The AI Agent AMA (Ask-Me-Anything): How to Actually Make AI GTM Agents Work in 2026

That's a good point. There's a, there's a few questions about our agents specifically, so I'm going to dive into a few of them. There's a couple on Artisan, there's a couple on AgentForce. So let me try and categorize the Artisan ones first. How do we set it up? Are you doing your campaign so I get better output? Okay, cool. Alright. If I miss anything, Nuance, just tell me in the chat. Okay. Let me start with Artisan because this is our oldest one. This is the oldest, I would say, outbound one that is sending email versus like, you know, the Jason AI, which is more advice and support focused. So You can see here just at a glance that this is my, this is my campaigns. They call it campaigns. I sometimes call it subagents, to be honest, because it's a better way to think about it. I think part of Artisan is they do make it really easy to get started and use it. But I think sometimes they make it a little bit too simplified. Whereas like I've given this warning to other folks like that have talked to me about using Artisan specifically. I'm like, think of your campaigns as sub-agents. If you don't think of them, if you think of it as like a marketing campaign, you're probably gonna set yourself up for failure, to be honest, because each, each one I think of as a sub-agent because you'll see I'll click in to two different ones here. Can I new tab? No, I'll just have to go one by one. That's okay. I set them up extremely differently, so they're all trained differently. They're all different contacts. There was a question in the chat of, am I using their, um, what lead source do I use? I primarily use our contacts. Listen, even if it's been 6 months, our database is huge. Like, I'm literally never going to get through all these people. So I, I don't— even with all 3 agents and all subagents, I'm never going to get through everyone in the database. As much as I would actually like to. So I use CSV upload of folks we have, and I break them up. Obviously we can't do our whole database. My rough rule of thumb, and this is true for any agent, never put more than 1,000 people into your sub-agent or sub-campaign or whatever you want to call it. For whatever reason, that seems to be a limitation today. I will say the agents have gotten better at this. I used to say like 400 was like the limit of where I would start to see it kind of get a little wobbly of like the output started getting bad after like it was very varied after like 400 people, or sometimes it would hallucinate after like 400 people. And so I used to keep it tailored down. Now I can safely do about 1,000, so I try not to go over 1,000. Just as a— that's not a— I'm sure everyone, every vendor will tell you otherwise. That is just an Amelia rule that I see tend to work the best. My also 1,000-person rule at the moment is that I am putting people into very distinctive buckets because then, yes, you'll see in the next ones, my pitch and outreach and workflow are very different per person. So in this case, this is one for, let me see what the name of this one is. So these are people who went to one of our Digital AI Days and then it is getting them right now to try and attend the next SaaStray Annual. That's in May. And so how I have this campaign set up is obviously I CSV uploaded all the people with the, within that bucket, which was I think 1,000 people in North America. So these are all like the North American contacts. This you can ignore honestly. I know you can generate a pitch and like have it ingest your website. I would write it yourself. I wouldn't even ask Claude, like this is, you know, this is just using Claude for OpenAI, like, the best copy will be copy you already know works. Like, I literally took, like, our best performing marketing emails and stuck that copy in here. So, again, not that you need to rely on the AI for everything to be magic. I was like, let me go find what's worked already. Let me find what ads have performed. Let me see what people have clicked on. Let me see what led people to already buy a ticket. Okay, now we'll put that stuff here. It's that I literally just adjusted our website. I actually ignore that one. That's why it actually says sponsor even though this is a ticket one because that's a weird default that it just grabbed off one time I did it and it wasn't that good and then I just wrote over it. So that's one thing to note there is I take the best performing things, but then obviously you have to write it for the LLM to understand, right? So in this case of ours, and you do proof point and then solution. So you kind of have to like I have all the solution data points, right? But then you have to like work backwards to the problem. Okay, what's the, what's the pain point? Knowing, here's the other thing, you have to be very intentional about this because it's gonna use all this. This is basically what you're telling the agent you want it to say for this email. The, you are, this is the context of the email and you wanna give it enough pain points and then you wanna give it enough proof points. So you can see there's quite a bit here. There's quite a bit there so that no matter— it has an almost endless combination that it can then contextualize to this person and send them an email. So that's, that's again, a little bit of the nuance there. You wanna pick things you know people have responded to, things that are driving value, things that are driving action. Coaching points are more so universal. So this is less so per subagent or per campaign. This is more things that are like best practice for your agent. So again, things like, you know, I put the dates in here so it doesn't hallucinate the dates. If you have anything timely, you should always reference that. You'll see I'll switch to a different campaign in a second. It's a little bit different of a nuance. In this case, again, just on this ticket thread, I've done, you know, the calls to action are a little bit different. So in these calls to actions, right, these are all like, okay, do you want to attend Gastrianual, and then this is the, this is what it's gonna personalize to them. So again, this is pretty much on the max. I click everything except mutual previous job 'cause I've only worked at like 3 companies. I was like, that's not gonna happen. So I've put, you know, I've put, I think the default resource age is like 8 or 10 months. I always go down to 6 'cause I'm like, past 6 months, probably not relevant anymore, to be honest. And then you can go into your workflow. So then you decide from there, okay, this is an all email campaign with 4 steps, you know, some wait steps in between. You can pick the date times, whatever, in your settings. And then it's like, okay, who's sending this? It's me, but I'll show you a different campaign because you'll see what I mean by treating them as subagents. This is one I sent to a portion of our renewing sponsors. So folks that worked with us in 2025 about our upcoming AI Day that we're doing at the end of January. You all should come. It's, it's going to be on a Wednesday, so they're free. We've done these a couple times now this year. We'll do another one in Jan. The pitch again is a bit different this time. So instead of it being pain points and proof points about SaaStr attending SaaStr, this is about, you know, generating leads, pipeline, getting value as a renewing sponsor because these are folks who have sponsored annual previously. And so this is what I mean by training all your subagents. Again, I call it subagents. I don't call it campaigns because it's very intentional that I'm retraining each and every one of these campaigns, subagents on something else. Like this is a very specific audience, very specific CTAs and proof points that is only relevant to these people, right? Like these people are not going to get that SaaS or annual call to action. Those SaaS or annual call to action people to buy a ticket are not going to get this email because it doesn't respond, like it doesn't correspond to them. And so that's where this piece is a bit different. You can see Twina Voice is a little bit different because AI Day's in January. So I'm like, you don't have that much time if you wanna do this. And so the calls to action are a little bit different. The resources are about the same. This one I let be a little bit older, right? Because these are folks that we know. And this is how we're getting that output at scale. But I think an important nuance there I hadn't really shown before is just thinking of each thing as You know, if you need to think of it as a person, you can, but just if you're onboarding somebody new to do renewals, you're gonna train that person a lot differently than somebody who's gonna do outbound, right? You know, for something that's a lower ASP like tickets or a higher ASP like maybe your software is. And so that's a big nuance to remember there. Looking at the follow-up questions. Yeah, review somebody. So yeah, make sure you review the output. Yeah, I would say too, the other thing I'll show you real quick just in this and then I'll go into some of our other agents is, this is my inbox. You can see I respond. This guy says, is this AI? Which I'm gonna respond, it's both. That's my usual response. This is my inbox of— you can see people literally, some of these are not purple anymore because I responded this morning. Some of these I've been on a couple calls I need to respond to. So each one though, I hand respond to. So I respond personally. Like this one was a sponsor. I just did the call actually with this person this morning. This is somebody who got that sponsor campaign I just showed you guys. Added in their team, want to book a call. I literally chatted with them this morning. And so there's different nuances like that, right? But then I have another one here where this person was asking about a ticket discount and it was a little bit nuanced. So I answered myself like all of these. I don't know how many are in this anymore. It's not gonna load all the way. 6 months back. All of these I have literally responded to myself. So there's a lot in there. All right, let me see if I answered all of your— hold on. Someone said, to what extent have you used agents to review other agents? We kind of touched on that, but I'll show you something interesting, actually. Because it will tie into my next little review. Used, not used. Okay, I think I hit that. Asking you to say if it's AI, we can cover that. Oh, there's a— sorry, there is a relevant question here. I'll show you an artisan. There's a question in the chat. Do you— what are the nature of the messages, which I've just shown a bit, which is In this case, it's outbound in different buckets. Do you see more success in some call to actions or types versus others? So it's a good question. See, let's see if the flows— oh, this is the last 30 days. Let me try and do all time. Obviously, this is live, obviously, because my Wi-Fi is a little slow. Okay, so this is not made up. This is how many messages Artisan has sent. So out of that 50,000, it sent about half. That makes sense. Um, Agent Force is our newest. Qualified does inbound. So that's pretty much like if you think of it right now as a split of, maybe not half, it's like a 40, 40, 20 split. Um, This is about 40% of it. You can see I used to do— they have a feature you can do LinkedIn messages. I turned it off. I mean, it's not that it didn't work. It actually worked too well where I don't— I'm not— I'm seriously not great at checking LinkedIn. As some of you will know from asking me questions there and I respond like a week later. I've tried to get better at it. I'm getting better at it. It's a human flaw, I know. But it was too much on LinkedIn and so I turned it off because I'm much better responding to email. I also saw on LinkedIn, because of how we were prospecting people in these two buckets, people would ask really weird questions on LinkedIn. So, I went away from it 'cause I was like, it's just opening too many cans of worms. So, many of these, so you'll see, you know, 22,000 messages across 8,400 people. How many responses have we gotten? So, a little bit more than half are positive, which is what you would expect. That's true, I think, of most outbound. Some negative. Fine. I will say some of the negative, it's a little weird how it skews it because sometimes it will skew it negative if it's something like, hey, I can't come to SaaStr, and then they say something like, oh, I can't come because I'm not in that geo, or I can't come because I already have something on that event date. For us, it's counting that as a negative, so just keep that in mind. For me, that's not always negative, like they did respond. I would say probably I probably have had like 30 actually negative responses of folks that are like, this is a terrible AI. And I'm like, okay, that's okay. But I'll look at their email to see if it was really bad or not. Or I get folks who are like, I don't, I just don't wanna take part in this whole AI thing. And that's fine. Or they roast it and that's fine too. I'm like, you can roast me. My agents are always trying to get better. It's good to tell it so that it learns. But okay, just in the, in the question there of how I seen some campaigns do better than others. Yeah, but it's also, again, somewhat intentional. So all these people that you're seeing here that are just in, you know, this, let's say, outbound general AI SDR agent, are warm outbound. So these are people that, you know, did something with us previously. So the one I just showed you guys a deep dive on, on renewal sponsors. Obviously those are folks and companies that know us already and did something with us big in 2025. The AI Day attendees, this one's a little bit in the middle but still warm of like, okay, they were newer to SaaStr when they came to AI Day in October, the last one we did. These are US North America attendees because we're doing SaaStr Annual next. But SaaStr AI Day is free, right? So it's kind of a mixed bag sometimes on like, the quality of those attendees and those names. And so these are folks who are kind of newer, like warm, but not as warm as the Renewal sponsor. So it's actually interesting to see that this one is doing the best right now out of our positive response rates. Then you go to things like this where like, this one is 7%. So that one did like, I don't think 7% is bad. I think average for outbound is like 3 to 5. So 7% is not bad. Our overall, you know, average is 5%. So I'm still happy to see these numbers. But again, I'm very intentional in what contacts I put into these campaigns, how much I train each of these sub-agents. I do reply to every single one of them to keep it top performing. So I definitely did not just like, again, I do think one of the like double-edged swords on Artisan in particular is It's almost too easy sometimes to like look at it like a marketo or an outreach and say, okay, I can just load up my contacts, put them into like a couple different campaigns and let it run. Like, that's not gonna work for a lot of reasons. So a lot of this takes time. Like, I spend more time outside of Artism and any of our agents organizing our data than I do in the platform. Like, usually by the time I get to the platform, I already know what I wanna say to them. I know what value I wanna add, or I know which CTA I wanna try and push them into. And then I go from there. So it's really like almost the last step by the time I actually get into the platform. There's another question here on like, how much have you used your agents to review other agents? So a couple interesting things on that. So in this, let me just go back real quick. Do you give me Let me pick one that's not running. Or do I want to pick one that's running? Let me pick one that I think is kind of in both. I don't know if that one's a previous sponsor one. I'll just pick this one. Or— okay, so I'll just pick this one. So there's a couple of things here. Obviously, this is all the training and tuning. I put into this subagent. Okay, I'm just gonna quickly scroll through them. So this one is about coming back as a sponsor to our SaaStr London event. Obviously this is not on anymore because SaaStr London's over. But an interesting nuance is some of these things like these, how this prompt was written, I put into a I rewrote the prompt and put it into Agent Force. I'm gonna show you guys that real quick. So let me go to— oh, it's 10:43. Damn, sorry, I gotta go faster to answer all the questions. Uh, hold on, let me pop over to my screen. I have a lot of them open. Okay, because one thing I'll show you, I have a— okay, so like similar to Artisan, I just showed you guys like this is the Artisan, right? I have a lot of— they call it campaigns, which I always have trouble going back to. It's the rocket. All right, so you can see I've got however many campaigns that is, 20+ there. I took some of that verbiage because Agent First was our most recent agent that we built and deployed. I took some of these prompts in here that were relevant. I'm like, okay, I've already figured out the pain points. I've already told AI, you know, I'll just click into this one real quick. I've already told the AI, you know, some coaching points of always put SaaS in the subject line, how to start the emails, how to break up the paragraphs. Like these prompts I literally copied over. So like, you may not need to copy over because you may not have multiple agents. But it's just an interesting thing of like once you nail the foundations, it's going to get easier to train your other agents.

Prompts are much more portable is the learning.

Prompts are very portable. And I also did the same thing with qualified.

I was like, it makes sense if you think about it. I mean, you can do it between Claude and ChatGPT and Gemini, but once you know, once you know something that works, that works well for the agent, just copy it to other agents.

Yeah. So like, if I can move my Zoom controls, I'll show you my AI Studio. Same thing here. Some of those prompts that are in there are in here. This one has a lot more because it's on sash.com, right? So like the other ones are doing very specific campaigns. The agent-first ones are doing very specific sponsor campaigns too. And so you don't have to copy every prompt, but I do think it's a good idea to have Like that stuff I just showed you guys in one, that is also in here. I have it just as like a fucking Word doc in my Google Drive of like master SaaS prompts that I put into any and all the agents that we use, especially the ones that are, you know, public facing. Because again, a lot of the times if I update one, well, we have a lot of agents, so I have to update all of them. But it is true that you can you can prompt from one or the other. So Qualified's also pulling in like all of our content. So it does this, I think a couple times a day. So you can see it did the 15th, the 11th. So yeah, a couple, every couple of days it's like refreshing all of our content that's ingesting to keep it also fresh. But things like snippets, these are things I've put in. So again, things I put in our article and I copy over here. Questions that people start to ask it, I start to put into, you know, all 3 of our agents, just frequently. It's also good— I think that's a, that's an interesting one I haven't really touched on yet because we, we haven't gone maybe this deep in the weeds of, um, I think it's really good to give any of your agents an FAQ. Like, just give it— like, maybe you don't have one yet. I didn't really have a formal one before all this started, um, but I made one for this and I started to put it in. Like, this one, you have to break it up per question. That's fine. So Um, but you have to— an FAQ is such a good idea, uh, because it will— you can tell your agent to always refer back to that. Like, you could say, hey, if you don't know the answer, always refer back to the FAQ. Versus if you don't put something like that into a prompt, it will make it up. I've— because ours has done that before. I did like FAQs. So I think it's interesting. I updated a lot. Um, it's something I update across, you know, all of our agents, but It is something where as I see people interacting with our agents too and asking questions and we start to get similar questions, I'll throw that into like its knowledge base so that I can always refer back to that. Like right now we get a question a lot about, you know, how to get a visa invitation letter if you're coming internationally for SASRA Annual. So now it has like a talk track for that versus it used to be, oh, just email SASRA Events. Now it's like, hey, email this information specifically to SASRA Events. Like they're going to need your full name, passport number, your ticket number. For us to actually do it versus it just being like more of a genero answer, which the genero answer wasn't wrong, right? Email SaaS for events. Yeah, we'll get you a password visa letter. Fine. But now it's just a little bit more detailed, which is kind of nice. So I do think it's worth having just like a master doc of the, of the prompts that you use in all of your agents because they're very portable. And then also having just like a general FAQ knowledge base that it can answer questions so that you can tell it to always refer back to that versus making it up. Because sometimes if it doesn't know, it's literally just gonna go out to like Claude or OpenAI and try and find the answer, which sometimes is right, sometimes is wrong. So it's always better if you tell it. You also have to treat it sometimes, I know it's a little funky, sometimes you have to treat it either like not really a child or like an SDR fresh off the street. But I do think it's also worth telling it things about your company just in case people ask it. 'Cause again, I'd rather have an answer set with my agents versus it try and make one up, which there are times I'm still guilty of this. Like people started to ask very nuanced questions for SaaS renewal and I'm like, oh, I have to like literally train our agent that into our agent so that it knows how to answer. And so that's part of it, right? We're always training and tuning. But just in the, to a long-winded answer to your question of prompt portability, let me see if it's this one. I want to show the other one. Uh, content, I think it's my other tab. Well, I'll show you guys some things here and then I'll show you the prompt builder. Yeah, which is here. So this is kind of two-part, which is interesting. So I think the interesting nuance on AgentForce is I don't have to tell it as much manually as I do things like Artisan Qualified because it has all of our Salesforce data. So like it does save me time and I'll show you an example of this. I've got a few different campaigns running. This one is for, we've talked about this use case a little bit. These are people who previously inbounded for SaaS renewal this year and our sales team never followed up with. So I gave it to our agent plus agent. So you can see here again, topic configuration is gonna look a little different but similar. So you're gonna see some of the instructions I put in here are the same instructions I put into our other two agents. So a lot of portability here of literally, I just, some of these are default. So you can see like this is a default Salesforce one. If you use the sales agent, it does have default instructions. They all do. All the agents will have default instructions. The instructions you're giving it though, again, those are fairly portable between platforms if you're using multiples like we are. So you'll see a lot of these look like a lot of these coaching points, and that's 'cause they reported over. Back one. So that's an interesting one. You know, if you have more robust like data libraries, it can pull that in. Ours is like not that deep. You can set the language fine. Events, you don't have to worry about that too much. And then I'll come into here. These are like the rules. So we named our first agent force Agent Hexi. He has his own mailbox. You know, same thing like Artisan. You're just gonna tell him how many times do you, how long do you want the sequence to be? How many days? Number of replies for us is 0, and that's because we reply. So you can see here there is a checkbox that it's literally recommending to me to turn on so that the agent can reply itself. And I'm like, no, I'll I'll, we'll reply. It's fine. You can change time zones here, right? Actually, Sundays usually are better days. So I can like add that one in. And then I know somebody asked this previously on the livestream is like, how do you assign leads in AgentForce versus something like an artisan or a qualified? And you could do leads or contacts again the same way you do for most things in Salesforce. So it's pretty easy. So you could just there's an AI agent field here. So when you update it, it'll trigger it getting assigned to your agent. That's that. And then I'll show you the actual prompt that we did to resolve. Yeah. So this portion, can I hide this for a second? Yes. Okay. Show you guys these two. So this is a random person I literally just picked. Before I loaded up the meeting. And it's interesting, this random person, I don't know, I, I was like, I just did my one-on-one with David, so I'm gonna name, I'm gonna pick somebody named David. So you can pick any recipient that's in your Salesforce, which is anyone in your Salesforce. What's nice about this is I didn't tell the agent about David. It just generated this response. So David happens to work at Freshworks, which is like a former sponsor of ours. And it actually brought in the context of, hey, Freshworks would be a returning sponsor to SaaStr versus it didn't call him an attendee. It didn't call him somebody fresh and new to SaaStr. It called him correctly as a renewal sponsor and then highlighted the context of renewal sponsor, like engagement and like year over year ROI. The CTA, I don't have the meetings booker turned on quite yet, because again, this is our newest agent. But you can see here, it's like the CTA is the agent to ask for a meeting, and then we would book it ourselves, just because that's how I have it set right now. But I think it's pretty cool because I'm like, in all that nuance, I have to give something like an artisan, right? Where like, I have to break everything into 1,000-person chunks. I have to tell it specifically, okay, these people attended, these people came, these people did this, these people did that. Like, use that context. I didn't have to tell this agent that. So that was kind of nice. Like it just knew. So that piece I would say is pretty nice because it has all of our Salesforce data. So I'm like, Salesforce already knew that this company was a sponsor. Actually, it knows how much they sponsored for, which is not on the email, but it knows all this stuff, right? So I was like, okay, well that's nice because I didn't have to take as much time to do that. And then a lot of this prompt, you'll see a lot of this is default. So don't get scared about the amount of text here like I was the first time. A lot of this is default. Depending on how you're setting it, this is all just fields, right? So this is just for the agent to call your Salesforce data. If you want it to reference more Salesforce data, you can. We prioritize obviously like, are they a sponsor? Have they attended? And that's the two, or have they reached out to us previously? That's like the three priorities for us. And that's how we prompted it here. It has some international versioning in here, which the others don't actually. So that's kind of nice. And then we have our target personas, right? So in the way that I had to break this out into specific campaigns in Artisan, this you could just do one agent and load people into because the agent will context— will like figure it out. It does like reasoning. It does a little bit more reasoning where like I can just put in, okay, previous sponsors say this, if they're a new person and they're CEO or founder, CFO, CCO, you know, some of our core personas, CMO, CTO, marketing director. So like there's obviously a lot more text here cuz it's all in one versus everything in like my other agents was a little bit more broken out, but now it's all in one, it's all here. And then the agent can use its logic and reasoning to say, okay, what are my instructions? Right? And I'm like, okay, you're gonna craft an email. You're gonna use some of this context that I'm giving to you below. And then here are some of those same instructions, right, that we put into our other agents. You know, always put SaaS in the subject line. Don't write me an essay, which I always tell all my agents, don't write me an essay. Like, it's, that's like the number one giveaway of using an AI agent. They want, like, it writes a really long message of why you should be at SaaS or why they should buy your product. I'm like, no one's gonna read that. You know, keep it short, be intentional. This one we say to tell people that you're SaaSer's AI agent. So that's a little bit more clear. Again, just something we're testing. And so you'll see that prompt here. This is saying, you know, identify yourself as AI. You know, use the right pronoun. Again, some of these are already in here. And then here are the KPIs we put into reference. And we say, you know, decide as needed. So it's just gonna decide based on their company and their context. What to say to this person. Again, don't be overwhelmed by this wall of text. A lot of this is just because it's all in one. And then you'll see things like, again, a lot of these are already in here. Keep it short. You know, again, same thing I put to our other agents. Here are the dates. Like, don't make these up. You know, be confident that we've helped sponsors in this case see ROI, generate the warm interaction. Cool. So that's all that. This is the response it generated, which again, is pretty cool because it's, you know, Putting in my instructions of putting the company as SaaStr in the subject line. That's for us because we have a really great brand. Always putting SaaStr in the subject line helps. And then I didn't say, you know, David is part of a campaign for renewing sponsors. It already knew that. So even though it takes a little bit more time, I would say, to put all your contacts into one, now it's faster because like now I've duplicated this agent to do a couple different things. And add more contacts, and it's a lot easier because it's just all in one and it can reference our stuff. So I know it looks a little bit different. It looks a bit more technical. Their team did help me set this up. I think they're helping everybody set it up if you get on Agentforce. So that's something to note too, is like once you get this all in, it goes fast. So pretty cool there. Okay, it's 10:59. Let me see what last questions I can answer. Jason, anything you want to add before I do some last questions?

I think it's good.

Okay. Someone said, how do you measure ROI on all this investment? Do you measure cost or human capital? I measure, I would say both, either formally or informally. Because whatever we do is, yeah, we didn't even get to RevOps, but we use Momentum, Read, Attention, like we have a few other RevOps things that are plugged into our meetings, like if somebody books a meeting, let's say on Saastrannual.com with like Amelia AI, we tag that as inbound AI SDR. Like it has its own category that we made in Salesforce. It has its own little path. Like it is tagged in our system as being separate. So it's something we do tag distinctly. But I think it's important to note too of like, it's not something that we're doing on its own. And so I think between that and plus like any of the, let's say, ticket discounts, which is more frequent for us, like coming to SaaStr London, coming to SaaStr Annual, each of our agents is like, they have again, very maybe similar prompts and then some nuanced training, but their CTAs, like if Artisan gives you, if Artisan, if you reply on Artisan, I respond to you and I give you a distinct code that is artist intact. If you talk to Amelia A on the website and she gives you a ticket code for, let's say, SAS Re:ANNO upcoming, which a lot of you have done for either London or SAS Re:ANNO upcoming in May, that's tagged a different way. And so we can see the actual ROI, but again, it's by design. Like we've set those all up distinctly so that anytime there's a CTA and people take an action through our agents, whether it's human-assisted or not, we're tracking that all distinctly. And so we're tracking all that ROI, all the meetings booked from any of our agents. Any like next step. And not everything will close. Like it's, it's something where like, you know, okay, do I wish everything would close because it came through AI and sometimes we have like a lot more context on these people? Yeah, but it's still like, it's, dude, it's 2026. Some people are going to say no. Like it's not everything will close from our AIs. And so we track that too, as much as we did with the humans. So. We track that all, and then obviously we are tracking our spend across our platforms, um, distinctly. On the—

I know we're over time— on the ROI question, if I understand it right, or maybe, maybe I can answer from a different perspective. You know, I've talked about this, maybe toward the end of next year more self-serve tools will work for these agents. We haven't found any that really work. They all require FTEs, they all require training, they all require an onboarding, and they're all $30, $50, $70K and up to start. Okay, these are not— they're I know other agents are gonna say, hey, we work great, we're $9 a month. We haven't found one that works like this for $9 a month. Okay? So that's something to think about. If you had a fully, if you had a great sales team and a great marketing team and everyone was terrific and they wanted to add a $100K tool, that's a lot of money. And you might compare that, hey, we're already paying for marketing automation in HubSpot. We're already paying for other tools. We already built this. You might, you know, it might be work to justify the budget here. You can justify it by saying, hey, we closed a couple extra six-figure deals. Instant justification. For us, and I think this is why so many, so many of the ones that are exploding so far have exploded. For us, honestly, we replaced 4 humans with these agents. So yes, the cost matters, but the ROI is obvious. We're actually— to summarize all of this, I've said it before— we're actually not doing any better with these agents than we were before. We're doing about the same with 4 less humans. And we're doing— we have higher quality of life because we don't have turnover. They work 24/7. We can do more. We're not doing— we might be doing better. Times have changed. It's hard to say. It's possible we're doing better. Like, give us another 6 months to really tell you. Right now, I would say quantitatively we're doing about the same with 70% less humans. And that's where you— one of the ways you have insane budget. It's why Sierra and Decacon and Supporter are taking off because you can do— sometimes it's better than a human for support. But honestly, an AI agent that's 80 to 90% as good as your best human and not having to find another best human, that's impossible. That's high ROI. So that's, that's the world we're in today. But if your team is perfect, they're all S-tier, they're all 10s out of 10s. They can—

you—

there's no prospect that has never been called back within minutes. There's no email that wasn't sent. Then these products are expensive. Maybe we could break on that. But yeah, for us, It's not— it is the cost of not having humans and running 24/7 is the ROI for us at the end of the day. At the end of the day, we could— we ran campaigns before this. They weren't some— they were almost as good. They weren't as personalized. They were okay, but they required a lot of humans to run. Right now we can run 10 times the volume with consistent quality. That's a simple way to think about it. So we'll, we'll dig into more of the ROI over time, but I think for GTM, The biggest ROI for a lot of you will be we just can't find these people. We just can't find enough SDRs that want the job. We can't find enough AEs that want to close $2K or $3K deals. Everyone wants to close million-dollar deals. We can't find folks to work on Saturdays or to work nights or evenings or folks that, you know, we can't find folks that don't want to take the holidays off. Okay, maybe having an agent just running the next 2 weeks rather than humans that don't show up, that's high ROI. So it's really doing— I've written this on SaaStr, but this is our meta learnings. AI agents today, they're the number one lowest hanging fruit is doing the stuff your team doesn't want to do or the people you can't find to do it. That's such high ROI. The folks just don't want to do it anymore. So maybe we're out of time, Amelia, but thanks for running this.

No problem. Yeah, we will be— I'm Amelia the human is taking some holiday time. So we will be, I'll be working with my agents, but I won't be coming to Workshop Wednesday the next 2 weeks. So, uh, we'll be back in the new year. We'll do AI Day at the end of Jan just to do a full day. Don't worry, we'll— Jason and I will come back, we'll do the kickoff probably, and then we'll bring in some other folks that I know you guys want to hear from, uh, on how they're using AI too. So we'll do that at the end of Jan. Um, SAS Re:Annual in May. But in the meantime, like I said, my agents