How Agentforce Handles Candidate Inquiries on Autopilot

Eric Housh and Zack Terry walk through a real Agentforce implementation at a talent placement firm drowning in routine candidate inquiries. Status update emails were scattered across inboxes, LinkedIn, and phone calls, leaving the team no time to actually place candidates.

The solution was a service agent built to retrieve live application data from Salesforce, translate internal status codes into candidate-friendly language, and flag priority candidates for human follow-up. All without exposing internal data or requiring a human in the loop.

For the full episode video, show notes, and related resources, visit: https://loom.ly/ltU25Ns

Transcript

SPEAKER_00 0:09

AI in Action is brought to you by Fast Slow Motion. Our team helps growing businesses put AI to work with practical, scalable solutions. To learn more about how we can help you implement AI in your business, visit FastSlowmotion.com/slash AI. Welcome back to another episode of AI in Action. I'm Eric Hush, joined as always by my good friend Zach Terry, Director of AI here at Fast Slow Motion. Zach, we talk a lot about big, complex transformations, but today we want to back it up really to that crawl phase, that foundational work that actually makes a team's uh day-to-day life better. Absolutely.

SPEAKER_01 0:48

Yeah, I think we uh all want shortcuts to a fully automated business or, you know, something, something that's automated, right? That's that's really where we see a lot of gains, but it's hard to automate something if we can't articulate what it is we're automating. I mean, we talk about this all the time. And so uh, you know, sometimes the biggest win with automation can just be clearing something repetitive off of somebody's desk. So, in this particular scenario, the story that we will talk about today is about a talent placement firm that we worked with. They actually ran two brands out of a single Salesforce org, and they were just drowning in this whole where is my application types of questions and emails and calls and very routine questions, just quite a big volume of them.

SPEAKER_00 1:33

And you can imagine with a talent placement firm, especially if they're running two brands, that volume can just be overwhelming. So these particular clients caught themselves answering the same five questions over and over and over that and they that volume was so overwhelming they didn't have time to actually do their job, which is placing candidates in the pipeline.

SPEAKER_01 1:53

You basically had a talent manager who was personally fielding, I think they said something like 20 emails a week just for status updates, which doesn't sound like a lot, but you start to realize that those inquiries were scattered across personal emails, general inboxes, LinkedIn, sort of any of these surfaces that they have when when they're doing this type of work. And so uh it really stacks up over time.

SPEAKER_00 2:17

And you can imagine just with the tyranny of the urgent, the data in Salesforce probably isn't kept clean, isn't kept up to date. So the candidates are getting late or robotic or incorrect responses anyway.

SPEAKER_01 2:30

That's a very good description of what was happening here. They had this massive database, something uh approaching close to 60,000 contacts, uh, which again, that's that's not huge when you're talking about massive enterprise scale, but for a small business, that is that is a lot, right? Especially when you're dealing with those contacts being active and actually reaching out and communicating with you. So because all of their permissions were set up in the system, those records were sometimes invisible to certain search tools or to certain users that might need to answer those questions. So in some cases, they were flying a bit blind and they had this big bench of talent that they couldn't easily surface. So you can see how this can kind of all compound to really create a bottleneck and a bit of a problem for the business and the people that are working with these customers.

SPEAKER_00 3:19

So they come to us uh and and immediately we start thinking about how we can apply AI to this problem, how we can apply agent force to this problem. Uh, we ended up building a service agent uh for them. So let's let's zoom in there for just a second. How do we deploy that agent to make it more than just a glorified FAQ page?

SPEAKER_01 3:38

It wasn't even focused on solving the volume problem internally. It was focused on a good user experience for the candidate. And so if we think about some of these questions, when you're applying to be considered for a job when you're working with a recruiter, you are submitting an application and you're probably eagerly sitting there waiting to determine if you've been approved and which businesses are interested in speaking with you. And so um, you can you can imagine kind of being on your tiptoes, kind of waiting for something to come through. And so a lot of times they're reaching out by email if they haven't heard anything saying, what is the status of my application? And what's interesting about this business, and I think a lot of businesses will resonate with this, but they don't just store like human-friendly responses about these applications. They have like codes associated with them. So you might have something like stage 4B, and that could map over to something like they're waiting in the pipeline, they're still in consideration, but they haven't been accepted to anything yet. So they have 20 or 30 of these sort of status codes with certain messaging. Sometimes the messaging was even, they are more or less fired, right? Like they're not being considered for any job applications, right? But you don't want an AI agent to turn around and say, uh, hey, you're fired for this process, right? So you you really need something to be customer friendly. It needs to be externally facing. And so for first, we built a system that stored translations of their internal codes in a customer-friendly, external facing format, right? And so these are essentially saying, all right, this stage 4B that maps to this customer-friendly phrase. And so that might be you're still in process, but you can expect to hear from us in another week or two, right? Something that's really friendly, answers the question, reassures them that their application hasn't gotten sort of lost in the system, right? So that was kind of the first step. And then when we actually hooked up the AI agent to their system, not only was it able to go and retrieve the information about the applicant, which is stored in the CRM, it would take that status code, cross-check it against this translation in the background, and then return that result back to the user. And so you had two things there. One, you're getting an answer, right? You're not waiting for an email to get turned around. You have real-time access to the status of your application. And then from a business perspective, two, you have protection on the communication for how that information is going to be returned. And so we built in this uh sort of layer where that AI is not able to immediately respond with the hard-coded status value, the internal value. It has to go and get the translation. And the only thing that it's able to surface back to the user is that externally facing translation. So it gives a little bit of protection to the business and it gets an answer immediately to the candidate who's looking for the status of their application. So that was sort of the biggest piece that we built. And then we also built in the ability to flag a candidate as, let's say, a priority. So maybe they've been on the bench for a while. They've been waiting around for the right opportunity to line up with the businesses that this company works with. And if they happen to reach out, maybe they want to flag that candidate as priority or that they need to be followed up with using an actual human person, right? Who can go out and spend the time to speak with them and let them know that they haven't been forgotten, that they're still in the system and they're working to find them a good match. And so we gave the agent the ability to not only go and retrieve information, but it could also go and make updates to that CRM record in the system, which could then be used to flag a person to go in and understand that they were marked as priority and that they could go in and actually follow up with that person. And so really it doesn't take it's not super complex, right? Like this is this is ultimately a pretty simple use case. But what's unique here is the ability to translate some of the internal messaging to external facing communication, and then also having that agent have the ability to go in and actually make updates to records and flag when a human needs to get involved.

SPEAKER_00 7:55

Zach, let's talk about some of the decisions that made this project powerful. Why did we focus on status retrieval instead of just going the path of AI chat?

SPEAKER_01 8:04

Things like document retrieval and being able to answer questions, sort of an FAQ agent. And that's helpful, but for this particular client, the friction point wasn't the documentation. The friction point was these inquiries about candidate application statuses and other things specific to that application. So the short answer is it's because that was the problem, right? The problem that they were facing was specific to these inquiries and the volume that they were getting and the ability to handle those. And so we built them a system that happened to be an AI system that had the ability to do some of those other things, but specifically was able to handle these requests in a sensitive way with the ability to escalate it to a human when necessary. So the agent has the relevant context. It's a natural solution to reduce the volume of questions that they were having to handle manually through emails and phone calls. And it allows them to do that on a 24-7 basis. And so it's not just when business hours are active, it's able to answer those questions at any time. So if someone goes and asks about their application at three o'clock in the morning, they can get an answer instead of having to wait for someone to get into the office and open their email inbox. So I think really it allowed us to help protect the team's time while also giving candidates an immediate answer to questions in a matter of seconds, as opposed to waiting around for a day, maybe two days, depending on you know how backed up the team was, in order to get an answer to that relatively simple question.

SPEAKER_00 9:33

Zach, talk for a minute about we internally, you know, we we build a lot of Salesforce projects here. We advocate the crawl walk-run approach. In this case, this was very much a crawl uh project. Uh, this was a crawl phase that we rolled out here. What makes something like that successful versus a project that might balloon and just kind of get stalled out?

SPEAKER_01 9:52

They were realistic about what these tools can do. So they weren't trying to have this AI agent perform the entire recruiting cycle as an example. Instead, it was really just there to inform a candidate about the process. So, yes, they had structured FAQ documents that were ready to go, things like policies and benefits and office locations and questions that candidates might ask. Um, and they knew exactly which of those internal statuses were confusing to outsiders, which is why we're able to uh create a translation to provide the right wording. And they knew exactly what data was going to be required for the agent to perform. They could tell us where it was located and what it meant. And so all of that combined to really expedite the planning and the build phases. And so it's really a combination of a realistic expectation of what these tools can do in a certain timeline, along with the business preparation. They had the details ready. They knew how to answer the questions that they were being asked. They knew where the data was located, they had that solid foundation set up to where we could really take advantage of an AI solution to help them with this business process.

SPEAKER_00 11:08

And I guess the question that's probably in a lot of folks' minds out there is did it work? Did the email onslaught actually stop? Right? We're always gonna continue to get emails.

SPEAKER_01 11:15

But for this process, yes. They went from dozens of these manual status emails a week to I won't say zero, but a significant reduction. And so immediately after implementing this, once it was live on the portal and people were able to start using it, once people realized it was there, they definitely started to have a reduction in the number of emails that were actually going to their team. Because now the process for getting in touch with the team was no longer to send them an email or to try to pick up the phone and call them. It was to use this AI chatbot to get those answers. And then, of course, they can still escalate if they need to. But what we found is that most didn't need to because they just needed to get an answer to the simple question. Now, for the few where they had a more complicated question, absolutely. We're able to create a case in the system. We can have a human agent actually go in and receive that case and get in touch with that customer. So it's a little bit more high touch. But the main thing there is those simple inquiries. I mean, if if we think about, you know, if they're doing 50 of these a week, that's 50 potential interruptions that the team no longer has to deal with. And so now they're they're able to focus on, you know, the fewer problems that require more effort that come in, which leads to less backlog and the emails that they're having to deal with. And so um, you know, you asked me a very simple question. I gave you a very involved answer, but the answer is yes, it did reduce the number of emails for these simple inquiries and allowed the customer to get an answer uh quite a bit more quickly than they used to.

SPEAKER_00 12:43

Let's uh zoom out for a minute because it feels like the applications here go beyond just recruiting or talent placement. Who else should be thinking about this sort of agent for setup?

SPEAKER_01 12:52

A high volume, low complexity problem, right? And so if if you have situations with an external party and you're receiving some kind of high volume, be it an inquiry or a request, whatever that is, but it's relatively low in complexity, that's that's essentially exactly the problem that we solved here. So we weren't trying to solve low volume, high complexity issues. That doesn't mean that there's not a place for AI to do that. However, it takes a lot more planning and a lot of considerations around how you want to design something like that, right? So in this case, because it was very high volume but very low complexity, we were able to pretty quickly implement something. And again, I'll I'll mention why was this client so prepared? Well, they already knew how to answer the questions. They could point us to exactly where the data was. They didn't have huge gaps in their customer base in terms of um missing data in the system. They were very diligent about making sure that they captured this information. And so um, yeah, I would say, you know, in anything, any business dealing with something that's high volume, low complexity, maybe you are a logistics company and you get questions around, hey, where's my shipment? Or maybe you're a medical clinic who has to answer questions about lab results, um, there's an opportunity for, you know, something like a system like this that can go in and help with that. Now, of course, there are different considerations, different compliance issues to think about. Not every business and industry is the same, but I think if we distilled this particular solution down to the building blocks, it's the high volume, low complexity problems that we were able to tackle really quickly using this type of solution.

SPEAKER_00 14:34

Yeah, the if the data is there, right? If the data is in the CRM, there's no reason that humans need to copy and paste that into an email. We can easily connect um the uh consumers with that info.

SPEAKER_01 14:44

Fast, slow motion. This is where we focus on the slow part. So you really have to slow down enough just to map those internal statuses and clean up your permissions. And so for them, it it wasn't about just launching immediately and having the AI answer the questions exactly as their data was. We took some time to slow down, think about it, make sure that we understood how this was going to be used. And uh, you know, before we were able to really go fast, uh, this allowed us to slow down and make sure that we understood the problem in its entirety and then build a solution to tackle that specific issue.

SPEAKER_00 15:19

I love it, Zach. Love this story because it's a great reminder to us all. Uh AI doesn't have to be this groundbreaking, confusing, uh massive thing to be successful. It just has to solve a problem, it just has to be useful.

SPEAKER_01 15:31

I I don't know if you've heard this, but I think a lot of people will say, Well, I really want AI that can help me do my laundry or something like that, right? That that's kind of what we're talking about. This is like the laundry of the business. This is a chore. So automating a chore is a real win.

SPEAKER_00 15:43

Anyone that's out there listening, if you're ready to clear the chores off of your team's plate, want a partner who knows how to handle the technical heavy lifting, we would love to meet you. Head over to fastflowmotion.com slash AI. And uh, we'll see you next time. Thanks, Zach.