Episode 593 - Pricing AI Credit: Activity-based costing in a hoodie

Speaker A: Like a chrysalis, we're emerging from the economy of the industrial revolution, an economy confined to and limited by the earth's physical resources, into the economy in mind, in which there are no bounds on human imagination and the freedom to create is the most precious natural resource. Welcome to Thresholds, the soul of enterprise business in the transformation economy. Charting the new economy where human flourishing is the bottom line. I'm at class with my friend and co host and co founder of Threshold, Ron Baker. And on today's show, folks, AI credits activity based costing in a hoodie. Ron, at the end of last week's show with Deborah, uh, I say, you know, what do we got coming up next week? And you just like, pretty much exploded AI cost. I mean, it was, I, we were thinking about replaying it, but it was very intense for the last segment of a show. Um, and it couldn't have come in a better week. I opened in west side Story, uh, this weekend and as you could probably hear in my voice, it's, it's a little stressed and strained. I've got my, my Ricola, um, on the side here. Um, so I'm just going to get out of the way here is really what I'm going to do. We're going to talk about AI credits and what companies are adopting them and how is this good or bad or not for customers and what should they do instead? But just a little quick background on this, Ron. Uber, as a company, burned through their entire budget of AI for 2026 in April.

Speaker B: Uh, what's that tell you, Ed? Well, I mean, set aside that budgets are BS and elastic, right? But the fact that they had a budget and they blew through it in four months of the year, something's wrong with this pricing model. It's not the customer's fault, it's the company's fault. But keep going.

Speaker A: Yep. Yep. So this is a process known as token maxing, where companies show off AI use on the, on even dashboards. You know, they're like saying, oh, look. Well, yeah, you used it to rewrite an email know in the active voice. Okay, yeah, that's great for you. Um, according to ramp, um, AI overall spending is up 13 fold in the past year. I mean, it, this sort of makes sense in the, in the fact that it's really just continuing to explode. But you know, this sounds more like a cash casino than a cash register when you randomly don't know what the price is going to be, you know,

Speaker B: and it's also cost per employee, Ed, so it's not Just that it's used by more people. It's. Each person's cost is going up in an uncertain way.

Speaker A: Yep. And, and you know, we've been talking a little bit about this on, on different shows. In fact, the last pricing update we did, we talked a bit about it and we don't like it, Ron. I mean it's, it's, when it really comes down to it, the, it's the tokenization of, of something that's already tokenized, which is money. Right. It's already tokenized. So why, why do we have to put this. It, it sounds like it's. This is classic price obfuscation, but not in a good way. This is in a really bad way. I mean sometimes we'd use price ob can be a little bit of an advantage and really the right strategy. But this just seems to be. They, they just want to make sure that they're covering their costs. And man, if there's anything you and I have railed against for 30 years, it's cost plus pricing. And how is it that the most advanced technology on the planet is now using the, the, the most unadvanced method of, of pricing? It just doesn't make any sense to me.

Speaker B: I know, uh, it really doesn't. And you know, we title this show AI Credits. But people do conflate this idea tokens. And they are different. They are completely different. Uh, and we'll get into that. But think of credits at, as casino chips. You know, for AI right. You go into a casino and you buy chips. Now casinos, part of the experience is to be uncertain. But uncertainty when you buy stuff is not a great experience. It may be fun in the casino, but it's not fun in the grocery store when you get wildly different prices for the same item at the same time in the same store on the same day. Uh, that's kind of what we're seeing with this. And it doesn't make a lot of sense. I mean, books have been written, papers have been written about how to prompt it and, and save on tokens, save on compute time and all this stuff. Algorithms and prediction things have been written. There's shadow pricing in this market. All of these things lead me to say, gee, this is just like compensation consultants and advertising agencies or the law firm auditors that went through the billable hour reports finding errors or you know, duplicated meetings and all of that. When your pricing model gets so friggin complicated that there is a shadow market, you're out of control. And this is out of control. Now the Good thing is only 30%, I think, of AI companies or they say, they keep saying new AI companies are using this model, but a lot of different companies, from Adobe to, you know, uh, all sorts of other AI driven companies are using it as part of their pricing options. So it, you know, and, and there's been professional pricers writing on this. We're going to talk about Mark Stiving, who has been on the show several times. We have very high regard for him. He just wrote an article on LinkedIn justifying, uh, token pricing, but with a heavy heart. He doesn't like it. He doesn't think you should stay there. This is just temporary, so we'll get into that. But, you know, tokens, so, so let's distinguish between credits and tokens. Yeah, really, this is really important. Uh, you know, when you buy credits, you're not buying AI directly, you're buying credits, which is a layer that the AI company has put on top of money. Created their own basic. Basically they've created their own currency. And tokens are the raw unit of AI consumption. Think words, characters, text, what you're asking it to do. The machine measures token use, but credits are. Instead of saying, hey, this costs 27,000 tokens, this costs 143 credits. Because a credit equals so many tokens. Right. It's just another layer, like you said, of obfuscation. But it's a completely redundant layer because we already have credits. They're called dollars. We already have a measuring rod for this. So what got me started on this was, uh, just all the things I was reading about. Credit pricing, token pricing. I'm going to use the two because they flip back and forth. Nobody makes the distinction. But they should. But Michael Mansard, who is the principal director of subscription strategy at zora. Uh-huh. So, you know, Teenso's company, uh, wrote a white paper, and that white paper is called AI Credit Monetization From Hype to Concrete Blueprint. And it's his compass framework. So he's got this elaborate framework to see if token pricing is right for you and to be fair to him, and I will be, because then I'm gonna, I want to tear him apart. But to be fair to him, he does say this at the start of the paper. Without a, uh, total operational overhaul supported by Fit for Purpose Systems, credits will brutally erode customer trust and crush your teams. No kidding. That's a complete understatement. Because when you start moving into this area of pricing uncertainty, like the billable hour we've seen, we've Already seen usage and consumption. It's called the billable hour. In professional world, it's a terrible customer experience. I figured that out in 1989. People before me had figured it out, but it didn't. It took a long time to diffuse, but it's a crappy customer experience when a customer doesn't know what something's going to cost until after it's done. Uh, there's not many providers that do that. People point to the medical market. Well, that's because there's third party payers and you don't really care what the price is because you're not paying the bulk of it. But when you go to a vet or something that's not covered, you darn well care, you darn well shop, you put in the research, you know, to find the best price for an MRI or hip surgery or whatever it may be. And so this is just violating every rule we know about pricing and what we've learned in the last 50, 75 years about how consumers make choices. And I can't believe there are pricers that are bought into this or that are advocating it. And I'm not talking about Mark, I'm talking about other people. I've read on this, right. And it's like the professional pricing community grew up railing against cost plus pricing and now we're seeing it in action. And this is like the, the cost accountants are back, or even worse, the engineers are back. This is why, you know, we call this, um, ABC and a hoodie. This isn't so much the finance or the cost accountants, these are the engineers. Well, we just don't know what it's going to cost. Well, too bad. That's not the customer's problem, that's your problem. And my contention, Ed, is that it's easier for AI companies to spread risk across their portfolio than it is for a customer to assume that risk. So we've got to bring in some actuaries here, actuarial principles that there's, hey, there's no such thing as a bad prompt. There's only bad pricing, you know, of the prompt.

Speaker A: Yeah, I mean, to me, this is a little bit like a joke that we often, uh, it's said in jest, I should say. It's not really a joke. We say ingest. We don't allow pilots to, to price the, the air, the, your, the plane trip. Uh, but now what we're saying is that we know we have to let the engineers involved, involved in the pricing just because they're the smartest guys in the all right, it doesn't make any sense. Now, I will say this. We use some AI tools, um, on this show, one of them being OPUS clips, which is what we use to put our clips out on LinkedIn. And it is a credit system. We buy a certain number of credits for the year. But I'm not sure why. I mean, I guess the way that it effectively works is it is a credit per minute, it's a credit per minute of video and plus one credit per post. So we do, we roughly take 60 minutes of video every, every month. I'm m sorry, every week. We then make about five posts. So we're burning 65 credits per week, give or take. Right. Sometimes a little bit more, sometimes a little bit less. So I quickly figured that out. I did the multiplication, multiplied it by 52 or whatever, and I'm like, all right, the plan that we need for us is this. But I'm not quite sure why we, you know, why we have this credit system in place anyway. There would be a lot easier way for them to do this than to introduce this notion of credits. The other thing that I think is partially the problem is like something like Opus and something. And even LinkedIn and all of these companies, they don't have really native AI. They are using others, the backend AI, the large language models on their side. So I think that that's partially also a, uh, disconnect. There's is like they're, they're using these utilities, if you will, and they're not. And the utilities are not pricing them appropriately. And I think that's it. And so it's become a downstream problem. That's, that's what I'm trying to express.

Speaker B: Yep. Yep. If your costs are uncertain from your providers, then you're gonna.

Speaker C: Correct.

Speaker B: You know.

Speaker C: Correct.

Speaker B: The natural tendency is to pass that along to the customer, but the customer's lost in this whole thing.

Speaker A: Correct.

Speaker B: Nobody seems to give a crap about the customer experience. And I'm telling you, the first AI company, company that figures this out is, is gonna, is gonna steal market share. You know, and we haven't even got into the fact that what's happening when these AI companies is they're climbing down the learning curve. And as you know, with learning curve or experience curves, every doubling of the production reduces the cost to serve by 30 or 40%. Now that's gonna happen dramatically with AI as the, as we move from more GPU chips to ASCII chips that are more specialized, they're greater efficiency, they have greater capacity, they can do things faster. And also, uh, bringing nuclear power online I think is going to make a little bit of a difference, even though electricity costs are only about 2 1/2% of the cost of running these things. So it's not the electricity generation that that is the big driver here. It's really the silicon. It's the chips that are driving the uncertainty in price. But Ed, uh, there's so much more to say about this, but unfortunately we're up against it. And folks, if like to remind you, if you want to contact me or Ed, send us an email to ask tsoeah thresholdnow.com do check out our Patreon Patreon channel where you can get access to our bonus content. You can do that@patreon.com TSOE and now we want to hear from our sponsors, including Rippling.

Speaker A: The best firms don't just advise on strategy. They help clients build better operations. Rippling is built for exactly that one unified platform for HR, payroll, finance, and IT that any business can run on. And Rippling AI cuts administrative overhead by

Speaker C: taking action across payroll, HR, finance and

Speaker A: IT in real time on your actual data. Visit rippling.com for more tax season and

Speaker B: an accountant working from home can't get into their software your IT guy isn't picking up. Sound familiar?

Speaker A: Summit was built for exactly this a secure, fully managed cloud workspace that hosts QuickBooks, tax software and any desktop application

Speaker B: your firm depends on, accessible from anywhere with the security and compliance your clients expect. And when something needs attention, our team picks up fast. 15 minute response 24, 7 with experts who actually know your software.

Speaker C: One predictable monthly price. No surprises.

Speaker A: Find out what your firm has been missing@SummithQ.com In a world moving faster than

Speaker D: ever, who helps make sense of what comes next? At Cal cpa, we bring together the brightest minds in accounting and finance, professionals shaping how businesses grow, adapt, and lead with integrity. From advocating for sound policy to advancing financial literacy, CALCPA is more than an association. We're a voice for trust, clarity, and the future of the profession. Discover how we're making sense of what comes next@Calcpa.org if someone tells you that

Speaker C: you need to productize your services, you're turning your wisdom into widgets. But that's not progress, it's a U turn. At Threshold, we help professionals escape the factory floor of formulas and checklists and enter the domain of the transformation economy. You weren't born to sell products and services. You were called to guide transformations. If you're being told to productize, call us instead because your highest value isn't an SKU, it's a shift. Visit thresholdnow.com to learn more.

Speaker E: You are tuned into the Soul of Enterprise with Ron Baker and Ed Klass. To find out more about our show, visit us on the web@thesoloventerprise.com youm can also chat with us on Twitter using AskTSOE. Now back to the Soul of Enterprise.

Speaker B: Welcome back everybody. We're talking about AI credits, activity based costing and a hoodie. Um, Ed, this really is a cost accounting, engineering, uh, mindset that we're dealing with here. But I want to get into Mansard, Michael Mansard's uh, white paper because the 75 pages, it's quite long, but it's quite good. Um, even though I'm totally befuddled by his Compass framework, it seems really over complicated to me. But put that aside. He gives a great history. He gives you concrete examples of how firms are doing this and, and there's quite a few of those. And he, he understands the risk, he understands this is not a good thing for the customer, that it's a crappy experience. But giving the strongest version of his argument, he says AI creates four genuine pricing problems. The first is variable cost. Some prompts cost pennies, others cost dollars. Some agents may take hours. Okay, we get that. Um, diverse use cases, you know, one gen, one user generates a summary and another performs a due diligence review. Right. Different people are using it at different levels. Okay, great. You've got this dispersed portfolio, we understand that. Unpredictable consumption, Nobody, nobody knows usage patterns yet. Now I gotta believe there is some data on this. They've got to know which customers are doing what, which ones are the heavy users. There's gotta be ways to segment this. The other thing is margin protection. Unlimited subscriptions can create enormous losses. So his conclusion is credits become a commercial abstraction layer between volatile costs and customer consumption. Um, and he likes it because, and this, this goes back to the 1980s. Telecoms use this. Remember the prepaid phone calls? You'd lock into a price, you'd buy a number of minutes. Bloomberg did the same thing with its terminals. It used it. To this day it still uses a credit system. You buy at a certain price now you have the right to use and you can track your usage. So you kind of know that just like with a prepaid car, if I call, call you and spend two minutes wishing you happy birthday as opposed to, you know, making a uh, 20 minute business call, I have a feel for about what it's going to cost. And I can at least make a value assessment, even though I might not know the exact cost of each call. Um, but the point is, I think this whole debate, it's not verse, it's not credits, it's versus subscription. It's who bears the uncertainty. Are we going to put that uncertainty on the vendor or the customer? So Mansard's best case, he says, um, AI is different from sas. We understand that the marginal cost for the most part, isn't zero anymore. Um, and he talks about different, um, prompts require vastly different resources, from a quick summary to a video generation to agentic workflows to autonomous coding. Very different cost. A flat subscription is dangerous. Credit is a mechanism to manage the uncertainty. It's a legitimate concern. Um, and you might ask, well, why don't you just pay for tokens? Why don't you just go right to the raw computer? Well, nobody wants to buy tokens. Imagine restaurants charging for the grams of food. That's kind of what tokens are, right? Or a law firm not even charging by the hour, but charging by the keystroke. Um, and at least credits translate into something the customer can understand. Um, and nobody likes the idea of just buying tokens. Um, so does this sound reminiscent of the billable hour to you lot?

Speaker A: Reminiscent, yes. It's chapter and verse.

Speaker B: And when, when you, when you have the billable hour, when, like you said, when the cash register becomes the slot machine and you don't actually know what things are going to cost, you start rationing and optimizing fear or, I'm sorry, optimizing for fear. I'm not going to call my lawyer because I don't want to pay $200 for a phone call. I'm not going to use a, uh, ticket on this tech question because I don't want to waste a ticket. Right, and it makes you reluctant to seek out the help you need, rather than asking, can AI help me solve this problem? How many credits is this going to cost? And you have to go through these gyrations in your head. Terrible customer experience. Um, it's just, you know, ABC tried to say, oh, we're going to look at the activities and we're going to cost, uh, the offerings based on that. Well, first off, ABC is just cost accounting and drag with its own wacky assumptions that have nothing. It's all bad math. But the second thing is, ABC was just a new way to be wrong. That's all it was. And that's why I call this ABC and not just cost accounting, because they're trying to make it look sophisticated, like we're actually looking at the tokens that we're using to do whatever it is your prompt, ask, video, whatever. And it's like, uh, you're, you're, you're thinking that. I think the logical fallacy is we have to make a profit on every prompt.

Speaker A: Yep.

Speaker B: Same thing as we have to make a profit on every hour, every job, every customer. And that's just if portfolio and insurance companies taught us anything, is that that's ridiculous. And Lost Leaders in grocery stores, by the way, teaches us that that's fallacious thinking.

Speaker A: I'm, uh, trying to express this as best I can, but we talked about this project that we've been undertaking to get all of the transcripts of previous Soul of Enterprise into a standardized format that we could then query against in AI. And I developed an agent, my first agent. I'm very proud of it. But one of the most confusing aspects of it was deciding which m. Which GPT did I want to use from Chat GPT, because they're all available. Like you can, you can pick, you can. In the code, you can write, which one do you want to select? And it was so complex this, this first time through because you're going through, and I'm trying to understand, like, what, what does it mean to have, you know, X. X number of, uh, you know, not only tokens, but, um, how, how long it's going to take the, the prompt to come back. And it was a combination of how long is this going to take? Because ChatGPT4, when I, um, were the different variations, was taking much, much longer. It was taking like 20 minutes per transcript to do it. But ChatGPT 5.5, the highest version, was costing like $3.50 per transcript. So I had to figure out by using the different models, like five or six different models, wanting one thing at a time through what's my optimum? And I got it down to where I was getting. Be able to get throughput of four to five transcripts per hour at a price of 60 cents a transcript. But that was completely left on me to do that. And I got to tell you, I don't know whether I fully optimized it or not. I just got to the point where, like, that is just an acceptable price for me, right? Because I knew the price was as much as $3 and as low as 40 cents. So when I got to 56 cents and at a decent speed, I'm like, all right, that's fine. But it was all on me to do this and not only that, but as I spent more money with OpenAI, uh, it changed the math on it. It would say, oh well now you have access to this many tokens per hour. And I'm like, does this make, can I run them through faster? Is it what's. And. And I would, I would have had to have retested the code each time, one transcript at a time. And I'm like, I don't want to do that.

Speaker B: Mhm.

Speaker A: Right. So again, you know, now I'm using it as, as the, as a vendor in the back end, not just a regular chat GPT. I'm remember I'm on their console now and I'm, I'm doing the stuff in the back end. But this is what all of these companies I think are experiencing with the open AIs of the worlds and the anthropics is that this. Again I brought this up earlier, but this disintermediated layer where they're like, we really are very uncertain about what this is by the one who's providing it for us. So it's almost like if you had a variable cost per hour and your employees at an accounting firm said, well today I'm um, 350 an hour and if you use me more I'm going to be 150 an hour. And then the accounting firm would then have to then figure out the price on the other side. So it's a double abstract layer, I think.

Speaker B: Yeah. And CFOs and companies can't budget or plan for it depends. It just doesn't work. So then Mark Stiving comes out and he writes uh, in LinkedIn every Monday, usually with a nice post and they're incredibly thought provoking. And this One was from June 15, the case for token based pricing. And he starts out by saying this, that he said, I've spent decades arguing against cost plus pricing and token based pricing is simply cost plus. So when I say token based pricing makes sense right now, I want you to understand how much it pains me to type that and I get it, Mark, I'm right with you. Um, he says value based pricing requires three things. The seller needs to know their costs. Um, the seller needs to know what the solution is worth to the buyer and the buyer needs to believe that value is real. He says when any of those things are missing, value based pricing isn't principled, it's a guess dressed up as a strategy with AI agents. Today all three are missing. Sellers don't yet have a clear handle right on their costs. So he actually Says this, he says, um, a contractor charging by the hour on a novel project isn't being lazy, they're being honest about cost uncertainty. Now uh, he says that shifts risk to the customer, but he says the vendor keeps the margins low, so that's acceptable. So I guess the point is that the AI tokens or credits are keeping the margins low. Now first off, I'm not sure about that and I'm not sure if Mark is sure about that. How do you know? You know, um, cost plus pricing is uncertain all the way around. Yes, we say it's suboptimal. We say you can make more money with value based pricing, but that doesn't mean you're making more money on everything. We've never said that. We've said on some things you do you might lose money and you probably deserve to because it's not valuable. You know, a CPA standing at a copy machine is not valuable. Um, and so he says either way, um, it might be the right place to start, but it shouldn't be the place that you stay. Which I think is kind of an understatement. I don't think you should even start there. I think if you put in some thought to this, you could find ways around it and it might actually be, at least for a while, a competitive advantage. Ed, I think we're going to look back on this and maybe two years, five years, not sure but and say this is just like when AOL charged by the hour.

Speaker A: Mhm, mhm.

Speaker B: It's going to be as silly as that because as costs come down as we, as we just you know, glide through this experience curve and it's going to happen, I think this is going to look ridiculous.

Speaker A: Well, Google had a huge impact on the world. Obviously there's an understatement. But I remember being in a presentation that they were giving this is back in an ITA conference maybe 15 or maybe 20 years ago and the, the, the person from Google said look, we develop everything. This is back when like Gmail was first coming out. We develop everything and assume infinite bandwidth. Mhm, Assume infinite bandwidth. And there was an infinite bandwidth at the time. It was a, it was a big deal. Bandwidth was. But, but they said we assume that going forward. And I think that's one of the things that drove so much innovation in that company is because they assume it. So I think what you're saying is the AI company that assumes infinite infinite tokens, infinite whatever, that they might have the competitive advantage to figure this out because otherwise, I mean it's almost as if our old joke, which is you don't want to be wheeled into the surgery and you hear your surgeon say, never, Never seen this before. Is every prompt being submitted and effectively the AI agent is going, I've never seen this before.

Speaker B: Yeah, and this is my retort to Mark when he says your contractor isn't being lazy when he says this is just too uncertain to give you a price. Got the wrong contractor. I'm sorry, but if a company can't give us a price, I'm done, I'm out. I got the wrong guy. I mean, give me the surgeon that's done this 10,000 times. Don't sit there and whine about your cost to me. That's not my job.

Speaker A: His retort may be this is you have we. Nobody's ever, nobody's ever done this operation before. So you can't. You're not going to be able to find somebody with experience. I mean, I think that that would be the retort.

Speaker B: Yeah, But Ed, open AI has what, 800 million users last time I looked. Or 900 million. And, and I don't know, anthropic and perplexity. They've got to have some data on this, you know, and actually run it through AI. Yeah, an actuary can determine an insurance rate for a satellite that they've never insured or JLo's butt or whatever it is. And they make these kind of things, uh, guesses all the time. So I just, uh. This is just lazy. This is just friggin lazy. And they're putting the burden on the customer and they need to take the burden because they have the portfolio. I don't. I only deal with you one AI company at a time. They're dealing with hundreds of millions of customers.

Speaker A: Yep. All right, Ron. Well, we're past time for breaks. I want to remind our listeners they can get hold of Ron or me by sending an email to ask tsoe thresholdnow.com the website is the soul of enterprise, where you can see show notes as well as previews to upcoming shows. But right now, a word from our sponsors, including Summit hq,

Speaker B: tax season and an accountant working from home can't get into their socks. Software your IT guy isn't picking up. Sound familiar?

Speaker A: Summit was built for exactly this. A secure, fully managed cloud workspace that hosts QuickBooks, tax software and any desktop application your firm depends on, accessible from

Speaker B: anywhere with the security and compliance your clients expect. And when something needs attention, our team picks up fast. 15 minute response 24. 7 with experts who actually know your

Speaker C: software one predictable monthly price. No surprises.

Speaker A: Find out what your firm has been missing@summithq ah.com the best firms don't just advise on strategy. They help clients build better operations. Rippling is built for exactly that one unified platform for HR, payroll, finance, and IT that any business can run on. And Rippling AI cuts administrative action overhead

Speaker C: by taking action across payroll, HR, finance and IT in real time on your actual data.

Speaker A: Visit rippling.com for more.

Speaker D: In a world moving faster than ever, who helps make sense of what comes next? At uh Cal cpa, we bring together the brightest minds in accounting and finance professionals shaping how businesses grow, adapt, and lead with integrity. From advocating for sound policy to advancing financial literacy, Cal CPA is more than an association. We're a voice for trust, clarity, and the future of the profession. Discover how we're making sense of what comes next@Calcpa.org if someone tells you that

Speaker C: you need to productize your services, you're turning your wisdom into widgets. But that's not progress, it's a U turn. At Threshold, we help professionals escape the factory floor of formulas and checklists and enter the domain of the transformation economy. You weren't born to sell products and services, you were called to guide transformations. If you're being told to productize, call us instead. Because your highest value isn't an SKU, it's a shift. Visit thresholdnow.com to learn more.

Speaker E: You are tuned into the Soul of Enterprise with Ron Baker and Ed Class. To find out out more about our show, visit us on the web@thesoulofenterprise.com youm can also chat with us on Twitter using AskTSOE. Now back to the Soul of Enterprise.

Speaker B: Welcome back everybody. We're talking about AI credits, ABC in a hoodie, ABC being activity based costing. I think you've got the impression that we don't like this pricing strategy and I certainly don't. Um, but Ed, there's also another big error, uh, built into this. And that is that costs determine price. And I can't believe how many. Even in the professional pricing community, I think we are outliers on this. Oh, we are, because we're total Austrian economists that say, no, no, value determines price sets the upper boundary of price. Because price, after all, is just how you divide up value. So obviously it's value that determines everything and price justifies cost. And nobody said this better than Thomas Nagel and Reed Holden in the the classic book the Strategy and Tactics of Pricing. Now this is out of the third edition, but I Need to read this because even though it's a bit long, it's three paragraphs. It's so important that I think it needs to be drilled into the head of every pricer, um, that costs don't determine price. Shut up about your costs. I don't care. If cost determine price, then no business would ever fail. Every business would make a profit. No restaurant would ever go out of business. Every government contract, by the way, would be successful. Including the bullet train here in, uh, California that will never be built. So here's what Nagel and Holden say, and these guys are legend in the pricing community. The problem with cost driven pricing is fundamental. In most industries, it is impossible to determine a product's unit cost before determining its price. Why? Because unit costs change with volume. That's the point. Nobody get. And that is true in the professional firm too. Your unit cost changes with the number of customers you serve. And that's changing all the time. Customers coming in and out, blah, blah, blah. Uh, so the cost change occurs because a significant portion of costs are fixed. Now, Reginald would say they're cash costs, right? M put that aside and must somehow be allocated. And they even put allocated in quotes because they know those allocations are completely arbitrary, right? How do we allocate the toilet paper to each customer? It's completely, completely arbitrary, um, to determine the full unit cost. Unfortunately, since these allocations depend on volume, which changes with price, that's the other dynamic people miss. Lower the price, you're gonna have higher volume. Right? On average, unit cost is a moving target. Then they say pricing affects sales volume and that volume affects costs. Cost plus pricing leads to overpricing in weak markets and underpricing strong ones. Exactly the opposite direction of a prudent pricing strategy. So. So for your weak users, the people that just want summaries or help me write an email, you know, we're probably overpricing them and we're underpricing the people that are using it to say, hey, give me a Pixar, you know, grade animated video on this. Right? Uh, it just, it's crazy. This is what professional pricers help the business world see about cost. Plus, it's suboptimal. It's not that it can't be profitable. Of course it can. I know Costco uses it, I know AWS uses it. But it's a suboptimal pricing strategy. Then the last paragraph is, the job of financial management is not to insist that prices recover costs. It is to insist that costs are incurred only to make offerings that can be priced profitably, given their value to customers. That, that needs to be up on every engineer's wall. And every cost accountant, every CFO needs to stare at that every day. Because I'm telling you, as Henry Ford said, he said no one knows what a cost should be. And if we've got the experience of the learning curve, the experience curve, whatever. This doubling of volume drops cost by 30 or 40%. That's consistent across everything. Bain has proven this. McKinsey's proven this. It's true in egg production, airplane production, everywhere. It's certainly true in AI. I mean, costs have been coming down, right? Uh, and it's going to happen. So these companies should be planning on it. Um, now the, um, electric companies, you know, use kilowatt hours, right? But everybody kind of understands that. I mean we have smart meters now that can give you a running total. Right. I've got a water meter that can tell me, you know, in real time what I've used. So, so we already know that. Um, and then so I asked GPT, I said, why can't you give me a price? Or why can't AI companies give me a price before I run whatever prompt I put in. Right. And now it could be hallucinating, it could be blowing smoke up my skirt. It knows I don't like this form of pricing.

Speaker D: Sure.

Speaker B: But it said we can act. He says we can actually do that. He says, I, I kind of know based on the prompt. He says we can estimate the complexity, the runtime, the model requirements and the tool usage. So why can't we estimate a price?

Speaker A: Mhm.

Speaker B: And then he said, you know, uh, there's certain ways you could do this that are actually really, really interesting. I'll give you some examples. He said for small customers, just put them on subscription. For mid market customers, you know, I guess like medium or maybe large users thinking T shirt sizes, you could have subscription plus. You should, you could put some usage bands in there. Right. Just this is kind of like nature of work, right? Oh, you want to do this more complicated thing. You have to step up to this higher tier. I thought this was an interesting option. You could have the enterprise option, um, which would be a, you know, a subscription plus outcome base metrics, of course. And then you could have frontier Compute, you know, these frontier models that are at the leading edge. There's just one from China that came out that I want to ask Greg about. But um, so you could have, um. It actually really loved the idea of pricing based on speed. It could come back to you and say I can do this immediately for three bucks, I can do this within an hour for a buck. Or I can do it tomorrow morning by a quarter. Or I could do a weekend batch and it would be included in your subscription. It thought speed was a great thing. Now it did push back on me. It said speed doesn't solve everything. Uh, GPT said this, but he did, it did say it's very interesting because that's Q pricing rather than consumption pricing. Right. Just like FedEx, you know, and all of that. So um, it also said, um, you could do this. This request exceeds normal usage. Um, standard processing processing to do this would be included. If you want high compute processing, say $12. And if you want enterprise rendering, $45. In other words, they could tier right at the prompt level. They could give you choices.

Speaker C: Mhm.

Speaker B: Depending on the complexity. And to me that would still just be for the outlier case, you know, but if they're really worried about these outliers, if it's really true that 1% of the users are driving, you know, 80% of the costs, then okay, let's segment those users and start giving them options.

Speaker A: Yep, yep. And would it, would it shock you to find out that I had this conversation with GPT when I was doing the project that we were working on? I mean, I actually, I had IT do the analysis and that's where I arrived on the ChatGPT version 5. 4. Because that's where it. But it told me ahead of time. Okay, right, it told me ahead of time. But I. Here is the point. I had to be smart enough to prompt it to tell me that.

Speaker B: And Ed, when you're staring at a KPI from your boss that says Use AI, uh, use AI, you're not. You don't care. It's like going to the doctor and just paying your $5 copay. You don't care what crap costs because you're not paying for it. Mhm. So let's get the, you know, ask the AI. Now, GPT was emphatic that it could tell you a cost. Maybe that's because we were both using OpenAI.

Speaker A: Yes, very possible. But it also, it also gave me the trade off against how long is this going to take too. And that's because I said, hey, listen, I am willing to pay more as long as this can be done faster. I don't want this project to extend over weeks, which is what would happen if I had used the older models.

Speaker B: And see, this is why the FedEx model was so revolutionary at the time. Sure. You had cargo rates by distance and blah, blah, blah. But they put speed, they introduced a, uh, speed factor to it. That was, that was brilliant, absolutely brilliant. And AI should be going down this road, I think.

Speaker A: Well, one, one last quick example based on your example of kilowatt hours. And I know you know this part of the story, but when I got a Tesla, I contacted the electric company that I'm with and for one bizarre reason, uh, the electric place, the company that I'm in is not txu, which is what Power Encore that it powers most of Texas. I'm part of an electric co op which was part of the new deal, um, Collin Grayson County Electric Co Op. And when I called them and asked them about rates, hey, I'm, um, thinking about putting in a Tesla. They said, well, you know, many people who call us and ask us when they're putting in the Tesla say that take advantage of our free overnights and weekends plan. And I'm like, excuse me. And they're like, yeah. So what you can do is you pay, you're going to pay more per kilowatt hour during the day. During peak time. It was like, let's call it 1.5 cents. I don't. I forget the numbers.

Speaker B: Sure.

Speaker A: Versus they're going to pay 1 point cents per kilowatt hour if you're, if you're a. But you're going to get free nights and weekends. Well, I can plug, I can plug the car in and say don't start charging until 10:15 at night, which is going to be free. So not only did I start plugging the car in and not. And have not been charged for it, I also turned down the air conditioning at night because I was. So my electric usage actually doubled. Like the, the kilowatt. The number of kilowatts that I consumed doubled. But my bill remained the same.

Speaker B: Same because you were doing it when it was off peak.

Speaker A: Off peak?

Speaker B: Yeah, off peak and on peak have been around forever as well. And they're completely reasonable and people understand it. Um, and I think people are really, uh, love speed. I mean we're, we always pay a premium for speed.

Speaker A: And that's, and that's the, the main point is, uh, just, you know, me doing this, this project and people will pay a premium for speed. What's wrong? Why not charge for that? Why not say, hey, this is going to take longer. All right, well we got to get out of here. Got to our. We're up against our Break One Reminder listeners and contact either one of us by sending one email and that is to ask tsoe@thresholdnow.com the website is the soul of enterprise, but also we'd love for you to check out our patreon channel patreon.com TSOE layer. You can listen to the show commercial free and also our bonus episodes that we do on most weeks. But right now a word from our sponsor including Cal CPA

Speaker D: In a world moving faster than ever, who helps make sense of what comes next? At, uh Cal cpa, we bring together the brightest minds in accounting and finance professionals shaping how businesses grow, adapt, and lead with integrity. From advocating for sound policy to advancing financial literacy, CALCPA is more than an association. We're a voice for trust, clarity, and the future of the profession. Discover how we're making sense of what comes next@calcpa.org the best firms don't just

Speaker A: advise on strategy, they help clients build better operations. Rippling is built for exactly that one unified platform for HR, payroll finance, and it's that any business can run on. And Rippling AI cuts administrative overhead by

Speaker C: taking action across payroll, HR, finance and

Speaker A: IT in real time on your actual data. Visit rippling.com for more tax season and

Speaker B: an accountant working from home can't get into their software your IT guy isn't picking up. Sound familiar?

Speaker A: Summit was built for exactly this a secure, fully managed cloud workspace that hosts quick tax software and any desktop application

Speaker B: your firm depends on, accessible from anywhere with the security and compliance your clients expect. And when something needs attention, our team picks up fast. 15 minute response 24, 7 with experts who actually know your software.

Speaker A: One predictable monthly price.

Speaker C: No surprises.

Speaker A: Find out what your firm has been missing@summithq.com if someone tells you that you

Speaker C: need to productize your services, you're turning your wisdom into widgets. But that's not progress, it's a U turn. At Threshold, we help professionals escape the factory floor of formulas and checklists and enter the domain of the transformation economy. You weren't born to sell products and services, you were called to guide transformations. If you're being told to productize, call us instead, because your highest value isn't an SKU, it's a shift. Visit thresholdnow.com to learn more.

Speaker E: Uh, you are tuned into the Soul of Enterprise with Ron Baker and Ed Class. To find out more about our show, visit us on the web at the soul of enterprise.com. you can also chat with us on Twitter using AskTSOE. Now back to the Soul of Enterprise.

Speaker B: Welcome back everybody. We're talking about AI Credits ABC See in A hoodie. And I think the other piece of this puzzle, too, is looking at the portfolio. You know, the actuarial, uh, axiom that there are no such thing as bad risks, there's only bad premiums. And when I hear stories about, well, you know, the contractor can't give you a price because the building's uncertain. It's not even a pricing issue to me. It just. It's a expertise issue. I got the wrong contractor. He's never done this before. Screw it. I'm not going to let him build my house based on the billable hour. That's not a lazy issue. That's an expertise issue, at least from the customer's vantage point. And if some of this is really, really risky, if there's customers out there that are really, you know, uh, eating into the costs, then great, have a tier for them. Like the black card.

Speaker A: Yep.

Speaker B: You know, and, hey, do whatever the hell you want, but you're going to pay a fortune. You know, it is what it is. And, um, I. I don't know, maybe that's. Maybe that's not doable, but my question is, why not? I mean, the concierge doctors figured it out.

Speaker A: Yeah. And we use another piece of software, uh, oftentimes here at the Soul of Enterprise, Camtasia, which is what I do any of the video editing on. And, and one of their got, uh, an email from them last week where they said, hey, listen, we're not charging you for this AI it's unlimited. If you do stuff with AI, it's included in your subscription. So there are people who are figuring this out. So that's the good news, uh, because

Speaker B: when you really start thinking about the portfolio approach, the portfolio takes care of a lot of outliers and people using different levels at different. Because not everybody's going to be a heavy user all the time. I mean, maybe, but, you know, the portfolio is going to work that out, and the map works. I mean, Hollywood figured this out. Venture capitalists figured it out.

Speaker A: Um, why can't you just make the blockbusters, Ron?

Speaker B: Yeah, let's just do the blockbusters. Um, the other thing, Ed, was there's a great article in the. The Dispatch, um, and it's called the AI Race isn't About Electricity. And it's by Michael Hochberg from June, I think, 19th. Um, and who is Hochber, Ed? Ah, do you know him?

Speaker C: He is.

Speaker A: I do not know.

Speaker B: He's a visiting scholar at the center for Geopolitics at Cambridge University, a California Institute of Technology, trained physicist And a serial semiconductor entrepreneur. This is incredibly intelligent article. Um, love Greg's take on this, but he says electricity is a real near term operational bottleneck. The electricity problem is going to be solved in the chips. And then he says the real constraints are not electricity, they're cash and silicon. And he gives this as an example. He says take Nvidia chips. Um, a gigawatt worth of Nvidia GPUs cost roughly $50 billion to deploy. That's amazing. He said of this, the GPUs and the network that connects them are about 35 billion. That's about 70%. The rest is infrastructure, cooling, electrical conversion, which is not the same thing as electrical generation. Uh, and the rest of the physical plant electrical generation is about two and a half billion of the 50. About two and a half percent. Okay, so it's not electricity now. He said specialized silicon is about to change the shape of the industry. The cost of being Flexible is spending 100 to a thousand times the power and silicon area to do the same computation. Flexible meaning the GPUs. Right. Because they're general. He's arguing for. What is it the, I don't know the acronym here. Ascii. As I C. The um. Asic, yeah, asic, sorry, asic. Um, which is the um. I've got it somewhere here.

Speaker A: Application specific integrated circuit. Yeah.

Speaker B: They're more specialized.

Speaker A: Correct.

Speaker B: And he's saying that will cost will fall, efficiency will rise and capacity will expand. And he said this is exactly what happened with semiconductors storage and bandwidth. And he said, you know, the learning curve is undefeated. So these companies should be planning on future costs going down. And as we know with one of the, one of the problems with cost accounting, it's about past cost. What matters to a professional pricer is planned cost or future cost. This is why when you know the straight of Hormuz closes the, the gas station instantly raises the price and people get all pissed off. But you didn't pay the call, you know, but it's like yeah, but I have to replace what's in there. Mhm. It's future costs that matter, not backwards cost and future costs, at least according to this guy, are going to come down in this world. So they should be building that into their models. And I think with portfolio they can do that.

Speaker A: Yeah, I think people missed that the GPUs were intended as graphics processors. That's what the G stands for, Graphic processing units. And it just so happened that they were good for AI as well. And when the application hit. So now yeah, the revision is, hey, let's redesign the chips to take advantage specifically of the AI and pull the stuff out that we don't need because of the graphic side of things. I mean, this is why. I don't know if you remember, Maybe it was 10 years ago that it was illegal to, to send over, like, the latest version of the Xbox or Nintendo to, to China. Um, and it was not because they were going to, you know, use it to, to, to play games on it. It was because they were going to take the chip and they were going to repurpose it to do AI.

Speaker B: Right, right. And, and, and Nvidia was a game company, right?

Speaker A: Yes. They made, yeah. And they made graphics boards and graphics processing units. It was just to make the games faster.

Speaker B: Right. So I thought that was a really interesting, uh, point that he made in that article. Um, so I don't know, is this the future of this type of pricing? I. I don't know. It's, it's certainly. I think we're going to have to live through it for a while. But you know what? They're fueling uncertainty and they're creating a really crappy customer experience. All these headlines about Uber, there's been others, too. J.P. morgan Chase, I think, think it ran through their budget and they figured out some employees are using, you know, $10,000 a month of tokens. There's that hysterical video that Greg sent us about, you know, $300,000. What's this?

Speaker E: Yeah.

Speaker B: Uh, so I don't know. I, I think, I think these AI companies should create a better customer experience and capture customers first and optimize the economics later. It's not like they're not sitting on a ton of money. And Anthropic and all these other companies that are going to go public pretty soon, um, you know, they, I mean, Uber. Look how long it took Uber to turn the profit. Amazon drug companies.

Speaker A: Well, let me just. I just. And I. We should probably should have clarified this earlier on, but there's two levels here. It's there, yes. There's the AI companies. Anthropic, uh, OpenAI and their pricing. But the pricing that I think a lot of people are talking about in this space is the pricing of the people who are using AI in their tool. Right. So, uh, say QuickBooks puts builds AI into their. QuickBooks into QuickBooks. So they're pricing that AI differently than they're pricing their actual product. And that's part of the issue.

Speaker B: Right? That's a good point. Yep. Um, so I, you know, I just I just think there's better ways to, uh, to deal with this. And I do think we're kind of going to look back on this time as why. Why do we have to go through that? I mean, the strides that we've made in pricing, psychology. We know price isn't about numbers and math. It's about psychology. Why are we going down this road? I just think they're, they're in the cul de sac. Yeah.

Speaker A: And despite what you say, I still think that there's the finance people involved and there's people with spreadsheets going. We're going to manage these costs.

Speaker B: I know.

Speaker A: So it's not just the engineers. I'm not going to spare, uh, the spreadsheet, uh, jockeys, the spreadsheet bandits.

Speaker B: Yeah, I agree. Yeah, I agree. GPT pushed back on me and it did make a good point. Said this is engineers that are doing this for the most part. And I thought, well, okay, engineers invented cost accounting. They share some of the blame. But, yeah, uh, it's the bean counters that think they have to make a profit on every prompt. And man, the world just doesn't work that way. It just doesn't. And that, See, this is why theory is so important. You get the theory wrong, everything else doesn't matter. No matter how optimized everything looks.

Speaker A: Yep. Absolutely. Absolutely. So, all right, Ron. Well, next week is. Yeah. Uh, next week is July 3rd. It is also the start of our 13th, let's call it season 13th year of the Soul of Enterprise. So that, that's, that's exciting news and we plan to talk a little bit more about America at 250, our semi quincentennial.

Speaker B: Excellent. All right, I'll see you in 167 hours. Foreign

Speaker C: this has been Thresholds the Soul

Speaker A: of Enterprise Business in the Transformation Economy. Charting the new Economy where human flourishing is the bottom line. Join us next week at, on Friday at 3pm Eastern. That's noon Pacific. In the meantime, feel free to Visit us at www.thesoulofenterprise.com.

Speaker B: sam.