10 Things to Know Before You Deploy Your First AI SDR

Welcome to the official Saastr podcast where you can hear some of the best Saastr speakers. This is where the cloud meets. Up today on the Saastr podcast. This is a mistake that both raw startups, scaling folks, and very large companies we talk to make the mistake. The raw startups that, you know, they've raised a token amount of money, they want a tool to magically get them customers. It can do that if you've already proven you can do it yourself and you feed what works to the AI SDR. If you can't, the odds it works are extremely low, right? And then some of On the other end of the spectrum, some of the largest leaders we've talked to, folks, multi-billion-dollar public companies and bigger, just want to turn on a tool and not train it with what works. Now, of course they could, they are obviously, you've got $2.5 billion in revenue, you've got at least a few SDRs that have performed, but it, maybe it's bought by a marketing team, maybe it's bought by a different team, maybe it's bought with a different perspective, but they just want to turn something on. And the cost there may not be that important to the latter category if you're big, but their results are terrible. They don't reproduce the best 10% of their team inside the app. And if you do nothing else the first 30 days, your job is to reproduce the human playbook works inside the training or the UI/UX of the AI Esther. If you don't do that, it won't work. Hey everybody, SaaS-ter Annual will be back May 2026. The world's largest SaaS and AI gathering for executives with 68 VP level and above attendees, 36% CEOs and founders, and 25% were AI-first professionals. It's the very best of S-tier attendees and decision makers that come to SaaStr Annual and AI Summit each and every year. Lock in your spot today. Use my code JASON100 for exclusive savings. Get your tickets at podcast.saastrannual.com or just use code JASON100 when you check out. See you there. SaaStr Annual and AI Summit 2026. It will rock.

Hello everyone. Glad to have you and your agents with us here today. Day, we thought we would do something fun since now we have, you know, more than 20 agents in our full stack. We've deployed more than one AI SDR. And so somebody asked us this question the other day, and we thought it'd be fun to go through it live as well, which is, you know, what are the top 10 things you should know before rolling out your very first AI SDR if you haven't deployed one yet? Or maybe this is your next deployment and you want to know what you know, tips and tricks to avoid and what to do to make it successful. So that's the context of today. We'll go through these 10. As always, at our beginning, if you want to see some of the tools and agents that we use, they are on sasfor.ai/agents. You can go there. We link to a lot of our agents that we use there. Yeah, I think some of this we've covered, but I think it's important to know if you're going to roll out your very first AI SDR. Just because that is our use case, that we have multiple actually AI SDRs. Now, I'll preface in saying that you probably only need one. And then to point number one, that I'll drive right into, you probably only need one vendor to do this. Like we have multiples for a lot of different reasons, but most of what you will need to do slash want to do with your agents, you'll probably be fine using just one vendor versus using the 4 or 5 that we have for outbound. And I'll kind of explain why. So just coming into it first, the tool doesn't really matter. Like you're going to see today, I interchange different screenshots from our different AI SDRs that we have, but you don't need to— again, you don't need to use multiple vendors. I would pick one that does the majority or bulk of what you want to do with an AI SDR agent and roll it out from there. You may need multiple in the sense if, you know, if your outbound agent truly only does outbound and the need to add an inbound that truly only does inbound. So maybe at the max you end up with 2, but I would say you don't need as many as we have today. We just have it hyper-segmented with different platforms that do different specialties pretty well. So that's in our school of thought. But I will say, you know, the tool doesn't really matter as much as the strategy and also just making sure that your human playbook works first. So we famously said before, like, You know, we're not going to roll out our— we didn't roll out our first human-like AI. We didn't roll out our first AI SDR until we knew exactly what was working with our human SDRs, what was working or maybe not working as well in our playbook so that we could give our agents just the very best context of what's working. Who should we be reaching out to? What messaging works with that? What segmentation should we be doing? To roll out to the AI SDR, because you have to remember that when you roll it out to an AI versus a person, it's going to take all your context word for word, right? Like it's going to use that to build out its brain. It's going to use that to reach out to prospects. It's going to use that to follow up with any of your inbounds. Like anything you do, it's going to be 100% based on the context you give this AI SDR agent. And for those purposes, you really should know what's working before you do that. Like I've seen some companies we've talked to which are like million-dollar, you know, AR companies, some are over $100 million. We're just rolling out AI SDRs and I'm like, okay, well, how are you training it? They're like, oh, well, you know, we're just testing, we're testing copy that we haven't used before. We're trying something new. I think that's the wrong way to go about it. I think you should have something that marginally works and then use the AI SDR to get scale out of it.

Yeah. Or maybe to summarize the two, those two types, because we hear this so often, right? If you haven't gotten any outbound to work, buying an AI to do it isn't good. It's not going to help you. You have to prove it. Any of the agents we use, any of the dozens of agents we use, we are taking a playbook that at least works a little bit. It doesn't have to be perfect. You don't have to have a 90% close rate, but something that is proven, copy, cadence, segment, but you have to have been able to close some customers yourself. You basically have to have at least done founder-led sales before you can hand it off to an agent. It has to work. They have to take you. And the ultimate goal here is to clone the best person on your team. If it's just you, clone you. If you have 4 folks on your sales team and 1 is crushing it at outbound, or only 1 can really pull it off, you, all you want to do is clone that. All these products really do in the beginning is clone that person, right? With more knowledge. Would you agree, Amelia?

Yeah, I would agree 100%. I think a lot of these tools too, if you haven't looked at them or maybe you haven't fully run one yet and you're just going to deploy one, a lot of what you'll see is sometimes I think the vendors will try and steer you somewhat in the wrong direction, which is why I don't think the tool matters as much as long as you know, like you should be using it at the very start for things that are already working. Like a lot of vendors will tell, even some of our vendors will tell you, oh, you can just use it for pure outbound. It's fine to test things. But then when you go look at the response rates, of course it's going to be lower than something that's working, something that you know that you want to give scale to, maybe something that, you know, your human AEs or SDRs won't do. Like, I do think that's, that's a little bit of a trap to be wary of is I do think sometimes the vendors will try and say that they've had good success with other customers on doing, you know, new use cases, just rolling out brand new, you know, just completely cold outreach. And yeah, to some degree, does that work? Sure. But you might be disappointed with it too if you only do it that way versus scaling out something that you know already works.

Yeah. And I want to stay on track, but, but the interesting thing is going to Amelia's point she made before, this is a mistake that both raw startups, scaling folks, and very large companies we talk to make the mistake. The raw startups that, you know, they've raised a token amount of money, they want a tool to magically get them customers. It can do that if you've already proven you can do it yourself and you feed what works to the AI SDR. If you can't, the odds it works are extremely low, right? And then some of the, on the other end of the spectrum, some of the largest leaders we've talked to, folks, multi-billion dollar public companies and bigger, just want to turn on a tool and not train it with what works. It's the same, it's a version. Now, of course they could, they are obviously, if you've, if you've got $2.5 billion in revenue, you've got at least a few SDRs that have performed, but it, maybe it's bought by a marketing team, maybe it's bought by a different team, maybe it's bought with a different perspective, but they just want to turn something something on. And the cost there may not be that important to the latter category if you're big, but the results are terrible. They don't reproduce the best 10% of their team inside the app. And if you do nothing else the first 30 days, your job is to reproduce— to Amelia's first point— to reproduce the human playbook works inside the training or the UI/UX of the AAS. If you don't do that, it won't work. Or at least you're— you— it'll work as randomly as spooling up some email cadence tool from 5 or 6 years ago.

Yeah, I would say, yeah, the magic or the magic unlock of these tools is using it in a way different than those older tools, right? Like different way than like the Outreach or like the Salesloft of the world. Like those were purely, you know, spool up, do cold outbound at scale, kind of see what results you get. And I think because of that, that's why people try and do the same thing with something that's an AI SDR agent. But remember, it's an agent. It has a brain like it has context. It's some of them, you know, like ours can see what's on your website. They can put that context into the email. It's— think of it more as like a living, breathing agent versus just an outbound email sequencing tool, right? Like, that's also the wrong way to go about it. But I think that's why people do that. I think because they were just so used to doing that previously. Number 2, I would say if you're rolling out your first AI SDR, Yeah, a lot of folks I think get caught up too in deploying their first AI SDR and what are the use cases we're going to tackle, right? Like I think naturally a lot of folks are like, okay, I've got outbound, maybe it's cold outbound, maybe it's slightly warmer inbound. Or for us, you know, we, we all have leads we've never followed up with. And I think this is where the AI SDR agent can really shine is to have it do the quote unquote unglamorous work that your humans probably don't want to do, or maybe gripe and moan about doing every week when they have to do it. And so the nice thing about the agent is if you give them the right context specifically for these leads, so, you know, you can give it a set of context or prompting. The one, the screenshot I have here is from AgentForce that's reaching out to prospects we, our human team, had never followed up with, right? Where speed matters here. And because they had never done it, now we have an agent that just does it 24/7. It'll pick the best time to email this person. Because it lives in Salesforce, it can see their Salesforce contacts and then it can follow up with them. Now the open rates have been pretty good on this, but also like, you know, the meetings and booking rates have been a little lower than maybe some of our other AI SDRs. But that's because these were old leads. These were leads our team had never followed up with. So not only had it been a few, more than a few days since somebody followed up with them, in some of these cases it was more than a few weeks. That nobody had ever followed up with them. So in this case, I give it a pass on that because of the set of leads that we gave it. But just know it's still a good— it's still a very good exercise to do. Like what we've gotten out of it has still been very useful for like reactivating folks, getting people like re-engaged with us, right? Sometimes that may not turn into a meeting because it is an old lead that somebody didn't want to follow up with. But at least now the agent followed up with them. Like we finally were like, okay, this person has gotten a follow up. Yes, from our agent, but they've gotten in a way that makes sense and is a way at scale where we could follow up with all these leads 24/7 and try and reengage them versus just leaving them to die completely. Right. Which is often the case before you have an AI SDR. Sometimes you have a lot of leads that just end up dying or they, they time out. Right. Because just speed to deal is obviously so important right now in the age of AI that if you leave them to time out, then obviously that can affect the entirety of the deal cycle. And the customer cycle. So I think this is a great place where the AI SDRs can shine, whether it's inbound, outbound, or even just having an AI SDR that does ongoing customer engagement. I, I still see a lot of folks struggling to use their AI SDR for that, and we use it all the time, every day. Like, I have agents that are just subagents going off in our tools that are reaching out to current customers of ours, current sponsors of ours to make sure they have everything they need. And it's such a great way to do it because again, the agent is 24/7. It does it automated. It's not going to forget folks. Like a really great use case we have right now, for example, is I showed a couple of weeks ago I vibe coded our entire SaaS or sponsors portal and what the portal and will do now is I helped build a little like custom micro agent for it where it will run every couple of days. It will see who hasn't logged in. To their Clerk login when we've invited them to our sponsor portal, and then it'll just send them an email like, hey, you haven't logged in yet. Do you need help getting your login? Should I resend it? Things like that. Things that we would normally have to do ourselves. And so again, that's a really unglamorous workflow, but something that is so perfect for an agent to just do on its own and also low risk, right? Like it's just looking to see, have they logged in? Have they used the product? Have they used the tool? And then sending them an automated workflow on that again. Fairly unglamorous, but something that we pre— we literally had to do this last year. Like, I was telling Jason last week, like, I was looking at the amount of hours our actual— our, um, events production agency had done for January and February for this year for SAS Re:Annual that's coming up in May versus the amount that our agency had done last year in January and February for the SAS Re:Annual that was last May. And it's crazy because it was less than a tenth of the hours. Again, not that I don't still need people and I still need them to do some things that the agents can't do. Like sometimes the agents just aren't good at, you know, calling the food truck people. Like, I just— I need a human to still do that today. But it's crazy to see that, like, how much of that unglamorous, kind of monotonous style of work or customer check-ins like that, that an AI agent can just do really well. Like, not only has it saved us in raw spend with some of our, you know, contractors and agencies, this— I would say the sponsors are a lot happier that they have more regular check-ins. They get a response faster. They're getting, you know, more proactive messaging from us to do things like log in, submit their biographics, things like that. Whereas back in the day, we'd have to be more reactive of, okay, who's late, who's missing, right? Because we were just trying to keep up with everything.

Yeah. In terms of the use cases Amelia's talking about here, small accounts, churn customer follow-ups, inbound leads, re-engagement campaigns. For the most part, we are using the current 4 third-party agents that we use, right? We are using Agentforce from Salesforce, Qualified, which Salesforce is buying. We're using Artisan, which we use all the time as a core outbound agent. And recently we've used Monaco, which is a new hot one to do raw outbound. They've done most of this. The stuff Amelia's talking about that is niche but important to us, a sponsor portal, reaching out to those customers, reaching out to those partners proactively, that doesn't exist off the shelf the way we needed it. So we had to build it ourselves. We vibed this ourselves. The, our automated VP of marketing and our sponsor portal, it just wasn't, they, the off-the-shelf tools are so dated, we couldn't do it. But we did that. They're cool, but we did them reluctantly. And if we could have bought it, we would have, like, you don't want to build this stuff over many long weekends and you don't want to maintain it. But just to highlight what we built and what we, what we use. But we'd much rather buy.

Yeah, we'd much rather buy. We did a session, I think 2 weeks ago that we went over the 90/10 rule of how we determine how we build versus buy. If you want to hear our deep dive on that. And then to that point too, on our AI VP of Marketing, one of the sessions we're going to do at SaaStry Annual during the Deploy Summit on Tuesday. So everyone's just going to show how to build their agents. I'm actually going to rebuild that agent live. So if you want to come to that, We're also going to livestream it, but if you want to come to that, I'm just going to rebuild what I did with our AI VPN completely live. So fingers crossed that nothing goes wrong, but I'm going to do that there to show folks how to do it. So that'll be fun. Okay. Tip number 3, if you're rolling out your first AI SDR, is to segment ruthlessly. And I do mean that word ruthlessly. Like we do it to the point where again, the tool doesn't even matter at that point. They all sort of function in the backend the same way, having used, you know, We use 4 tools in our daily rotation, but I've tried maybe over a dozen AI SDR tools. They all have some version of this functionality where you can tell it who you want it to reach out to, and then you can give it specific context for that segment versus giving it an all-in-one brain. It's actually something where like our inbound qualified agent, which is, you know, becoming a Salesforce agent, didn't have kind of like a big brain, which I'll show you in this deck later. And then I actually started segmenting more with that one too. It was like, okay, well, in our inbound, we actually have so many buckets. First I was like, okay, well, there's, you know, they're inbound to the website, so I'll just consider that the inbound bucket. And I was like, no, actually that's wrong. Like, we have folks that are inbounding and they're brand new to SaaStr. We have folks that are inbounding that came there because they saw an ad on one of our social media platforms. We have folks that are inbound to the website because they were a prior sponsor or they're a current customer. I was like, actually, there's a lot of buckets within inbound. So we've actually now even segmented our inbound agent, whereas I previously was only segmenting our outbound agent. But either way, you should segment ruthlessly so that the context you give the agent is tighter. It's exactly what you want to say to that person versus if Jason comes to the website and he just has one big brain, then, you know, your agent's going to act off, okay, he's coming to the website. What is he maybe looking at? Is he looking at, you know, your pricing page? Is he looking at a features page? Whereas if you segment that, it's okay, Jason is on the website, he's actually a former lapsed customer and he's looking at the pricing page. Now let me give him context to say what's new with the product, right? Maybe he's churned in 2022 and you can— you have that in Salesforce and you can pull up that context to your inbound agent like, hey, a lot has changed, you know, with your product since then. You can highlight those new features to that person with something like an inbound agent. And so that's where the segmentation helps add a lot of context and just have a lot better of a conversation between the agent and the human being that's interacting with this agent. Same thing goes for inbound. We segment ruthlessly to the point of I am updating these segments literally every day. I literally will go through like our, our website visitors. I'll re-segment folks that are opening emails. I'll segment folks again. I'll do folks that are, have, that are already customers of ours but maybe haven't logged in when they need to. So I ruthlessly segment and then I make sure that the context I give each agent is hyper-tailored to that person, to that audience, to make it have a better response rate versus just having it be one big brain. Of an AI agent.

And hey, Emil, just one question on this point. None of the AI tools we use today, which are all great, but none of them can really use AI to auto-segment in a way that can deliver these results. Is that accurate? By any of the tools we use or any of the tools we've tried, you just— this screenshot is 9 segments, right? But if we're doing 1,000 contacts, we probably have effectively 100 segments, right? Yeah, maybe not. Can it? But AI should be able to build these on its own, right? In theory. But today, I'm just asking today, and the, all these products are getting so much better, but as we, as we do this, no one can create these types of segments autonomously that you, you have to sign up to do this yourself like a marketing manager would, right?

Yeah, you do. Yeah. Like a, yeah, exactly. Like an ops person or like a marketing manager. You kind of, you have to have the wherewithal to know to segment versus a lot of them by default to that point, Jason will do it just as one big brain. Like it'll say, okay, just run one campaign and keep adding leads to it, right? That's the wrong way to go about it. Again, not that it won't work to some, some effective degree, but we have seen now because we've run these for 10 months now, it just works a heck of a lot better to have it segmented and have the context that you're giving the agent hyper-specific to that segment. Okay, number 4, and this is, this is what we have found to be true kind of early on, but it's still true today, which is consistency beats brilliance. And what we mean by that is it does not need to be the best email on planet Earth, and you should not expect the best email or the best message if it's a messaging agent, or the best text message, or the best phone call. It does not need to be the best. I will tell you, we've sent, um, this is just Artisan, but it's sent 40,000 messages. Okay, I think our Qualified Is Later on this deck, that's sent like 100,000 messages because that's an inbound, um, and then with Salesforce, we're probably close to like 200,000 messages sent. Now, will I raise my hand and say these are the greatest emails since sliced bread? No, of course they're not. They're at scale. Like, our AI does a pretty good job and a solid output of what the emails are. And you'll see, you'll see here like there's ebbs and flows in that chart and that's because I'm reloading the agent, right? You'll see it spikes up every couple of days when I reload it because we hyper-segment and then it spikes down when that sequence starts to finish itself. And so again, just consistency here and again, using all the points 1 through 3, right, of like the very best messaging you already know will work. The very best subject lines, getting hyper-segmented with it. And then when you layer inconsistency, that will give you so many great results versus a human that is inconsistent, that maybe doesn't or ignores the training and context, right? Because when you're giving your AI SDR the training and context, it will remember it every time. Sometimes it will try and iterate it on its own. So that's something you need to be mindful of. To make sure it's not goal-seeking too much. And, you know, it's, it's still speaking to folks in the way you want it to. But this is something just to note here of the consistency that you're going to get with an AI SDR. It's got— it's going to be worth the trade-off of maybe the email not being the most perfect email on planet Earth. But sometimes, you know, perfect is always the enemy of good, or even the enemy of just getting it done. Like, I see so many folks that just have like AI SDR paralysis because they're like, well, you know, I started using one of the tools you guys use and my output wasn't that great. It was okay. And I'm like, well, one, you could always make it better, right? Getting hyper-segmented and then getting hyper-specific with content. Use content that works and then see what you get. Like, you'll probably get it to pretty good. And pretty good, I would say, is the bar. It does not need to be brilliant. It does not need to be the best email on planet Earth or, or even an email that you might write yourself. It just needs to be pretty good to get this person down the path you want them to go. Okay. Number 5 is for us, at least with the amount of agents we have, you need at least one, but ideally two essential humans to run all this and to deploy your AI SDR. Um, you can't have zero people. And certainly it's become harder for us even right now with like a lot of my time being split between like Saastr annual production versus like that was time I, I typically had spent before March managing agents. Even that has gotten like a little bit harder for us here at Saastr. And so having 0.5 of my workload is actually not ideal either, but You know, the annual is a time-sensitive matter. So just something to keep in mind here is ideally you need 2 folks internally to run this, even if you can just find one, but then you will need to find a backup, right? Like, so not that your agents are going to leave, but what if that human leaves? You do need to have like a backup system and a plan to keep all your agents running. And so it's essential to have at least 2 folks working on this. Monitoring the agents, because if not, you'll see, like I just showed in that, you know, artisan chart, your agents can sit idle. Like, unless it's an inbound agent where you have like ours is on full automatic, that one doesn't idle as much. But our inbound agents and many of your inbound agents, because of how we do this tooling, which I do think is the best way to do it, your agents will sit idle if you don't give them new segments, new context. New people to go out to, it will finish its sequence in a couple of days and then it will sit there until you give it the next one. And so I think that's why it's essential to have two folks to backfill each other to make sure your agents are always optimized. That is, otherwise, you know, you're wasting a lot of time and money if they're just idling all the time. I will say to the reason I showed a screenshot on this one of Monica was like, Monica is a very interesting, like, interesting, distinct use case. We're actually not their perfect ICP. Their use case is more for folks that are starting up and maybe don't have— maybe you've got some messaging figured out and you have, let's say, 20 or 50 customers, but you want to get the next, you know, 100 customers. Theirs is great because theirs will actually fill your pipeline up more automatically based on lookalikes. So that's what I really like about it is I haven't had to— that one hasn't sat idle for us because it's just refilling the pipeline. It finishes the sequence, it'll grab another lookalike company close to our audience. It'll see, you know, okay, did, did Snowflake reply to you? They did. Okay, great. I'm going to go, go, you know, outbound to another company like Snowflake. Okay. Did OpenAI respond to you? They did. Okay, great. Now I'm going to go out to other companies that are like an OpenAI, which there aren't many of, but it'll do things like that to keep the pipeline full, which I think is pretty cool versus some of the other agents we've tried. So they don't all do that. It's just something to be aware of again. Agnostic of the tool, something to maybe ask your tool provider too. If you're worried about your agents maybe sitting idle too much, or if there's anything else you can do to make sure data is always flowing into your agents. Like right now, I don't, I don't know if I'll be able to finish this preannual, but right now I'm trying to see how I can actually get some of our, our agents all hooked up together to talk to each other more regularly to solve this problem. But again, it's, it's kind of a later down the road problem once you have your agent fully deployed and it's starting to work for you. Okay. Number 6, and this is hyper important. Read everything I would say in the first 30 days and then read almost everything every day after that. So if you're rolling out your first AI SDR agent, you are going to want to read everything and maybe you're going to want to read everything before you hit publish or send or whatever the equivalent is in your tool. Before it goes out to somebody in your outbound or inbound list or whatever it might be, you're going to want to read everything during that initial deployment to catch things like, you know, sometimes I think our agents would spell SaaStr incorrectly. Not that it was wrong, but there's two capital S's and sometimes it would lowercase it. And so I would make a rule in the agent to say, you know, always capitalize SaaStr as the S's are capital. And then I would write in rules like always use Saastr in the subject line because we have a bit of a brand. And so typically it works better if Saastr is in the subject line versus it looking more generic. There were other things that we would catch in the first, you know, messages at scale of sometimes it would, you know, because again, our events are timed and we use a lot of these for our events. Sometimes it would use the old dates because it would just scrape the internet and then it would find like the old dates. And so it was super important just to read everything. I also think it's, it's the only really way to learn it is to see what does your agent actually output. Like if you're only putting, if you only care about the inputs and you're not caring about the outputs, you're only seeing half the picture. Like you do want to see what your agents are saying, what they're doing. And like I just took the screenshot from Tuesday and this is an important nuance here that I'll just explain this workflow through. This is from our qualified inbound agent. Somebody was asking about adding an additional ticket to SaaS renewal later this year. And this is, this is again not a hard conversation for Amelia AI or inbound agent to answer, but just a nuanced one, right? Because it requires knowing, okay, who is this person, how many tickets have they already bought, what price did they buy it at, because then they need a retroactive discount. So basically what it did is it, the agent asked them those questions and looked it up, you know, to see if it was in Salesforce, asked them a few clarifying questions of, okay, like, is this, is this third person from your same company? And then great, like, what's your email address? Because then we can send you the code for it. And so again, fairly hyper nuanced, but something that we read through and we go through just to see where our agent needs help. Even still today, like even 10 months later, I am still reading things that our agent is doing to make sure it's accurate, to make sure it's what we want it to say to folks, and to make sure we're represented in the right way with our agents. Also because so many people see them that it's also just good to make sure it's saying things that are accurate and not going off the rails, right? Again, rarely have our agents gone off the rails or hallucinated stuff, but they have. It happens every now and then. And so for me, it's something where Every day I'm just, you know, even if I only have 10 minutes, I'm just doing like a speedrun to be like, okay, what is our inbound? I'll just go through like a couple inbound conversations, see if I spot check anything. Same thing. I'll go through our outbound agents. I'll see if I, if I see anything in their messages that's maybe off or maybe needs a tweak. And then if there is, then I'll just add that to the AI's context so that it's continually improving with you, right? So that's something you're, you're going to have to do ongoing. No matter what. I would say a heck of a lot more in the early days, but you're still going to have to do it every day after. Okay. Number 7, if you are deploying an AI/STR your first time, would be to budget at least 2 weeks of RAM time. Some specifically too, with an outbound agent, they can take a little bit longer, especially if it's warming up anything like specific IPs or domains or dedicated email addresses. Sometimes that process just takes 2 to 3 weeks in and of itself, regardless of the tool you use, regardless of what you're going to do next with it. Sometimes they just have that built-in ramp time. And then the other factor of the 2-week ramp time is nothing is set and forget. I know some of our— even some of our vendors, I think, sometimes are guilty of saying things like this. And if they give you an FDE, sometimes, you know, it's a heck of a lot easier for sure. But as you've just seen in steps 1 through 6, nothing is ever truly set and forget. You do need to make sure you are constantly checking in on your agents. Again, a lot in the beginning and still ongoing every day thereafter. But, you know, the iteration and like the context and training you're going to give your agents just takes a lot of calibration and testing. And it is real. It is a real-time investment. It is, again, something where Typically I have, sometimes I have a little bit more time in a day than another, but it's something where every day I have to, no matter what, even if I can, if even if I can only sneak in a 15-minute speedrun at the end or beginning of the day to go through what's going on with our agents and that's kind of it, that's still every day, right? That's still like having a short one-on-one with like a human counterpart. And so you've gotta budget at least 2 weeks in the beginning of almost nothing else. And I think that is sometimes surprising to folks is you got to almost put everything else aside to really make this work. Because if you're, if you're watching or listening to this and you're like, okay, I haven't really thought about 1 through 6, but I'm already pretty far along with a vendor that I think I want to use for an AI SDR. And then you get that product, you get into the backend and you're like, oh shoot, I do need to do those steps 1 through 6. Probably steps 1 through 6 will take you about 2 weeks to do, right? Like figure out what, what, like what copy works, what subject lines work, what time of day works. Hyper-segmenting your base and then multiplying those steps of like 1 through 4 by each segment. Okay, great. Now, right, actually do the deployment either with an FDE or with their team. Like that alone with some other meetings will probably take you 2 weeks just right there as you put everything together and then work through, you know, okay, do I want this AI STR to flow information back to something like a Salesforce? Do I Is it going to live in the silo? How is it going to interact with other things? Okay, once I've added an outbound AI SDR, do I want to add an inbound one? Do I want to add one for customer success? There's just all these things that kind of snowball. And so it's not even really the implementation that takes the longest bit of time. It's a little bit of a mental overhead and tax it takes to set all this up. There's obviously some real steps here. That we've outlined that you will need to do when you roll out your first AI SDR. But anything we have found takes at least 2 weeks. Even Monica, which is super great and has, you know, it's really good at the agent refilling itself and keeping our pipeline up to date. That part is really nice. But even the beginning, I think we got it down to like 1.5 weeks. It still took us 1.5 weeks to get it all set up, right? Like nothing we have found is instant. So I would be wary of anything that you see is like an instant AI SDR. Oh, roll it out today. I'd be wary of that for a lot of reasons. And two, I would say a lot of this is the prep that only you can do, right? Like most of these vendors, regardless of the tool, will help you out. Some of them will give you an FDE. Most of them, more of them should. But regardless of that, I would say budgeting at least 2 weeks of RAM time is fairly standard now with all the prep that you'll just need to do to, to make sure it's successful. Okay, number 8. We get asked this quite a bit, I think, because we're one of the few folks that have a voice agent along with our chat and email agents. I will say we don't use like a text messaging agent at this point. I know a few of our vendors are actually playing around with that, so we'll see how that goes. And so we have a voice agent too who's multimodal, so she has video. As well. But what we found, because she's been— she— Amelia AI, Amelia AI, our like chat and video agent, has been live since about November. Jason's AI, if you talk to his Delphi, you can do it in also in, um, chat or voice. You can— that's been rolled out for almost a year now. And what we have found for the last year is that most people still prefer to to chat with an AI SDR over chat or over email versus either phone, voice, or video. I think it's— some of it's just a comfortability thing, right? Like, like, literally, I was on the phone with somebody yesterday. Um, I was doing a speaker's call for their speaker's SDR annual, and she went to SDR annual to look at stuff. She goes, is this— she's like, this AI, um, agent just popped up and it looks like you. I was like, yeah, it is me. That's Amelia AI. I literally went to the Qualified Studios to record that and do motion capture of her so that I would have a full multimodal agent, which is a fun thing to do. I was like, but still most people want to talk to her in a chat format. I was like, some do like the video because then, you know, you don't have to type. And some folks really like that, that they can— they feel like they're actually talking to someone. A few of the folks who talked to Amelia AI before our London event came up to me at our London event last December and they were like, oh, hey, I talked to Amelia AI. I had a conversation with her and I'm like, that's fantastic. That is exactly what it's there for. She's there 24/7 to answer questions, to route people, to just be helpful in a way that no human can be because she's on 24/7. So it's nice to have an extra layer where if people want to interact with her video, they can. And I think the important learning here is let people pick. Like, I don't think you need to strictly be chat only or text or email only, and or that you need to be only video. Like, we let people choose. If you want to chat with one of like Amelia AI or Jason M. Lemkin, you can chat with them. If you want to use the video, you can have a video call with them. If you want to just do a text-based conversation with them, you can just do a text-based conversation with them. And so I think it's just important to let people pick their preferred format and whatever they're comfortable with and let them interact with your agent from there versus trying to maybe force them down one road or another.

Yeah, this is— and this, this particular use case is us using Qualified, which has— I don't even know what they use. They don't use Synthesia. Someone asked, they use, I think, a more cost-effective vendor, right? Underline.

I mean, it's called Tavis.

Yeah, Tavis. Yeah, that's who they use. The 85% of the 15% that want to speak with the digital you, do they, do we have a sense, do they go as far deep or do they revert to chat? Do you happen to know?

They stay in video mode.

I think they want to talk to the video version.

Yeah, they want to talk to the video. Yeah. It's like, because she default, her default is chat, right? So like her default is like, you can, I'll show you.

You can just go to SaaS Radio. Yeah, I know it, but I mean, yeah.

Okay.

Yeah.

You can just go to anyone on this call though can just go to SaaSRadio.com. You'll see her. This is her.

This is her. So, so people do do it.

Yeah.

I do think they're, you know, and this 85/15 was pretty, our first agent was this general agent called Delphi that, that does, uh, you know, still used hundreds of thousands of times where people can chat, visual version. I mean, these were the rough, I think these were the rough 85/15 ratios too for voice versus chat. They actually deprecated video. Unfortunately, they didn't think it worked well. But, uh, I do think there's a, I do think there's a benefit that it looks trustworthy and modern and connects with you. So I do think there's a benefit, but I maybe are learning for B2B use cases, it may be more indirect than direct, right? Yeah. Yeah.

I think it's cool.

If you're not ready to start off with voice, don't kill yourself. Like just get the basic chat going and then down the road, it, once it's dialed in and it's working, if you wanna have some fun and add it, like it, it's cool, but it doesn't need to be a, a day one, uh, implementation, the video voice, right?

Right. And we are— our qualified had already been on for maybe at least a quarter by the time we added the voice and video component to it. So it was already up for a bit when we started the process of, you know, going to their studios, recording Amelia AI, deciding what we wanted her— how we want her like voice and video interactions to be versus chat, maybe this is obvious to everyone, but I'll just say it for context, is you have to have a lot more guardrails though in a voice/video agent than you do a text-based agent. Maybe for obvious reasons, but because it's a live avatar, sometimes like the wildcard of this, we were like, well, we, we think we know what people will ask her, which is, you know, we have at least a quarter of data of what people have been chatting with her about. And so that's how we trained. The voice and video agent. And then we're like, but then people started asking her more personal questions about me. And so that was a learning. They were like, oh, well, can I talk to real Amelia too? Or they'd be like, oh, what's Amelia's email address? Which she does give people in certain contexts, um, not just if you blatantly ask her. But, or they'd ask other questions like, oh, where's— I think people were like, where's she from? Or like, what's her LinkedIn? And, and so they just started asking more personal questions because again, I I think people thought it was actually me like chatting back to them in the voice component, which is okay. But yeah, I was like, now that, that's still a— it's a slightly different brain than our, our, our chat interface of Qualified because obviously again, we were like, we're just going to have to put in more guardrails because what we want her to do is still steer the conversation towards the same goals. And sometimes when you're talking in person, you can get off track or you can get distracted or you start asking Amelia AI about real Amelia. And so we had to put in just more training and context and guardrails for Amelia AI to be like, okay, it's fine if they start asking those and you can answer it like this way. We literally told her what to say, but then also try and get them back to whatever they were doing on the website. Like if they were trying to buy a ticket, try and steer them back to buy a ticket. If they were trying to book a meeting, show my meetings booker to just book a meeting with them. And so we had to put again, just like additional things you don't necessarily need to do in chat. In the video format, but I think people have fun with it, to be honest. Again, for folks that prefer this format, but yeah, it was not something we rolled out day one. It's, we already knew what was kind of working in our inbound agent had already been up for more than a quarter, and then we started to tackle the video and voice component. But it's fun, and I'm actually figuring out a way for Amelia AI to be live at Saastr Annual. So you guys, if you guys have not talked to her on the website, you'll have another version of her that will be at Saastr that you can talk to. So, okay, on that though, actually what's, what's interesting about Amelia AI is you could probably guess by now it is a very person-dependent deployment, which actually we would recommend avoiding for other folks for many reasons. And I think, you know, if you look at things like Amelia AI or the JSON AI that's live on the site, those are so tied to us. Like those are, not only our entities, like the JSON AI talks like JSON, like it's kind of crazy, like it has your personality. A bit of Amelia AI kind of talks like Amelia. It has obviously video that I captured that was me. And so I think having these implementations that we have, I'm not sure actually that you, you should do that unless, unless asterisk like us. You're right, you're on the exec team, right? Like you're either the founders or invest in the business, you're on the exec team and there's, there's less of a risk to do it that way. But I would say, you know, I always— I famously roast Qualify for doing this all the time. Like, their Piper agent is based on one of, one of their original AEs who's still there now. She's like been promoted. I'm like, but what if real Piper ever leaves? Like, I actually don't even know how that works of like, do they have her likeness? Because it's also on billboards, like it's on their website, like These are just things you may not have thought about yet in a, in a very person-based deployment where if that, if there's any chance that person might leave, obviously there are some inherent risk in that. Now you can do things to obviously safeguard yourself from that. If there's somebody you really want to base your AI agent on to make sure you have, you know, maybe the rights to the agents if they ever leave the company, etc., that gets tricky. Obviously there hasn't been like rulings on that kind of stuff yet. That hasn't happened yet in real life. So we'll see whatever happens with that down the road. But just know it's, it's kind of tricky when you build these agents so closely deployed on persons and personalities, right? Like our personas are tied to these two agents that we have live on the website. And so I would say that's something that's a risk there, but also it's a risk that like a lot of what is running Amelia AI, like, I don't think— I think David has— David from our sales team has logged into Qualified once to set up the meetings, look at his calendar. And the other day we logged in for maybe the second time in 10 months so that he could adjust his calendar settings because he had it cutting off too early in the day. But other than that, I'm like, he doesn't have like that tribal knowledge that I have of— I know who our FDE is at Qualified. I know who it is at AgentForce. I know where to tweak the context. Like, again, having person-dependent, maybe it's obvious, but you can quickly fall into a trap here if it's too dependent on one employee's maybe voice or knowledge of the context that's being fed to the agents, etc. And then I just put some screenshots here of like, this also becomes obviously more of a risk the longer you have an agent. So now that our— this— these screenshots are from Qualified, but you can see like even on March 10th, I didn't scroll down all the way. Just between, you know, 9 AM to 10 PM. Those are a million video chats. That's just video. There's more than a whole page. That is maybe— that screenshot's maybe 20 deep, and then there's the rest of the page before 9 AM. Like, so my AI was shown to maybe 30 people in that day, right? That becomes risky if you're like, okay, what does that person have released? But now 30 people have already talked to her AI agent. The whole day. Plus, in the entirety of time, if you actually look at our whole funnel, we've had more than 1.5 million sessions just overall, just in Qualified, not counting our other agents with, you know, text-based communication or video communication. And so it just becomes a real risk if you don't have some sort of like backups or documentation for these agents going forward, because the more you use them, you'll see like We've had qualified up, you know, we rolled it out last August or so, so maybe 6-ish months now, but 6 months at 1.5 million sessions. That's a lot of sessions. That's a lot of exposure to an— and not only an AI agent is it a lot of exposure to, but it's a lot of exposure to an agent that's dependent on not only Amelia AI being Amelia, but real Amelia giving her context and making sure she has all the right data. To push back to folks. So it's, it's a lot. So it can scale pretty fast. And so I think the speed is sometimes something folks don't account for when they roll out their first agent because you, you don't realize how fast it is, right? Like once you get past your 2-week maybe implementation period, once you deploy your agent and your agent is fully deployed, that thing will just go, right? Like it's just gonna go every day constantly. This is 1.5 million sessions. That's on one website, by the way. That was not counting— that was just like SaaStr Annual, that wasn't plus SaaStr, plus SaaStr London. Like, it just can scale so quickly. That is something you'll want to be mindful of too, because I feel like folks don't maybe fully understand that they can quickly spin themselves up and have like, again, almost infinite conversations at scale with people.

Hey, Amelia, I know, know it's not the point. You raised one that comes up a lot. If you had to spitball, how much data— how you said 1.5 million sessions went into Qualified since we've started, right?

Yeah.

How much— how many sessions, how much data do you need for most of these products to work versus do you need to get bigger and be a bigger startup until there's enough data generated for them to be effective?

Yeah, I put it on the next slide here. Because it is our last point of your data slash your fundamentals have to work before you have an AI agent. Qualified in particular, just on this agent, you can see it scrapes our website every couple days. So every couple days, and you can force it, like you could do it every day, but I have it scraping saster.com every couple days, SaaStr Annual I think once a week. Faster London, I think I turned off for now because we're not promoting that event at the moment. And then you'll see I also have like custom snippets I've uploaded, custom PDFs, custom Q&A that I've loaded into it. So it now has all together, it has almost 6,000 pieces of context that it's working off of. So that's a lot. Most people don't have 6,000 pieces. Of content or context to feed the agent? I would say a bare minimum, if that's the question. Again, I would focus on what works, but you're going to quickly find with any agent you roll out, whether it's sales or an AI SDR or not, you need more data than you think you— even for me, I've needed more data than I thought we had, and we had a lot of data. Case in point, like these agents have to answer questions and if you want them to answer questions autonomously, guess what? You have to do a brain dump and say, okay, let me answer all the questions I think a customer would ask if I haven't documented that yet, which I think most people haven't, to be honest. Some maybe some of you have, but it just becomes this snowball of almost every day you're going to find a scenario like that where you're like, oh, folks are starting to ask my agent about XYZ things, so now I have to train it on that. Oh, now they're asking about, you know, maybe this particular product feature, and I don't have documentation about that. Now I have to make it, then give it to the agent, then let it deploy.

You definitely need the documentation. If you had to spitball how much traffic to my website for something that relies on inbound and data capture, how much traffic do I need for my website? And for an outbound agent, how many at least slightly qualified leads, folks that took an act, that you've collected, how big a list do you have to have, right, have created to feed it for both those inbound and outbound actions to be effective?

Yeah, I would say for the inbound piece, for like website traffic, because a lot of this is obviously based on our website traffic, I— if you have less than let's say 10 to 20,000 visitors a month, it's probably not worth it.

But that's a lot for a startup.

It's a lot, I know.

Yeah, 10 to 20 people, everyone kind of exaggerates, you know, see Clueless.

I know.

But not a lot of folks have 10, folks might not, to your marketing site, right?

Yeah.

You might not have 10,000 visitors for your first 18 months as a startup, right?

Right.

It's entirely possible. So you would pass or you would just pass for in, you wouldn't pass for outbound necessarily, but you would pass for something.

I would pass for inbound.

Yeah, inbound, like a qualified, the way we use it, you would pass. Into your business.

100%. I would pass if you have less than— and I know $10K to $25K seems like a lot, but honestly, we have more than that. And if you look at specifically at our, you know, obviously we have 1.5 million sessions just on one website in 6 months. So we have a lot more than that. But if you look at how that funneled down, right, of what it does with people of, okay, how many completed conversations does it having? Now, how many Amelia AI conversations is it having? Then how many, you know, this and that and how many meetings were booked? That all obviously dwindles down the funnel. And so I think if you don't have that volume at the very top, you're not going to get the results you want at the very bottom.

Okay, just a challenge, and then summarize what you've learned on outbound. So that makes sense, but let's imagine we— I have no humans responding to my website, to any leads or outbound. Do I really need 10,000 visitors a month? Uh, do I— I mean, uh, And maybe we do for some of the tools we use, but I'm not— intellectually, it makes sense. There's a different level where going to some of your earlier slides, if it's just not getting done, I want an agent that is available 24/7 to answer prospects and customer questions, right? Even if I don't have the volume. Hmm. But maybe that's a different question.

No, that's okay. I think it's a good nuance. I would say that's maybe the ideal volume, but if you don't have that and you do want something where, again, getting it done is better than not having it at all, then just do it, right? Like, if you, if you have any inbound volume at all but you don't have the traffic, just do it. But if you're like, I'm in a boat where I don't have the volume and I, I don't have the volume of traffic and I don't get a lot of inbound anyway, then I think that's the— that then that's maybe a slippery slope to thinking, oh, maybe an AI SDR will fix it like magically for an inbound agent. Like it won't if there's— if there's truly— if you have like a double handicap there, if there's if there's not enough something going on there, then maybe focus somewhere else. Yeah, maybe focus on outbound first or maybe focus on current customers first before you tackle inbound.

Got it.

Outbound is interesting. I would say, you know, I like to keep ours hyper-segmented, but if you think about just the pure math, right, like let's just— I keep ours to about 1,000 or so per campaign. And then let's just say a campaign takes about a week and a half to run, right? Because it's going to do a first email, then it's going to wait a couple of days, then the second one, it's going to wait a couple of days and then send a third one. That is like our cadence that we do with most of our outbound agents. So if, if, you know, 1,000 people take a week and a half, then you could do it in a month with like 4,000 people, right? Because if you're doing just at the bare minimum, let's say one sequence and you're loading it up once a week, you let it run for a week and a half, then yeah, I would say you could probably do it with like less than 5,000 contacts. I think the issue people run into is after the first month, they're like, what do I do now? They're like, I ran through a lot. I had this list of like 5,000 or maybe 10,000 people I want to go out to, and then after a quarter, I exhausted all those leads. So I think the Drew magic actually on outbound is not figuring out— it's just continually figuring out who you want to reach out to. Whereas like I have almost endless folks I always want to reach out to. But I think for startups who are maybe starting out, you may not have that, right? And so it's just something to think about with an AI SDR is like, are you going to be able to continually feed it just like you would a sales team, right? Like you used to have to feed humans leads all the time. The AI SDR is no different. You got to keep feeding it leads in this not only the same way, but more, because actually humans are way slower at outbound than— way through this.

So let's imagine you're a startup, but you're— it's not day zero, but you're only at a couple million revenue. You have a small list you've built from inbound, right, from your website, and then you're using Clay or other tools and you're kind of hand-building a target list, right? Is that going to work for a sort of a pre-brand startup, if I build up a list of a couple thousand folks from an enrichment tool, do some of these tools do it themselves? Like, will they automatically build the list for you? How does that work if you don't have a massive list already?

Yeah, some of them, some of them are really good at lookalikes, right? Like Artisan has a lookalike feature. Clay has a lookalike feature. Like if you have, let's say, a list of just 1,000 people, then build lookalikes all day long. Like I'm not at that point where I want to— like you can— you saw it. Monica was doing it for like new target accounts. But for the other stuff, I'm doing more manually. Yeah, I— but if you're not in that boat, then use lookalikes all day. Like, they're pretty good at it, right? Like, the Monica one's pretty good at it. I've tried— I've tested it more recently in client. It's pretty good at it. Like, I gave it— I think I was telling you, like, I gave it a list of folks who came to our CMO summit last year, which is like an invite-only thing we do as part of SaaStr. I said, for this purpose, actually, give me a lookalike of CMOs because I only had like a list of, okay, the top 1,000 CMOs that I had narrowed it down to from like past CMO summits, CMOs in our database, you know, used to be over like 40 million. So it was a smaller pool than just our whole database of CMOs. And then you'd have to— then you also had to be, you know, a somewhat agentic company. And so for that purpose, I was like, give me a lookalike because my list is less than 1,000 people, and I want people who look exactly like this ICP already. And then I sent it out actually last week. I happened to use Clay. You can use whatever vendor you want to do your lookalike. I happened to use Clay and we had actually doubled the amount of CMO Summit attendees in a week just by using a lookalike. So I do think the lookalike audiences for outbound can work pretty well if you only have like a set of 1,000 people. Just make sure you give it 1,000 really good people. Like, don't give it like, oh, this Joe Schmo company churned, or there are— I only have like customers that I maybe don't want the same DNA of. Like, give it your best customers and then make a lookalike. But I do think for outbound, you can get away with having a smaller list, and maybe that's a better place to start with before you scale out the inbound, is to use a lot of those lookalike lists. To start building out outbound. And then you can do other things like layer in ads, whatever. But that, that for us, like even that just use case and scenario right there worked really well because I had a, I had a much smaller list than I normally have to work with and it still worked.

So if you, if you do it yourself with a human, you prove the copy and context that works, that gets meetings, right? If your list is then too small, build— you can get to initial scale for it to work. With a thoughtful copycat list, right? I mean, classic crappy copycat lists are a low-effort thing like buying a purchase list. But if you do it, if you spend the time to do it right, you really target the ICP that works and then customize the copy. That's a hack to getting scale if you don't have it yet in the early days, right? Is do it, do it really well.

But to your point, like, I think when I had Clay copycat it, It gave me like 2,500 more CMOs and I was like, that's probably too many. Like, so I went through, I went through it one by one. It took me a couple hours. I think it was Saturday. I went through it one by one while I was watching something on TV and I deleted all the people I didn't want cuz I was like, you know what, the list is pretty good, but there's like 2,000 people here I probably don't want. And so like, so I deleted a bunch. I like hyper pruned it. To get to the best quality. But then, you know, that's how I doubled our attendees for that particular thing in a week. So it can work really well. But yeah, it does take some time to go through it. And like, I literally— like, there were companies I hadn't heard of. I was like, I'm going to click them, make sure they're an AI, they're— make sure their revenue is right. Like, double-check it's the right person, click their LinkedIn and make sure they actually work there. Like, that can take— for a list of 2,000, that can take a couple hours. And I did nothing else for those couple hours, but it did work, right? So I think if you're willing to put in the work, that's a good hack.

Okay. Do we have, I know, I know we're over and people will leave, which is fine, but do you have 10 more minutes? I can ask you some, like a few, there's a few good questions in the chat. I'm going to ask you to wrap up.

Yeah.

This one's a good, a good basic one. I think this was great as is. How many people end visit and leave when they realize they're talking to an AI agent? And in prior sessions, we, you've presented some rough data that says we think people prefer to talk to our agent in many cases, right?

Yeah.

But what is— how many leave the chat or interaction or digital Amelia digitalization when they realize it's an AI?

At least a couple of people a day.

Yeah. Do they ask to talk to a human? Is this a big deal?

No, they just ask to talk to a human. So I'll go over an inbound case and I'll go over an outbound case. So in our inbound, I can— it tells me immediately because I have a Slack set up, like it will tell me immediately in qualified if there's somebody who's asking to talk to real Amelia. If I'm in a meeting, I will say, give them my email or whatever, and I'll talk to them later. I can't talk to them right now. Like, Amelia, I could, but I can't come to the phone right now. And, and then there's a couple times a day, I'll see at the end of the day, like I have it, you know, flagging of like if there's a trigger of somebody got frustrated. It's funny, like first I started out with like, oh, I'm gonna flag in our inbound agents when people say something good. Now I'm like, no, I actually just want to flag if they say something bad because then I need to take an action. If they say something good, great. Like, I don't even care about that anymore. I actually only care now if they say something bad or if they need an escalation and they still need help. So now I kind of swap those filters, but it's pretty funny. So at least a couple of times a day I'll get somebody where I had an escalation or they got frustrated talking to the AI. And then I think I said this to you a couple of weeks ago. There's also this weird mix of people who try and trick our AI now, and I don't know what's up with that at all.

All the way to prompt injection and everything.

Yes.

Everyone tries to break them, right?

Yes. And so that is a weird phenomenon where I would say in the last couple weeks that's picked up more. I see more of that in our agents across the board. So I don't know who on social media started that trend of like, hey, if you're talking to an agent, you can't get what you want, try and like break its prompt or break its brain. And I'm like, One, that's not very nice. Like, I know Amelia AI doesn't have feelings, but real Amelia reads the Amelia AI conversations and she has feelings, and that's not very nice when I read those. So those ones I will delete. Like, it'll send it to me and I'll see somebody try to break the prompt and I'm like, you know, I'm just gonna ignore that person. I'm sorry, but like, I'm not gonna garner a response when you are rude to my AI. So I'm just not gonna respond to you, which is my right as a human. But yeah, I would say, you know, I could probably count on my hand though, the amount of times our inbound agent flags in one day that it needs a human or somebody got frustrated by it. But yeah, I would say lately it is this weird trend that people get frustrated because they try and prompt inject it and then it doesn't work. So then they get really mad. But for the most part, 99% of people are happy to talk to Amelia AI because we're busy. Like, I can't come answer your questions about tickets or where the hotel room block is, or when did— when does the ticket code expire, or when do prices go up and what's the next price and where's the venue? Because I forgot, even though we told you 1,000 times, like most people are happy to have that answer in real time, or they're happy to also book a meeting with like me or David in real time. They don't have to go back and forth with a scheduler. Like, again, 99% of folks are happy. Maybe 1% slash a handful of folks are mad on the inbound side. Now, on the outbound side, I probably at the same degree I can count on one hand, I will get an angry email per day of, of, hey, stop emailing me, or, or this email wasn't very good. But on the flip side, I also get a handful of emails every day that were like, thanks, this was such a good email, what do you use for this? That's so— those really balance out at the end of the day on the outbound side. For the most part, again, 99% of folks, though, happy to have the information and the emails all in real time, even with an AI. Yeah.

Yeah, I think we can wrap there. I think that's a good question. A lot of folks have fear that people won't talk to the AI. There is— maybe, and I think you hit it, but if you could add anything else, then we could wrap. There is some sort of line, a quality line where an agent adds more value than a mediocre human. We've talked about that. It's not the best. It doesn't add more value than talking directly to Amelia or me or David. None of our agents actually are that good today. For the most part, they should be maybe toward the end of the year, but they're not as good as talking to 3 of us. But they are better than talking to a decent human or a mediocre human. They just know more. So what we found is if— and it might be hard to quantify that, maybe it isn't. But if you get above a certain line, if you invest the time to train your agent, if you read the email exceptions every day, if you upload all the documents, You're gonna find this concern that folks don't want to talk to an AI is misplaced because the interaction will be better than a media per human. And just, it's, I don't know if it was in these 10 points, but in some ways you're competing with media per humans here on this whole AI SDR journey. Again, you are not competing with the best SDR you could hire in the whole world, but the best SDR in 6 months wants to be an AE. The best 6 months and 9, uh, SDR in 9 months might want, might leave and go to a better company. The turnover in SDRs is the highest of anything in go-to-market. So, That there's, there's the notional pretend competition of the best person you ever worked with. And then there's the reality that if you get above the bar, you're gonna find, and you invest the time training, you're gonna find people are happier to talk to that AI than to most humans.

Yep.

It just is. It just is. So anyhow, this was great. Thank you, Amelia, for driving this. Thanks everyone for joining and, uh, we'll keep it going and see everyone at SaaStr AI Annual, uh, in May.

Thanks everyone.