WEBVTT 00:00:00.000 --> 00:00:04.559 So everyone is playing with AI right now. Yeah, 00:00:04.700 --> 00:00:07.139 but almost no one is actually making real money 00:00:07.139 --> 00:00:10.359 from it. Why is that? Because they're selling 00:00:10.359 --> 00:00:13.220 the technology itself. They're not selling the 00:00:13.220 --> 00:00:16.120 actual outcome. Right. Welcome to the Deep Dive. 00:00:16.260 --> 00:00:18.500 I'm glad you're here with us. Today we're unpacking 00:00:18.500 --> 00:00:20.820 a really specific framework. We're looking at 00:00:20.820 --> 00:00:24.699 how to transition from just from staying busy 00:00:24.699 --> 00:00:27.219 with AI tools to actually profiting from them. 00:00:27.600 --> 00:00:29.719 Exactly. We've got a lot of ground to cover for 00:00:29.719 --> 00:00:31.640 you today. We're going to explain why prompt 00:00:31.640 --> 00:00:34.020 packs are a complete dead end. Yeah, they really 00:00:34.020 --> 00:00:35.579 are. And then we'll show you how to structure 00:00:35.579 --> 00:00:38.619 your inputs using the MAPS framework. Plus, we'll 00:00:38.619 --> 00:00:40.719 get into the rule of R for automation. And we 00:00:40.719 --> 00:00:42.520 also need to talk about the warning signs. You 00:00:42.520 --> 00:00:44.859 know, what happens when you scale up with AI 00:00:44.859 --> 00:00:46.820 agents. But I think we have to start with the 00:00:46.820 --> 00:00:49.399 biggest trap in the AI space right now. Oh, absolutely. 00:00:49.560 --> 00:00:52.939 Which is this... this deep obsession with the 00:00:52.939 --> 00:00:55.460 tools themselves. It's everywhere. I mean, people 00:00:55.460 --> 00:00:58.560 are obsessed with the shiny new toys, and they 00:00:58.560 --> 00:01:00.820 completely miss the business value. It's like 00:01:00.820 --> 00:01:03.270 they're hypnotized by the tech. They really are. 00:01:03.329 --> 00:01:06.170 You see everyone trying to sell these like massive 00:01:06.170 --> 00:01:10.909 prompt packs or basic chat GPT tutorials. Even 00:01:10.909 --> 00:01:13.049 automation templates kind of fall into this trap. 00:01:13.209 --> 00:01:15.129 I mean, technically there is a market for those 00:01:15.129 --> 00:01:18.450 things. Sure. Yeah. But the ceiling is brutally 00:01:18.450 --> 00:01:21.409 low. Like competition is just overwhelming because 00:01:21.409 --> 00:01:24.129 everyone has access to the exact same information. 00:01:24.709 --> 00:01:28.010 The real money always sits one level deeper than 00:01:28.010 --> 00:01:30.329 the tool itself. Right. It's about finding a 00:01:30.329 --> 00:01:33.510 broken process. Precisely, because business owners 00:01:33.510 --> 00:01:35.469 don't care about language models. They have a 00:01:35.469 --> 00:01:38.349 costly process, right? Say they're losing warm 00:01:38.349 --> 00:01:40.310 leads because their follow -ups are terrible. 00:01:40.489 --> 00:01:42.510 They're bleeding money. They just want the bleeding 00:01:42.510 --> 00:01:45.709 to stop. Exactly. Whoever fixes that specific 00:01:45.709 --> 00:01:48.250 outcome gets paid. And they can get paid well, 00:01:48.450 --> 00:01:50.469 regardless of the tools they use. The tool should 00:01:50.469 --> 00:01:53.689 be a... Totally invisible. Yeah. I like to think 00:01:53.689 --> 00:01:55.989 of it's like stacking Lego blocks of data. The 00:01:55.989 --> 00:01:57.870 client doesn't care about the plastic bricks. 00:01:58.030 --> 00:01:59.930 They just want to buy the finished castle. That 00:01:59.930 --> 00:02:02.290 is a perfect analogy. Let's look at IG Group 00:02:02.290 --> 00:02:05.290 from our source material. Massive global financial 00:02:05.290 --> 00:02:07.989 services company. And finance means heavy compliance. 00:02:08.490 --> 00:02:10.770 Incredible compliance, yeah. They had strict 00:02:10.770 --> 00:02:13.909 regulatory requirements and they needed multi 00:02:13.909 --> 00:02:16.490 -language content fast. They were paying insanely 00:02:16.490 --> 00:02:18.710 high agency costs just to keep up. So what did 00:02:18.710 --> 00:02:21.719 they do? They deployed Claude. quietly across 00:02:21.719 --> 00:02:25.120 their teams. And the results were wild. The analytics 00:02:25.120 --> 00:02:28.599 team saved 70 hours a week. Marketing saw triple 00:02:28.599 --> 00:02:31.639 -digit speed -to -market improvements. Wow. Yeah. 00:02:31.979 --> 00:02:35.520 Full ROI in under three months. But here's the 00:02:35.520 --> 00:02:38.000 kicker. Claude was just the invisible engine. 00:02:38.379 --> 00:02:40.379 They didn't sell AI to their customers. They 00:02:40.379 --> 00:02:42.659 just delivered better financial services. So 00:02:42.659 --> 00:02:44.560 looking at that, what are the three conditions 00:02:44.560 --> 00:02:47.159 for spotting a real opportunity? You're looking 00:02:47.159 --> 00:02:49.719 for a task taking skilled people a long time. 00:02:49.840 --> 00:02:52.159 The final output has to be very high stakes, 00:02:52.539 --> 00:02:54.840 and the business must already be spending money 00:02:54.840 --> 00:02:57.259 to solve it. Look for manual high stakes tasks 00:02:57.259 --> 00:02:59.919 where businesses already spend money. Exactly. 00:03:00.099 --> 00:03:01.919 Right. So once you identify the right high stakes 00:03:01.919 --> 00:03:04.080 problem, you actually have to get the AI to do 00:03:04.080 --> 00:03:06.919 the work. And if your output is generic, you 00:03:06.919 --> 00:03:09.400 know, the model isn't broken. Your prompt is. 00:03:09.500 --> 00:03:11.539 Yeah, your prompt is definitely broken. Vague 00:03:11.539 --> 00:03:13.860 prompts just give you vague answers. Which brings 00:03:13.860 --> 00:03:16.460 us to the MAPS framework. Let's break this down 00:03:16.460 --> 00:03:19.719 for you. M is for mission. You have to give the 00:03:19.719 --> 00:03:22.180 AI the actual business goal, right? Right. Not 00:03:22.180 --> 00:03:24.599 just the task. So instead of just saying, find 00:03:24.599 --> 00:03:27.180 me leads, which is terrible. Because it has no 00:03:27.180 --> 00:03:30.500 context. Exactly. You say, I need 30 customers 00:03:30.500 --> 00:03:33.620 to hit my $15 ,000 revenue goal. That's the mission. 00:03:33.680 --> 00:03:37.949 It grounds the AI. Beat. Then A is for ask. Make 00:03:37.949 --> 00:03:40.370 one specific request. Right. Don't say, help 00:03:40.370 --> 00:03:43.949 me with sales. Say, give me a list of 40 U .S. 00:03:44.229 --> 00:03:46.810 leads with name, company, email, and LinkedIn. 00:03:47.320 --> 00:03:50.599 Precision is key here. It is. P is for parameters. 00:03:50.759 --> 00:03:53.039 This is the background context. Your ideal customer, 00:03:53.199 --> 00:03:55.759 your constraints. And honestly, a massive time 00:03:55.759 --> 00:03:58.780 saver here is using voice input to just dictate 00:03:58.780 --> 00:04:00.800 those parameters. You can just talk it out. I 00:04:00.800 --> 00:04:02.919 have to admit something here. I still wrestle 00:04:02.919 --> 00:04:05.000 with prompt drift myself when I rush and skip 00:04:05.000 --> 00:04:07.419 the parameters. It's really easy to get lazy. 00:04:07.740 --> 00:04:10.360 Oh, we all do it. But skipping parameters guarantees 00:04:10.360 --> 00:04:12.759 garbage output. The context is what makes it 00:04:12.759 --> 00:04:15.469 work. Yeah. And the last one is S shape. Tell 00:04:15.469 --> 00:04:17.990 the AI exactly what the output should look like, 00:04:18.089 --> 00:04:21.170 like format it as a table. Right. And I have 00:04:21.170 --> 00:04:24.850 to ask you, why is shape the step everyone seems 00:04:24.850 --> 00:04:27.470 to forget? Skipping it means wasting 20 minutes 00:04:27.470 --> 00:04:30.189 reformatting a useless wall of text. It really 00:04:30.189 --> 00:04:32.689 does. It's such an unforced error. Absolutely. 00:04:33.170 --> 00:04:34.310 We're going to take a quick break here. Don't 00:04:34.310 --> 00:04:37.290 go anywhere. Insert a clear break here for the 00:04:37.290 --> 00:04:39.910 mid -roll sponsor read. All right, we're back. 00:04:40.370 --> 00:04:43.569 So MAPS can use you perfect single outputs. But 00:04:43.569 --> 00:04:45.910 for tasks that happen every single week, running 00:04:45.910 --> 00:04:49.329 manual prompts just, it isn't enough. You need 00:04:49.329 --> 00:04:52.129 automation. But only if it passes the test. Right, 00:04:52.269 --> 00:04:54.870 the rule of R. We use tools like Zapier or Make 00:04:54.870 --> 00:04:57.730 for this, which is basically digital glue that 00:04:57.730 --> 00:05:00.300 automatically moves data. between your different 00:05:00.300 --> 00:05:02.319 apps. That's exactly what it is. So the first 00:05:02.319 --> 00:05:05.399 R is repetitive. Does the task happen regularly? 00:05:05.759 --> 00:05:07.759 Like manually checking ConvertKit every Monday. 00:05:07.819 --> 00:05:09.759 Yeah, and then pasting those email stats into 00:05:09.759 --> 00:05:12.040 Notion, that's highly repetitive. The second 00:05:12.040 --> 00:05:14.879 R is rule -based. Meaning the logic is perfectly 00:05:14.879 --> 00:05:17.079 predictable. Right. Say a Typeform lead comes 00:05:17.079 --> 00:05:21.000 in, it goes straight to your HubSpot CRM. A CRM 00:05:21.000 --> 00:05:23.439 is a digital filing cabinet tracking all your 00:05:23.439 --> 00:05:26.319 customer relationships. Right, so it hits HubSpot, 00:05:26.519 --> 00:05:29.040 then triggers a ConvertKit email. The logic never 00:05:29.040 --> 00:05:32.899 changes. And the final R is return. Will it save 00:05:32.899 --> 00:05:35.600 you at least one hour a week? Because if yes, 00:05:35.740 --> 00:05:37.980 the setup effort pays back in a month. Which 00:05:37.980 --> 00:05:40.459 is huge. So what's the danger of ignoring the 00:05:40.459 --> 00:05:43.180 return rule? You'll spend days building a machine 00:05:43.180 --> 00:05:46.480 nobody needs. It's a trap. Spending 60 hours 00:05:46.480 --> 00:05:48.379 to automate something that saves two minutes 00:05:48.379 --> 00:05:52.399 is a bad trade. An incredibly bad trade. So what 00:05:52.399 --> 00:05:55.040 happens when a problem is too complex for a single 00:05:55.040 --> 00:05:58.000 prompt or, you know, rigid automation? That's 00:05:58.000 --> 00:06:00.459 when you graduate to AI agents, systems that 00:06:00.459 --> 00:06:02.959 can actually make decisions mid -task. They adjust 00:06:02.959 --> 00:06:05.360 on the fly. Exactly. Look at Klarna. They built 00:06:05.360 --> 00:06:08.560 an OpenAI customer service agent. In its first 00:06:08.560 --> 00:06:11.379 month, it handled 2 .3 million conversations. 00:06:11.620 --> 00:06:15.899 Whoa. Imagine scaling to 2 .3 million queries 00:06:15.899 --> 00:06:20.040 in a single month. It's mind -blowing. It cut 00:06:20.040 --> 00:06:22.699 resolution time from 11 minutes to 2 minutes, 00:06:23.060 --> 00:06:25.500 and repeat inquiries dropped by 25%. That saved 00:06:25.500 --> 00:06:27.519 them a ton of money, right? Service costs per 00:06:27.519 --> 00:06:30.920 transaction fell from 32 cents to 19 cents. But 00:06:30.920 --> 00:06:33.860 there is a real warning here. Klarna eventually 00:06:33.860 --> 00:06:37.420 had to rehire human agents. Prioritizing cost 00:06:37.420 --> 00:06:39.990 over quality really backfired on them. Right, 00:06:40.129 --> 00:06:42.189 because agents can hallucinate. You need humans 00:06:42.189 --> 00:06:45.149 in the loop to review drafts. How do we prevent 00:06:45.149 --> 00:06:47.829 these long agent workflows from hallucinating 00:06:47.829 --> 00:06:50.430 or going off track? Well, you have to engineer 00:06:50.430 --> 00:06:53.350 the process carefully. Build checkpoints where 00:06:53.350 --> 00:06:56.129 a human or second agent reviews the work mid 00:06:56.129 --> 00:06:58.949 -process. That's exactly it. A second agent checking 00:06:58.949 --> 00:07:01.810 the first agent is a game changer. It's a fascinating 00:07:01.810 --> 00:07:05.089 architecture. So you have this toolkit now. Claude 00:07:05.089 --> 00:07:08.199 for Thinking with MAPS. Zapier for scheduled 00:07:08.199 --> 00:07:11.019 tasks with the rule of R, and agents for complex 00:07:11.019 --> 00:07:14.259 workflows. The final piece is knowing how to 00:07:14.259 --> 00:07:16.339 charge for this superpower. Yeah, and this is 00:07:16.339 --> 00:07:18.279 where people mess up. You have to become the 00:07:18.279 --> 00:07:20.339 orchestrator. You know which tool fits which 00:07:20.339 --> 00:07:22.639 problem. But there's a big pricing mistake happening. 00:07:22.779 --> 00:07:25.879 Massive mistake. When AI makes delivery faster, 00:07:26.100 --> 00:07:29.120 people instinctively drop their prices. But a 00:07:29.120 --> 00:07:32.319 lead gen service worth $5 ,000 before AI is still 00:07:32.319 --> 00:07:35.100 worth $5 ,000 after... Because the outcome hasn't 00:07:35.100 --> 00:07:37.839 changed. Exactly. Klarna didn't pass all their 00:07:37.839 --> 00:07:40.220 40 % savings back to the customer. They kept 00:07:40.220 --> 00:07:43.480 the margin. You anchor your price to the client's 00:07:43.480 --> 00:07:45.980 outcome, not the hours you spent running, Claude. 00:07:46.319 --> 00:07:48.980 Let the technology stay invisible. Totally invisible. 00:07:49.199 --> 00:07:51.199 Why is it so hard for creators to keep their 00:07:51.199 --> 00:07:54.500 prices high when AI does the heavy lifting? Because 00:07:54.500 --> 00:07:57.139 of guilt. Honestly, it feels like cheating. They 00:07:57.139 --> 00:07:59.240 wrongly tie their self -worth to hours worked 00:07:59.240 --> 00:08:01.920 instead of the final results. Exactly that. You 00:08:01.920 --> 00:08:04.560 have to break that mindset. Let's synthesize 00:08:04.560 --> 00:08:08.360 this. To actually win with AI, stop selling your 00:08:08.360 --> 00:08:11.740 knowledge of the tools. Find expensive problems. 00:08:11.980 --> 00:08:15.339 Master your inputs with MAPS. Filter your busy 00:08:15.339 --> 00:08:17.819 work through the rule of R. Use agents for big 00:08:17.819 --> 00:08:20.319 workflows, but keep humans in the loop. And above 00:08:20.319 --> 00:08:23.040 all, price for the outcome. I want to leave you 00:08:23.040 --> 00:08:25.860 with a final thought to mull over. If the ultimate 00:08:25.860 --> 00:08:28.680 goal is to let the AI technology remain completely 00:08:28.680 --> 00:08:31.360 invisible to the client, What happens in five 00:08:31.360 --> 00:08:33.860 years when every single business has the exact 00:08:33.860 --> 00:08:37.379 same invisible AI capabilities? What becomes 00:08:37.379 --> 00:08:39.740 your ultimate differentiator then? That's the 00:08:39.740 --> 00:08:42.379 real question. Thank you for joining this Deem 00:08:42.379 --> 00:08:44.960 Dive. Look at your own workloads today and just 00:08:44.960 --> 00:08:47.600 find one task that passes the rule of R. Start 00:08:47.600 --> 00:08:49.580 there. OUTRO music.