You're probably paying too much for AI

00:00:00,100 --> 00:00:27,820 [Eric] [upbeat music] Welcome back to the Token Intelligence Show, where we help you and ourselves think through how to use AI well, uh, in this new era that we're in with this new technology. So John, the question you asked me earlier this week- 00:00:27,820 --> 00:00:29,080 [John] At lunch. At lunch 00:00:29,080 --> 00:00:30,370 [Eric] ... at lunch today actually. 00:00:30,370 --> 00:00:30,380 [John] Or earlier too. 00:00:30,380 --> 00:00:30,700 [Eric] Yeah, yeah, 00:00:32,040 --> 00:00:32,400 [Eric] is 00:00:33,660 --> 00:00:39,690 [Eric] whether AI is actually too expensive for some businesses. 00:00:39,690 --> 00:00:39,720 [John] Yeah. 00:00:39,720 --> 00:00:42,430 [Eric] Which I think is a great question because 00:00:43,940 --> 00:00:47,210 [Eric] many, many people are trying some form of AI, 00:00:48,460 --> 00:00:59,780 [Eric] but if you ask the average person trying to implement AI in a business, "Do you feel like you're getting unbelievable ROI when you look at the actual cost at the end of the day?" 00:01:01,360 --> 00:01:03,640 [Eric] Not everyone can answer immediately. 00:01:03,640 --> 00:01:12,180 [John] Yeah. Yeah. I think, and I think there's two ways to look at the problem. One is the current economics, and one is the rapidly approaching future economics. 00:01:12,180 --> 00:01:12,470 [Eric] Yep. 00:01:12,470 --> 00:01:22,220 [John] 'Cause this came up in my mind 'cause there's been a lot of news this week around OpenAI and Anthropic making moves where clearly they're prepping for IPO. 00:01:22,220 --> 00:01:22,360 [Eric] Yep. 00:01:22,360 --> 00:01:34,320 [John] And once that happens, most people believe that, that the current subsidies where they're, where they're... have these plans that are unlimited where you use thousands of dollars potentially of, of, of credits- 00:01:34,320 --> 00:01:34,330 [Eric] Right 00:01:34,330 --> 00:01:36,399 [John] ... like that goes away or changes at least. 00:01:36,400 --> 00:01:37,100 [Eric] Yep. 00:01:37,100 --> 00:01:46,020 [John] Um, so then businesses are, you know, w- gonna be left with, especially the mid-market businesses, 'cause a lot of enterprise maybe is already paying the, the true cost. 00:01:46,020 --> 00:01:46,260 [Eric] Yep. 00:01:46,260 --> 00:01:51,520 [John] But especially mid-market's gonna be left with, "Oh, like, can we afford this? Like, does this make sense?" 00:01:51,580 --> 00:02:02,400 [Eric] So today we're gonna answer that question. Let's figure out if AI is worth the cost, and we're gonna do that by evaluating really three major 00:02:03,500 --> 00:02:14,720 [Eric] vectors through which you can determine whether or not you might get high ROI. So we'll start with number one, and I'm gonna ask you this 'cause you have done this for your business. 00:02:14,720 --> 00:02:15,460 [John] Right. 00:02:15,460 --> 00:02:23,120 [Eric] And you are implementing this for other people's businesses. The first thing, and this is interesting, I see this a ton, 00:02:24,280 --> 00:02:28,160 [Eric] is, uh, ability to use AI well. 00:02:29,260 --> 00:02:34,880 [Eric] Anyone can open an AI tool and ask it a question, 00:02:35,900 --> 00:02:42,459 [Eric] right? But to actually be an advanced user is pretty hard. Speak to that a little bit. 00:02:42,460 --> 00:02:47,820 [John] Yeah. I, I think the, the range of, of people's abilities is huge. 00:02:47,820 --> 00:02:48,800 [Eric] Mm-hmm. 00:02:48,800 --> 00:02:52,820 [John] And the range of people's perceptions of their own abilities is also huge. 00:02:52,820 --> 00:02:53,470 [Eric] Yep. 00:02:53,470 --> 00:02:56,860 [John] Right? So like, yeah, use AI, AI all the time, right? 00:02:56,860 --> 00:02:57,380 [Eric] Mm-hmm. 00:02:57,380 --> 00:03:04,080 [John] Or, or I don't. Um, either way, it, it... the g- the value might not be there. You could be using it all the time- 00:03:04,080 --> 00:03:04,090 [Eric] Right 00:03:04,090 --> 00:03:05,400 [John] ... not getting v- very much value. 00:03:05,400 --> 00:03:05,420 [Eric] Right. 00:03:05,420 --> 00:03:06,620 [John] So it's not just a usage problem. 00:03:06,620 --> 00:03:07,440 [Eric] Yep. 00:03:07,440 --> 00:03:16,560 [John] And, and I think that we've covered this before, but the biggest gap is, um, are you using it to help you think better? I think the best people, like that's how I'd categorize- 00:03:16,560 --> 00:03:16,780 [Eric] Mm-hmm 00:03:16,780 --> 00:03:25,680 [John] ... like how they're using AI, um, because that's your leverage, right? If you, if you think better and you approach the problem better, you can solve the problem better and get the highest leverage. 00:03:25,680 --> 00:03:25,940 [Eric] Yep. 00:03:25,940 --> 00:03:27,060 [John] Whereas if you, um, 00:03:28,080 --> 00:03:37,840 [John] are just either, one, using it like Google and just asking it, you know, a question that you could Google, or two, maybe trying to tell it to do something for you. 00:03:37,840 --> 00:03:37,960 [Eric] Yep. 00:03:37,960 --> 00:03:49,799 [John] Check my email. Um, that, that there's some value there, but, but the leverage is, is how many steps back can you take from that to think through, you know, bigger and bigger problems and solve them. 00:03:49,800 --> 00:03:49,940 [Eric] Yep. 00:03:50,950 --> 00:03:51,329 [Eric] Yep. The... 00:03:52,560 --> 00:04:03,020 [Eric] Going one step deeper, actually, I think there are some techniques that are extremely powerful that very few people actually use. 00:04:03,020 --> 00:04:03,030 [John] Yeah. 00:04:03,030 --> 00:04:05,600 [Eric] And I'll give you one example that I've seen. So 00:04:06,900 --> 00:04:07,760 [Eric] you can 00:04:08,920 --> 00:04:15,200 [Eric] use AI tools in a, uh, connected to folders that are on your computer- 00:04:15,200 --> 00:04:15,260 [John] Yes 00:04:15,260 --> 00:04:26,800 [Eric] ... or your desktop, and so I'll use Claude Cowork as the example. So normally, you know, if you go to chatgpt.com, if you have an account, you log in, and then you can ask it questions, you can upload files, et cetera. 00:04:28,020 --> 00:04:28,360 [Eric] And 00:04:29,540 --> 00:04:37,240 [Eric] the... We now have the ability to connect to a folder on your computer, and so Claude Cowork does this really well, but a, a lot of tools do it. 00:04:38,360 --> 00:04:38,760 [Eric] And 00:04:39,960 --> 00:04:59,080 [Eric] it can be way more effective for analyzing a large number of files, analyzing a bunch of disparate files, um, and it's faster because it's running on your local machine and reading those files locally as opposed to having to translate that into an API call, you know, and hitting a service in the cloud. Um, 00:05:00,220 --> 00:05:04,040 [Eric] you know, and, and processing the entire file as part of that payload. And so 00:05:05,060 --> 00:05:13,620 [Eric] th- those types of things can make a big difference, but I think that most people, average users, are not using those. Do you agree? 00:05:13,620 --> 00:05:23,760 [John] Yeah. I mean, yeah, we were, I mean, we were just talking about this, where th- there are even some ways that the tool like leads you down as far as how to use it that aren't necessarily the best. 00:05:23,760 --> 00:05:24,650 [Eric] Hmm. 00:05:24,650 --> 00:05:35,880 [John] 'Cause if I were onboarded, I mean, I haven't been 'cause I've been using it a long time, but the onboarding flow of like you created an account, you chat with the thing, or these little pop-ups of like, "Add a connector," or add a whatever they're called. 00:05:35,880 --> 00:05:35,920 [Eric] Yep. 00:05:35,920 --> 00:05:39,900 [John] Integration. And then, you know, Google Drive might pop up or whatever- 00:05:39,900 --> 00:05:39,910 [Eric] Mm-hmm 00:05:39,910 --> 00:05:46,220 [John] ... whatever common tools that, that people use, and you add all those things, and then you just assume like that's the best way to use it. 00:05:46,220 --> 00:05:46,479 [Eric] Right. 00:05:46,480 --> 00:06:01,000 [John] But that's a perfect example of if you're trying to do a really major project to consolidate tons of information, the actual optimum way to do that is to have the physical files, PDFs, text files- 00:06:01,000 --> 00:06:01,440 [Eric] Yep 00:06:01,440 --> 00:06:10,552 [John] ... um, Word docs, whatever they are, on your computer, and then have the AI access it via your computer. It'sYou have to wait. It, it works better and it's faster. 00:06:10,552 --> 00:06:10,932 [Eric] Yep. 00:06:10,932 --> 00:06:12,792 [John] Um, and that's just a very simple example. 00:06:12,792 --> 00:06:20,772 [Eric] Yep, yep. But there are lots of, there are lots of examples of-- there are lots of similar examples of different techniques and, you know, it sort of varies by discipline. 00:06:20,772 --> 00:06:20,952 [John] Right. 00:06:22,032 --> 00:06:26,422 [Eric] What-- So in the default state, I would actually say 00:06:27,532 --> 00:06:28,262 [Eric] if you 00:06:29,292 --> 00:06:33,152 [Eric] get a cheap GPT or Claude plan- 00:06:33,152 --> 00:06:33,972 [John] Mm-hmm 00:06:33,972 --> 00:06:35,192 [Eric] ... for people in your business, 00:06:36,492 --> 00:06:42,952 [Eric] let's just say it's the $20 a month plan, I think you get a lot of ROI probably- 00:06:42,952 --> 00:06:43,072 [John] Yeah 00:06:43,072 --> 00:06:46,472 [Eric] ... from the standpoint of personal productivity. Right? 00:06:46,472 --> 00:06:59,452 [John] I think there's a potential for a lot of ROI, but not necessarily 'cause I would imagine that like most software, like how do SaaS companies make money? They make money for the-- at any given point in time, only a fraction of the people are using the software, right? 00:06:59,452 --> 00:06:59,602 [Eric] Hmm. 00:06:59,602 --> 00:07:00,752 [John] That's part of the making money. 00:07:00,752 --> 00:07:02,992 [Eric] Oh, interesting. It's adop- yeah, it's an adoption challenge. 00:07:02,992 --> 00:07:03,532 [John] Right. 00:07:03,532 --> 00:07:03,572 [Eric] Yep. 00:07:03,572 --> 00:07:09,912 [John] So and then, then you buy a corporate software, like, "Okay, we need an AI thing. We'll pick this one," you know, maybe Claude or ChatGPT. 00:07:09,912 --> 00:07:10,612 [Eric] Yeah. 00:07:10,612 --> 00:07:13,652 [John] And then you buy it for 1,000 people. 00:07:13,652 --> 00:07:14,372 [Eric] Yep. 00:07:14,372 --> 00:07:18,172 [John] And like who uses it every day? Like what fraction? 00:07:18,172 --> 00:07:19,302 [Eric] Yeah, that's interesting. 00:07:19,302 --> 00:07:19,332 [John] 10%? 00:07:19,332 --> 00:07:19,642 [Eric] Yeah, that's such- 00:07:19,642 --> 00:07:20,522 [John] Who uses it once a week? 00:07:20,522 --> 00:07:21,632 [Eric] ... such a good point. Yep. 00:07:21,632 --> 00:07:31,472 [John] Yeah. And then I think you just have this long... And then, and then if we looked at-- And then eventually IT comes in and audits, like, "Oh, all these people haven't used it for 90 days. We're taking your license away." [chuckles] Type of thing. 00:07:31,472 --> 00:07:32,422 [Eric] Right, right, right. Yeah, it happens. 00:07:32,422 --> 00:07:33,052 [John] But yeah. 00:07:33,052 --> 00:07:35,692 [Eric] Yeah. Uh, I think it's a great point. Uh, the-- 00:07:37,052 --> 00:07:39,092 [Eric] So I'll-- this is my takeaway 00:07:40,232 --> 00:07:46,932 [Eric] from, you know, actually becoming a power user of AI and that creating a higher ROI- 00:07:46,932 --> 00:07:47,292 [John] Yeah 00:07:47,292 --> 00:07:56,112 [Eric] ... is that you can't just drop AI into a business and expect people to create a huge amount of leverage for it. It's probably going to be a smaller number of people 00:07:57,252 --> 00:08:06,242 [Eric] who are power users because they naturally gravitate towards that. What I've seen is that generally those are people who are power users of other types of software anyways, right? 00:08:06,242 --> 00:08:10,112 [John] Well, and the fun thing is that OpenAI and Anthropic agree with you. 00:08:10,112 --> 00:08:10,572 [Eric] [chuckles] 00:08:10,572 --> 00:08:13,692 [John] 'Cause they've started consulting arms in the last few weeks. 00:08:13,692 --> 00:08:14,711 [Eric] Yeah. Yeah. Yeah. 00:08:14,712 --> 00:08:15,012 [John] Right? 00:08:15,012 --> 00:08:15,791 [Eric] Yeah. Totally. 00:08:15,791 --> 00:08:16,271 [John] So- 00:08:16,272 --> 00:08:16,662 [Eric] Okay. So takeaway- 00:08:16,662 --> 00:08:17,652 [John] I think that's very true 00:08:17,652 --> 00:08:24,752 [Eric] ... takeaway one is you actually have to be proactive, do some training, actually have an adoption program- 00:08:24,752 --> 00:08:25,032 [John] Mm-hmm 00:08:25,032 --> 00:08:29,672 [Eric] ... in order t, you know, to, for people to leverage the tool. 00:08:29,672 --> 00:08:40,192 [John] And one more point on that. There's been a lot of criticism in the tech world, and this may spread outside of that, for incentivizing people to basically use the maximum amount of AI possible. 00:08:40,192 --> 00:08:40,712 [Eric] Hmm. 00:08:40,712 --> 00:08:43,872 [John] Um, token maxing is kind of the tech world term for it. 00:08:43,872 --> 00:08:44,272 [Eric] Yep. 00:08:44,272 --> 00:08:44,472 [John] Um, 00:08:45,712 --> 00:08:48,732 [John] and valid criticism 'cause people are doing wasteful things with it. 00:08:48,732 --> 00:08:49,592 [Eric] Yep. Yep. 00:08:49,592 --> 00:09:03,611 [John] But I think it's a little unfair, um, and businesses still have to think about how to incentivize people to use it. Maybe there's a better way, but I would say some way of incentivizing is probably better than none. 00:09:03,612 --> 00:09:04,092 [Eric] Yep. 00:09:04,092 --> 00:09:15,572 [John] Um, which is why I think people are doing the, like they're doing leaderboards of like who's using it the most. They're doing these different things that just incentivize the behavior that obviously can be wasteful or people can be, can game the system. But that doesn't- 00:09:15,572 --> 00:09:15,692 [Eric] Yep 00:09:15,692 --> 00:09:19,792 [John] ... mean that like, like there, there isn't a reason to encourage people to use it. 00:09:19,792 --> 00:09:31,952 [Eric] Yeah. I think one of the learnings here, in addition to you have to be proactive, is that it's not magic. You still have to learn how to use a tool, and people still have to change their behavior- 00:09:31,952 --> 00:09:32,772 [John] Right 00:09:32,772 --> 00:09:35,172 [Eric] ... which is, which is hard for anything. 00:09:35,172 --> 00:09:35,592 [John] Right. 00:09:35,592 --> 00:09:36,912 [Eric] Uh, I think once people, 00:09:38,432 --> 00:09:44,432 [Eric] once people generally f-- like once they see the l- you know, see the light, we'll say, or sort of feel the magic of- 00:09:44,432 --> 00:09:44,442 [John] Mm-hmm 00:09:44,442 --> 00:09:46,932 [Eric] ... wow, this used to take hours and now it takes minutes, 00:09:48,332 --> 00:09:49,792 [Eric] that can help accelerate 00:09:51,052 --> 00:09:55,572 [Eric] adoption because people, you know, actually feel that they're more productive, uh- 00:09:55,572 --> 00:09:55,602 [John] Yes 00:09:55,602 --> 00:09:56,792 [Eric] ... you know, and they are more productive. 00:09:56,792 --> 00:09:56,892 [John] Right. 00:09:56,892 --> 00:09:58,611 [Eric] But it's not just flipping a switch to get there. 00:09:58,612 --> 00:09:58,752 [John] Yep. 00:09:59,832 --> 00:10:06,332 [Eric] Okay. The second area here is actually measuring ROI. 00:10:06,332 --> 00:10:06,372 [John] Yes. 00:10:06,372 --> 00:10:17,052 [Eric] So let's say that you, you know, buy a license for your business or buy licenses, you know, for the people in your business, and you do a training program, so you're onboarding people. 00:10:18,392 --> 00:10:33,472 [Eric] That is a foundation for, you know, creating leverage through this because people are actually adopting it. But then you get to the point where you have to, you actually have to [chuckles] make a call on are we getting leverage, which is an ROI calculation. 00:10:33,472 --> 00:10:33,812 [John] Right. 00:10:33,812 --> 00:10:45,032 [Eric] Which you would do with any technology. We adopt this technology, it costs us some amount of money, and then we have to decide are we getting ROI from that spend, right? So what's your- 00:10:45,032 --> 00:10:47,232 [John] And over what time period is acceptable. 00:10:47,232 --> 00:10:53,872 [Eric] Yeah, exactly. Right? So I mean, if you're switching ERP systems, you have a longer time horizon. 00:10:53,872 --> 00:10:54,212 [John] Mm-hmm. 00:10:54,212 --> 00:11:00,372 [Eric] Right? Of course, because there's gonna be a big migration and, you know, and you may actually regress for some amount of time- 00:11:00,372 --> 00:11:00,702 [John] Right 00:11:00,702 --> 00:11:11,212 [Eric] ... in order to get the future gains, right? As opposed to, you know, adopting a project management system, you would hope that in a quarter, you know, you do the migration and things are getting better. 00:11:11,212 --> 00:11:11,752 [John] Right. 00:11:11,752 --> 00:11:15,392 [Eric] Uh, h- what is your mental model though for 00:11:16,512 --> 00:11:20,892 [Eric] ROI with AI or technology in general? 00:11:20,892 --> 00:11:27,872 [John] So with ROI, I, I think most people jump to cost, but obviously there's two components to ROI. 00:11:27,872 --> 00:11:27,992 [Eric] Yep. 00:11:27,992 --> 00:11:29,292 [John] Revenue and cost. 00:11:29,292 --> 00:11:29,772 [Eric] Yep. 00:11:29,772 --> 00:11:30,092 [John] Um- 00:11:30,092 --> 00:11:35,512 [Eric] Which are, just explain those at a basic level, like if you're, I mean, if you're buying software for your business. 00:11:35,512 --> 00:11:35,632 [John] Mm-hmm. 00:11:36,732 --> 00:11:40,812 [John] Yeah. Right. So, so at a basic level and using AI specifically, 00:11:41,872 --> 00:11:54,172 [John] um, cost, like for services business is, is, um, some reduction... [chuckles] Well, actually, ironically, um, it's complicated actually for services 'cause if it, if it's a reduction in hours- 00:11:54,172 --> 00:11:54,182 [Eric] Hmm 00:11:54,182 --> 00:11:57,172 [John] ... and the service is some kind of retainer thing, there's a value. 00:11:57,172 --> 00:11:57,262 [Eric] Right. 00:11:57,262 --> 00:12:00,592 [John] But if it's reduction in hours and it's hourly, like it's confusing. 00:12:00,592 --> 00:12:00,712 [Eric] Yep. Yep. 00:12:00,712 --> 00:12:14,152 [John] Um, but generally, like let's say you're, you're selling a physical good or you're, um, have some, you know, some kind of incentive for more efficiency, um, which mo- every business does, like, you know- 00:12:14,152 --> 00:12:14,162 [Eric] Yep 00:12:14,162 --> 00:12:16,728 [John] ... one way or another.That, that's the easy part. That's the cost- 00:12:16,728 --> 00:12:16,908 [Eric] Yeah 00:12:16,908 --> 00:12:17,788 [John] ... reduction part. 00:12:17,788 --> 00:12:23,507 [Eric] A, a very simple example of that is expense tracking or time tracking- 00:12:23,508 --> 00:12:24,548 [John] Yeah 00:12:24,548 --> 00:12:27,908 [Eric] ... or other things like that where in almost any type of business 00:12:29,268 --> 00:12:33,907 [Eric] you would adopt that software in order to more closely track expenses- 00:12:33,908 --> 00:12:34,158 [John] Right 00:12:34,158 --> 00:12:39,648 [Eric] ... to implement accountability or to understand is there slippage in time, you know? 00:12:39,648 --> 00:12:39,788 [John] Mm-hmm. 00:12:39,788 --> 00:12:43,528 [Eric] In, in the time or, or where are employees allocating time. 00:12:43,528 --> 00:12:45,878 [John] Yep, or make it easy for employees to allocate time. 00:12:45,878 --> 00:12:46,498 [Eric] Right, exactly. 00:12:46,498 --> 00:12:46,508 [John] Yeah. Yeah. 00:12:46,508 --> 00:12:48,618 [Eric] And so, and so there's a clear cost savings there, right? 00:12:48,618 --> 00:12:48,768 [John] Right. 00:12:48,768 --> 00:12:56,288 [Eric] Like, okay, we expect that we're gonna pay this amount for this expense tracking thing, and we're actually going to be able to hold people accountable- 00:12:56,288 --> 00:12:56,318 [John] Right 00:12:56,318 --> 00:12:57,868 [Eric] ... or whatever it is. 00:12:57,868 --> 00:13:07,708 [John] Yeah. Yeah. So that, that's... Yeah, that's the cost component. The revenue component as far as we made more revenue from AI is, is interesting. 00:13:07,708 --> 00:13:08,108 [Eric] Hmm. 00:13:08,108 --> 00:13:13,648 [John] Um, for two reasons. One, there's a ton of AI FOMO out there. 00:13:13,648 --> 00:13:13,948 [Eric] Hmm. 00:13:13,948 --> 00:13:17,268 [John] Right? Of like, "Oh, well they're using I- AI to do this thing." 00:13:17,268 --> 00:13:17,608 [Eric] Yep. 00:13:17,608 --> 00:13:19,168 [John] Um, and underneath that FOMO 00:13:20,288 --> 00:13:24,088 [John] is what I'm gonna call fuzzy attribution. 00:13:24,088 --> 00:13:24,448 [Eric] Hmm. 00:13:24,448 --> 00:13:28,528 [John] So, you know, headline story, whatever company, 00:13:29,708 --> 00:13:33,807 [John] you know, typically it's saving money. You know, saving money is the easier measurable- 00:13:33,808 --> 00:13:36,968 [Eric] The, this goes back years, but the first major example of this was Klarna- 00:13:38,148 --> 00:13:38,318 [John] Yeah. Yeah 00:13:38,318 --> 00:13:39,968 [Eric] ... saying, "We, we let go of-" 00:13:39,968 --> 00:13:40,807 [John] Customer support, however many people 00:13:40,807 --> 00:13:42,658 [Eric] ... this huge number. Hundreds of people. 00:13:42,658 --> 00:13:43,728 [John] Yeah. And then they hired them back for 00:13:43,728 --> 00:13:44,348 [Eric] And then they hired. 00:13:44,348 --> 00:13:44,378 [John] Yeah. 00:13:44,378 --> 00:13:45,718 [Eric] Yeah, I don't think all of them, but yeah. 00:13:45,718 --> 00:13:45,728 [John] Yeah, yeah, yeah. 00:13:45,728 --> 00:13:47,448 [Eric] But I mean, it was kind of one of those. 00:13:47,448 --> 00:13:47,688 [John] Right. 00:13:48,808 --> 00:13:48,988 [Eric] Yeah. 00:13:48,988 --> 00:13:51,428 [John] But that's on the cost side. I think there's a revenue example too- 00:13:51,428 --> 00:13:51,438 [Eric] Mm-hmm 00:13:51,438 --> 00:14:00,148 [John] ... where like let's say, you know, our salespeople were able to double their outbound calls or, um, qualifying sales- 00:14:00,148 --> 00:14:00,158 [Eric] Right 00:14:00,158 --> 00:14:02,048 [John] ... like whatever you... Something, something in that- 00:14:02,048 --> 00:14:02,167 [Eric] Mm-hmm 00:14:02,167 --> 00:14:04,468 [John] ... you know, sales arena, which is real, 00:14:05,528 --> 00:14:07,428 [John] but, but there's a lot of components where, 00:14:08,528 --> 00:14:11,928 [John] where like was AI truly the full attribution there? 00:14:11,928 --> 00:14:12,027 [Eric] Hmm. 00:14:12,028 --> 00:14:18,748 [John] Or is it because we really had to hone a process to let AI do it, for example? 00:14:18,748 --> 00:14:18,808 [Eric] Yeah. 00:14:18,808 --> 00:14:27,608 [John] Like, there's all this like process internal business work that like we say AI, but what maybe we actually did was just carefully map the process. 00:14:27,608 --> 00:14:27,617 [Eric] Hmm. 00:14:27,617 --> 00:14:29,388 [John] And we do have this thing that probably, 00:14:30,528 --> 00:14:31,368 [John] probably saves time, 00:14:32,788 --> 00:14:34,518 [John] but part of it was the business process thing. 00:14:34,518 --> 00:14:34,528 [Eric] Right. Right. 00:14:34,528 --> 00:14:36,048 [John] We didn't need the AI. 00:14:36,048 --> 00:14:37,568 [Eric] Can you think of an example where 00:14:38,588 --> 00:14:40,228 [Eric] if you used AI to do 00:14:41,288 --> 00:14:49,068 [Eric] something that would traditionally have been difficult or, you know, w- that is a clear path to revenue? Is there any example that comes to mind around- 00:14:49,068 --> 00:14:49,258 [John] Yeah 00:14:49,258 --> 00:14:51,728 [Eric] ... a direct relationship there? 00:14:51,728 --> 00:14:56,397 [John] Yeah. Um, we just did an episode on this. AI is really good for research. 00:14:56,397 --> 00:14:56,868 [Eric] Hmm. 00:14:56,868 --> 00:15:00,668 [John] So researching prospects is a great use of AI that, that- 00:15:00,668 --> 00:15:00,737 [Eric] Oh, yeah 00:15:00,737 --> 00:15:05,328 [John] ... legitimately saves time and can le- I guess leads to cost reduction potentially- 00:15:05,328 --> 00:15:05,338 [Eric] Yeah 00:15:05,338 --> 00:15:13,568 [John] ... but also can kind of lead to revenue because leads to revenue, my argument there is like you probably just weren't gonna do it, were you? Um, and if you weren't- 00:15:13,568 --> 00:15:13,578 [Eric] Yeah 00:15:13,578 --> 00:15:14,628 [John] ... and then those calls go better- 00:15:14,628 --> 00:15:14,718 [Eric] Yep 00:15:14,718 --> 00:15:18,308 [John] ... or you reach out to be- more of the right people, then that is revenue. 00:15:18,308 --> 00:15:20,168 [Eric] Yeah, you can make more calls, right? 00:15:20,168 --> 00:15:20,348 [John] Right. 00:15:20,348 --> 00:15:21,928 [Eric] More calls means- 00:15:21,928 --> 00:15:22,368 [John] Right 00:15:22,368 --> 00:15:24,598 [Eric] ... you know, it's a numbers game, and so there's more at the bottom of the funnel. 00:15:24,598 --> 00:15:28,048 [John] So that, I think that is the most obvious- 00:15:28,048 --> 00:15:28,708 [Eric] Yep 00:15:28,708 --> 00:15:30,957 [John] ... one is... Yeah. 00:15:30,957 --> 00:15:39,228 [Eric] Uh, the, uh, A/B testing, um, ads or creative content or, you know, publishing pages- 00:15:39,228 --> 00:15:40,168 [John] Okay. Yeah 00:15:40,168 --> 00:15:41,928 [Eric] ... could be another example that comes to mind for me. 00:15:41,928 --> 00:15:46,508 [John] Yeah. I, I, I think a lot of companies can really up their, their creative game. 00:15:46,508 --> 00:15:47,038 [Eric] Yeah, exactly. 00:15:47,038 --> 00:15:49,948 [John] Like the average company's Facebook ad- 00:15:49,948 --> 00:15:49,968 [Eric] Mm-hmm 00:15:49,968 --> 00:15:55,138 [John] ... like a display ad or like a, like a lot of people wouldn't have done video at all 'cause it's too expensive. 00:15:55,138 --> 00:15:55,188 [Eric] Yep. 00:15:55,188 --> 00:15:58,668 [John] Now, like maybe you can do an AI video and, and do better- 00:15:58,668 --> 00:15:58,798 [Eric] Yep 00:15:58,798 --> 00:16:04,168 [John] ... than you would have, and then probably even, depending on the company, up their game as far as a graphic ad. 00:16:04,168 --> 00:16:04,268 [Eric] Yep. 00:16:04,268 --> 00:16:05,408 [John] Sure. 00:16:05,408 --> 00:16:09,948 [Eric] Okay. I have two takeaways, and then I'm gonna add one additional thought to this. 00:16:11,328 --> 00:16:12,408 [Eric] So in terms of ROI, 00:16:13,428 --> 00:16:14,428 [Eric] you have cost and revenue, 00:16:15,608 --> 00:16:30,608 [Eric] and the takeaway on cost is I think just do the math and, and try to evaluate AI as, as you would any other service, right? It costs you a certain amount of money. Is it actually going to save you some amount of money, right? 00:16:30,608 --> 00:16:30,838 [John] Right. 00:16:30,838 --> 00:16:32,308 [Eric] And I think defining the metrics there is important. 00:16:32,308 --> 00:16:38,778 [John] To, to be measured in hours or potentially reducing usage of other software. 00:16:38,778 --> 00:16:38,818 [Eric] Yes. 00:16:38,818 --> 00:16:40,248 [John] Are probably two of the big buckets. 00:16:40,248 --> 00:16:50,618 [Eric] Yep, yep. So let's say that you can wire, you know, an AI interface up to like a service, whatever it is, a CRM, a w- 00:16:50,618 --> 00:16:50,618 [John] Mm-hmm 00:16:50,618 --> 00:16:56,588 [Eric] ... whatever that is, a database or something, and so you actually have less human users of that- 00:16:56,588 --> 00:16:57,428 [John] Mm-hmm 00:16:57,428 --> 00:17:02,048 [Eric] ... you know, service, right? Or you have an agent that they can interact with that's the primary user of whatever- 00:17:02,048 --> 00:17:02,268 [John] Right 00:17:02,268 --> 00:17:03,348 [Eric] ... that service is, right? 00:17:03,348 --> 00:17:03,508 [John] Yeah. 00:17:03,508 --> 00:17:06,508 [Eric] So instead of paying for 20 seats, you pay for two seats or something like that. 00:17:06,508 --> 00:17:13,768 [John] Yep. Yep. Yeah, I mean, the, the easy one is, yeah, reduce the seat, reduce seat count until the, until SaaS companies catch on. 00:17:13,768 --> 00:17:13,918 [Eric] Yeah. Yep. 00:17:13,918 --> 00:17:25,548 [John] And then on the other one is the human time. So if it used to be 20 hours of an analyst and 20 people had to have seat counts, and we went to like 10 hours of an analyst and 10 seat counts, then there's savings on both sides. 00:17:25,548 --> 00:17:25,667 [Eric] Yep. 00:17:25,668 --> 00:17:26,168 [John] Which is real. 00:17:26,168 --> 00:17:26,278 [Eric] Yep. 00:17:26,278 --> 00:17:27,708 [John] And that happens today. 00:17:27,708 --> 00:17:29,368 [Eric] Yep. And then 00:17:30,928 --> 00:17:43,108 [Eric] the takeaway that I heard on the revenue side is really pick use cases that AI, uh, especially starting out, where there is a pretty clear line to revenue. 00:17:43,108 --> 00:17:43,428 [John] Right. 00:17:43,428 --> 00:17:44,588 [Eric] And it's not as blurry. 00:17:44,588 --> 00:17:45,508 [John] Right. 00:17:45,508 --> 00:17:56,988 [Eric] Um, so that would be things like is it increasing throughput in the sales and marketing funnel? Is it increasing the amount of iteration you can do on the creative side? Uh, you know, those sorts of things. 00:17:56,988 --> 00:17:58,048 [John] Yep. Agreed. 00:17:58,048 --> 00:18:02,448 [Eric] Um, are you developing prototype features and getting them in front of customers faster, right? 00:18:02,448 --> 00:18:03,708 [John] Yep. Oh, yeah. That's another good one. 00:18:03,708 --> 00:18:03,948 [Eric] Um, 00:18:05,228 --> 00:18:05,808 [Eric] the- 00:18:06,908 --> 00:18:07,348 [John] Yep 00:18:07,348 --> 00:18:07,948 [Eric] ... other- 00:18:07,948 --> 00:18:14,708 [John] Customization in the... And I have to bring this one up 'cause it's so big. In the sales cycle, because AI- 00:18:14,708 --> 00:18:15,228 [Eric] Hmm 00:18:15,228 --> 00:18:30,536 [John] ... is so fast, especially if you're in software... you know, world. To be able to show something, like, with dummy data of their thing, like more personalized and, and customized for, like, the deal and sales cycle, I think that's a huge one. 00:18:30,536 --> 00:18:31,436 [Eric] Yeah. Totally. 00:18:31,436 --> 00:18:32,236 [John] Yeah. 00:18:32,236 --> 00:18:33,676 [Eric] The one thing I'll add is 00:18:34,716 --> 00:18:35,536 [Eric] I think that, 00:18:37,056 --> 00:18:40,476 [Eric] I, I think that you can be too rigid 00:18:41,656 --> 00:18:43,256 [Eric] with trying to measure ROI- 00:18:43,256 --> 00:18:43,346 [John] Yeah 00:18:43,346 --> 00:18:44,776 [Eric] ... with, like, a hard cost or a hard revenue. 00:18:45,956 --> 00:18:52,736 [Eric] Because one thing that AI is wonderful for is helping people generate interesting ideas. 00:18:52,736 --> 00:18:53,346 [John] Mm-hmm. 00:18:53,346 --> 00:19:07,056 [Eric] 'Cause you can build things, you can research things, you can explore things way, way faster than you would before. And so you can actually generate ideas or explore different things or make really interesting comparisons far easier than you could in the past. 00:19:07,056 --> 00:19:07,536 [John] Mm-hmm. 00:19:07,536 --> 00:19:20,816 [Eric] And it sort of augments that. And so I think there's actually a creativity and innovation element that's really hard to measure, but that is really powerful because when you feel that spark of an idea inside of you, 00:19:21,916 --> 00:19:26,136 [Eric] interacting with AI is a really good way to fan that flame. 00:19:26,136 --> 00:19:26,336 [John] Yep. 00:19:26,336 --> 00:19:30,756 [Eric] You know? And so there's sort of an idea generation element to it that's pre-prototyping- 00:19:30,756 --> 00:19:30,956 [John] Yeah 00:19:30,956 --> 00:19:31,216 [Eric] ... or pre, 00:19:32,256 --> 00:19:45,356 [Eric] you know, um, creating a bunch of different ads, right? Um, which is, which is kind of interesting. Okay. The third, the third vector here is the actual hard cost of AI. 00:19:45,356 --> 00:19:46,116 [John] Yes. 00:19:46,116 --> 00:19:53,275 [Eric] And it was subsidized in the early days, and by when I say early days, it's, you know, not that [chuckles] long ago. 00:19:54,676 --> 00:19:57,056 [Eric] But it wasn't too long ago that you could buy 00:19:58,296 --> 00:20:03,076 [Eric] a relatively inexpensive subscription, so let's say a couple hundred dollars- 00:20:03,076 --> 00:20:03,276 [John] Mm-hmm 00:20:03,276 --> 00:20:12,296 [Eric] ... you know, to your major AI provider of choice, and essentially feel like you almost got unlimited usage of the, you know, the latest, greatest models, 00:20:13,516 --> 00:20:20,796 [Eric] and that has, uh, tightened up. Uh, and there are a number of reasons for that. There's compute capacity, right? 00:20:20,796 --> 00:20:21,096 [John] Mm-hmm. 00:20:21,096 --> 00:20:23,796 [Eric] So they're just running out of GPUs 00:20:24,876 --> 00:20:29,336 [Eric] and experiencing downtime, which is why X and Anthropic, you know, sort of- 00:20:29,336 --> 00:20:29,406 [John] Mm-hmm 00:20:29,406 --> 00:20:38,296 [Eric] ... made strange bedfellows in the deal with Colossus, where, um, these two theoretically opposed, you know, companies that were ... Or competing- 00:20:38,296 --> 00:20:38,425 [John] Competing 00:20:38,425 --> 00:20:39,425 [Eric] ... companies- 00:20:39,425 --> 00:20:39,425 [John] Yeah 00:20:39,425 --> 00:20:40,976 [Eric] ... um, actually strike a deal. 00:20:42,096 --> 00:20:43,296 [Eric] Very interesting. 00:20:43,296 --> 00:20:43,616 [John] Yep. 00:20:43,616 --> 00:20:43,806 [Eric] Um- 00:20:43,806 --> 00:20:48,316 [John] Now Colo- Colossus is the, for those that don't know, the data center- 00:20:48,316 --> 00:20:48,826 [Eric] Yes 00:20:48,826 --> 00:20:48,966 [John] ... that 00:20:50,176 --> 00:20:51,856 [John] I guess SpaceX- 00:20:51,856 --> 00:20:52,096 [Eric] Yes 00:20:52,096 --> 00:20:52,956 [John] ... owns right now. 00:20:52,956 --> 00:20:53,696 [Eric] Elon Musk. 00:20:53,696 --> 00:20:53,716 [John] Yeah. 00:20:53,716 --> 00:20:54,916 [Eric] It's Elon Musk's- 00:20:54,916 --> 00:20:54,946 [John] Elon Musk. There you go 00:20:54,946 --> 00:20:55,436 [Eric] ... data center. 00:20:55,436 --> 00:20:56,496 [John] There you go. [chuckles] 00:20:56,496 --> 00:20:59,726 [Eric] And Anthropic was running out of compute ca- compute ca- 00:20:59,726 --> 00:20:59,796 [John] Mm-hmm 00:20:59,796 --> 00:21:02,956 [Eric] ... capacity, and so they struck a deal, right? Um, 00:21:03,996 --> 00:21:06,986 [Eric] but also these companies are looking to go public. 00:21:06,986 --> 00:21:07,006 [John] Yeah. 00:21:07,006 --> 00:21:19,316 [Eric] And so tightening cost, and so subsidizing user acquisition or usage or whatever the things that you subsidize sort of on your, on your, you know, um, unbelievable unicorn growth trajectory. 00:21:19,316 --> 00:21:19,396 [John] Right. 00:21:19,396 --> 00:21:20,636 [Eric] Hyperscaler growth trajectory. 00:21:20,636 --> 00:21:21,656 [John] Right. 00:21:21,656 --> 00:21:26,205 [Eric] Um, and so now the actual cost of this is becoming more apparent, right? 00:21:26,205 --> 00:21:26,236 [John] Right. 00:21:26,236 --> 00:21:32,216 [Eric] The plans have gotten way less generous, and the def... This is what I, this is what I wanna know. 00:21:33,336 --> 00:21:38,636 [Eric] You use AI every day. You use it to run your business. You implement it for other companies. Do you think it's expensive or not? 00:21:40,196 --> 00:21:41,116 [John] That's a good question. 00:21:43,336 --> 00:21:46,196 [John] I mean, obviously the answer is it depends. Um- 00:21:46,196 --> 00:21:46,736 [Eric] Right 00:21:46,736 --> 00:21:48,356 [John] ... there are still tiers of usage 00:21:49,856 --> 00:21:52,436 [John] that are highly subsidized. 00:21:52,436 --> 00:21:53,296 [Eric] Yep. 00:21:53,296 --> 00:21:58,936 [John] ChatGPT's tiers not on company plans are highly subsidized. Same in Cloud- 00:21:58,936 --> 00:22:03,556 [Eric] At an ind- it is the, the individual level is way more subsidized, for sure. 00:22:03,556 --> 00:22:05,116 [John] Right. Right. Um- 00:22:05,116 --> 00:22:05,956 [Eric] Especially the middle tiers. 00:22:05,956 --> 00:22:09,176 [John] And, and businesses are totally using those, small businesses, right? 00:22:09,176 --> 00:22:09,596 [Eric] Right. 00:22:09,596 --> 00:22:10,016 [John] Yeah. So- 00:22:10,016 --> 00:22:11,176 [Eric] Because it... Yeah, because it's- 00:22:11,176 --> 00:22:12,496 [John] So there's that weird small- 00:22:12,496 --> 00:22:13,176 [Eric] More bang for your buck 00:22:13,176 --> 00:22:15,176 [John] ... small business prosumer- 00:22:15,176 --> 00:22:15,276 [Eric] Yep 00:22:15,276 --> 00:22:22,436 [John] ... stuff is still highly subsidized. Enterprise is getting less and less, if not, like, mostly not subsidized- 00:22:22,436 --> 00:22:22,506 [Eric] Yeah 00:22:22,506 --> 00:22:23,356 [John] ... at this point. 00:22:23,356 --> 00:22:23,796 [Eric] Yep. 00:22:23,796 --> 00:22:24,586 [John] Um... 00:22:24,586 --> 00:22:30,286 [Eric] Which is difficult because increasingly you really need the enterprise controls around security and- 00:22:30,286 --> 00:22:30,516 [John] Right 00:22:30,516 --> 00:22:38,245 [Eric] ... single sign-on and other things like that, right? I mean, if you're any sort of mid-sized business, you should be using the security features 00:22:39,615 --> 00:22:41,916 [Eric] of an actual business plan as opposed to having- 00:22:41,916 --> 00:22:42,176 [John] Right 00:22:42,176 --> 00:22:44,306 [Eric] ... a bunch of individuals on an individual- 00:22:44,306 --> 00:22:44,396 [John] Right 00:22:44,396 --> 00:22:47,635 [Eric] ... plan use, you know, have your company data in their personal plan. 00:22:47,636 --> 00:22:53,876 [John] What's the, what's the old saying? Um, something like if the product's free, you are the product? 00:22:53,876 --> 00:22:54,146 [Eric] Yes. 00:22:54,146 --> 00:22:55,195 [John] Something like that. [both chuckling] 00:22:55,196 --> 00:22:55,776 [Eric] Yeah, exactly. 00:22:55,776 --> 00:23:00,916 [John] And then, like, you know, the point of the free plan is the reason they do it is not just for adoption. It's, like, so they can have all the data to- 00:23:00,916 --> 00:23:00,986 [Eric] Yep 00:23:00,986 --> 00:23:01,315 [John] ... train on. 00:23:01,316 --> 00:23:02,666 [Eric] Exactly. Exactly. 00:23:02,666 --> 00:23:02,856 [John] Um, yep. 00:23:04,086 --> 00:23:14,256 [Eric] Yep. So, okay, so there is still value in sort of what sounds like a small slice. You can still get, you know, you can still get a lot of leverage on sort of the prosumer plan. 00:23:14,256 --> 00:23:19,336 [John] Yeah. I, I think, I think the small businesses that haven't used it can use it on prosumer 00:23:20,796 --> 00:23:22,116 [John] who knows how much longer, like- 00:23:22,116 --> 00:23:22,336 [Eric] Mm-hmm 00:23:22,336 --> 00:23:24,436 [John] ... a little while longer, maybe another year or two. 00:23:24,436 --> 00:23:25,096 [Eric] Mm-hmm. 00:23:25,096 --> 00:23:33,916 [John] Um, and get a lot of value and, and be creative and do that. I, I think increasingly to enterprises like, let's call it pretty much at full price- 00:23:33,916 --> 00:23:33,946 [Eric] Mm-hmm 00:23:33,946 --> 00:23:38,226 [John] ... minus the incestuous deals where people are trading compute for, [chuckles] for tokens. 00:23:38,226 --> 00:23:38,256 [Eric] Yeah. 00:23:38,256 --> 00:23:40,716 [John] But, you know, normal enterprise, full price. 00:23:40,716 --> 00:23:41,436 [Eric] Yeah. 00:23:41,436 --> 00:23:45,996 [John] Then you got the stripe underneath there where it's like, you know, I think, I think kind of mostly there. 00:23:45,996 --> 00:23:46,376 [Eric] Mm-hmm. 00:23:46,376 --> 00:23:49,616 [John] Or at least, like, lightly, kinda lightly subsidized, then you have prosumer- 00:23:49,616 --> 00:23:49,726 [Eric] Mm-hmm 00:23:49,726 --> 00:23:50,706 [John] ... which is highly subsidized. 00:23:50,706 --> 00:23:50,736 [Eric] Mm-hmm. 00:23:50,736 --> 00:23:51,916 [John] Especially ChatGPT. 00:23:51,916 --> 00:23:52,516 [Eric] Mm-hmm. 00:23:52,516 --> 00:24:00,236 [John] Which is that, um... But yeah, the, the point being, 'cause back to your actual question is, like, is it e- is it expensive? Is it worth it? 00:24:00,236 --> 00:24:00,396 [Eric] Mm-hmm. 00:24:00,396 --> 00:24:08,616 [John] Is what people are trying to figure out. And, and in the use cases we listed, I think it often is. Like, all right, we're gonna work on revenue- 00:24:08,616 --> 00:24:09,296 [Eric] Mm-hmm 00:24:09,296 --> 00:24:14,356 [John] ... and we, we, we think we can move this metric by doing account research and whatever. 00:24:14,356 --> 00:24:14,816 [Eric] Yep. 00:24:14,816 --> 00:24:26,136 [John] And then on the other one, it's even easier. And on the cost piece of, like, there are a lot of SaaS subscriptions or software... that is licensed per seat. 00:24:26,136 --> 00:24:26,676 [Eric] Mm-hmm. 00:24:26,676 --> 00:24:31,436 [John] And there are people that use the software that just wanna log in and look at a dashboard- 00:24:31,436 --> 00:24:32,196 [Eric] Mm-hmm 00:24:32,196 --> 00:24:36,656 [John] ... or need some very basic information on a screen somewhere. 00:24:36,656 --> 00:24:37,156 [Eric] Yep. 00:24:37,156 --> 00:24:41,856 [John] That, like, there's an obvious cost saving to say, "Hey, just talk to an agent about that." 00:24:41,856 --> 00:24:41,866 [Eric] Totally. 00:24:41,866 --> 00:24:44,956 [John] "You didn't wanna log in anyways, did you?" Like, "No, I didn't." Um, so- 00:24:44,956 --> 00:24:45,496 [Eric] Yeah, exactly 00:24:45,496 --> 00:24:45,616 [John] ... yeah. 00:24:45,616 --> 00:24:46,376 [Eric] Just ask Claude. 00:24:46,376 --> 00:24:46,996 [John] Yeah. 00:24:46,996 --> 00:24:47,146 [Eric] Uh, or ask GPT. 00:24:47,146 --> 00:25:00,696 [John] And then there's another class, which we can't get into right now, of which software is either so simple that, like, you can make it yourself, or it's not simple, but the thing that you need is actually pretty simple. 00:25:00,696 --> 00:25:00,765 [Eric] Yeah. 00:25:00,765 --> 00:25:02,066 [John] And you can make it yourself or 00:25:03,416 --> 00:25:08,256 [John] find somebody that decided to make it for themselves, make it good, and is now selling it for a fraction of whatever you're- 00:25:08,256 --> 00:25:08,286 [Eric] Yeah 00:25:08,286 --> 00:25:10,016 [John] ... paying for your current thing. 00:25:10,016 --> 00:25:10,216 [Eric] Yep. 00:25:11,456 --> 00:25:18,756 [Eric] One thing that I'm interested in your perspective on is the model that comes out of the box with a lot of these services. 00:25:18,756 --> 00:25:18,816 [John] Yes. 00:25:18,816 --> 00:25:19,016 [Eric] So 00:25:20,376 --> 00:25:21,096 [Eric] generally, 00:25:22,316 --> 00:25:28,156 [Eric] uh, you know, if you just log into ChatGPT, you have a prosumer account, or you have, you know, an inter- a business- 00:25:28,156 --> 00:25:28,336 [John] Mm-hmm 00:25:28,336 --> 00:25:30,196 [Eric] ... you know, a business tier plan or enterprise tier plan. 00:25:31,516 --> 00:25:37,476 [Eric] The service defaults to their latest model. Now, it may not have everything turned all the way up. 00:25:37,476 --> 00:25:38,176 [John] Mm-hmm. 00:25:38,176 --> 00:25:49,416 [Eric] You know, so you, so you can toggle, um, you know, things like deep research or, um, you know, advanced reasoning. They have a bunch of different creative names for these. 00:25:49,416 --> 00:25:49,636 [John] Mm-hmm. 00:25:49,636 --> 00:25:51,506 [Eric] You know, turbo, power mode. 00:25:51,506 --> 00:25:51,516 [John] [chuckles] 00:25:51,516 --> 00:25:52,556 [Eric] You know, whatever you wanna call it. 00:25:52,556 --> 00:25:53,255 [John] Yeah. 00:25:53,256 --> 00:25:53,956 [Eric] Um, boost. 00:25:55,056 --> 00:26:02,856 [Eric] Uh, those n- may not necessarily be turned on by default, but the model is generally the latest and greatest, which is- 00:26:02,856 --> 00:26:02,946 [John] Mm-hmm 00:26:02,946 --> 00:26:04,936 [Eric] ... also the most expensive. 00:26:04,936 --> 00:26:05,156 [John] Yeah. 00:26:06,436 --> 00:26:09,216 [John] And, and it's interesting because they don't, there's not a uniform 00:26:10,456 --> 00:26:16,796 [John] plan here. So if you looked at Gemini, Google's product, versus ChatGPT versus, um, 00:26:18,416 --> 00:26:19,216 [John] uh, Anthropic- 00:26:19,216 --> 00:26:19,516 [Eric] Mm-hmm 00:26:19,516 --> 00:26:25,956 [John] ... you know, Claude, it's not uniform. It's not like there's three tiers and each of them has, like, an aligned, you know? 00:26:25,956 --> 00:26:26,096 [Eric] Right. 00:26:26,096 --> 00:26:28,496 [John] Like, they've, they're each kinda doing their own thing. 00:26:28,496 --> 00:26:28,736 [Eric] Right. 00:26:28,736 --> 00:26:29,396 [John] Like recent- 00:26:29,396 --> 00:26:30,556 [Eric] Yeah, it's, it is. Yeah, yeah. 00:26:30,556 --> 00:26:37,656 [John] Yeah. Like recent, recently Google's been really focused on, like, speed and, like, just massive scale c- 00:26:37,656 --> 00:26:37,866 [Eric] Mm-hmm 00:26:37,866 --> 00:26:41,756 [John] ... 'cause the, 'cause their product goes into the world's most popular search engine- 00:26:41,756 --> 00:26:41,806 [Eric] Yep 00:26:41,806 --> 00:26:44,176 [John] ... that has more users than any of these things- 00:26:44,176 --> 00:26:44,276 [Eric] Yep 00:26:44,276 --> 00:26:44,976 [John] ... still. 00:26:44,976 --> 00:26:45,396 [Eric] Yep. 00:26:45,396 --> 00:26:49,876 [John] Um, and then Anthropic's been more focused on enterprise, right? 00:26:49,876 --> 00:26:50,096 [Eric] Yep. 00:26:50,096 --> 00:26:53,985 [John] And then ChatGPT, you know, everything. [chuckles] 00:26:53,985 --> 00:26:53,985 [Eric] Yeah. 00:26:53,985 --> 00:26:54,656 [John] Still, you know. 00:26:54,656 --> 00:26:54,896 [Eric] Yeah, yeah. 00:26:54,896 --> 00:27:14,536 [John] Enterprise consumer. They're, they're trying to do it all. Um, but as far as comparing them i- in general, on any of the paid plans, like, your default is not the absolute, like, smartest version of it, but it, it is typically one of the best models that on, on a, you know- 00:27:14,536 --> 00:27:15,325 [Eric] It's in the, it's in the 90th- 00:27:15,325 --> 00:27:16,446 [John] It's in the ballpark, yeah 00:27:16,446 --> 00:27:19,576 [Eric] ... 90th cost percentile of what you can pay. 00:27:19,576 --> 00:27:21,336 [John] Right. And, and there are typically, 00:27:22,636 --> 00:27:27,556 [John] um, orders of magnitude cheaper options with any of those. So if you're- 00:27:27,556 --> 00:27:27,716 [Eric] Yes 00:27:27,716 --> 00:27:38,736 [John] ... working with ChatGPT or Anthropic or whatever and, and you're at the spot as a business where you're having to pay for usage, not just seats, because a lot of the business models ends up seats plus usage- 00:27:38,736 --> 00:27:38,976 [Eric] Yep, yep 00:27:38,976 --> 00:27:45,396 [John] ... is, and, and that's, I think, gonna be a thing going forward. Um, you don't have to use it on the default- 00:27:45,396 --> 00:27:45,556 [Eric] Right 00:27:45,556 --> 00:27:46,396 [John] ... setting. 00:27:46,396 --> 00:27:46,456 [Eric] It often- 00:27:46,456 --> 00:27:48,696 [John] Which w- I mean, it's orders of magnitude. 00:27:48,696 --> 00:27:49,556 [Eric] It is. It's, it's- 00:27:49,556 --> 00:27:50,016 [John] Yeah 00:27:50,016 --> 00:27:51,576 [Eric] ... it's wild. Um, 00:27:53,196 --> 00:28:10,906 [Eric] I think that's a huge takeaway for this, so I'm gonna summarize the takeaway from this, from this vector. So is it expensive? In the default state off the shelf, it can get very expensive, and I... It doesn't have to be that way. I think one of the big takeaways 00:28:12,136 --> 00:28:19,116 [Eric] is that especially if you're doing more basic tasks, reading a PDF- 00:28:19,116 --> 00:28:19,236 [John] Mm-hmm 00:28:19,236 --> 00:28:26,876 [Eric] ... or summarizing something, um, or doing basic search, you can put it on a way cheaper model. 00:28:26,876 --> 00:28:27,086 [John] Mm-hmm. 00:28:27,086 --> 00:28:30,116 [Eric] And it is, it is way, way cheaper. Um- 00:28:30,116 --> 00:28:34,176 [John] Well, the other thing is, especially on business plans, the cost isn't apparent to the employees. They don't know. 00:28:34,176 --> 00:28:35,136 [Eric] They don't know. 00:28:35,136 --> 00:28:35,386 [John] So, yeah. 00:28:35,386 --> 00:28:44,395 [Eric] Yeah, and the, yes, the, the form factor of the apps themselves, you have to click many, many times in order to actually get in there to figure out what it's costing. 00:28:44,395 --> 00:28:44,955 [John] Sure. Right. 00:28:44,955 --> 00:28:49,456 [Eric] But I think that's, I mean, that's something that I've seen, and actually with my team that we've implemented, is 00:28:50,476 --> 00:28:58,595 [Eric] there are certain things where we may be analyzing a really large amount of disparate types of information- 00:28:58,596 --> 00:28:58,896 [John] Mm-hmm 00:28:58,896 --> 00:29:05,166 [Eric] ... or files or context or hitting different APIs to pull information from different systems, and 00:29:06,416 --> 00:29:16,666 [Eric] for that job where it's doing the initial crunch of all of that, we sort of turn everything on and it's very expensive, but it may run for 30 minutes at the most expensive setting. 00:29:16,666 --> 00:29:16,696 [John] Mm-hmm. 00:29:16,696 --> 00:29:19,736 [Eric] And then we'll actually bump it down to a way cheaper model. 00:29:19,736 --> 00:29:20,856 [John] Mm-hmm. 00:29:20,856 --> 00:29:30,016 [Eric] We're not even necessarily thinking about cost, but actually for certain tasks, like, the cheaper models are, are better because they're not trying to do as much. That's not- 00:29:30,016 --> 00:29:30,035 [John] Yeah 00:29:30,035 --> 00:29:44,146 [Eric] ... that's kind of a weird way to put it 'cause that's not necessarily true under the hood. But you, they work completely fine, and in some ways are better for certain tasks, um, you know, than the latest and greatest- 00:29:44,146 --> 00:29:44,156 [John] Yeah 00:29:44,156 --> 00:29:44,796 [Eric] ... model, right? 00:29:44,796 --> 00:29:51,586 [John] And a, a silly visual for this is thinking about, like, a, a framing hammer versus a sledgehammer. Like- 00:29:51,586 --> 00:29:51,586 [Eric] Mm-hmm 00:29:51,586 --> 00:29:54,456 [John] ... sledgehammers are big and fun and can do a lot of, like- 00:29:54,456 --> 00:29:54,656 [Eric] Yes 00:29:54,656 --> 00:30:01,236 [John] ... you know, work, but, you know, the, there's a, there's a non-practical use for a sledgehammer. 00:30:01,236 --> 00:30:01,986 [Eric] Yes. Yep. 00:30:01,986 --> 00:30:09,556 [John] And, and it's a little different in that the smartest models can do all the things, like, pretty much just as well, but there's a cost component and they're slower. 00:30:09,556 --> 00:30:10,016 [Eric] Yes. 00:30:10,016 --> 00:30:10,146 [John] Like- 00:30:10,146 --> 00:30:11,136 [Eric] They are slower. Yep. 00:30:11,136 --> 00:30:12,306 [John] And, you know. 00:30:12,306 --> 00:30:26,116 [Eric] And I think the answer is if you're gonna, you know, build a, if you're gonna build a shed in your backyard, you know-And you're gonna, like, pour concrete and, you know, or hammer a four by four into the ground. 00:30:26,116 --> 00:30:26,816 [John] Right. 00:30:26,816 --> 00:30:29,915 [Eric] You need a sledgehammer, and you need a framing hammer, right? 00:30:29,916 --> 00:30:37,136 [John] You need, you need both. And, and thinking... And the cost of operation analogy is perfect because think about your physical effort in using a sledgehammer. 00:30:37,136 --> 00:30:37,996 [Eric] Mm-hmm. Yeah. 00:30:37,996 --> 00:30:45,176 [John] Like, that mirrors the cost of operation for the largest, best models, and then your physical effort in using a little framing hammer, like, it's just a lot easier. 00:30:45,176 --> 00:30:50,996 [Eric] Exactly. Yep. So, and this actually comes back full circle, which is a great place to land, in that, 00:30:52,256 --> 00:31:05,976 [Eric] that me- the methodology of deciding when to use which model and then cost optimizing that for a, you know, a business where you have a portfolio of different use cases and you're trying to calculate ROI, 00:31:07,116 --> 00:31:09,276 [Eric] is something that has to be taught. 00:31:09,276 --> 00:31:09,715 [John] Yeah. 00:31:09,716 --> 00:31:18,416 [Eric] The, the, the providers do not lay that out for you in a way that is straightforward, right? They sort of just default to something expensive. 00:31:18,416 --> 00:31:34,496 [John] Yeah. And I was telling you about this earlier. This is so interesting. Um, there's a set of decently complicated tasks I wanted done. Worked fine with, like, the most expensive model. And then for fun, I was like, "Well, I wonder if it'll work on the cheaper model." Tried the exact same ins- instructions. 00:31:34,496 --> 00:31:34,516 [Eric] Mm-hmm. 00:31:34,516 --> 00:31:44,256 [John] Changed nothing. It did not work. Um, but then I asked the smarter model, like, "Hey, can you make a call to the cheaper one and optimize it to work?" 00:31:44,256 --> 00:31:44,836 [Eric] Hmm. 00:31:44,836 --> 00:31:45,426 [John] And it did. 00:31:45,426 --> 00:31:45,436 [Eric] Yeah. 00:31:45,436 --> 00:31:46,076 [John] And it worked. 00:31:46,076 --> 00:31:46,296 [Eric] [laughs] 00:31:46,296 --> 00:31:48,235 [John] So those are the really interesting, like- 00:31:48,236 --> 00:31:48,456 [Eric] Right 00:31:48,456 --> 00:31:57,076 [John] ... meta things going on here, where, where because these are, um... because, because these tools can, like, talk to each other- 00:31:57,076 --> 00:31:57,196 [Eric] Right 00:31:57,196 --> 00:32:00,186 [John] ... you, you can actually, like, use the smarter one to fix the one that's- 00:32:00,186 --> 00:32:00,186 [Eric] Yeah 00:32:00,186 --> 00:32:01,656 [John] ... less smart, and then just use the less- 00:32:01,656 --> 00:32:01,796 [Eric] Yeah 00:32:01,796 --> 00:32:02,496 [John] ... smart one. It's- 00:32:02,496 --> 00:32:02,836 [Eric] Absolutely 00:32:02,836 --> 00:32:03,955 [John] ... interesting. 00:32:03,956 --> 00:32:04,656 [Eric] Okay. So 00:32:05,716 --> 00:32:06,436 [Eric] our takeaways 00:32:08,316 --> 00:32:11,936 [Eric] are can... I- is AI too expensive for the average business? 00:32:13,136 --> 00:32:23,196 [Eric] Of course, the answer is it depends. But I think we've landed on if you are proactive about it, if you're proactive about training- 00:32:23,196 --> 00:32:23,856 [John] Mm-hmm 00:32:23,856 --> 00:32:27,136 [Eric] ... if you really define your ROI cost clearly- 00:32:27,136 --> 00:32:27,536 [John] Mm-hmm 00:32:27,536 --> 00:32:30,585 [Eric] ... or if you define your ROI equation clearly- 00:32:30,585 --> 00:32:30,716 [John] Right 00:32:30,716 --> 00:32:34,146 [Eric] ... whether that's cost or revenue, or whether it's intangible 00:32:35,436 --> 00:32:40,876 [Eric] with something like, you know, uh, enabling more innovation in the company- 00:32:40,876 --> 00:32:41,296 [John] Right 00:32:41,296 --> 00:32:47,796 [Eric] ... you know, trying to bring more ideas to life. Uh, and then if you learn how to cost optimize model usage- 00:32:47,796 --> 00:32:47,805 [John] Mm-hmm 00:32:47,805 --> 00:32:50,976 [Eric] ... so that you're not just paying the highest cost off the shelf- 00:32:50,976 --> 00:32:51,156 [John] Mm-hmm 00:32:51,156 --> 00:32:54,256 [Eric] ... you're in a way better place to get ROI. 00:32:54,256 --> 00:32:59,296 [John] Yeah, if you do all those things, I don't think there's very many businesses that, that I would say it's too expensive. 00:32:59,296 --> 00:33:04,726 [Eric] All right. There we go. We answered the question. All right. Thanks for joining the Token Intelligence Show, and we will catch you on the next one. 00:33:07,896 --> 00:33:13,256 [Eric] [outro music]