins do a lot of product managing. Is there any advice that you would give that you’ve learned in your year or so at Salesforce that would help them manage the onslaught of new features or how you balance priorities?
Liz Awad:
Definitely. I think the biggest skill that’s important in being a great PM is just communication, and over communication. And ensuring that everyone is aligned on not only what the priorities are, but why they are prioritized how they are. Because when you explain, “This is the use case, this is the problem, and this is the solution, and this is why it’s important,” then you really get that alignment.
As far as how to understand what’s being released at Salesforce and all the new features, that is a beast within itself. But that’s why we have release notes. That’s why we have all of the help docs. I would encourage you to even engage with your AEs and your SEs to learn more and inquire. As product managers on Prompt Builder and Copilot, we are actually aiming to talk to our customers as much as we can, because we learn from you all. Because you all are the ones that are engaging with the tool every day. That’s also what helps us prioritize, is engaging with customers.
Mike:
Yeah, absolutely. Too often, as an admin, I couldn’t communicate enough with all the features that were coming out.
Thanks for coming on the podcast, Liz. I will end by saying it’s always fun to know, outside of managing Prompt Builder, which is what admins love, if there’s anything fun you do as a hobby?
Liz Awad:
I’ve recently picked up, not recently, but in the past year or two, pickleball.
Mike:
Okay.
Liz Awad:
But I’m based in New York, so that requires me to fight over pickleball courts as there’s limited space. It’s actually turned into quite an expensive hobby recently, because they charge per hour and per court. But I love it. I grew up playing competitive tennis. So this has just been my outlet as I can play tennis and racket sports, but not be too harsh on myself.
Now my brother thinks that pickleball is an abomination. There’s really two groups of thought on that coming from tennis players. But I love it and it’s been really fun.
Mike:
It’s about the size of a tennis court, right? From what I know.
Liz Awad:
No, it’s actually much smaller.
Mike:
Okay. But about?
Liz Awad:
About, about, about.
Mike:
But much smaller. Okay. Okay.
Liz Awad:
Small enough that-
Mike:
My neighbors set up pickleball in their driveway.
Liz Awad:
Yeah. If anyone wants to challenge me to a pickleball game, I’m ready.
Mike:
You should set that up. Do a pickleball tournament in New York. That’s what we should have is a pickleball court at Dreamforce.
Liz Awad:
That would be amazing. Maybe I should go talk to the event team.
Mike:
That would be low impact sort of thing. But there’s shoes, right? There’s pickleball shoes. I know that you have to have a special paddle. Is it called a paddle?
Liz Awad:
Yes.
Mike:
Or is it called a racket?
Liz Awad:
I believe it’s called a paddle.
Mike:
Okay.
Liz Awad:
I’m actually not sure. Yeah.
Mike:
It kind of looks like a ping-pong paddle, just a little bit bigger.
Liz Awad:
Exactly. It is called a paddle.
Mike:
Okay.
Liz Awad:
You can get fancy with it. But you can go out in sneakers and be totally fine. You just have to watch your ankles.
Mike:
Sure. Yeah. Boy, I don’t know. Well, good luck with that because unfortunately, in New York, it’s probably also hard to have an apartment any kind of size at which you could practice pickleball. You should take up ping-pong. It’s New York size friendly.
Liz Awad:
Yes, exactly. Well, thank you so much for having me on.
Mike:
Absolutely. Thanks for being on, Liz.
Okay, I feel like I’ve been on a lot of television shows lately that have featured pickleball. Am I right or am I wrong? It’s just everywhere. I think there was even some sort of deal on pickleball equipment on one of the morning shows I was watching. Good for people playing pickleball. I don’t know, do you play pickleball? I don’t. But my mom wants to play pickleball. We’ll see how that goes. I do think it would be fun to have a pickleball court at Dreamforce. Don’t you? We should ask for that.
Anyway, I thought that was a great, fun discussion with Liz. I think it’s pretty cool, the stuff that admins can do without writing a single line of code in Prompt Builder. I wish I had this when I was a Salesforce Admin. Boy, it’d be really cool. Anyway, if you enjoyed this discussion, I did, I’d love if you could just share it with one person. If you’re listening in iTunes, or if you’re listening in Apple Podcasts, or any of the podcast apps, usually all you have to do is click the three buttons. There’s three buttons somewhere, you get a little arrow. You could share it on social, you could share it with your friends. You could text it, post it to Facebook, LinkedIn. I would appreciate any of those to help spread the word of the podcast. Of course, you can always give me feedback on the podcast if you go to iTunes. I’d love to hear what you think.
Now if you’re looking for any of the resources, links to anything that we mentioned, or just great blog post reading and podcast listening in general, that is on admin.salesforce.com. Including a transcript of this show. You can join with other admins in our Admin Trailblazer community. That, of course, is in the Admin Trailblazer group. I’ll include the link to that in the show notes. Until next week, we’ll see you in the cloud.
Today on the Salesforce Admins Podcast, it’s another deep dive with Josh Birk as he talks to Bobby Brill, Senior Director of Product for Einstein Discovery. Join us as we chat about how you can use Model Builder to harness the power of AI with clicks, not code. You should subscribe for the full episode, […]
Today on the Salesforce Admins Podcast, it’s time for a deep dive with Josh Birk, who talks to Raveesh Raina, Principal Solutions Engineer at Salesforce. Join us as we chat about what Prompt Builder can do and how to write effective prompts. You should subscribe for the full episode, but here are a few takeaways […]
Today on the Salesforce Admins Podcast, we’re replaying our episode with Sarah Flamion, Research Architect on Salesforce’s Research & Insights Team. Join us as we chat about what recent advances in generative AI mean for admins and the Salesforce Platform. You should subscribe for the full episode, but here are a few takeaways from our […]
