The Economics of AI Implementation  🔗 **Source:** [The Economics of AI Implementation](https://share.snipd.com/episode/76f4f688-88a2-440b-826f-d7d71308051b) ## Highlights - **AI Acts Like A Person Not A System** - AI behaves like a person not a system, so ROI comes from changing how people work rather than swapping software. - Connor Grennan explains AI is closer to change management and should be measured by KPIs tied to people and roles, not product replacement. Transcript: Connor Grennan I mean, so I think I started to talk about this on the last podcast. Maybe I didn't. But anyway, so I was at Morgan Stanley just recently and really working with a big group over there about like roi and what does roi look like because i think that this is one of the big things And we so at amines that we get to you know roll out these enterprise partnerships with you know big organization like fortune 100 organizations like you know 25 35 000 people at a time We kind of like roll out this course with and one of the big questions obviously they have jaden is like well how do we measure ROI? And so we have like a few ideas on this, but like our thinking has evolved. So I'm going to kind of take us through just like a couple of ways of thinking about this. But then at the same time, Katie George, who I'm a big fan of, she was over at McKinsey for a long time and is now over at Microsoft, wrote that piece in Time about like ROI. And first of all, I really agree with sort of like what she was saying about this. But I'm also, you know, what she was sort of saying is essentially like people keep looking at this as if it's, you know, a new system or a new product that you put in place and all of a sudden Your ROI jumps. And she's like, it's just not like that. It's about how changing how people work. And we're like, that's exactly how we think. And so Gene, just to sort of kick off this conversation that we've, we think about this is I'm going to give two ways that we think about this. Number one is we're already measuring ROI now every time we give somebody a salary, right? If you give somebody a salary for $200,000 in your organization, you are counting on the fact that there's an ROI on that person in that role paying that amount of money to. Now, everyone knows that there's not the same ROI in every salary. You know that sort of like people can be making the same. And on some, like the ROI on that salary is like, you know, 10x. And on some, it's like, oh, it's like, you know, 1x or 2x or something like that. You're basically getting your money's worth. Why? Because behind that ROI, behind that salary is a human that has to work better or work less. So if that's the ROI and AI is intended to augment how people work, now you're talking about X-ing that ROI of that salary. It's not like, and Jane, we do some work with hedge funds at AI Mindset. And listen, God love you for a hedge fund out there. You're not the easiest people to work with. I think you know that. I think you're, no, I'm not telling you anything you don't know. But the thing about hedge funds is that like, they're thinking often in terms of like, you know, algorithms and things like that. Like, how can we, you know, just sort of like get a slightly better return on investment, but on huge investment on, you know, billion dollar investment. And I get that, but that's not AI. So where does the disconnect come in? It comes in because AI looks like a technology, but acts like a person, right? What AI, we know when we do this with big companies, what we say all the time is like, this looks like a digital transformation, but it has nothing in common with a digital transformation. It has everything in common with change management. The digital transformation comes when you switch out your old system for Salesforce, and then you have to burn down your old CRM, and then you move everybody over to Salesforce. That's a digital transformation. You've moved people from a less efficient system to a more efficient system. AI is not that. AI is how do you augment how people think and work. And so when we think about ROI, the first one, as I said, is can you look at the KPIs that you expect out of people and teams? And are those things going up? And by the way, it doesn't have to be AI. Like I don't care if somebody shows up to work and they took their car and got here in 10 minutes or if they walk two hours, like they still show up at work. So in a way, AI isn't that important. You know what I mean? Like the AI part is not as important as what drives value, right? So and then the second one that I think about, Jayden, is the innovation quotient. And we do this actually at AI Mindset. We have a whole framework around it. But the idea behind that is, and we did this with Walmart actually, we find that productivity tends to be the enemy of innovation. And what I mean by that is like an hour of productivity gain. Let's say you've kind of like gotten 10, 20% more productive over an hour. Look, that's gonna add up. That's great. That's awesome. That feels like everybody has the end state, but that's not how we think about ROI. We think about ROI as like in terms of innovation. How do you actually transform? If you spent that hour and just like, you know, really focus on how you would transform your role and use AI to lift the cognitive load off you so you don't have to figure all that out. Now you're talking about ROI. That's not 10%. It's like 10 X. And I'm not even exaggerating because we see this with companies like Anthropic. This is possible, but that's when you focus on innovation. So ROI can mean a lot of different things, but what I urge people to think about is not like we have now have AI. What can we expect from our company? I'm like, it's not what you expect from your company or product. It's what you expect from your people. ([Time 0:01:38](https://share.snipd.com/snip/aa8b6df4-7b87-457d-a114-2f4956e74c01)) - **Measure AI By People KPIs Not Tool Metrics** - Measure AI ROI by the KPIs you expect from people and teams, then assess whether those outputs improve. - Grennan points out salary already encodes expected ROI, and AI's impact is effectively multiplying that return by augmenting people. Transcript: Connor Grennan Number one is we're already measuring ROI now every time we give somebody a salary, right? If you give somebody a salary for $200,000 in your organization, you are counting on the fact that there's an ROI on that person in that role paying that amount of money to. Now, everyone knows that there's not the same ROI in every salary. You know that sort of like people can be making the same. And on some, like the ROI on that salary is like, you know, 10x. And on some, it's like, oh, it's like, you know, 1x or 2x or something like that. You're basically getting your money's worth. Why? Because behind that ROI, behind that salary is a human that has to work better or work less. So if that's the ROI and AI is intended to augment how people work, now you're talking about X-ing that ROI of that salary. It's not like, and Jane, we do some work with hedge funds at AI Mindset. And listen, God love you for a hedge fund out there. You're not the easiest people to work with. I think you know that. I think you're, no, I'm not telling you anything you don't know. But the thing about hedge funds is that like, they're thinking often in terms of like, you know, algorithms and things like that. Like, how can we, you know, just sort of like get a slightly better return on investment, but on huge investment on, you know, billion dollar investment. And I get that, but that's not AI. So where does the disconnect come in? It comes in because AI looks like a technology, but acts like a person, right? What AI, we know when we do this with big companies, what we say all the time is like, this looks like a digital transformation, but it has nothing in common with a digital transformation. It has everything in common with change management. The digital transformation comes when you switch out your old system for Salesforce, and then you have to burn down your old CRM, and then you move everybody over to Salesforce. That's a digital transformation. You've moved people from a less efficient system to a more efficient system. AI is not that. AI is how do you augment how people think and work. And so when we think about ROI, the first one, as I said, is can you look at the KPIs that you expect out of people and teams? And are those things going up? And by the way, it doesn't have to be AI. Like I don't care if somebody shows up to work and they took their car and got here in 10 minutes or if they walk two hours, like they still show up at work. So in a way, AI isn't that important. You know what I mean? Like the AI part is not as important as what drives value, right? So and then the second one that I think about, Jayden, is the innovation quotient. And we do this actually at AI Mindset. We have a whole framework around it. But the idea behind that is, and we did this with Walmart actually, we find that productivity tends to be the enemy of innovation. And what I mean by that is like an hour of productivity gain. Let's say you've kind of like gotten 10, 20% more productive over an hour. Look, that's gonna add up. That's great. That's awesome. That feels like everybody has the end state, but that's not how we think about ROI. We think about ROI as like in terms of innovation. How do you actually transform? If you spent that hour and just like, you know, really focus on how you would transform your role and use AI to lift the cognitive load off you so you don't have to figure all that out. Now you're talking about ROI. That's not 10%. It's like 10 X. And I'm not even exaggerating because we see this with companies like Anthropic. This is possible, but that's when you focus on innovation. So ROI can mean a lot of different things, but what I urge people to think about is not like we have now have AI. What can we expect from our company? I'm like, it's not what you expect from your company or product. It's what you expect from your people. ([Time 0:02:46](https://share.snipd.com/snip/fa239101-d44c-4da8-9d71-ae2a7b185df7)) - **Productivity Gains Can Kill Innovation** - Productivity gains often reduce time but can block transformative innovation; shifting cognitive load unlocks step-change value. - Grennan recounts Walmart work showing an hour repurposed from routine tasks to redesigning the role can produce 10x impact vs incremental speedups. Transcript: Connor Grennan And we do this actually at AI Mindset. We have a whole framework around it. But the idea behind that is, and we did this with Walmart actually, we find that productivity tends to be the enemy of innovation. And what I mean by that is like an hour of productivity gain. Let's say you've kind of like gotten 10, 20% more productive over an hour. Look, that's gonna add up. That's great. That's awesome. That feels like everybody has the end state, but that's not how we think about ROI. We think about ROI as like in terms of innovation. How do you actually transform? If you spent that hour and just like, you know, really focus on how you would transform your role and use AI to lift the cognitive load off you so you don't have to figure all that out. Now you're talking about ROI. That's not 10%. It's like 10 X. And I'm not even exaggerating because we see this with companies like Anthropic. This is possible, but that's when you focus on innovation. ([Time 0:05:08](https://share.snipd.com/snip/45d02cb0-8777-4cf4-b238-702f73bb96ad)) - **Redesign Workflows Around AI Capabilities** - Reimagine processes rather than merely speeding them up; aim to replace the horse with a car by redesigning workflows around AI capabilities. - Jaeden Schafer illustrates this with agent-mode and Claude's background scripting that solved tasks by editing files instead of clicking UIs. Transcript: Jaeden Schafer A hundred percent. And I think one interesting thing about all of this is, you know, you can, you could be like, okay, how do we optimize what we're currently doing? Okay. We're using AI. We're getting, you know, 10% better, 20% better, 50% better, and all of the same processes. But the way you use AI can or what AI can do can fundamentally shift how your entire company operates. It's like, how can I get my horse to go like 20% faster, 10% faster? Don't don't use the horse, go to a car, you're in a completely different league, you're doing things in a completely different way. It's interesting. And this is probably a terrible analogy here. But it's kind of like, for me, the way I think about this, when I first started using chat GPT's agent mode, which was like maybe like a year ago, I had to pay like $200 a month to use it. And everything it would try to do on my screen was doing it the way I was used to doing things. So take a screenshot of my screen, move the mouse, click a button, take a screenshot, move the mouse, click a button. It was very slow. It had glitches. It had issues. Then came along Claude and the way Claude's co-work, which has absolutely taken over and they went up 80X in like, you know, usage is because with code and co-work, what it does is it sure It might look at your screen. It might take a screenshot. It doesn't try to click the button. It runs a Python script in the background that extracts everything on the webpage and figures out where the buttons are and how everything goes. And then it formulates a plan and does like logic and reasoning to think about what you're trying to do. And then it's like running scripts to do things. I told it to edit a video on CapCut once thinking it was gonna open up the CapCut app and click around the UI and edit the thing. And first it tried that for two seconds, said this is too slow. Close the CapCut app, opened the file it had created, edited the code on the file and said, okay, your video's edited. I opened it up and it had edited everything for me. This was a completely different way of approaching the problem. If I was like, how much faster can I edit a video in CapCut and Claude opened up the screen and it clicked, you know, 20% faster than me on the edits I needed to do. Well, whatever it got 20, but it did it in a completely different way. So anyways, this applied to every business is I think a big thing that we should be thinking about. ([Time 0:06:08](https://share.snipd.com/snip/39405fb2-7567-4f9c-bbd8-51ca564248c7)) - **Claude Edited My CapCut Video By Modifying Files** - Jaeden describes using Claude Co‑Work to edit a CapCut video by running background scripts instead of automating the UI. - Claude closed the app, edited the video file directly, and returned a finished video far faster than clicking through the interface. Transcript: Jaeden Schafer And this is probably a terrible analogy here. But it's kind of like, for me, the way I think about this, when I first started using chat GPT's agent mode, which was like maybe like a year ago, I had to pay like $200 a month to use it. And everything it would try to do on my screen was doing it the way I was used to doing things. So take a screenshot of my screen, move the mouse, click a button, take a screenshot, move the mouse, click a button. It was very slow. It had glitches. It had issues. Then came along Claude and the way Claude's co-work, which has absolutely taken over and they went up 80X in like, you know, usage is because with code and co-work, what it does is it sure It might look at your screen. It might take a screenshot. It doesn't try to click the button. It runs a Python script in the background that extracts everything on the webpage and figures out where the buttons are and how everything goes. And then it formulates a plan and does like logic and reasoning to think about what you're trying to do. And then it's like running scripts to do things. I told it to edit a video on CapCut once thinking it was gonna open up the CapCut app and click around the UI and edit the thing. And first it tried that for two seconds, said this is too slow. Close the CapCut app, opened the file it had created, edited the code on the file and said, okay, your video's edited. I opened it up and it had edited everything for me. This was a completely different way of approaching the problem. ([Time 0:06:40](https://share.snipd.com/snip/ef5ae9cb-7d79-460c-9f4f-27a4c17473ce)) - **Thumbnail Costs Nearly Broke A News Side Project** - Jaeden built AIChatDaily and discovered image generation costs $0.50 per thumbnail, costing ~$50/day at scale for 100 articles. - He solved it by pre-generating and cycling stored images per person/company, cutting costs to near zero. Transcript: Jaeden Schafer This is probably a smaller thing, but I just have to share. I'm here for this. It was killing me the other day. I have my new site, which is AI chat daily.com. And it basically scrapes 20 different sources and writes up, consolidates the news and writes really great articles on everything happening in the news every day in AI. So something that I was doing was I needed to generate like really good thumbnails or like hero art for every article that was being generated. And I was using open AI 1.1 or open AI to image model to do this. Now, I had these really amazing prompts. I had this amazing prompt cycle where I had like 20 different prompts and would cycle through them. And it was making these images that just looked incredible. And it was like grabbing them from Wikipedia and then editing. Anyways, I had this like whole process that was amazing. The problem was every image was costing me like 50 cents. This is a brand new site that wasn't making any money. I was making like 100 articles a day. And so I mean, it was like 50 bucks a day that all of a sudden I just started having to pay for all of these images. And I was like, I was like, just because AI can do this, and it looks amazing. And I like I'm getting people on LinkedIn that are like posting like, Oh my gosh, have you seen this new website? It's like so good for AI news. It's like I'm like, wow, this is cool. Like people are finding this. I was thrilled. But like, it was costing me 50 bucks a day for like a brand new project doing nothing. And I'm like extrapolating like how am I going to spend on this in a year? Anyways, at the end of the day, I was like, I basically just changed it so that every time it made an image, it automatically categorized into a category. And then I made a folder for every single company and person in the world. And then I just put, you know, 10, 20 images that had been generated of every person in the world into these folders. And then it just automatically picked one and cycled through. So every 20 articles, it kind of cycles through them. Okay. All of a sudden I went from spending X to spending free basically on this, just because AI can do something doesn't mean AI should do it. You have to look at the ROI. And so anyways, I get really excited about AI, but even I sometimes realize you got to optimize this stuff. ([Time 0:08:12](https://share.snipd.com/snip/752b1e33-48f2-4c83-a69d-17af23dd2a76)) - **Prompt Models To Invent Alternatives Not Just Steps** - Ask your model to 'think about reinventing this' or 'take a breath' to get higher-level alternatives instead of stepwise automation. - Connor Grennan calls this the 2023 prompt trick and says it surfaces elevator solutions the user wouldn't have thought of. Transcript: Connor Grennan So this is a tip. So I don't usually do a lot of like tips. Right. But I'll just tell you something that works really well for me jade remember when um before like chain before the reasoning models came out the way you could sort of like hack that was To sort of say like uh it's like the chain of thought prompting it's like hey walk me through this hey think deeper about the whole tip about take a deep breath and everybody's like why Does that work it's because ai understands what you mean by that it means to think deeper. And so like what I do all the time is I'm sort of like thinking like, Hey, I want to do this. And I'm kind of like faster horsing it. Right. I'm sort of like kind of working my way up and it's getting better and better. I'm like, this is great. And then I always do the same thing. And I, it's usually in cloud Gemini, something like that. And I'll say, Hey, listen, you know what I'm trying to do here, right? Can you just think like, what would you say if you're coming in and like reinventing this because you're sort of like trying to answer me and do what I'm doing? And Claude will often say, it's like, yeah, listen, like, let me just take a second and think about it. So it's, it's a weird sort of like old timey 2023 trick, but it works unbelievably well. Cause what it does is it's Jaden, if you and I were sort of like building something, it's sitting back, like hey i guess we're going down this road but what are we actually trying to accomplish Do you know what i mean like and so and usually we're taking all these steps but if the if the trick is to get this end state and claude will sometimes literally say yeah connor listen i've Been doing it this way because you've been asking me to do it this way my thought is like if you really want to get this why don't we do it this way? And there's no way, Jayden, because of my cognitive limitations, I ever would have thought about that. So guys, just an easy trick on this is just to sort of like pause, get that 2023 tip handbook out here and just be like, hey, listen, Gemini, Claude, chat, whatever you're using, just take A breath for a second. This is what we're actually trying to accomplish. We are trying to 10x our revenue and I'm taking you down this path and you're trying to help me build this little staircase over here. But is there an elevator that you could just like build, you know, sort of like over there in the background? ([Time 0:10:41](https://share.snipd.com/snip/9dfc488a-55f8-49e6-b8d4-1b58a3f48e1f)) - **Train People At Scale To Capture AI ROI** - To scale AI adoption in enterprises, train entire departments rather than only deploying tools; roll out seats and courses organization-wide. - Jaeden recommends Connor Grennan's AI Mindset course as a method to level up people across large organizations. Transcript: Jaeden Schafer Thank you so much for tuning into the podcast today. If you enjoyed Connor's breakdown here, the number one thing I would absolutely recommend you have to do is go check out Connor's AI mindset course. He gets into a lot of these kind of bigger questions, you know, what people are getting wrong about the ROI with AI, a lot of ways to fundamentally shift how you're thinking about AI and How you apply it. And the place that I think this is the most valuable because he's telling me about, you know, different organizations that will buy like 10,000 seats for their whole organization to Get this rolled out internally. That's what I would really recommend is bring this to your department, bring this to your boss, bring this to your own organization. And roll this out for everyone in there to really be able to level up themselves with AI, how they're approaching AI and what they're doing. So there's a link in the description to Connor's AI mindset course. The number one course I would recommend to anyone on AI. Absolutely phenomenal, especially in enterprise. ([Time 0:12:39](https://share.snipd.com/snip/1bffe37e-c046-47d6-9cb9-0671bf6d0f5d))