WEBVTT 00:00:00.000 --> 00:00:02.279 Let's just start with the fact that really changes 00:00:02.279 --> 00:00:07.480 everything. A staggering 95 % of corporate generative 00:00:07.480 --> 00:00:12.980 AI pilots are failing. 95%. That's not a typo. 00:00:13.039 --> 00:00:15.599 It's not. And it means businesses are officially 00:00:15.599 --> 00:00:17.559 done experimenting. They've burned through those 00:00:17.559 --> 00:00:19.859 early budgets. And now, well, they're demanding 00:00:19.859 --> 00:00:23.500 answers. And that simple phrase, we do AI automation, 00:00:23.940 --> 00:00:25.899 it's already a dead language in the enterprise 00:00:25.899 --> 00:00:28.929 market. Welcome to the Deep Dive. That failure 00:00:28.929 --> 00:00:30.870 rate you mentioned is basically a clearing function. 00:00:30.969 --> 00:00:33.689 It's changing the entire landscape for AI services 00:00:33.689 --> 00:00:37.170 as we head into 2026. It really is. Our mission 00:00:37.170 --> 00:00:39.390 today is to unpack what that means, whether you're 00:00:39.390 --> 00:00:41.689 running an agency or you're on the other side 00:00:41.689 --> 00:00:44.210 trying to hire one. We're seeing this rapid transition 00:00:44.210 --> 00:00:48.380 away from that. crowded, kind of low value, generic 00:00:48.380 --> 00:00:51.420 automation vendor era. The AI automation bro 00:00:51.420 --> 00:00:53.659 era. Right. And we're moving very decisively 00:00:53.659 --> 00:00:55.679 toward this high ticket transformation partner 00:00:55.679 --> 00:00:58.600 model. That's where the real value is. Exactly. 00:00:58.780 --> 00:01:00.979 So today we're going to cover that central tension, 00:01:01.179 --> 00:01:04.900 the pivot to demanding measurable ROI. We'll 00:01:04.900 --> 00:01:06.579 look at the new entry point, the paid audit, 00:01:06.719 --> 00:01:09.239 and we'll talk about why fixing messy data is 00:01:09.239 --> 00:01:12.519 suddenly like. 80 % of the actual work. And then 00:01:12.519 --> 00:01:14.620 we'll wrap up with the strategic end game. This 00:01:14.620 --> 00:01:16.819 is where the smart agencies start owning their 00:01:16.819 --> 00:01:19.519 own IP, turning all that service work into scalable 00:01:19.519 --> 00:01:22.120 software. Okay, let's get into it. Let's do it. 00:01:22.180 --> 00:01:27.129 So that MIT report stat, 95 % failure rate. It's 00:01:27.129 --> 00:01:29.250 just wild. When you hear that, you have to ask 00:01:29.250 --> 00:01:31.969 what really went wrong. It wasn't just bad tech, 00:01:32.030 --> 00:01:34.489 was it? No, it rarely was the tech itself. The 00:01:34.489 --> 00:01:37.310 failure was almost entirely about poor strategic 00:01:37.310 --> 00:01:40.329 execution. Scope creep. Companies got really 00:01:40.329 --> 00:01:42.530 seduced by the idea of large language models, 00:01:42.709 --> 00:01:44.769 you know, the algorithms that can generate text 00:01:44.769 --> 00:01:46.870 or code. But they never actually integrated those 00:01:46.870 --> 00:01:49.310 cool projects into their core business. It feels 00:01:49.310 --> 00:01:51.510 like they prioritized innovation theater over 00:01:51.510 --> 00:01:55.299 just... Real operational efficiency. Precisely. 00:01:55.299 --> 00:01:57.700 And that has caused this massive behavioral shift. 00:01:57.799 --> 00:02:00.900 It's no longer, hey, let's try some AI. The mindset 00:02:00.900 --> 00:02:03.200 is now, prove this saves us X amount of money 00:02:03.200 --> 00:02:05.459 or Y amount of time or just don't even bother. 00:02:06.519 --> 00:02:09.099 Clients are deeply skeptical now. And, you know, 00:02:09.159 --> 00:02:11.500 they should be. And this is why that generic 00:02:11.500 --> 00:02:13.919 automation agency label is suddenly so meaningless, 00:02:14.080 --> 00:02:16.800 isn't it? Completely. The survival requirement 00:02:16.800 --> 00:02:20.219 now is to niche down hard. You have to stop leading 00:02:20.219 --> 00:02:23.340 with the tech. Don't say, we use LLMs and vector 00:02:23.340 --> 00:02:26.060 databases. That's just feature talk. Instead, 00:02:26.259 --> 00:02:28.759 you need to lead with outcomes. Quantified outcomes. 00:02:28.939 --> 00:02:31.819 Yeah. Say, we help e -commerce teams cut support 00:02:31.819 --> 00:02:34.819 costs by 30 % in 90 days, or something just as 00:02:34.819 --> 00:02:36.900 specific. You need those bulletproof case studies 00:02:36.900 --> 00:02:40.439 with real numbers. Like, we saved Company X $180 00:02:40.439 --> 00:02:43.180 ,000 a year by automating their intake process. 00:02:43.560 --> 00:02:45.879 It has to be about money saved or revenue gained, 00:02:46.020 --> 00:02:48.719 not about features installed. So if the market 00:02:48.719 --> 00:02:51.199 is demanding that kind of concrete value immediately, 00:02:51.520 --> 00:02:54.800 what is it that specifically lets a smaller agency 00:02:54.800 --> 00:02:58.039 command higher fees and become that trusted long 00:02:58.039 --> 00:03:00.659 -term partner? By offering comprehensive guidance, 00:03:00.939 --> 00:03:03.800 partners eliminate the chaos of managing a bunch 00:03:03.800 --> 00:03:06.039 of small vendors. Which brings us to the next 00:03:06.039 --> 00:03:08.520 big thing, because the fundamental problem for 00:03:08.520 --> 00:03:11.759 clients is just confusion. They know AI matters, 00:03:11.939 --> 00:03:13.879 but they don't know where to start. And they 00:03:13.879 --> 00:03:15.819 definitely don't trust a random vendor to tell 00:03:15.819 --> 00:03:18.819 them. Exactly. The solution to that is the paid 00:03:18.819 --> 00:03:22.719 AI audit. It's really becoming the new universal 00:03:22.719 --> 00:03:25.939 entry point. It reminds me of the early 2000s, 00:03:25.939 --> 00:03:28.360 right? Every business needed a website, even 00:03:28.360 --> 00:03:30.979 if they weren't sure why. Now, every business 00:03:30.979 --> 00:03:33.680 needs an AI roadmap. And this model works because 00:03:33.680 --> 00:03:36.219 you charge, say, $5 ,000 to $15 ,000 just for 00:03:36.219 --> 00:03:38.280 the diagnosis, depending on the company's size, 00:03:38.360 --> 00:03:40.960 of course. You're getting paid to figure out 00:03:40.960 --> 00:03:42.759 the problem. Not to build the solution right 00:03:42.759 --> 00:03:45.159 away. Right. You map their existing, usually 00:03:45.159 --> 00:03:48.000 messy processes. You find every single friction 00:03:48.000 --> 00:03:50.560 point and you identify the key opportunities 00:03:50.560 --> 00:03:52.979 that will give them a fast ROI. And then you 00:03:52.979 --> 00:03:55.020 present this really clear roadmap. It's a strategic 00:03:55.020 --> 00:03:58.599 document. And that audit, it naturally upsells 00:03:58.599 --> 00:04:00.500 into the big implementation packages. We're talking 00:04:00.500 --> 00:04:03.400 $10 ,000 up to $100 ,000 or more. Yeah, I saw 00:04:03.400 --> 00:04:05.740 a stat that one program's average audit deal 00:04:05.740 --> 00:04:10.139 was $11 ,400 just for the initial diagnosis. 00:04:10.500 --> 00:04:13.280 That money immediately covers your discovery 00:04:13.280 --> 00:04:16.040 time and it just builds trust from day one. The 00:04:16.040 --> 00:04:19.199 strategic value there is just huge. Once you 00:04:19.199 --> 00:04:22.480 diagnose what's broken. you automatically become 00:04:22.480 --> 00:04:24.480 the most knowledgeable person in the room. You 00:04:24.480 --> 00:04:27.139 stop selling. You just continue the work. So 00:04:27.139 --> 00:04:29.800 what's the core reason that initial diagnosis 00:04:29.800 --> 00:04:31.959 is more valuable than just trying to build a 00:04:31.959 --> 00:04:34.199 pilot project right away? Understanding what's 00:04:34.199 --> 00:04:36.720 broken makes you the expert, and that eliminates 00:04:36.720 --> 00:04:39.620 the vendor -client friction. Okay, so let's get 00:04:39.620 --> 00:04:41.540 into this uncomfortable truth, which I think 00:04:41.540 --> 00:04:44.399 explains a lot about that 95 % failure rate. 00:04:44.620 --> 00:04:47.300 Most businesses, they're just not ready for advanced 00:04:47.300 --> 00:04:49.779 AI. Not even close. There was a Cisco report. 00:04:50.250 --> 00:04:52.850 It said 92 % of companies still don't have AI 00:04:52.850 --> 00:04:55.350 ready data. Yeah, their systems are fragmented, 00:04:55.410 --> 00:04:58.089 scattered. They're held together by like duct 00:04:58.089 --> 00:05:01.949 tape and spreadsheets from 2015. That is the 00:05:01.949 --> 00:05:04.029 messy reality we're dealing with. So if you're 00:05:04.029 --> 00:05:05.649 selling a transformation package, you're not 00:05:05.649 --> 00:05:07.689 really selling AI. You're selling digital plumbing. 00:05:08.319 --> 00:05:11.420 Exactly. And that means 80 % of the actual work 00:05:11.420 --> 00:05:13.399 is going to be general software development. 00:05:13.620 --> 00:05:16.000 You know, building internal tools, dashboards, 00:05:16.060 --> 00:05:18.600 cleaning up systems before the AI can even do 00:05:18.600 --> 00:05:20.720 anything useful. You're stacking Lego blocks 00:05:20.720 --> 00:05:23.660 of clean data. Perfect analogy. But here's the 00:05:23.660 --> 00:05:26.540 upside we can't ignore. Building that foundational 00:05:26.540 --> 00:05:29.540 stuff is so much faster and cheaper than it used 00:05:29.540 --> 00:05:32.759 to be. Oh, way faster. AI has basically cut the 00:05:32.759 --> 00:05:34.819 cost of general software development by like 00:05:34.819 --> 00:05:37.699 five to ten times. Developers using tools like 00:05:37.699 --> 00:05:41.000 Copilot are reporting speed ups of... 20, 30, 00:05:41.079 --> 00:05:43.980 even 55%. I saw one user who said they built 00:05:43.980 --> 00:05:47.379 30 working apps in 30 days, something that would 00:05:47.379 --> 00:05:50.300 have taken months to do manually. Whoa. I mean, 00:05:50.300 --> 00:05:52.360 just imagine scaling a business that serves a 00:05:52.360 --> 00:05:54.819 billion queries based on that kind of build rate. 00:05:54.899 --> 00:05:57.759 The speed at which you can fix the mess is just, 00:05:57.959 --> 00:06:00.680 it's astonishing. Okay, so that foundational 00:06:00.680 --> 00:06:02.620 layer has to be addressed. But we also have to 00:06:02.620 --> 00:06:04.879 talk about what you call the adoption tax. I 00:06:04.879 --> 00:06:06.699 love that framing. What does that mean exactly? 00:06:07.069 --> 00:06:08.829 It means that all the potential time savings 00:06:08.829 --> 00:06:11.350 from a new system often get canceled out by user 00:06:11.350 --> 00:06:15.629 distrust or just complexity. Think about Apple's 00:06:15.629 --> 00:06:18.509 Newton message pad. It was amazing tech for its 00:06:18.509 --> 00:06:20.689 time, but people didn't trust the handwriting 00:06:20.689 --> 00:06:23.910 recognition, so they just stuck with paper. So 00:06:23.910 --> 00:06:26.649 a great system can completely die if the team 00:06:26.649 --> 00:06:29.029 doesn't trust it or know how to use it. That's 00:06:29.029 --> 00:06:32.399 it. So education stops being optional in 2026. 00:06:32.639 --> 00:06:35.079 It has to be built into every single proposal. 00:06:35.420 --> 00:06:37.879 Agencies actually need to add a mandatory, say, 00:06:38.060 --> 00:06:41.459 $15 ,000 to $20 ,000 education and adoption package 00:06:41.459 --> 00:06:44.339 to every project. That covers training, documentation, 00:06:44.879 --> 00:06:47.220 all of it. Yeah. And, you know, I still wrestle 00:06:47.220 --> 00:06:49.939 with prompt drift myself sometimes. Oh, yeah. 00:06:50.000 --> 00:06:51.639 That's when the instructions you give the AI 00:06:51.639 --> 00:06:53.720 just kind of lose their effectiveness over time. 00:06:53.899 --> 00:06:56.879 So I know from experience that professional training 00:06:56.879 --> 00:06:59.279 is absolutely essential for the client's team 00:06:59.279 --> 00:07:01.540 to succeed. And then there's the iteration imperative, 00:07:01.839 --> 00:07:05.399 the idea that the real money isn't in V1, it's 00:07:05.399 --> 00:07:07.779 in what happens after the launch. That post -launch 00:07:07.779 --> 00:07:10.620 refinement is everything. There was a case study 00:07:10.620 --> 00:07:12.779 where continuous testing on a client's app over 00:07:12.779 --> 00:07:15.379 18 months resulted in user retention peaking 00:07:15.379 --> 00:07:17.439 at nearly three times what it was at the start. 00:07:17.560 --> 00:07:20.300 And they used clear metrics to prove it was working. 00:07:20.500 --> 00:07:23.019 So treating V1 as the start, not the end, is 00:07:23.019 --> 00:07:25.660 what turns those one -off projects into reliable, 00:07:25.800 --> 00:07:28.540 ongoing revenue. That's the key. So if iteration 00:07:28.540 --> 00:07:31.920 is how you stabilize revenue, what does the truly 00:07:31.920 --> 00:07:35.639 ambitious partner aim for beyond that? And I 00:07:35.639 --> 00:07:37.519 think that gets us to the strategic endgame, 00:07:37.699 --> 00:07:41.870 owning IP. Yes. Let's shift to that. The ultimate 00:07:41.870 --> 00:07:44.029 goal should be owning your own intellectual property. 00:07:44.250 --> 00:07:46.949 You have to use your client work as a live R 00:07:46.949 --> 00:07:49.389 &D lab. So if multiple clients are paying you 00:07:49.389 --> 00:07:51.170 to solve the same problem. That's your signal, 00:07:51.269 --> 00:07:53.449 a huge signal. You've got a scalable software 00:07:53.449 --> 00:07:55.810 asset on your hands. The classic case study for 00:07:55.810 --> 00:07:58.149 this is group talent, right? Exactly. They were 00:07:58.149 --> 00:08:00.410 a recruitment agency. They built some internal 00:08:00.410 --> 00:08:02.569 software to fix their own sales process. And 00:08:02.569 --> 00:08:04.850 that tool eventually spun out to become Outreach 00:08:04.850 --> 00:08:08.889 .io. Now a $60 million plus business. Service 00:08:08.889 --> 00:08:11.350 work becoming a scalable asset. Yeah. With much, 00:08:11.370 --> 00:08:14.610 much higher margins. Which leads to the next 00:08:14.610 --> 00:08:18.470 prediction. The partner race. Agencies need to 00:08:18.470 --> 00:08:20.850 focus on the mid -market. $50 to $500 million 00:08:20.850 --> 00:08:22.649 in revenue. They have budgets. They feel the 00:08:22.649 --> 00:08:24.550 pain. And they can actually move fast enough. 00:08:24.790 --> 00:08:27.269 They're too big for freelancers, but too small 00:08:27.269 --> 00:08:30.290 for McKinsey. It's the absolute sweet spot. And 00:08:30.290 --> 00:08:32.970 the race is on to lock down long term relationships 00:08:32.970 --> 00:08:36.129 with these companies right now. OK, but if 92 00:08:36.129 --> 00:08:39.809 percent of companies have messy data, isn't the 00:08:39.809 --> 00:08:42.429 mid market just a minefield of legacy systems? 00:08:42.490 --> 00:08:44.669 Yeah. That seems like a contradiction. That's 00:08:44.669 --> 00:08:47.299 a fair point. It is messy, but that's why we're 00:08:47.299 --> 00:08:49.879 seeing the rise of Prediction 9, the forward 00:08:49.879 --> 00:08:53.080 deployed engineer, the FDE, this new hybrid role. 00:08:53.320 --> 00:08:56.000 Yes. Half consultant, half developer. This person 00:08:56.000 --> 00:08:58.340 goes on site, they interview staff, and then 00:08:58.340 --> 00:09:01.039 using what we're calling vibe coding, they build 00:09:01.039 --> 00:09:02.960 working prototypes right there in front of the 00:09:02.960 --> 00:09:05.620 client. And vibe coding is just using AI tools 00:09:05.620 --> 00:09:08.259 to build functional code really fast based on 00:09:08.259 --> 00:09:10.379 verbal feedback. Right. And companies will pay 00:09:10.379 --> 00:09:13.519 a huge premium for that immediate visible value. 00:09:13.879 --> 00:09:16.879 Then there's prediction 10, the AI executive's 00:09:16.879 --> 00:09:18.940 assistant offer. This is a high -end service 00:09:18.940 --> 00:09:21.340 for leadership. Yeah, building systems that pull 00:09:21.340 --> 00:09:23.320 critical insights from across the whole company 00:09:23.320 --> 00:09:25.639 for planning and decision -making. This is a 00:09:25.639 --> 00:09:28.639 strategic partnership. A $50 ,000 to $150 ,000 00:09:28.639 --> 00:09:32.000 annual retainer opportunity. You become a strategic 00:09:32.000 --> 00:09:35.139 advisor, not just a chatbot builder. And finally, 00:09:35.240 --> 00:09:39.340 prediction 11, the Google effect. We have to 00:09:39.340 --> 00:09:41.159 think about the threat of big platforms like 00:09:41.159 --> 00:09:44.139 Google or Microsoft launching these sophisticated 00:09:44.139 --> 00:09:47.899 business setup wizards. That could vibe build 00:09:47.899 --> 00:09:50.679 a basic tech stack for a small business instantly. 00:09:51.210 --> 00:09:54.269 When that happens, the agency role shifts. You're 00:09:54.269 --> 00:09:56.610 not building from scratch anymore. You're focusing 00:09:56.610 --> 00:09:59.250 on integration and facilitation, connecting those 00:09:59.250 --> 00:10:01.850 powerful platform tools to the client's messy 00:10:01.850 --> 00:10:04.549 legacy systems. Okay, let's recap the big moves 00:10:04.549 --> 00:10:07.950 for 2026. Niche down hard. Focus beats broad 00:10:07.950 --> 00:10:11.230 every single time. Two, expand your scope. You 00:10:11.230 --> 00:10:13.149 have to add consulting, software development, 00:10:13.230 --> 00:10:15.909 and that non -negotiable education package. After 00:10:15.909 --> 00:10:18.820 the audit. Turn that $5 ,000 diagnosis into a 00:10:18.820 --> 00:10:21.559 $50 ,000 project. Invest in your internal infrastructure 00:10:21.559 --> 00:10:24.860 so you can iterate fast. And five, go after that 00:10:24.860 --> 00:10:27.080 mid -market now before your competition does. 00:10:27.240 --> 00:10:29.899 The gap between that cheap automation freelancer 00:10:29.899 --> 00:10:32.860 and the true AI transformation partner who guarantees 00:10:32.860 --> 00:10:36.139 ROI, it's just widening exponentially. The opportunity 00:10:36.139 --> 00:10:38.259 isn't disappearing at all. It's just evolving 00:10:38.259 --> 00:10:41.360 into a much more mature professional market where 00:10:41.360 --> 00:10:43.519 expertise and results are all that matter. So 00:10:43.519 --> 00:10:45.919 a final thought for everyone listening. If most 00:10:45.919 --> 00:10:48.200 businesses are still using those messy Excel 00:10:48.200 --> 00:10:51.080 sheets from 2015 to run core functions, what 00:10:51.080 --> 00:10:53.279 deep foundational layer of our current business 00:10:53.279 --> 00:10:55.720 infrastructure will we look back on in five years 00:10:55.720 --> 00:10:58.379 and realize was the single biggest obstacle to 00:10:58.379 --> 00:11:00.820 true AI transformation? Something to think about.