you2idea@video:~$ watch NSwl_E0WWKk [41:19]
// transcript — 653 segments
0:01 In this video, you're about to watch me sell a $6,000 AI agent. This close took
0:05 place over three separate calls back in December of 2024. So, it was when I was
0:10 pretty new to the space, but I still managed to close the call. So, today I'm
0:12 going to break down these calls and show you guys what I did well, what I could
0:16 have improved on, and how you can go ahead and close a deal just like this
0:19 one. So, let's get into it. All right. So, before we actually start watching
0:23 these calls, and I try not to cringe at myself as best as I can, I wanted to
0:28 real quick tell you guys my mindset when I'm hopping into these discovery calls
0:31 and I haven't met this person before. So, it's a discovery call, which means
0:35 it's for you to discover as much information as you can about this
0:38 client, their business, and what they want to do with AI automation. And in
0:41 order to make sure that I'm not rambling on and letting the client talk the
0:45 majority of the call, I use a framework that I call the LRP, which stands for
0:49 listen, repeat, poke. It basically makes sure that I listen to what he or she is
0:52 saying. I repeat it back to them to make sure that we're all completely aligned
0:56 going through the call and then I poke with a question and I prompt them for
0:59 more information. So, I just wanted to throw that out there. Throughout the
1:01 call, I'll stop and I'll mention some times when I did LRP and I'll show you
1:05 guys why that's beneficial, but let's get started. >> Almost invited my noteaker. Uh
1:11 >> you if you want to go for it, otherwise I always send you a copy of this one.
1:14 >> Uh no. So, it's funny actually. Um when I sent our own, I meant literally we uh
1:19 this is different. This is a separate conversation, but my company, we
1:23 actually are launching our own uh >> Oh, okay. >> Yeah.
1:26 >> So, just wanted to show you guys real quick that was like the first 20 seconds
1:29 of the call. And it's important because I spent the first 2 minutes just kind of
1:32 building rapport and realizing real quick that this client is very bullish
1:36 on AI. You can see he's innovative. You can see that they're doing things. They
1:39 have AI initiatives already in the company, which was a really great green
1:43 flag for me of someone that we would want to work with because they have
1:46 long-term goals with this and they have realistic expectations. A lot of times
1:50 you have people on the two ends of the spectrum which would be way too bullish
1:54 and they think it's magic and they have unrealistic expectations and you know
1:57 right away that throughout this discovery call you may have to try to
2:00 manage those expectations or people on this left side where they're very
2:03 skeptical. They need to still be convinced about AI automation and then
2:07 you know how to approach the rest of the call and sort of the questions that you
2:09 ask. So just want to throw that out there real quick. The next thing that I
2:12 try to do here is basically just open the floor up to him and say why did you
2:16 book in a call today? How can I help you? >> Yeah. Well, uh, yeah, if you want to
2:19 just tell me about sort of what you were looking I was reading your your
2:21 Calendarly booking link a little bit, but um, yeah, if you just want to sort
2:23 of let me know, you know, what what prompted you to book a call and what
2:26 you're what you're looking to do here. So, that's the way I wanted to open it
2:30 up, just lay the agenda and get him talking right away about what it is he's
2:33 looking to do. You can see I wasn't very like I wasn't speaking very confidently
2:36 there. So, that was a tough watch. But now, what I'm going to do is just listen
2:39 until there's an opportunity for me to repeat and then poke with another
2:41 question. >> Yeah, I understand. I understand. So, you know, we're looking at uh, you know,
2:49 email integration, Slack. Um, what about sort of your calendar sync? Would that
2:51 also be like a Google calendar or do you use Outlook? >> Uh, Google.
2:54 >> Okay. >> So, what just happened there was I listened for a while and he started to
2:59 break down his tech stack and what he was looking for this type of assistant.
3:02 And I repeated back to make sure that we were fully aligned on the different
3:05 tools. And then I threw out another question which wasn't a great question,
3:08 but just kind of like trying to get a better understanding of the tools and
3:11 asked about Google and Outlook, stuff like that. But what I missed here was a
3:15 big opportunity to highlight the value ad that this system would bring because
3:18 he basically just went through talking about the different tools he used and
3:21 what he does in those tools. And that would have been a great opportunity for
3:24 me to bring light to that and say, "How often are you doing this kind of stuff?
3:27 And how much time do you think per week are you spending in each of these tools
3:30 where this assistant would solve that?" And just by me asking that would get him
3:35 going more on about how much value this system will actually add. So something
3:38 that you should try to do. It's a great point you bring up because that's kind
3:40 of the way that we like to work as a you know a very custom agency is we we
3:44 really like to understand the the operations of the business, the goals of
3:48 the business and then um not just help you solve like one pain point. In this
3:51 example, you come to me and say like we're interested in this personal
3:53 assistant, but we you know we've seen a lot of use cases. We approach a lot of
3:57 businesses. So we like to be able to say you know this is the way that we think
4:00 you should start to integrate because um you know this is really just the
4:03 beginning of AI as far as like getting integrated into businesses and really
4:06 helping transform a business. >> Yep. So, you know, the earlier you
4:10 start, the better in six months, 12 months, like where is the space going to
4:12 be? Something that I think about every day. Um, and just being able to start
4:16 early and building up a foundation so that when these more powerful tools come
4:20 out, we already got, you know, a leg up on everyone else. So, um, yeah, we we
4:24 like to sort of that's the way that that to approach working with the client,
4:27 make seeing if it makes sense for both of us to sort of engage in like a
4:30 partnership, I guess you could call it, as we're we're sort of like a dedicated
4:35 team for you and your projects. Um, so I guess that leads into my question of,
4:38 you know, what are sort of some of your goals with AI integration and AI
4:41 automation within your business? Is it kind of like you want those personal
4:43 assistant and then that's kind of all you're looking for or do you want to see
4:47 like how we can continue to slowly chip away, chip away, save more time, um, and
4:50 help you grow at the end of the day? So, what I did right there was within the
4:53 first 10 minutes of the call, I wanted to make sure that he understood the way
4:57 that we work, which is not just a one-off project. We want to work on a
5:00 longer term partnership where we're actually deeply integrated into the
5:03 business. Like I said, understanding their goals, their operations, so that
5:06 we can help implement solutions where they make sense. If it wasn't clear,
5:09 this client found me through my YouTube video about a personal assistant. And
5:12 that's kind of the first functionality he was looking for. And what I ended up
5:15 asking as my poking question was, are you just looking for this? Are you
5:18 looking to have to see how we can help you chip away and chip away and save
5:22 time over the course of many months? And while that was a good question to ask,
5:24 it shows two things about the bullishness and also shows some of our
5:28 value. But I should have tied that a bit more to actual business outcomes by
5:32 saying after we nail this first phase with the personal assistant, where's the
5:35 next area where you'd like to be able to buy back 5 to 10 hours a week
5:41 >> at the William level? Uh on maybe I didn't mention I'm the founder and CEO.
5:43 So, >> okay. >> So, if you didn't hear that, he basically just said that he's the
5:47 founder and CEO of this company, which was great because I didn't even have to
5:50 ask about that, but it's something I needed to figure out if he was actually
5:53 in a position to make decisions within the organization and dedicate a budget
5:56 towards something like this. So now I've validated that he's qualified. They're
5:59 bullish about AI. I'm going to keep going on. Um but the idea that the
6:04 operations bot would be able to again take the next steps from our meeting
6:12 notes and either populate them inside of our notion space immediately or would be
6:16 able to recognize when that stuff is t you know has has been populated and can
6:21 then you know our operations team can say hey tell me all the tasks for Bob
6:28 give give me all of Bob's tasks that are due this week right?
6:34 Whatever that is. So on one hand there's workflow efficiency within operations.
6:40 Um there's again recall all of that just what you would expect but those are the
6:44 main domains at the moment. I don't see us having it directly link into
6:47 QuickBooks and send out invoices automatically or things like that. We've
6:50 got other systems to do that stuff. Does that give you a good picture?
6:55 >> Yeah. Yeah. Awesome. Um I mean you know typically the first thing we do is we
6:58 like to break down because obviously there's multiple pieces. Um and it seems
7:01 like the first thing here would you know the most important aspect is always how
7:05 can we get all the information that you have into the database so that the agent
7:09 has contact of everything that's going on. You know data and context are the
7:11 two most important things especially with something like a personal
7:14 assistant. >> Yep. So yeah, it would really be diving into
7:19 the different integrations you have all the data and then first aspect is can we
7:23 get that into some sort of database for you so that it's in you know in an
7:27 optimal way so that the agent's able to recall all that information whenever
7:30 it's prompted to um and then you know figuring out some sort of automated
7:33 system so that whenever you change information when you whenever you have
7:36 new stuff you upload in there that that process all takes place um you know in
7:41 the back end for you as well. So yeah, um another question I had was how big is
7:44 the organization and like what does that operational piece look like when it
7:48 comes to you know the next phases where we're trying to help automate some of
7:52 those um interactions as well. I don't love what I just did there. What
7:56 happened was he talked for a while about some of the operations or some of the
7:59 different departments and what he would be looking for out of his assistant. So
8:03 just dove more into the scope of this first phase and he talked about a lot of
8:06 things. He talked about a ton of different tasks that he would use the
8:10 assistant for. And rather than me poking more at that pain point. I basically
8:14 said, "Okay, I hear you." And then I tried to prove expertise. That's what I
8:17 was doing there is I tried to throw out that stuff about the vector database and
8:20 building that automated pipeline to convince him that I knew what I was
8:23 talking about and that I was the right fit for him. That was not a bad thing to
8:25 do. But what I should have done before that was say, "Okay, you just told me
8:29 all of these processes that you want this personal assistant to be able to do
8:32 for you. How valuable is that? How many hours is that going to save you?" And
8:36 what that's doing is it's making the ROI clearer. It's also quantifying his pain
8:40 and bringing more light to it. And then it anchors the price and allows me to go
8:43 even higher. Another good question would have been since you're saving 2 to three
8:46 hours a week with this new assistant, what can you dedicate those 2 or 3 hours
8:50 to? And by putting the client in that position where they're thinking about
8:52 that and they're imagining a world where they have five extra hours a week
8:56 indefinitely, it just opens up some cool ways to keep the conversation going. So,
9:00 you're pretty eager to get this get this out there. Um, yeah. And it's just it
9:04 just comes down to the integrations I had in my video was like it was Gmail
9:08 and calendar and that's pretty much it. I mean I had to send a Slack message,
9:10 but there's so many more actions in Slack besides just sending a message.
9:13 You want to be able to search through there, I'm sure. You want to be able to
9:17 recall certain information. Um and the same thing with Gmail. You want to be
9:20 able to not just send an email or get emails. You want to be able to um reply
9:24 to drafts or reply to threads, create drafts, label emails, delete emails,
9:27 move them around, um mark the spam, whatever it is. So, we want to be able
9:30 to make sure you have every single action and every action in, you know,
9:34 every action we add is another layer of in the prompting. We need to make sure
9:37 that the agent understands what's to do, when to do, when to do it, and how to do
9:40 it. So, yeah. Um, >> and the truth is really quick. I don't
9:46 actually need all of those. Um, like I've got a labeling system. I like kind
9:50 of use it probably use it better. And I know that there's a way to have the AI just kind
9:54 of like automatically apply labels based off of what it thinks, but that could
9:58 also be a little bit of danger. Um, you know, I think I'm a big fan of
10:02 artificial intelligence and we're, you know, we're in like we're in day one
10:09 thing is going to be. Um, but I also am not quite I think that anything that it
10:14 does needs to be vetted. Um, >> like it's only the lowest stakes emails
10:19 where I would say just respond to so and so, right? Like the
10:23 It's always going to be drop it and drops, right? >> Yes, you're 100% correct. I agree. And
10:28 um you know, I'm working through a project right now where, you know, we're
10:31 doing automated research, which is great. Um and then from there, taking
10:34 that research and creating like LinkedIn posts and blogs. And I I told the
10:38 client, I was like, "How about we drop all these into a spreadsheet for you?"
10:42 So you you're able to go in there and um you know, basically just every week have
10:46 like those those three automated posts, but we don't want to just post them on
10:48 your LinkedIn. You know, you just never know. uh I mean you know it's consistent
10:52 it's been cool and also one aspect that we're trying to add in is AI generated
10:56 images infographics for the LinkedIn post and especially with those you want
10:59 to make sure those are checked as well obviously we'll get to a space I don't
11:02 know if you played around with Sora at all but um that's been really cool but
11:06 you know it it just gets smarter exponentially every day which is cool
11:10 that's kind of cool to see because I remember thinking to myself I can't wait
11:14 till image generation gets good enough for being able to confidently make some
11:17 graphics to post on LinkedIn and stuff it wasn't there yet But that's beside
11:22 the point. What happened there was this also just validated to me that this
11:26 person is very very realistic about AI. And I really liked that because that
11:29 helps me not have to worry about managing expectations. I have to set
11:32 expectations and hold up to certain expectations, but we don't have to play
11:36 that game of making sure that we're bringing them back down to earth.
11:40 >> I guess I have two questions. One is um your buckets of hours, I guess. How what
11:44 are what are those tiers? Is it I don't know 20 for like $20,000? I Yeah. I
11:49 mean, I'm throwing numbers at it, but what does that what does that break down
11:51 to? >> Yeah. So, the first one is going to be 4 and a halfK for the 20 hours and then um
11:57 8K for 40 hours. >> You like that's something where you can
12:01 build out a version of your personal assistant AI provided we're able to get you, you
12:07 know, everything you need in a timely manner. You know, login information,
12:12 database API, I don't know, API links, whatever that is. >> Yeah. I like I it's impossible to
12:20 estimate but I I really I do think so. Um considering like I said the toughest
12:25 part here is going to be understanding where your data is coming from and how
12:29 to make that automated. Um that usually is the biggest chunk is setting up that
12:32 foundation because that's the most important part. Y >> and then from there once we are able to
12:37 connect all the systems which in this case um the integrations don't seem like
12:42 it'll be too challenging. It just seems more like the the end 20 piece 20% again
12:46 will be the prompting and testing where that's the only that's the only um area
12:51 where I could really see any spillover would be the the the basically refing to
12:55 make sure that this thing's 100% ready to go for you. >> Okay,
12:59 >> so I don't love what just happened there and let me explain why. Ultimately, when
13:02 you get to the end of that first call, the question is going to start to come
13:05 up in the client's mind of, you know, I wonder how much this is going to cost me
13:08 or how expensive it is to work with this person. And so the client asked me
13:11 basically like how we work and we talked about different models and we ended up
13:14 considering maybe just kind of working together for 20 hours. And so naturally
13:17 the question is well do you think you can get this done for me in 20 hours?
13:20 First of all I think that maybe you should only use an hourly model if
13:23 you're first starting out and you need to get experience and that is a safer
13:27 investment for the client. But like I've talked about many times before, you
13:30 don't want to be tying yourself to an hourly rate because that's just you're
13:34 putting yourself in a box. But pricing aside, what I should have done here, I
13:37 should have just not answered the question. I don't think there's anything
13:39 wrong with saying, you know, the point of this call and the next discovery call
13:43 is to scope out the build to a point where I feel comfortable with everything
13:46 that we need to be able to do for it. Because at this point, I have no idea
13:49 how much data he even has that he's expecting this agent to have access to.
13:53 And I've even said in this call probably two or three times, the data is the most
13:56 important part. That's what takes the longest is getting all of that right so
13:59 that the agent has all the context it needs to actually perform. And without
14:03 understanding what the data looks like, how much it is, the format, where it
14:06 lives, I should not have been saying anything about how quickly I could get a
14:10 PC spun up or if I could deliver in a month. Like I said, anything that uh
14:13 that that pops into mind before our next call too, we can bring there. And um
14:17 yeah, I'll just dive into it and we'll we'll that way we'll be able to bring
14:21 you like a very outlined build strategy, a very outlined wireframe of what we
14:24 think the build will look like. That will also contain um our estimated
14:28 timelines on that for each each step of the the build. And uh the main thing I
14:32 want to talk about for the next one will be um you know besides those
14:36 integrations will be what your data looks like, where it's kept and how we
14:40 want to get it in there. So um yeah. >> Yeah. Well, so let's quickly what about
14:46 um yeah, what's your Friday look like? >> Okay, so at this point I basically
14:50 concluded by saying what I should have said earlier, which is we still need a
14:53 lot of scoping around the data. So you can see I ended off the call that way
14:56 and now what we're doing is we spend that last minute coordinating our
14:59 schedules and booking in a time. So, that's a great thing you want to try to
15:01 do is at the end of the first discovery call, just book in a call while you're
15:04 both still on the call. And then at the end of the second call, try to book in
15:07 the third call while you're both still on the call. But that's kind of how I
15:10 wrap this one up. And now I have some information to head into the next one
15:13 with, but I don't have enough yet to give a price or like I said, realistic
15:16 timelines. >> How are you? >> I'm doing well. Doing well, actually.
15:22 Just uh just got into uh I just drove like 3 hours in the snow. I was I was I
15:27 was thinking I might have to uh email you guys and say, "Hey, um I might be a
15:30 few minutes late because I didn't expect the roads to be so so bad." Especially I
15:34 got up and and left at like six, but I made it just in time. So, doing well.
15:36 How to you guys? >> Classic Nate trying to be charming,
15:39 trying to build rapport. But now, let's actually get into the meat and potatoes
15:44 of the strategy call. The process of still scoping out a build. It's about
15:48 the questions you ask and about the the way we talk about um just like the
15:52 goals. So, I'm glad that you have that that sort of mindset here. Um,
15:55 >> awesome. >> And so, as long as that goes too, I
15:58 don't know. So, I don't I'm not a database guy. I don't know.
16:02 >> I don't know how that works. >> Mhm. >> Period. Like, I I learned about a vector
16:07 database from your videos. >> Rats tells you uh and and don't get me
16:10 wrong, your videos are great, but like I still don't know what the a vector
16:13 database is. In short, >> it's they're confusing. Yeah. And
16:17 honestly, like this is something that I tell people. um I feel like I have a
16:21 good a good grasp on it, but there's still so much um under the hood that I
16:26 don't get yet. So much math that goes into it. And basically what I tell
16:31 people is I understand maybe about 80%. And that 80% allows me to get your
16:35 agents to work, but I don't need to understand the formulas and the math and
16:39 all of the deep the deep down embeddings. Um, but I I think I
16:43 understand them well enough to be able to pull back metadata, filter through
16:46 things, understand how to do it optimally, and like you were talking
16:50 about, have a big database, but maybe not having every agent have access to
16:53 the entire thing, but pieces of it, and be able to filter and retrieve through
16:56 certain section, which is super cool. So that's um I've been pretty loyal to Pine
17:00 Cone with that right now, but um I've been exploring some of the other vector
17:05 databases like Superbase and Quadrant um because yeah, you know, it's it's those
17:08 things where right now I've been loyal to OpenAI, Naden, Pine Cone, but that
17:13 could change. You know, Vappy has been also our voice voice uh agent goto.
17:19 >> Okay, so that was a bit of a good and bad mix. I was able to relate to the
17:23 client and we kind of had a laugh about the technicalities of vector databases,
17:27 but I don't love the rest of the piece of what I said because that could
17:31 instill some doubt in the client. So, what I should have done was taken a
17:34 technical concept like the things that happen under the hood of a vector
17:38 database and just transformed that and communicated that into business value.
17:42 So, instead of saying, I understand 80% which is enough to get your agents to
17:44 work, I should have said, yeah, completely hear you there. Vector
17:47 databases are pretty technical, but here's why it should matter to you, and
17:51 here's like what we need to do with it. It's going to let your agents be able to
17:54 accurately search through tons and tons of stuff quicker. It's going to be
17:58 cheaper than if we basically made your agent read through all of your documents
18:00 every single time. And so, that would have probably clicked with the client a
18:03 lot better. >> On a personal basis, probably is that's
18:06 where I think reality is going. But certainly in sci-fi dystopia, right,
18:12 each each mega corporation has its own AI Um, and I bet you where we're going in
18:22 this reality is that the AI are just going to start talking to each other.
18:24 We're already seeing that in social media anyway. >> Completely agree. And um, one way that I
18:29 talk to my partners about this kind of stuff is like we're basically trying to
18:35 create like interns where you can tell them like what to do sort of thing. They'll compile stuff.
18:39 It's not always going to be perfect, but they they get you started. you're able
18:42 to give feedback and they'll kind of go back and do it again and talk to these
18:46 other people about it. Um, and yeah, like you said, agents that can talk to
18:50 each other and talk to each other hands off is the goal. And you know, that's
18:53 the kind of stuff that happens in workflows already is like just agents pass data between each
18:58 other. And something that I've been doing a lot recently with my agents is I
19:03 have them talk to each other and like if if it's not good enough, they'll they'll
19:07 send it back. So some like this agent will send over like a newsletter article
19:10 to this other guy and then this agent will read it and then send um like some
19:13 information to a different one that will like check it and then this agent if if
19:17 it's not good enough it'll send it back to start and they'll just continue down
19:20 that process until the final checking agent said it's good enough to go and
19:23 then that's when I get to speed that output which is really cool.
19:28 >> Let's bookmark that. That's fantastic. Um, so one space,
19:34 one space where I'm a bottleneck is in our marketing space. >> Okay, that was pretty cool because what
19:40 I did was I listened to him and I basically like threw in some insight
19:43 from some other products that I've worked on with having agents kind of use
19:46 like an LLM as a judge and he loved it. You could see he said, "Let's bookmark
19:49 that." Um, and he was very excited about that. But then I miss out on an
19:53 opportunity here because he literally explicitly said to me one area where we
19:56 have a bottleneck and then he starts talking about it. But I never go back
20:00 and address that bottleneck by saying how many hours a week are you losing
20:03 because of that bottleneck or what would the value be if we could actually use AI
20:07 and use automation to get rid of that bottleneck and then once again we're
20:10 just poking the pain points and we're shedding more light on them so it
20:13 becomes very clear for the client having agents check at every step before it
20:17 passes through uh it really opened my eyes and then I saw just the power of I
20:21 think that's where agents are able to get smarter too because they have that
20:24 memory and they can see what didn't work and what did pass and just continue to
20:27 get smarter and smarter. smarter as we sort of adjust the prompts and all that.
20:32 So yeah, um definitely cool stuff that we we could work on um in in the future
20:36 as this kind of stuff gets gets smarter. It's just so exciting. But um yeah,
20:41 >> well I wanted to one thing I wanted to dive into today would be specifically
20:46 the inputs that you're looking to have into the the big overall database. um
20:53 what those inputs look like specifically the integrations and then like how often
20:59 do we need to be able to feed in you know dynamically is it every day will we
21:02 have new stuff to feed in every week every month what does that sort of look
21:05 like so what I did right there is I tried to bring the conversation from
21:09 kind of a fun use case which actually is the bottleneck that I maybe should have
21:12 been poking at a little more but I tried to bring it back to the initial
21:15 conversation where I needed to really scope out how much data that he has and
21:19 the frequency of updates and deletions and stuff like that so out. I let him
21:22 talk for a while. I listen and then I follow up about that. >> For the most part, it's Dropbox, Notion,
21:30 Slack, Gmail. Um, and and yeah, otherwise AWS. Uh, so you know, at some point I think it'd be
21:39 great to have Ari's output just like a web scraping, you know, if we can have it listen to a
21:47 YouTube video and pull a transcript and summarize for example or if it can, you
21:51 know, scrape a website or and say like, oh, hey, this is this is what Deote says
21:55 about themselves, right? And we can figure out because then obviously we can
22:00 do something with that information. So some capability to interact with the
22:04 website. This isn't a tool, but it is the last thing that's on my mind is like
22:08 citations that link to where it found that stuff. That would be really
22:13 just really helpful from >> Yeah. Yeah. 100%. That's uh that would
22:17 be possible. One of the products I'm actually working on right love that he
22:21 just spent 10 minutes talking about the data sources and transformation that
22:25 they're looking to automate. He ends his kind of breakdown with literally saying
22:29 that would be really helpful for me. And I say, "Oh, okay. Gotcha. This is
22:32 something else I'm working on. I don't know what I'm thinking there. That is
22:35 one of the times where he says, "It's helpful for me." I say, "How helpful? If
22:38 we can automate this for you, we could save your ops team five hours a quarter.
22:42 That's probably low, but more than that, what could we actually save you if all
22:46 of that process is taken care of now?" So, learn from that lesson after a big
22:50 breakdown like that. Relay it back, poke with another question, but also try to
22:54 make the value clear as you're going through the scoping. I consider that a
22:57 win because I just got tons of information about where the data sources
23:00 are and what we need to do. But once again, I should have just related back
23:04 to business value because the first phase will be a data ingestion and that
23:09 is what I imagine taking the longest is basically just optimizing all of your
23:13 data in an automated pipeline to get it into your vector database. And that's
23:15 yeah, like I said, that's what's going to take the bulk of work. So, if we were
23:19 able to maybe start with one of those sources, get you something up and
23:22 running in Slack that you can see and test out and then it's like, okay, yeah,
23:25 let's start to add our notion now and now let's add Dropbox, that sort of
23:29 thing. So, okay. Yeah, I think that that makes sense and we'll we'll align in
23:33 early January and just uh just sort of have a conversation about it.
23:37 >> Sounds good. Um so really quick, when you say ingesting our data, how
23:42 what would that process look like? Um and how would that work and how is that
23:48 different from just like connecting the bot to one of NAD's um triggers or something
23:57 like that? N and N. I'm always saying that because I have that baseline from
24:01 the videos that I watch, but I don't say that from a space of knowledge. I'm
24:03 genuinely curious. >> Yes. Yeah. Basically, um, we want to get
24:09 everything in that source into a vector database. And it's not just as simple as
24:13 plugging in notion to pine cone. We have to optimize the data, create some sort
24:17 of pipeline that we can just basically pull everything. It'll run through this
24:21 logic to optimize it with metadata, um, with the right tags, um, the right
24:25 embeddings, and then put it into um, vector database. We have to basically
24:29 load it in correctly, embed it correctly or not embed it correctly, split it up,
24:32 chunk it correctly, get it in there. A lot of testing to make sure that it's
24:35 actually coming back accurately because sometimes with the vector database, it
24:38 can start to hallucinate a little bit if you don't sort of split it up right. And
24:42 after we get everything in there, we want to set up that automated pipeline
24:45 so that if you do add new tasks, new pages, they'll also go through and it
24:49 will, you know, it'll be a robust system. Um, and so that's basically that
24:54 process of getting it into vector database because it's going to
24:58 ultimately it's going to be able to run quicker. It's going to be run run more
25:03 efficiently. It's going to um help pull you more accurate results back when
25:06 you're asking like hey what was uh Bob's task on the 26 and um like what were the
25:11 action items and that sort of thing. Oh, um, it's all about the data and that's
25:15 the big chunk is how do we get that in there and how do we get that in a robust
25:19 system so that you can manipulate things in notion and not have to see it happen
25:22 in Pine Cone, but it is happening for you. Okay, so that was interesting. I
25:25 didn't really love the answer at first, but then it ended up being a little bit
25:28 better. And the reason I say that is because he asked what does ingesting
25:31 mean and what's the difference between doing that and just connecting the agent
25:35 to something like notion. He didn't care really technically what that means. I
25:38 didn't have to talk about embedding, splitting, chunking, all that kind of
25:41 stuff. What I should have said is it's just a different strategy to provide
25:44 context to your agent. Going through a vector database is going to let it be
25:48 cheaper, faster, more accurate compared to if we just have the agent searching
25:52 through notion which could be slower, more expensive, and all this kind of
25:55 stuff because that's what he wants to hear is cost and efficiency.
25:59 >> Well, really quick, looking at the calendar, are you thinking first is a
26:03 Wednesday? Are you thinking like that Friday like you know Thursday, Friday,
26:07 second or third or I'm kind of breaking for Christmas and all that kind of stuff
26:09 and we decided we're going to meet up after the new year. So we find a date
26:13 that works and now I'm going to head into that final call with this client.
26:17 Um I will go ahead and share screen go over this build strategy unless you guys
26:21 have anything you wanted to talk about real quick before we sort of dive in
26:24 here and go over some of the details. >> Uh nope just excited.
26:29 >> Okay. Well, we'll get forward whatever. >> Okay. we have a more detailed doc that
26:35 we would go over sort of um once this once we're actually really launching
26:40 this uh this project but obviously for now it's it's it's just going to be
26:43 focused on that sort of personal assistant that we had talked about. Um
26:45 we talked about some of your data sources like notion, Dropbox, um Slack,
26:51 Gmail, um managing information being a personal assistant for you and with the
26:54 long-term goal of potentially pushing this solution out to more employees
26:58 within Kexus. Um but the initial focus here is quick and easy access to your
27:02 data. Um being able to interact via Slack and um obviously having a a good
27:09 knowledge infrastructure of different tasks that is going to update
27:13 automatically. Um and just that sort of uh capability. So that's like the
27:18 general overview of course and this is what we looked at doing for the actual
27:22 build strategy. Like we talked about the first phase would be the data ingestion
27:25 getting everything into a vector database. Um, we talked about your
27:30 different data sources. So far, we we've been able to work with all these
27:31 sources. There's nothing too complicated. There's nothing that, you
27:34 know, doesn't have any endpoints we could connect to. So, we don't see any
27:39 major complications there. Um, it we would obviously have to get into the
27:42 details like what the data really looks like. I don't assume it's going to be
27:46 too complicated, but um that's obviously how we're going to start off and then we
27:49 will do a lot of testing before we really push into integrating with the
27:53 tools like um you know, Slack and your email. Obviously, a lot of testing is
27:56 important to see how the metadata is going to work, what type of embedding
27:59 and what type of chunking and splitting we should do to make sure that
28:03 everything coming back um is accurate. Um and then that way we're also going to
28:07 explore sort of doing some name spaces up front. That way when you do have um
28:12 wanting to push this solution to other employees, we'll already have like a
28:16 separated sort of system for information specific to each person or um you know
28:21 if you guys want to have access to all the data but not all the agents should
28:25 that sort of thing. Okay. So now we're at the point of the call where I have a
28:30 proposal and I'm talking about this is what we've talked about. Are we aligned?
28:33 Honestly, I would recommend that you bring something like this earlier than
28:37 the final call. It ended up being that we were able to close on this call. And
28:40 if you notice, what I did here was I have all of the text stack listed out. I
28:44 have all of the functionality and I'm reading it off to make sure that he
28:46 agrees with everything that I'm saying so that we can ultimately use this as
28:50 the functionality requirements and the definition of done for the system. And
28:53 the whole point of scoping is that you can get to a place like this where you
28:57 confidently understand exactly how you're going to build it and what the
28:59 connections are going to look like. And if you haven't sculpted out enough to be
29:03 able to say that, then you're not ready to close a sale because you're not ready
29:06 to talk about timelines and integrations or put a price tag on it. For the most
29:10 part, what this agent will be doing is waiting for commands. It's not going to
29:13 really be taking action on itself, right? So, um, you know, there there's
29:17 other integrations that we've done with basically like an inbox manager that
29:21 takes emails and can sort of send them around, label them, respond to certain
29:24 ones, but we're looking for more of like a you will be specifically prompting
29:29 this when you want it to do something. >> So, I think one tool integration to add
29:33 to this is search, you know, like being able to search Google or something like
29:36 that. Uh, I that is going to be a thing that would be very useful for us to have
29:41 up front. Uh and partly for instance whether we're collecting blog post
29:44 information something like that we'll want to be able to just you know have it
29:48 have it search and support our efforts uh which also leads to the automated
29:52 element that like one thing that could be really useful is does a search and
29:57 collects information into a giant summary and sends it off and and shoots
30:00 it off internal to some database or something like that. So I do think there
30:06 are certain unprompted automations that we would want to add. Um, but in terms
30:12 of like proper autonomous agents, not yet. >> Okay. And so when you say like
30:17 searching, do you mean you would be able to prompt on like certain keywords that
30:20 it would go look for recent articles about? >> Exactly. Like hey, I go search Google
30:26 for or hey, pull this from Wikipedia or not even, right? Rather that it has the
30:30 ability to do whatever, you know, collect some information and give me a
30:33 document or you know, whatever, right? Um, and it does, you know, it knows that
30:39 Wikipedia is a pretty reliable source. It knows that like your times or
30:42 whatever else are like reliable sources, but then it's not afraid to search for
30:45 other things, but at the end of the day that, you know, maybe it's got a
30:50 preference list for sources. Um, >> okay. >> And it knows to or at the very least it
30:56 has the ability to go retrieve information from those sources online as
31:00 part of the request. >> Okay. >> Um, cool. And or that um we're we'll
31:05 also obviously be able to come back with like okay here's a a quick summary of
31:08 like what I found here are the links that I used that sort of thing okay
31:11 >> exactly >> so you can see what kind of happened
31:14 there was even though we had talked about this stuff on the first two calls
31:17 we saw that we weren't aligned on something already which was the ability
31:21 to have some autonomous web search which is fine to add in but the reason why you
31:24 want to list all of this stuff down as clearly as you can and avoid as much
31:29 ambiguity as possible is to avoid scope creep because you don't want to be 2
31:31 weeks into the build 2 weeks into the build and then the client wants to add
31:34 more things. So that's the kind of the whole point of getting all this stuff
31:37 written down as clearly as possible. That's why in-depth scoping is required
31:40 to understand every little nuance of this build that could potentially pop up
31:44 and then you can bring it like a proposal like this to a line on during
31:48 the call. Where I'm at right now is as far as next steps, um I know we've had a
31:52 lot of calls which has been good. We've gotten um to know each other a little
31:55 bit. We've gotten to understand sort of the way we work. Um, and so my
32:00 recommendation here, next call we would want to, um, it would be like a project
32:04 kickoff call. Um, we'd really get started from there. We would have, um,
32:07 one of my one of my devs would hop on the call, too, just to make sure that
32:10 we're we're both aligned so that we could pretty much kick off um, right
32:15 away after that call. So, yeah. Um, any questions that you guys
32:19 have about moving forward with this? Um, one thing that I'd want to throw out is
32:23 I would love to hear I know we talked a little bit about you know end of January
32:27 love to see something but um with the more calls things have been delayed a
32:31 little bit. I would like to know what you guys are looking at here as far as
32:35 the urgency um ideally what is a good timeline for this project look like for
32:39 you um and then any final thoughts or questions that you've got.
32:43 >> Honestly as soon as possible um is the as the timeline. Um there is no urgency
32:48 from a business perspective. Um we're not promising this to anybody. This is
32:53 totally internal. Uh but it's it's all direct related, right? The sooner we get
32:57 this, the faster we can begin to do other things. One to test, you know, two
33:01 to figure out like our own processes will be augmented by this. So I'm
33:05 looking forward to having increased automation uh in our own systems. Um but
33:10 in terms of you know, we've got no client recommendations or anything like
33:13 that. Like this is barely an internal effort. Uh >> right. expo as soon as possible is ideal
33:20 and um I recognize that that's a balance. So yeah, so I guess I would say
33:24 you know you let me know what you think is possible. Um if if end of January
33:27 still for some reason becomes possible because it's actually pretty
33:31 straightforward, great. You know like to not have this go into March, right? So
33:36 we could say like, you know, by end of February. Um, but if we get into
33:40 February and you're like, hey, actually, we need to add write capabilities to
33:45 Dropbox because Notion's a dead end and we need to be able to update the rag
33:50 database uh in, you know, the the Google spreadsheet in Dropbox, for example,
33:54 right? Like I get that too. So >> yeah. Yeah. >> Okay. Awesome. I think that that aligns
33:59 really well with what we were thinking. Um I don't see this pushing into March.
34:04 Um just as a a ballpark at the moment. Definitely don't see this pushing into
34:08 March. Um espec especially since if if we sort of agree on that then you know
34:13 we'll we'll pretty much we'll make it a priority to make sure that it doesn't
34:16 push into March. Um unless any unforeseen things pop up and it's just
34:19 like really complicated for some reason which I don't think will be but um yeah.
34:24 Okay. So that sounds good to me. Um that would give us sort of you know four or
34:29 five check-ins. um throughout the process um just to make sure that we're
34:34 moving forward as as planned. Based on estimations, I don't really see this
34:40 taking anywhere upwards of um I would say 40 hours max would be my
34:46 estimate right now. Um, and so what I was thinking is ballpark. I think that maybe doing
34:55 $3,000 upfront and then maybe a $3,000 upon completion, which would put it a
35:01 $6,000 total project. Um, that's sort of what we're thinking based on, you know,
35:05 like I said, just scoping out um all the integrations and the the data in piece
35:10 up front. I think that that makes sense as far as um the timeline and sort of
35:15 the resourc that is so tough to watch. First of all, I'm not speaking confidently at all and
35:20 that kind of stuff is important when you're trying to sell the stuff because
35:24 I don't have a sales background. So, I don't have these sales tactics and I'm
35:27 not a pushy salesman. So, speaking like that and then you've also got the aspect
35:31 of just being transparent. So, telling straight up if I think something will
35:34 work or if I don't think something will work. Just trying to build goodwill. And
35:37 then the final thing is being able to back stuff with facts because ultimately
35:42 most decisions in life come down to a math problem where if you have data you
35:45 can make a strategic data driven decision. And what you just saw me do
35:48 there is I threw out a price. I was probably sitting there with my feet
35:52 shaking and just waiting to see what he would say back. But what I should have
35:55 done is been able to justify it by saying we're charging you this much.
35:59 This is why we've done the math and we're saving you this much time over
36:02 these different departments which basically translates to this much dollar
36:04 savings and you're going to make your money back in 2 3 months if not quicker
36:08 and then over the course of a year or two years you're netting tens of
36:11 thousands of dollars and then you just stand on that because that's objectively
36:15 true. But what I did is I said I think that this makes sense based on timeline
36:21 and resources and that's just horrible. If you think about the way the client
36:24 now perceives me in the future, if we want to work on more and more, which is
36:27 the goal, it's hard to break out of that barrier of now being perceived as an
36:31 hourly worker. Even though I tried to set that foundation up front in the
36:34 first and second discovery calls about the way that we want to ingest in their
36:38 business as a long-term partner. So really, really don't like the way that I
36:42 kind of approached the pricing in this conversation. Cut me some slack. This
36:46 was in December. But the reason why this worked here is a couple things. What I
36:49 still did well throughout this process was I used LRP. I listened to everything
36:53 he was saying. I repeated everything back and then I asked more questions.
36:56 And because I was able to scope out the build so well, we now have a full build
37:00 strategy that completely aligns with what exactly he was thinking. And
37:03 through the other calls where we had kind of bonded and we talked about kind
37:07 of passion projects and other use cases of AI, we got to this place where trust
37:10 had been established. So, a lot of stuff happened in those three calls, and I
37:13 obviously couldn't show you guys everything, but what I wanted to do is
37:17 now kind of just recap with six key lessons from what we just looked at. So,
37:20 the first one is something that I didn't do great in these three calls, which is
37:24 to prove the value with ROI. Collect your data before and after. Without this
37:27 baseline data, you cannot actually prove ROI. Keep it simple. Focus on three to
37:31 five key metrics that align with business goals like hours saved, lead
37:35 conversion, or error reduction. translate your technical improvements
37:38 into financial outcomes because clients can actually understand that. So like
37:41 with that example earlier about the vector database, rather than me talking
37:45 about the metadata or the uklidian distance or whatever technical jargon I
37:48 was throwing around, I should have just translated that to the agent being able
37:52 to respond cheaper, quicker, and more accurately. And hopping right off of
37:56 that ROI stuff, you want to price based on the outcomes, not the effort. Value
38:00 is based on result. Value is not based on time. Never sell hours or number of
38:04 nodes or the complexity of the build. Sell business transformation. More
38:08 leads, less churn, reduced cost, whatever it is. And that's how you can
38:11 actually pitch out a number and confidently justify it rather than what
38:15 you just saw me do where I said, "I think maybe 6,000 makes sense. I think
38:20 that sounds good. That was just hard to watch." And once you start understanding
38:25 how to actually price based on value, it makes it clearer for you and you're not
38:29 going to undervalue or underpric your services, which is a big issue that we
38:33 had for a long, long time was severely underpricing tons of our builds. Master
38:37 consultative selling, not technical selling. Position yourself as a problem
38:41 solver, not a tool pusher. Ask deep questions in discovery, like what
38:44 business problems are we solving here? What result must this deliver? It's
38:48 important to be technical enough to understand how to scope out a build, but
38:52 speak not technical when you need to so that the person can actually understand
38:55 what you're saying and not get confused or overwhelmed. Use analogies and visual
38:59 aids instead of overwhelming the client. You can see in the call when I used that
39:03 intern analogy, he left it. Use a structured framework. And that doesn't
39:06 mean to enter the call with a script on your right or a set list of questions
39:09 that you're going to ask in order no matter what because you have to be fluid
39:12 and you have to be a human. But if you have a framework like LRP where you
39:16 listen as soon as you listen and it's your turn to speak you recap and then
39:19 you just poke another question that's going to help the conversation flow in a
39:21 way where they're speaking the most of the time which they should be in a
39:25 discovery call and where you're getting a ton of information that later you can
39:29 revisit the call, read the transcript and have it all there in front of you.
39:32 Turn every win into a growth engine. After showing that ROI, that's when you
39:36 can start to ask for video testimonials, case studies, referrals, stuff like
39:40 that. And that's going to help you in your discovery and your sales if you're
39:44 facing any objections or like you could see in these calls, I was able to talk
39:47 about other products I'm working on or other things I've done which help build
39:52 trust for this client in me and prove that I could do what he was looking for.
39:55 And then finally, just to maintain your sales rhythm, make sure you're
39:58 researching and you're prepping before a call, not just hopping into it a minute
40:01 late with no idea who the person is. Listen for 70% of the time, speak for
40:05 30% of the time. Focus conversation on business value, not technical details.
40:09 unless you're specifically asked because sometimes the clients do like to nerd
40:12 out a little bit as well. And then follow up with low pressure, highv value
40:16 touch points, insights, quick wins, or relevant examples. So, that's going to
40:18 do it for this one. What I'm going to do is drop this document as well as a
40:22 resource that pretty much broke down everything that I just broke down in
40:26 those three calls, and I'll drop that in my free school community. All you got to
40:29 do if you want to download that is hop in using the link in the description.
40:32 Like I said, it's completely free. And once you get in here, just search for
40:35 the title of this video or go to YouTube resources and find the post associated
40:39 with this YouTube video. And in there, there'll be a PDF just like this where
40:43 you'll have all of the resources that we just talked about. And if you're looking
40:45 to take your skills further and connect with people who are running businesses
40:49 using NADN, then definitely check out my plus community. The link for that's also
40:52 down in the description. Not only is this a great community of people who are
40:55 passionate about end and monetizing their skills, but by the time you're
40:57 seeing this, we would have just dropped this new course called Automation to
41:01 Monetization where you learn how to actually launch your AI freelancing or
41:04 consulting business. So, if you like this type of content that you've been
41:06 seeing from me on YouTube, this full course covers a lot more. Anyways,
41:10 that's going to do it for this one. If you learned something new or you enjoyed
41:12 the video, please give it a like. It definitely helps me out a ton. And as
41:16 always, I appreciate you guys making it to the end of the video. I'll see you on
$

How I Sold a $6000 AI Workflow to a Business (Live Sales Call)

@nateherk 41:19 4 chapters
[AI agents and automation][productivity and workflows][open source and self-hosting]
// chapters
// description

Full courses + unlimited support: https://www.skool.com/ai-automation-society-plus/about All my FREE resources: https://www.skool.com/ai-automation-society/about 14 day FREE n8n trial: https://n8n.partnerlinks.io/22crlu8afq5r Code NATEHERK to Self-Host n8n for 10% off (annual plan): http://hostinger.com/nateherk Hey guys, so in this video I'm gonna be showing me live selling $6,000 AI Workflow to a business. I''ll be reacting live, and diving deep into the good and the bad parts of these sales

now: 0:00
// tags
[AI agents and automation][productivity and workflows][open source and self-hosting]