How SalesMind AI Uses AI Memory to Achieve $9 CAC x Work Optional | Part 2

0:03 uh for salesmind 0:05 how often 0:07 does does your client update the package 0:09 if you will you know the training data 0:11 unit and and is it is it self-learning 0:14 with memory or is it clients coming in 0:17 who who does that 0:19 so before it was the client that had to 0:22 update and validate what we what the the 0:26 AI was generated um now we have started 0:29 to make the model learn by itself and 0:32 improve the machine over time based on 0:33 the result that it gets itself. This is 0:36 very the very beginning of that. Uh the 0:38 product needs still improvement and 0:40 iteration based on that and how we can 0:42 improve the the photo optimization but 0:45 this is very exciting to to see that now 0:48 based on the result the AI can can get 0:50 better by itself. 0:51 Yeah. Yeah. Okay. So now we get to sort 0:53 of the technology. This is the that I 0:55 think in in my view for for any any any 0:58 sales to any agent, the ability for the 1:01 AI to have memory, right? And and if 1:04 you're through sales man, if you're 1:06 doing 150 200 outreaches per day, you're 1:09 getting objections, you're getting 1:10 yeses, the ability to sort of learn from 1:13 this and then propose to the client how 1:16 to improve your product or your sales 1:17 pitch, that is just amazing, right? and 1:20 and and and again with the memory it 1:23 will become infinitely better than a 1:25 human just because humans cannot process 1:27 190 every single day. So talk to us 1:30 about that uh maybe maybe as you do that 1:32 in terms of as well how how your agent 1:36 is learning and improving 1:39 if not yet how you plan to do that and 1:40 then how does the text work. Basically 1:44 we build rag uh for some of the some of 1:47 the purpose and then we increase and add 1:51 the objection database of uh the 1:54 knowledge base of of the the agent that 1:57 have access to it for getting new 1:58 objection and then basically each reply 2:01 we get is a reassessing. Do we need to 2:04 reassess the strategy based on that 2:06 reply? And then the AI decide by itself, 2:09 yes, it is useful for me to understand 2:11 better the products, understand better 2:12 the the prospect, understand better the 2:15 lead itself or the segment that it lead 2:17 is belonging to and and then basically 2:19 it improves the knowledge base that will 2:22 have an impact on on the next 2:23 conversation and the next activities 2:26 basically. 2:28 And is that live? 2:30 We we at the very beginning of it. So we 2:33 have started to work on that. Uh this is 2:35 not released yet on production. This is 2:37 still on on beta testing from our end. 2:39 But we have started to work on that yet. 2:41 So at this point the current product if 2:44 I if I'm talking if I'm prospecting 100 2:46 200 a day Julian uh am I getting are you 2:50 giving feedback to me as a client? 2:53 Because you you have the transcript you 2:55 have the transcripts you have you have 2:56 the conversations right. So, so if it 2:58 doesn't automatically give me feedback, 3:00 are you able to download that into some 3:02 kind of Excel, feed it to some kind of 3:03 agent and then summarize it on a weekly 3:05 basis? Are you are you offering that 3:07 service right now or is that eventually 3:09 going to be offered as an autom 3:10 automated feedback system? 3:12 No, we do. What kind of feedback do do 3:15 you have in mind? For example, if if 3:18 salesmind is talk prospecting 100 people 3:20 and 50 came back saying they they like 3:23 this and they don't like this and 20 3:25 says you know the price is too high 3:26 whatever right I don't really have time 3:28 as a client to look at 100 transcripts 3:30 but you can summarize that for me right 3:33 you can say out of the 100 20% like love 3:37 the product but they would wish for an 3:39 average 20% discount or whatever right I 3:42 I as a client I would want that and it 3:44 seems to me the the the the the the data 3:48 can can be digested easily by your agent 3:51 to give me that feedback. I'm just 3:52 curious if that's offered whether it's 3:55 manual or it's coming. 3:58 So we do not have that precision yet. We 4:00 are able to tell you okay you have like 4:02 10 10 20 people that are interested 70 4:05 people that are skeptical and and 4:08 whatever. So we are able at that point 4:10 to do that. 4:11 But the the feedback about 4:13 of the out of the interested people like 4:16 5% were expecting a lower price them 4:19 expect we're not having that yet. But I 4:22 take the note and as a and as a a good 4:25 product manager I'm going to add that to 4:27 the road map at some point. 4:29 Okay. Okay. I mean it just seems to make 4:32 sense, right? Especially as you scale 4:34 this up. Um. Gotcha. Can you tell us 4:37 about the tech technology the tech stack 4:39 what what are you using what's 4:40 proprietary how defensible it is 4:44 the way how we plan to achieve that is 4:47 by starting getting integrated in people 4:50 workflow and and that why we're 4:52 switching slightly the audience that we 4:54 are targeting is because we know that 4:56 agency B2B marketing agency and lead 4:58 generation identity they will start 5:00 putting workflow in their client uh 5:02 processes and that where we want sales 5:05 mind to to start being not only an 5:07 interface, not only the agent that can 5:09 perform the task on your team, but being 5:11 integrating on workflow and interacting 5:13 with the application and the knowledge 5:16 base that we have about people in the 5:18 company. So the technology that we're 5:20 using is fairly simple, right? Node GS 5:23 and react to to build the thing. Uh 5:27 but my CTO can can talk much much more 5:30 about that. So I I don't have a very 5:32 comprehensive 5:34 overview of what are all the tools that 5:37 we're using all the the languages that 5:39 we are using for all the application we 5:41 we have among 10 different microservices 5:44 and so on. The the tech team is is 5:46 really working hard on that. 5:49 How easily replicable is it you know if 5:53 somebody were to basically come out with 5:56 a competing product? what what what's in 5:57 your sauce technology or distribution or 6:00 something that would prevent somebody 6:02 else from offering something that's 49 a 6:05 month. 6:06 Um the way how we see that is by having 6:11 a very comprehensive solution like 6:14 we see so many different use cases. We 6:16 even have like now client and and people 6:19 on gold that trying to ask ask us what 6:22 are the difference between myself and 6:24 and their self doing workflow and n and 6:27 and whatever and the answer is fairly 6:29 simple either you want to build and and 6:31 spend time to actually build your own 6:33 workflow and and put resources out 6:36 there. So it's actually very cost 6:38 intensive to have the knowledge have the 6:40 people inside and so on. So you can 6:42 start maybe in three weeks if if you are 6:45 fast enough or you can start with us uh 6:48 tomorrow. Now if a company wants to 6:50 start exactly the same as what we have 6:52 now um and and do the same product as a 6:55 product to resell and set that 49 that 6:58 will be um very challenging because we 7:01 we currently have over 10 million 7:03 profile in our database that we 7:05 understand how to communicate 7:07 efficiently and every single messages 7:08 that we having with those person make 7:11 the system smarter. So each message make 7:13 the the system and the agent more likely 7:16 to succeed the next time you send a 7:18 message. So the data that we processing 7:21 about people and profile in in the 7:23 business environment make us a little 7:25 bit smarter all the time and and I 7:26 believe that what going to make make 7:28 hard for people to catch up uh from from 7:32 this 7:33 interface perspective. 7:36 Next will be to get integrated in people 7:39 workflow like when you when you plug 7:41 your CRM for something and whatever. Um 7:44 like sending directly the work to the 7:47 human where salesman say hey I have a 7:49 good lead for you. It need like it need 7:51 a human intervention. He needs to get on 7:53 a call or whatever like we aim to get 7:57 irreplaceable by value as well. If if we 8:00 make your business survive or even 8:02 strive because we send you enough lead 8:04 to fill your pipeline and grow your 8:06 revenue, you you will never be able to 8:09 leave us. 8:10 So for the audience, uh we we love 8:13 featuring tools that actually work and 8:16 it it looks to me like giving you value. 8:18 You've done 3 million plus messages and 8:20 the value proposition to a a client, 8:23 Julian, and and feel free to correct me 8:25 is, you know, for less than $100 a 8:27 month, right? Uh you can with salesmind 8:31 uh uh maybe spend a few hours training 8:33 salesminds agent and they send out 150 8:37 200 uh almost 200 per day to prospects 8:41 and your history shows that you're 8:42 getting a 42% reply rate, right? Then 8:46 you can then converse and maybe take it 8:48 off LinkedIn to whatever you're selling. 8:50 It could be selling financial service, 8:53 selling a software, whatever it is. And 8:55 if you then divide that, you know, $100 8:57 a month by 150 a day times 30 days times 9:02 42%. 9:04 It's a extremely cheap, 9:07 painfree way of getting prospects, would 9:09 you say? Is that sort of the the value 9:11 ad? It's getting that prospect right. 9:14 It's extremely inexpensive and quite 9:16 automated. It's it's it's beautiful. 9:19 Amazing. I think you you you rephrase it 9:22 and you you nail it very effectively, 9:25 right? The cost of acquisition customer 9:27 acquisition cost for sales mind is 9:29 currently $9 per client. 9:32 $9. Okay. 9:33 So very sorry. So how did you get that 9:36 number? What do you mean by uh cost of 9:38 acquisition? Because a reply is not 9:40 acquisition. A reply is a reply. When 9:42 you say acquisition, did that mean 9:43 actually buying the service? 9:45 Yes. So customer acquisition cost for us 9:48 is $9 because we eat our own dog food. 9:51 We use sales mine to protect and 9:53 generate. 9:53 Oh, this is for sales mind. For sales 9:55 mine I see. 9:57 We cannot we cannot know that 10:00 effectively for all the the users but 10:02 for test mind we monitor that. 10:04 So you mentioned fundraising. Uh to the 10:06 extent you're comfortable have you 10:08 raised money? What round are you at? And 10:11 and and what's the plan once you've got 10:13 the money? 10:15 So we raised preed with a venture 10:17 studio. So basically they are providing 10:20 now the entire development team for 10:22 years for a year. So that has been 10:24 significantly uh improving the quality 10:27 of the product the fast uh the speed of 10:30 the delivery that we were able to do as 10:32 well. So we've we've passed that now 10:35 we're raising uh feed and the way how we 10:39 plan to spend the money uh from seed 10:42 will actually to scale what actually 10:45 works for us. So we need to increase the 10:47 number of account that we are having the 10:49 number of activities that we are doing 10:51 for ourself. So the the the the 10:55 main goal for that will be actually for 10:57 us to to scale. We have a little bit of 11:00 product um that needs to be to get done 11:02 like the email, the phone, WhatsApp will 11:05 be the easiest way to open a larger town 11:09 basically because you mentioned it 11:10 yourself right out of the 10,000 people 11:12 you need to reach maybe 8,000 of them 11:14 are not on LinkedIn. So we need to reach 11:17 them through other channel and and that 11:18 what we will be able to do with the 11:21 funding to improve the agent 11:23 capabilities other channel and ensure 11:26 that we are multi- channelannel and be 11:28 able to communicate through the 11:30 preferred channel of the lead. 11:32 I mean, you're one of many agents that 11:36 we're seeing that are solving very 11:37 specific business cases and there's a a 11:40 number of different sales products that 11:42 that we're we're invested in are also 11:44 doing. 11:46 How fast is this developing in your view 11:48 and how do you see this 11:51 uh enterprise agent space evolving over 11:54 the next three five years? 11:56 Wow, that's a tough question. 11:59 It is. It is. But I think I think you 12:01 probably know you have an opinion 12:02 because you're playing the space, right? 12:05 Yeah. 12:05 Yeah. We see the challenge is is going 12:08 extremely fast like from months I say 12:11 like four months to six months we see 12:12 new trends appearing um new challenges 12:16 that we need to face uh and so on. So 12:20 the the space is moving extremely fast 12:22 especially like Fman is in a competitive 12:24 landscape. uh the sales industry is 12:28 really big and and the time is just 12:31 enormous right like 57 billion has been 12:34 spent in sales and marketing tool last 12:36 year but people still have an empty CRM 12:41 so in the next three to five years where 12:44 I see that the sales agent will have the 12:47 most impact will be and at least the 12:49 vision that we are having at salesman 12:51 will actually make the funnel the entire 12:54 yourself funnel entirely personalized. 12:57 So it mean the first message is 12:59 personalized, the reply is personalized, 13:01 the seat that you're seeing is 13:03 personalized and the interaction you 13:04 have with the AI live avatar is 13:06 personalized to you only and your 13:09 specific need because you don't have 13:11 time to spend 10 minutes on a website 13:13 and on the person profile to try 13:16 understanding what is the offer, what is 13:17 the value, what is the benefit and so 13:19 on. We just need to show you what 13:21 matters to you. know 13:23 how far would it be for me to prompt GPT 13:27 plus connected to my LinkedIn and my 13:29 Dropbox and my sales kit to say do what 13:33 salesman does and is that a threat to 13:35 you? 13:37 It is not uh it is not a threat for us 13:40 because you can eventually do it for one 13:42 of your accounts but can you reliably do 13:47 it for 100 accounts of LinkedIn or 100 13:50 email. Um so this is the use case that 13:53 we we solve for agency uh and that 13:56 actually a great a great moment for us 13:58 to to switch if you if you notice that 14:00 chip will actually do that. I'm not 14:03 entirely sure they will allow you to 14:05 connect with with LinkedIn itself 14:07 directly to do what we do. Um but 14:10 eventually that we don't see that as a 14:12 threat because one person that is trying 14:14 to automate his own profile is not 14:17 entirely the the audience that we are 14:19 looking for. So you you you can do it by 14:22 yourself. You can play if you're a 14:24 builder and you likes it and and I know 14:26 you likes it, right? You're in the same 14:28 space. you have a sparkling mind and and 14:30 a beauty and and and great hand to to 14:32 build a product. Um so I believe you can 14:35 you can do that but again um it depend 14:38 of where is your focus is is it is your 14:41 focus on building things to make the a 14:44 little bit better or 14:45 just yeah I wasn't I wasn't picking on 14:48 sales mind necessarily Julian because a 14:51 GPT or Perplexely are obviously they're 14:55 all launching or have launch agents 14:57 right I I was playing with Perplexity 14:58 Labs this week and and it's not as good 15:00 as Menace or GSpark so my question is 15:03 less on sales but really saw these LLM 15:06 tools that are evolving into agents and 15:08 making easier uh you know there's always 15:12 that danger of them disrupting 15:14 especially a successful company like 15:16 yours right so I'm just curious how you 15:18 think about that 15:20 and how you defend against that 15:22 the the way how we think about that and 15:24 we defend about that is by the number 15:26 and the scaling uh capabilities that we 15:29 have I think if tomorrow open AI is 15:33 launching a system for replying your 15:35 email. Yes, it will be able to do it for 15:39 yourself that works. Uh but for the 15:42 entire um set of client that I didn't 15:45 see have or making it easy for the 15:48 entire company to actually integrate 15:50 that I think it will be a little bit 15:52 more challenging over the the the coming 15:55 years. Uh but that's only the very first 15:58 top of the funnel, right? Where you 16:00 build a little spark link. So where we 16:03 see that cesman will have a very big 16:05 advantage will be to actually have um 16:09 adaptive micro site where again the 16:12 messages is only a door for you to go to 16:15 the next step in the sales funnel and 16:17 having a way more personalized 16:20 uh experience from that. So we're not 16:22 just sending messages on LinkedIn. We're 16:24 just like moving the lead in the funnel 16:26 and and moving it down and qualifying 16:28 it. So this open will be able to send 16:30 the messages but will it be able to make 16:33 it move in the funnel? Good question. Uh 16:37 I hope no. Uh but if they do it 16:40 something different in that case. 16:42 Yeah, I think that's probably as good an 16:44 answer as we can give at this point. 16:45 It's moving so fast. Um a geopolitical 16:49 question. Uh 16:51 um and I I don't know enough about your 16:54 tech stack, but China obviously is 16:56 leading with quite a it's primarily open 16:59 source, right? 17:01 How do you see the use case of China 17:04 versus US technologies? 17:07 Um great question. Um we we see that the 17:11 models have different strength uh each 17:13 of them for different use case. So we 17:15 leverage I think five or six different 17:18 ones currently for all the process that 17:21 we have the qualification identification 17:24 matching and so on. Um, now the 17:26 competition between both I have no idea. 17:30 I have no idea. I love China so much. So 17:32 I have bias in in the balance. 17:35 It's also a lot cheaper, right? 17:38 Yes. Yes. And extremely good, right? The 17:41 reasoning model are amazing. So I would 17:44 say it is really surprising that I've 17:46 been able to release the model that fast 17:50 eventually. But we know China is really 17:53 leading the the innovation. So I'm 17:55 really confident that in the in the next 17:58 few years it's going to be everywhere. I 18:00 was in China a few few weeks ago and 18:01 actually I see that now in BU map 18:04 they're integrated deepseek. So when you 18:06 go on a business profile like Google 18:08 Google business right you can find the 18:10 the time when they open the address and 18:12 so on and you can like text deepseek to 18:15 actually answer the question from there. 18:17 So I think what are the the big 18:20 advantage China have in the in the race 18:23 is they have one one pass only. So if if 18:27 they decide to put deepseek in all BU it 18:30 will be instant. If they want to put 18:32 that in which one billion and half 18:35 people will have access to AI right away 18:37 right now. So the adoption rate will be 18:40 incredible for them. 18:42 I have a question about enterprise 18:44 adoption. I I put it in my initial Q& 18:47 list Q&A list for you. Um uh uh you know 18:50 we're we're spending some time with 18:51 enterprises. Um and 18:54 it's different than what you and I have 18:58 built at the startup world. It seems 19:00 larger companies they want a lot of them 19:03 see the the value of enterprise agents 19:05 right but the issue of governance and 19:09 ethics privacy and security those are 19:14 major issues to be solved right are you 19:17 talk are you talking to enterprises that 19:20 are thinking about let's say deploying 19:22 salesmind or similar agents and how are 19:24 they resolving these major issues that 19:28 frankly is very new, right? Like you 19:31 know if if you want to deploy a sales 19:33 mind and plug it into a a enterprises 19:36 database, what if the agent goes 19:38 haywire, hallucinates or agent decides 19:40 to steal your data or whatever, right? H 19:43 have you have you come across this 19:44 issue? I'm just curious. It's because 19:45 it's top of mind. 19:46 Yeah. Yeah, we do have enterprises 19:48 client actually and and the way how they 19:51 manage that that was funny because I got 19:53 in touch with them but the sales process 19:55 is just so much longer. I think we 19:57 started with one pilot, one person and 19:59 and one team, 10 people and and now like 20:02 80 or 90 people in in the company for 20:05 the sales. Uh but what changed is they 20:10 replaced the people that were like 20:13 manually sending the outreach because 20:16 that was the job of 15 people actually 20:18 to go on LinkedIn. Try to understand 20:21 what the person is doing. Try to qualify 20:23 the lead based on the information they 20:25 have. Start a message put the lead in 20:27 the in the in the in the Google sheet 20:30 and then report the Google sheet at the 20:31 end of the day. So the guy was lasting 20:33 not even 30 connection today. uh so they 20:36 had like basically a bunch of people 20:38 doing that. Now those people have been 20:40 that that team have been reduced. So I 20:42 guess those people now have other 20:44 attribution uh and higher 20:46 responsibility. Now what happened is 20:49 they have validator. So each time they 20:52 come up with a new agent, with a new 20:54 campaign, with a new mission and so on, 20:56 they have someone that is validating the 20:58 quality of the AI understanding and 21:01 fine-tune if needed because we give you 21:04 access to everything. We let you handle 21:06 the modification, if the AI hallucinate 21:08 a little bit or if is not entirely 21:10 correct or the way you will promote it 21:12 yourself. So they have someone that is 21:14 controlling that, but one person is 21:16 enough for the entire company. 21:18 Gotcha. Um so Gary Tan uh who's the Y 21:22 Combinator I think a few months ago said 21:25 in the latest batch of YC companies um 21:28 he surveyed the CEOs um and they said 21:32 that he's I think he mentioned like 90% 21:34 of the YC latest batch are using agents 21:38 to build to build software right so 21:40 instead of having 10 devs they've got 21:41 maybe uh like like two or three and 21:45 using agents um really uh uh you know if 21:49 you if you think about that it really is 21:52 um what I mean YC is selectively finding 21:56 entrepreneurs that are using agents to 21:58 build software right and so this is 22:00 really the the whole idea of a lean 22:02 startup and we we had a news newsletter 22:05 that talks about that 22:07 um how do you see that and and are you 22:09 guys AI native as well? 22:14 Yes. Yes, definitely. Uh this is one of 22:16 the first thing that we we do when we 22:18 onboard people. We give them a chat GPT 22:21 plus account and a curl account so that 22:24 we ensure that they they have at least 22:26 the bathing tool to to do what they need 22:29 to do. Um we actually migrate one part 22:32 of the application that could have take 22:34 weeks uh before with codeex this weeks. 22:39 uh so we leveraging significantly the AI 22:42 we have training internally about one uh 22:45 one champion that have identified the 22:47 right tool and the right capability of 22:49 the tool and the right use case. So he's 22:52 leading the adoption within the company 22:54 and within the other developer. So we 22:55 have like someone that is uh finding the 22:59 way and and showing the way for the 23:01 other collaborator to actually leverage 23:03 AI more in the code. But I've been 23:06 extremely surprised that Codex now is 23:08 able to send PR basically for you. 23:10 You're just giving a task and and it 23:12 send PR on on GitHub. 23:14 Do you guys have a goal? Are you guys 23:16 going to go to two million per employee? 23:20 My team was joking about we can we can 23:22 replace our people with Codex. What I 23:24 what I believe is we're going to have to 23:27 keep the people and just make them 10 23:28 times more productive. So I I don't 23:30 think uh I don't think having a small 23:33 team of 20 people will will help us 23:36 because eventually we need to build a 23:38 lot of different use cases but I I wish 23:40 the team that I have now is able to 23:45 help themsel with AI to get the 10 times 23:48 more productive. I think we're not at 23:49 the 10 10x yet 23:52 but we're working on that. So if if 23:55 gamma is 1 million per employee that 23:58 that may be the next our goal. 24:01 What my takeaway with this conversation 24:02 is 24:04 it doesn't really matter from a client 24:06 standpoint whether you're using deepseek 24:08 or using this text. What matters is that 24:10 you're specifically solving a problem 24:13 and the problem is 24:16 uh you you've trained an AI to 24:19 essentially learn the product or the 24:21 service and you you've now very smartly 24:24 integrated into LinkedIn which is as you 24:26 said is a billion people. So, you'll 24:28 find a lot of clients there. And you 24:30 you're now uh at least in salesminds, 24:33 each your own dog food, you're you're 24:35 acquiring new customers at $9 per per 24:38 client. Um I would argue um 24:43 is cheaper than that. um in the sense 24:45 that if we didn't have salesmind and 24:48 you're having people call right the 24:51 amount of time that you spend training 24:52 or retraining sales people turn over 24:56 uh at management time uh is that usually 25:00 isn't really boxed in. So if I were to 25:02 to compare using salesmind to do a sales 25:05 lead versus the traditional way it's a 25:08 it's a massive factor of cost and time. 25:10 Would you say that's and I haven't done 25:12 that but it just seems to me it's a lot 25:14 more intuitive because once you train it 25:16 it it scales up so much quicker right 25:18 and it gets smarter especially with the 25:20 memory piece. Yeah. So so to me the $9 25:23 is a fantastic number but compared to do 25:26 traditional way I'm sure if you box all 25:28 that in it's actually I think a much 25:30 more higher multiple of productivity. 25:34 I'm not I didn't say that very clearly. 25:36 Hopefully that makes sense to you and 25:37 and maybe you can comment on that. Yeah, 25:40 that make a lot of sense. What we see as 25:42 an average in the B2B industry is a lead 25:45 cost 200 to $250 25:48 as a cost of acquisition, right? 25:50 Wow. 25:51 With salesmind because we are using our 25:54 own solution, we we've done like $9. So 25:57 it's significantly lower like we we 26:00 could not actually achieve that with any 26:02 other uh way I believe. 26:07 Great. Great. Okay. Hey Julian, Saturday 26:10 morning. I'm sure you've got lots of 26:11 things to do. Do you do you want to tell 26:13 us how the audience can reach you? Our 26:15 sales mind. 26:17 Yes. Uh fairly simple on LinkedIn. We 26:19 are we are super visible. We're creating 26:21 more content about AI in sales and we 26:24 we're really evangelist about how AI can 26:28 make the the businesses 10 times more 26:30 productive as you I guess. So LinkedIn 26:33 will be the the best way. Julian Gada on 26:36 self mind or self mind. 26:38 on LinkedIn, sorry, or sales man on 26:40 LinkedIn and it would be very easy to 26:42 find. 26:43 Fantastic. Well, hey, thank you so much. 26:45 Uh maybe we can get you back on uh in a 26:47 few months to to see how you you guys 26:49 are growing and and uh see the ver Yeah. 26:53 Yeah. Fantastic. Thank you so much, 26:54 Julian.