So we can take it wherever you want to go. This doesn't have to be that formal. I'm only so smart, but you know, I have built a SaaS company to a couple hundred million in revenue from zero. I've invested in some winners and some that are very different than Mango. Some, like Owner, that a lot of you folks know that I'm on the board of, that in some ways have a lot of similarities. So that's interesting on business-to-business software.
And then I would say at the start of this year we were pretty behind on AI. We were pretty behind. I could talk the talk, I could read the tweets and talk about how great ChatGPT was, but I was behind. And we kind of went on a roll and we went from being pretty far behind in using AI ourselves on our little team to I think being at the cutting edge. We have 20 AI agents in production. We just today rolled out — and it'll be interesting, I don't have enough data to talk about today — we just rolled out Salesforce's Agentforce. And if you listen to Marc Benioff talk about everybody using Agentforce, but hopefully we're not streaming, we might be the only company that is smaller like ours that is in production. We might be the first. It's early, which we'll chat about. So maybe we have 21 AI agents in production.
And you guys are going to be doing a lot of this. You've got Connect that's coming out. You have more products that are coming. You will be on a similar journey where I think when I invested in Mangomint, AI was a discussion point. And I think next year it will be a deep part of what you do, both internally and externally, and for customers. So this is not the most profound stuff, but what I'll do is I'll quickly share what we're doing. And then maybe more interestingly, I'll share what we've learned having all these AIs in production. And we're not like you, but we have more AIs than humans now. It's kind of weird. There's not that many people that have more AIs than humans. But I'll walk you through quickly what we've learned.
Here's us on the left. I had Claude make this. But this is how our agent journey, zero to 20. I guess it's 21 with Agentforce. And the first one we started with, and this is pretty interesting, I'll share more with you. The first thing we started with was sometime around March of this year, I used a tool called Delphi. And Delphi, you can go to saastr.com if you want, on the bottom right. It's a general AI tool that's trained on 20 million words of SaaS content. I've been writing blog posts since 2012 about scaling B2B companies. We've had like a thousand speakers. We've done 10,000 people to 12 years of SaaStr Annual. And this is a particular AI that could easily ingest all of this and create a digital version of me. So I did it as an experiment.
A few other leading folks that have a lot of content did it. And I'm like, well, theirs are pretty good. And I thought, well, ours will be better. Not because I'm better, but we have so much content and it's all on B2B and a lot of it's on scaling revenue. So I bought this tool called Delphi. I turned it on and it immediately broke and the founder got mad at me. It immediately broke. And it broke because no matter what, everyone's kind of full of it in AI. And he said, 'Oh, we have customers much bigger than SaaS. We have folks that ingested two billion words of content.' I'm like, 'Oh, we only have 20 million words of content.' But the truth is, we had more YouTubes, tweets, blog posts than anybody did. So it broke their ingestion engine. After two hours, it went down.
I mean, I'm kind of self-sufficient. So I tried again. It broke again. It broke again. It's much better now. Finally, the engineers had to help us to work because they were full of it, like a lot of folks are, but it did work. And then it was magical. It was magical. So what was interesting — I might not have another slide on this — what is interesting, there's me, digital Jason. So I kind of compared this. Here's an app we use, a really dated SaaS app we use for our SaaS University called Mighty Networks to help. It's free. It's like SaaStr University has 20,000 people on it. Here's their support: 'We typically reply within a day.' Hooray.
Okay. We have four people on our SaaStr team. We've already had 130,000 chats on the right. 130,000. So here's an example. I just pulled it up to make the slide. 'When is SaaStr Annual 2026?' — and I misspelled it. 'This isn't that hard. This is the date, December 20th to 24th at San Francisco. It'll be even bigger than before and AI-focused.' 'Who were the top sponsors?' Google, Rippling, Artisan, G2. This is pretty good, right? And this is all automated. This was all ingested.
And so what was interesting when we did the first one is I built this tool so that founders and executive folks like you, anyone in this room, could come in and say, 'Hey, I'm at Mangomint. I'm coming up on 20 million. It's SMB SaaS for salons and spas. We're thinking about going up market this way. What should I do?' And it would give you — digital Jason would give you a pretty good answer. It wasn't designed for support. It wasn't designed to do inbound qualification. But when it was live, people started asking them these types of support questions. 'Where the hell is SaaStr?' Well, it's on the website. And they would come in and they would say, 'Am I the right sponsor? Like, can I get the right leads at your event?' And it was well enough trained that we ripped crappy old Intercom out pre-incident. And this became everything. It did everything.
Now, did it do it great? No. But the thing is, we don't have 20 people on our support team like you did. We had none people on our support team. When you went to our form, it would say we reply within a month. Okay. And so immediately it was better at everything than nothing. And it was an eye-opening moment for us. And so I uploaded our sales collateral and our sponsor perspectives. And it could answer. And immediately when I had that aha moment to upload our sales collateral, I noticed it was better than anyone on our sales team at answering sales questions. Not better than Amelia that runs SaaStr with me, but even better. I mean, our sales team is three people. It's pretty small. But we have one guy that's been with us for five years. He's probably closed $10 million or something. But it knew everything in the prospectus.
You know, you're going to — I mean, these are pretty banal questions. 'I want to have a booth at SaaStr Annual, a super-gold booth. How thick can the carpet be?' I mean, you think the sales rep knows this, but the AI does. It doesn't forget, right? And 'what do I get if I upgrade to this?' or 'what do I do?' And it would know everything. And then we noticed it was trying to set up appointments on its own. And we're like, well, maybe we should connect that to a human. And so at all this, this was our start. And going into May, we had one. This was it. And then we started adding to it. We're like, well, that went pretty well. Okay.
And then the lesson that will go through this whole presentation to the end — if you want to learn, if I had to distill all the learnings from 20 agents — and you will have to figure this out with your end customers too, one way or another, you have to train them. You have to train them up front and you have to constantly train them. And the amount of training will inevitably decline over time. But to get digital Jason, which was the easiest because we had 20 million words of content, I had to get it up and going. I had to test it and then I had to see where it made mistakes. So every day I would spend an hour in the morning looking at the answers and they would get better and better.
And even though it ingested content, there was a way in this app where I could upload correct answers. And it would get some things wrong, and mostly it would hallucinate things about the future. Like if you asked it when SaaStr 2027 was, which we don't know, it would make it up. Okay, that really sucked. Okay, it was funny for some things, but I had to find the worst hallucination. But anything that was in the prospectus, it would know. It wouldn't get that stuff wrong, but it would make stuff up that wasn't known, right? 'Oh, SaaStr Europa this year will be in Barcelona.' Well, that's terrible because Marshall's going to London. Like, it would make up the wrong location.
Okay, so every day for about a month, just me — and it arguably wasn't a good use — I would spend an hour looking at this. Now it's about an hour a week. Okay, but other folks would say the tool is terrible. 'It doesn't — I tried Delphi. It didn't work for me.' 'How much content did you ingest?' 'What do you mean ingest? What do you mean put content in?' 'How much time have you spent since you bought your digital clone?' 'I just bought — I paid — I just paid 200. I didn't train it. What do you mean? I didn't get it.' But one of the folks that was successful was Brian Halligan, chairman and founder of HubSpot. He's got a digital clone too. There's a digital Brian Halligan. You can go talk to him as long as you want.
And he started before me. He's like, 'Well, I trained it.' I'm like, 'What do you mean you trained it, Brian? I mean, you're chairman of a public company.' And he didn't have 20 million words of content, but he spent a week just talking to it for two hours a day. That's how he trained it. Okay? And so, in any event, to get any of these things working, you've got to train it. And I'm probably talking out of order. Let me go ahead. But we have — I'll talk about our whole stack — but we have three AI SDRs, now four with Agentforce, and an AI BDR called Qualified.
We can talk about these vendors, but here's an interesting thing. It's interesting and a little scary for both Artisan and Qualified, which have hundreds of customers probably for outbound and inbound. We're their number one user by performance, number one. And we're not far from the biggest. In fact, we're anomalously small. But we have a lot of data and we train them. We trained both of these for 30 days before we went live, and then literally every single day Amelia is working on improving these even today. Every single day improving them. And that's why they're great. 14,971 outbound emails.
I might or might not have the data. It might be here, but you can see just we had the most meetings and most engagements. This is on Qualified, of anyone on their whole platform. Anyone on their platform. So if you learned nothing else about these AI agents, you'll talk to — you'll especially see it on LinkedIn. 'Oh, I bought this AI SDR tool. It didn't work. I bought it just — AI SDRs are terrible.' 'How much time did you spend training them, Mr. CRO?' 'What do you mean? How much time did — and how many weeks and how much time did you personally look at it and how much time did you iterate with it every —' 'I don't — none of it — it didn't work.' Of course it didn't work.
And there's other reasons, like not to be nerdy, basically a lot of AI — and it benefited us — a lot of AI got a lot better in the beginning of this year. Too much information, but Claude, which is, you know, Anthropic, which is number two to ChatGPT, changed their model. They came out with these four models and everything got better. So like Qualified, that we use to qualify inbound, it's kind of old and it used to be kind of terrible. Okay, it's actually — you can try it. Go to saastrannual.com, our event site, and you can talk to Qualified. You guys have looked at it. Marshall tried. It looks like any old chat. It doesn't look that great. It was probably developed four years ago. But once they added the latest AI in it, plus training, magic happened. Magic happened.
Vibe coding blew up. No one could vibe code their own app last year. Then Claude 4 came out. Replit's a 10-year-old company and it went from 1 million to 150 million already this year because the AI got better. So we're doing that plus training are these magical moments and that's the journey we're on. And so if you work with these apps, they'll all get better. And internally, you guys for your own tools, you'll figure it out. Um, externally it's the biggest challenge for SMBs because if you like these tools like — sorry to jump around — but these tools that we have, I mean, we don't really pay for all of them, but basically they're all like 50 grand to 100 grand, that's what they charge, and it comes with like a month of forward-deployed engineering or training.
Okay, so you'll buy Qualified, it'll be 60 grand or $100,000 a year to help you qualify inbound. It works, but that's six figures, and then you've got to train — an average spa does — so I don't think they have $100,000 or the time to train their AI, right? So that is different. It's a challenge for you. But I will tell you, for what it's worth, I'm on the board of a company called Gorgias, which is about 100 million, and their AI support for e-commerce and Shopify. So it's a different ecosystem than you, but it's a lot of similarities. Your app costs about the same as they do, very similar things.
What they have though is they have all the data because they have all the merchant data, all the Shopify data, everything. So when they do support, it's really great. So their AI is really, really good. So it is possible. It's just the more SMB you are, the less humans there are to train this app, right? And so as you add more and more to Mangomint, your job is harder than an Artisan or a Qualified or a Salesforce or someone because you can't put a team of humans on site or training them. But that's the opportunity too. It just means actually SMB AI has to be better than enterprise because you can't have humans filling these gaps for training.
On the other hand though, as I'll show you, you don't need as much training as you think. You just need consistent training, right? And I know I'm jumping around in the deck, but maybe I'll get to it. But the other learning — so we launched the first one in March, this Delphi, and I ingested 20 million words of SaaS content, and it was great. And you could see founders like, 'This is the best AI I've ever used. It was like talking to a better version of you, Jason, except it's slower and you don't get tired and you don't get a little angry or grouchy.' It's like, great. And we have some founders and we had to turn it down. We literally were on it all day long. They literally have it in a window and they talk about their whole day with digital Jason. It's cool.
And I went around thinking, hey, I'm so great because we had 20 million words of content and I broke it, but I did it. And it automatically ingests every tweet I do every day and every video we do. And if we put this video on our YouTube channel, it'll automatically ingest it and it'll know what we're talking about. And I thought the reason ours is better than Brian Halligan's is the 20 million words. And then the founder of Delphi came to SaaStr Annual this weekend and he used my Delphi, Lenny Rachitsky's, Brian Halligan's, and Keith Rabois's — all of whom have a lot of content — and we dug into it. And he did, I think, two million simulated conversations with all of us. Two million — I think a million, two million. I just wrote it up. You can find it on saastr.com and a lot of learnings from it.
But for me, the number one learning is it wasn't the 20 million words of content that made ours better. It was having a number of really good recent pieces of content, a little bit in the long tail, and then constantly updating it made it better. I didn't need 20 million words of content to make it great. Maybe 20,000 would have been enough if it was exactly what I needed to do to get my salon up and running or to get more things. So a lot of folks poo-poo this like, 'Well, we can't do what SaaStr does. We don't have 12 years of content.' I thought that too. I was wrong. Okay, you need a small set of really good content and then constantly make it better when it makes mistakes. That's all you have to do and nobody does this. So I got that one really wrong.
Another product where we're number one on, I got that piece really wrong. And maybe a few other things, then I can go into the learnings, then we can answer questions. So, and again, another learning. Why are we on the bleeding edge here? It's just to learn from. We're not like Mangomint. We had an issue where folks — I mean, we're a tiny team — but folks would leave and I just got — I just don't want to do recruiting anymore. I'm not Marshall. I don't have that energy to constantly be out there recruiting. And we just got tired.
And then we hired two folks who literally — one of them never even learned how to sell SaaStr, and another one was a DJ most of their time and only helped us a little bit when they had time. And I just couldn't go through this one more time in life. I mean, as the intro said, I've been a three-time founder and investor. I just couldn't deal with the mediocre any longer. And I said, 'Listen, I don't need our AIs to be better than the best I've ever worked with. I just need AI to be better than the DJ that only worked for us five hours a week. That's all the bar was.' And we trained it and it is better. It is better than that. It is not better than the best we've ever had.
It's not so much that I think AI can replace a great human. It can't. We know this. Can it replace someone that doesn't know your product, doesn't know how to spell Mangomint, doesn't know anything? Yeah, maybe. Think about the worst person you've ever worked with in any functional area, sales. Can AI replace that? Yeah, it can. And so if you read this email, it's not even that great an email, but at least it knows what SaaStr is, and it's not full of typographical errors, and it has a couple interesting pieces of information about the prospect. It's not even that good, but it's better than someone phoning it in. That's the bar.
And so there's all different ways to do that. But I could imagine at the end of the day for your end customers, like as you're helping supplement back and front office at the product level, you don't have to replace the person that knows everything, that's worked at the salon for 20 years and knows it cold. You might just try to be able to supplement the person that works there for three hours and quits, or phones it in, or isn't available, or doesn't show up for work, or is sick half the week. Right? All you have to do is replace that person. This is what I've learned for the AI to be profoundly valuable. All right? So that's what we've done. We've replaced humans with AI and we're willing to live with some limitations so that I don't have to deal with the DJ drama. It's a true story.
And then this DJ quit — that made a lot of money here and quit to go full-time DJing and do something else. So even paying people doesn't — so we have three. We have one that does sponsorships. So any — we have three different ones if this is interesting to folks. We had one and then we've broken it up into three different agents because they have different personas and different goals. One does for us a bigger deal, a 50K to 100K deal to get a big booth at our events. That you need to know a lot of data about them. You need to know if they've sponsored before. You need to know where they worked before because CMOs have a lot of — CMOs are the buyers, and every year 50% of them work somewhere else.
So if you didn't know they went from Databricks to Google Cloud to Replit, your data better know that because you may think it's a cold email, but it'd be a lot better if you knew that Jason used to work at Salesforce and now is at Level — like, it's going to be much better. So that needed its own low-volume, hyper-personalized. Then we had one for ticket sales because we sell 10,000 tickets a year. But it still had to be smart enough to know if you've been to an event before. It's a very different pitch. 'Hey, we haven't seen you. Thanks for coming in 2019. You know, things are a little different since that little bug went around the world.' Like, it's a different 'come see us' versus a raw — or we need to know, are you in our database because you read our newsletter, but maybe you want to come in real person. So we need a certain amount of personalization and that had a lift.
And then we needed a super VIP one when we go out to like CMOs and CEOs to see if they want to come for speakers. Because I'll get to it briefly. We used to have a speaker agency that we paid about $300,000 a year mainly to do outbound and inbound for speakers. And it was great for a while, and then toward the end of last year they got tired of it and they wanted to charge us twice as much and not really do any of the work except sit in the green room at the events. Okay. So they wanted $600,000, $700,000 a year, and they actually didn't want to read the inbounds because it was too much work, and they didn't want to really do the outreach because it got harder.
So we let them go. First we just had our Delphi general agent do it, and then I built one to replace it, and then we added one to do outbound. So we've replaced — like, we used to have an A-tier resource when it became a C-tier resource. We had nothing. We're like, well, we'll just AI it. I would have rather have them when they were great, but when they decided they didn't want to do the work anymore, which is life, we did it. So that worked. The outbound thing worked because we had no one doing outbound after the DJs and the guy that lived in the van full-time. So that was an easy lift because it just had to — and I was just arguing with the AI. I'm like, 'Could you please — like, I know the CEO and like he's a super fan for SaaS for 10 years. Can you send him one email? He already told me in a DM he'll sponsor.' And it'd be like — maybe — I just couldn't take it anymore. So I could do better.
So then there's this app called Qualified. There's a lot of folks like this that was originally designed to qualify them, and we were reluctant to try it because pre the AI changes that early in the year, I just didn't think the app worked very good. I'd seen it in 2024. I thought it was good for marketers that maybe didn't know what they were talking about, right? To check the box higher. But finally we tried it and it was like somehow it got really good with Claude 4. And now it's pretty great. We've done 194,000 sessions when we did this, and Daniel and I were talking about this. It's not magical, okay? But what this does is you come to one of our sites for our events and it can figure out who you are.
Some of it's not even that clever. 'Be an event sponsor.' This is not the cleverest thing. It looks like stuff, but it has all of our data integrated. And so if you came from Google Cloud, and Google Cloud's been a sponsor for 10 years, but they haven't renewed yet, it will know that and it will reach out to you and we'll talk to you about it and we'll want you to come back and we'll tell you what's changed and that we have a vibe cafe and then it's better to connect with founders. And it will instantly set you up with an appointment when you're ready with someone on our huge sales team.
Now, could you do all that stuff before with, you know, a HubSpot link and a Chili Piper? Yeah, but it's just a mess right now. It talks to you. It qualifies you without you feeling qualified, which is always an icky feeling. It's not 'I'm talking to some 20-year-old SDR that decides whether it's worth their time for me to talk to an AE,' which is one of the worst experiences in SaaS sales. It just wants to know who I am. Have I been before? Do you know who SaaStr is? What's your budget? What are your goals? You know, if your budget's $1,000 and you need a million leads, it just tells you no. Like, don't do it. Like, if you've got budget and you want to reach B2B professionals, let me have you talk to Amelia or David or Brian, right? And that just removes so much friction, right?
And again, this works really well. We are the number one performing user of Qualified across 100 customers, 100 Artisan. It doesn't have to be amazing. It just has to be pretty good at knowing those things, knowing our data, and linking it together where there was no human doing this before, right? The bar is not that high. So that's the next one we've done. We talked about the fact I didn't even think this digital clone would do support. This is a fun thing about AI. If you train it, sometimes you don't know what it's going to do, right? Very versatile. And then we've done a few other things. I don't want to spend too much time on this. I'll wrap it up and I'll go to the learnings.
But it's just learn — like, we're also one of the top users of this tool called Gamma. Have anyone ever heard of it? They went from like nothing to 50 million. It's really a consumery tool to make cool-looking stuff. We use it for B2B, but we got tired of like people handing out our prospectus that was like two years old. Like they would send the prospectus to be for SaaStr 2024. Like we would delete the links, but they would have downloaded — they'd send the wrong — I mean, we know this problem. And so we kind of fixed that. And then we're like, but now Amelia has to like create a custom deck for everybody because the sales team couldn't do it.
So we just had now Gamma just automatically pulls all that data from our marketing automation tool and creates a custom deck for every single prospect. Custom. Here's the right thing for you. Hey, if you're DigitalOcean and you're like Google Cloud, here's how many leads Google got, so here's how many you should expect. It'll show the peers, it'll show why they should be there, how to do it, customize to them what their ICP, who's going to be there. It doesn't look as pretty now that our designer doesn't do it, but it just takes a couple minutes. So pretty cool thing that we do, and then it gets passed around far more than our other decks did.
This one I want to talk about. We all kind of copied Kyle. I was skeptical of using like Momentum and Attention because we had such a small team, but man, it's great. I've told the story before. We had someone that quit the day we rolled it out. Ultimately, if you want to know the life lesson, we had one person — wasn't the DJ or the guy living in the van — but he was the guy who would spell SaaStr with an E in the middle, S-A-S-S-T-E-R, in his collateral. And he would always send out two meeting requests, like with two different apps, one using Calendly and one using Mixmax. So the prospect would never know which one to click. So you get the story, right? I'd always be like, 'Kenny, what are you working on today at all hands?' 'It's like doing a lot of outbound.' Good. Two months, three months. Didn't really see any pipeline, but good.
And then we rolled out Momentum and then the gig was up because we could now see what everybody was doing. And he quit that afternoon. He just quit. He didn't even tell anybody. He just disappeared. But it has been great to have this transparency because the net learning from all this is just the pace of everything's faster with AI, right? Because people don't want to wait two days for the meeting and they don't want to wait for this. And if AI can find the data and surface the data, it should be acted on almost instantly.
And then a few more. We had this agency that wanted to go from 250 to 600 but didn't actually want to review the content because we get 2,000, 3,000 speakers that want to speak at our event. And so it's not that fun to read all of them, right? Especially because the real tough one is — even the terrible ones take time, and then the ones in the middle, like the maybe, you reach out to them and then there's like 28 emails after, and it was just too much. And so they just stopped reading them. So I just went into Replit, built my own reviewer that now just grades them in real time. So now it does it all automatically. They put in the data, I push it to the Anthropic API and it just grades it and it tells them instantly what their score is and gives them the brush-off if it's not good enough.
So this is not the learning from this, and this saves us so much time, right? Because now thousands are read and they're told their score. 'You got a 54 now.' It's nicer to them than what it tells us. It's terrible to them. It's like it always says 'probably not right for this event, but we'll let them know.' Okay, but this is what's interesting. It's a micro app. Like, who else will use this but us? But we could use it a lot. So AI lets you do some of these things. It's a micro — it's almost an agent for us. So valuable, right? I wake up each morning. It'll say we got one that's a 54. Here's a 20. Here's a 79. Take a look. Here's why. It's just amazing. So we'll see more and more of using like little bits of AI.
And sorry to ramble, but the other interesting thing about it — it's so good. It's so good. It doesn't need that much data. I told them, here's what a great presentation is like. Here's Marshall last year talking about 10 learnings from SMBs. It wasn't a commercial. We have a criteria for SaaStr speakers. So look up on the internet if Marshall's been to a hundred events on the same topic. Kindly tell her no. Has she not spoken before? Is it on point? Is she mainly promoting her own services? Is she a service provider or a leader? Like, it can figure out all this stuff and then compare it to all of our other SaaStr speakers for 10 years and rank it. That's not that hard with AI. So as you get smarter, you can think, what are some amazing things we can do if we just inject a little AI? You don't need that much data to make this awesome. Not that much data to make it awesome.
And then maybe I want to get to learning, so I'll just breeze to the last one. Then I got on a tear. I built some others. I built this calculator to value startups. It was 200,000 when I wrote this. We've already crossed 400,000 valuations. And then we added reviewing pitch decks. We've already done a thousand something in a week. So pretty cool stuff. But my learning, if you're curious on the last point, and I want to get to some of the learnings — like, and we were talking about it a little before — there's like so much stuff you might want to build with AI, but if you can buy it, just buy it. Like, don't build anything because it does take a lot of time.
Okay, let me flip through to a few other things that are not obvious. I think we hit some of this because I went through it, but I want to make sure I'll at least get this. Then we can open up to questions on anything you want. It doesn't even have to be on this, but this is the Captain Obvious one, but again, I think for applicability —
I'm only going to say it three or four more times before I end the presentation, but this is the key. And you will hear on the internet agents managing agents and you will hear about grading and all this stuff. It really doesn't work today. Not in a B2B application. You've got to constantly be training them. We don't — you a lot of agents, you're not really going to see hallucinations like you used to see in the early days, but you will see mistakes. The mistakes are just a different version of hallucination. You've got to constantly be training. There is no set-and-forget. And figuring out how to do this at scale is kind of the biggest challenge.
The other learning we had, and it's a version of the before, but this is the other mistake so many people make. Onboarding an AI agent is about the amount of work as onboarding a human. I mean, if you hire someone to Mangomint and you don't train them at all in sales, you don't give them access to anything engineering, you tell them nothing about how the project works, and you just say good luck — I mean, sometimes it works, but it sure is better if there's a hint of training, isn't it? I mean, maybe we're still too small to do any training. I hope these people get at least a little bit of help before they start. Right? So it's the same thing with the AI roughly. You've got to put in that amount of time.
Like, if you think about it, an AI agent — it is maybe even ironic — it's not a human, but you might have to spend the same amount of time in the first 30 days that you would invest in a human to have a successful outcome. Otherwise, it's going to fail. This one is — I kind of hit it, but this is the most obvious one, and it's a challenge for a lot of apps, but this data question, right? It's not all about volume, but the more you know about your customers and users, the far better the AI can be. It doesn't need 20 years of data like I thought, but three or four points of data is so valuable to give the AI to run on. That's what we've learned, and again, that's why we're number one on these platforms.
Okay. Number four. This is one hopefully we're getting past the debate on, but a lot of folks still challenge this. We have all the data for it. It's so interesting. Buyers don't care if it's an AI. They don't care. A lot of us have had, again, we're kind of thinking about ChatGPT in 2023 or some terrible support bot they used, okay, the other day. Listen, there's plenty of terrible support bots, even by leading AI companies. But when you have a well-trained AI like we do, people don't mind. It's a joy to not have to wait for Jason to call me back tomorrow. To not have to work on a human's terms. It is a joy. If the AI is good, it's fine that I know it's Jason Bot.
If Jason already says to me, 'Hey, I know you came to SaaStr last year. It's really exciting. Mangomint seems to be on a roll. Can we get you to come back here next year?' Do you really care if that's an AI? It turns out you don't, especially if it does something for you, if it adds value and it's fast and rapid. So people don't mind, right? And let me know if —
You're an agent. We found, and it's fun, because when we do these 16,000 emails, we would, you know, a lot of them are not to salons. They are to tech leaders and founders, and they would know it was an AI, and they say this is a great AI. They don't mind. So, don't fear it being an AI. In fact, disclose it. And yes, have some scar tissue about that crappy bot you used when you were trying to reschedule your American Airlines flight or whatever. We've all lived it. AI doesn't have to be like this in 2025. It doesn't. It can be, it can be empowering for your end users. And don't hide, don't try to pretend that it's not an AI. Force yourself to make it great, and they will love it. They will love it because, again, a well-trained AI can know more than most humans. It really can know more than most, most humans can.
Um, okay, Captain Obvious. This, this is the biggest, this is small, but it is interesting for like our little teams. It, it is exhausting that the AI is always on. It is exhausting. So the first thing that we, we still do, and we certainly had to do as we scale it, like 7 a.m. with the coffee, was going through all the AI output because if the AI is setting up meetings and appointments and doing all these things, you've got to be on it, and you do have to sleep. Like the AI agent doesn't have to sleep, we have to sleep, right? So I would say as a small team, we work more from 7 to 8 a.m. than we ever did before these agents because they've queued up so much work during the night. You, you, you got to do it. It's a different cadence. It's a different pace. And you know, the most obvious one is when you have like an AI BDR, you can't wait a day to get back to that prospect because the prospect's having an immediate conversation with the website. They expect the human to be almost immediate. And so that, that, it just, our day is just stacked before it even starts. And it's a little tiring and a little stressful, and it's just, it is just different, right? So, um, how to figure this out is, is to, we're trying to figure out a triage system. We try to figure out office hours, but I don't really have all the answers. Um, yeah. Here he is. The CEO of, of what did I say in the hat? CEO. Uh, they never sleep. What's that?
Yeah. Yeah. I, I kind of like my images that I make in, in Reev art. Yeah. He's just, he can't take it anymore, can he? He's not quitting, but he's pretty tired. He thought 7 a.m. was the time where he just did office chitchat. Now he's got to like spend an hour reviewing the AI agents. Um, you know, this is, this is interesting, and, and hopefully for your customers you can, you can, um, you know, you will, you will insulate them from it. But when you get so much information, when there's so much communication, it is so much to process. It's so much more to process, um, than it is from a, from a human, which is just going to surface, you know, you meet with your team once a day, they're going to surface a couple things. Your AI is going to surface everything, everything the minute you log in, right? Um.
This one I think I hit, um, but this I think I had Amelia put this. The first one was how many chats that our general AI did when we launched. 139,560. That's a lot. I mean, I, we were just talking with your head of support about how many you guys do. This is on that scale, right? With, with, with an AI, it might be more than you do in a month because you have a pretty good app. Um, this was pretty interesting. Our AI at this point sent 1,497 messages, um, and 56,070 were identified as leads. It's pretty good. Um, okay. This is just, okay, another interesting point. There's a lot of meta interesting points is, um, there are costs. There are costs. And, um, so we, we, I've never fired anyone in my entire career except for inappropriate behavior. I let people just leave or figure it out. I should get better at it. You may find this hard to believe listening to me, but I'm conflict averse. So, I, I, I let people go. I mean, I let them go when they're ready. I thank them even if they don't deserve it. Always enjoy the DJing. Um, send me the demo tapes. Uh.
But, um, but as folks left since the beginning of the year, we replaced them with an AI. As much to learn as anything else. Let's try, let's try to do, when our content team wanted half a million dollars a year just to sit in the green room, let's try to replace them with an AI. When the DJ left, we weren't sure it would work, but we thought we, of anybody, we should try. And it's interesting. We, we now, some of these, they are nice enough to comp us because we get them a lot of leads. But the nominal cost is we're about a half a million dollars into these vendors. About a half million a year. It's not nothing. It's not nothing. Um, most of them cost some, the cheaper ones are $200 to $300 a month. The Gamas and, and the, the general support. But the, but the other ones are all like $400 or $5,000 a month. It adds up, right, as you replace. It's, it's not nothing. Is it cheaper than a human? Sure, right? As this 8 to 12,000 versus 2 to 4,000. But it, it's, it's, it's something. Um, and the soft costs are significant, right, of, of monitoring them. Um, what's interesting that isn't in this slide, um, is most of our agents now plug into Salesforce. This is something to think about at a meta level from Manglement. So we only, now that we have all these agents rather than humans, we only need three seats of Salesforce. So how much do we pay for Salesforce? I don't, I mean, Salesforce just keeps going up, but 10 grand a year for those seats. 10,000. But the agent companies are getting $500,000 of value. 50x. 50x. So, my challenge to everyone, Daniel, I talked about, think about how you can get, what's your 500x at, at, at, at Mango Men, right? It's, it's, it's, it's, the answers aren't really obvious, but these apps that I talked about are for us, for, and it's different for large enterprises, are, are in any event creating and extracting far more value than Salesforce. That's, it's an existential threat, I think, to, not to Manglement, but to a lot of CRM and others, because it's not that I'm getting rid of our CRM, but when the agents do all the work and they're much more valuable, think about that ratio. It's crazy. 500,000 versus 5,000 or 10,000. That's a lot of value distortion. It's a lot of value distortion. Um, that's the main one. I'm not sure how I put this quote up, but when I, we did an interview with Marc Benioff on, on 20VC a couple weeks ago, maybe a month ago, and he said they had a thousand leads their human SDRs didn't follow up on. So, that was the first thing that he had their Agentforce use. So, I'm not sure exactly where I came. I'm not sure what the point I was trying to make, but, um, an interesting one. But that, but the, but, but, oh, the point is like, just, but it is work to fall below this thousand. Anyhow, this ratio thing. So like, I, you know, it, I'm just kid, maybe it's not my role, but you know, Mango Mint's a pretty cheap product in the grand scheme of things. It's a good deal. It's not inexpensive, like, I mean, it's not 10 bucks a month. And, and, and your customers are very cost-sensitive. They're small businesses. But getting that massive ratio by saving them over time massive amounts of labor, massive amounts of ROI. When you unlock that, you might unlock a mega, even 10 times bigger than it is today. And I, these artisans and qualifiers, they didn't intend to do that. They didn't know that they would be worth that much. I mean, Salesforce is 42 billion in revenue. So, it's pretty big. But this is kind of a scary ratio if you get it right, that we're spending 10 to 20 times more on AI agents around Salesforce than for Salesforce itself. I, I didn't know that. Um, and then the last two are fun, and then we can do any questions if we have time. You know, here's our office on the left with the agents. These desks are empty. Do you need any, anyone need a 32-inch monitor? We've got, got quite a few. We've got three, three video games there. No one's playing. Um, it's a little lonely. The culture is different. The team, the daily standups at 1:30 are a little different. The agents don't say much at the standup. They don't say much. Now, they do their own check-ins in the morning. They tell us how the day went. They push it into Slack, but they don't say much. Um, they have no need to come to the Omni. They don't do any bonding or any fall exercises. I mean, I'm used to it now, but it was a weird transition. It was a weird transition. And, and listen, dealing with people, people are tough. We're humans, right? There's conflict. There's issues. People yell. They get upset. They, they, they have opin, they tend to have opinions, especially more open companies like Mango. And that, that does burn me out sometimes, but I do miss it. I, I do, I do miss it. Not enough. Not enough to bring the DJ back or the content people that don't want to do any content. But, uh, it's a weird, it's a weird world. It is quiet. Um, and, um, what an office looks like when you have 10 agents. So I, I don't think this is really happening. We're pretty extreme. But, um, I think the learning is in the future, just realize it is going to be a different world culturally. And if we come to this in two years, this room will be bigger. Okay? But it might not be as much bigger as it would have been in 2019 going to 2021. And you will interact with your AIs differently externally and internally. It will be a different world, and embrace it. It is fun. It is more productive. But it, it is just because we're seeing an extreme version. Everyone's going to see some. I don't think anyone in tech is going to have no employees. And even these AI-native folks that have very few employees in the beginning, they catch up. Okay? Then they build out sales teams, then they go more enterprise, and then they, then they decide they need support. Okay? Um, so they catch up, but it's not ever going to be quite the same. So, so, so it's, it's just a learning. Um, and then the last one, this one I got to think about more. I don't know that I love the copilot term, but we're using it. But, um, this is the interesting thing. When we started seeing founders sit all day on our digital Jason, like all, like we had, we had to turn it back to like two hours a day. It's not that it costs that much. It just would create like little micro issues. Um, uh, because it would ask, if you're on it like two to three hours a day, you'd start asking weird things our agent doesn't know the answer to. So it would create a lot of little, little micro issues. But, um, but when they're talking, when they, when the agents talk, when they're talking to them all the time, or when they're really answering all their sales and support questions and they trust them, they, they are becoming team members, and it's getting like better, like we're constantly getting better. So I don't have all the answers, but people do talk with digital Amelia and digital Jason again and again and again. So, uh, I, I used to hate this term AI agents because I don't even know what it means, and there's a lot of nerdy people talking about it just to get our hands around it, but they're, they are becoming part of our team internally and externally. People know them. So, it's, it's a different world. And this part I'm, I'm still growing with, uh, on their early days. So, I think those are my 10 top learnings. I don't know if it's helpful or not. We can talk about anything that people want with our remaining time, but thanks everybody.
Should we wrap or do questions? What do we want to do?
