Transcript 0:00 Earlier this year, Uber blew through its entire LLM-supported coding budget by April. Microsoft pulled Claude Code licenses from thousands of their engineers once they realized the real AI usage costs. 0:13 Some of their engineers were running up to two thousand dollar LLM bills every month. 0:18 Now, if two of the most sophisticated tech operations on the planet are caught by the exact same thing, the fact that nobody set a large language model usage cap, you're about to get into trouble as well. 0:31 Here's why that happened. Most software you buy comes with a seat license or a flat fee per person per month, no matter how much you use it. 0:41 But when you're calling an AI large language model from another application, you're not paying for a seat anymore. You're paying for a smaller amount every time you call the LLM. It's like a metered utility bill. 0:54 There's no ceiling unless somebody builds one into your operations. And if you think that that's someone else's problem because it's a tech issue, I'd get comfortable being wrong. 1:04 In the next eighteen months, I think AI cost governance is going to land squarely on procurement's desk, and your IT category managers are gonna start working overtime because right now, the tools that you'd normally use to manage this kind of spend don't exist. 1:21 This week, the pricing playbook every major frontier large language model vendor is running and the questions you need to be asking before it runs into you. 1:32 [upbeat music] Welcome to ProcureTech Unpacked. I'm Joel Conney-Demers. 1:46 This is the show for procurement professionals who wanna understand the technology shaping our function. 1:52 Here's the pattern, and if you've been doing this job for more than one technology cycle, you'll recognize it immediately. A couple years ago, cloud storage looked cheap. Then the invoices started getting strange. 2:05 Fees for moving your own data, different charges by regions, and costs for data transfers that were never in the sales pitch. 2:13 A recent Backblaze survey of IT decision-makers managing large-scale cloud infrastructure and cloud storage in enterprise found that ninety-five percent of them had been hit with unexpected charges that they didn't see coming. 2:26 That wasn't a pricing accident. It was the plan, and we're watching the sequel right now. Same director, new actors, much bigger numbers. 2:36 This time, the product is large language models like Claude, ChatGPT, and Gemini, and the invoice approvals are already starting to get uncomfortable based on the examples I just shared with you. 2:47 One quick term, since you'll hear it in every AI vendor conversation from here on out. The term is a token. A token is the unit that an LLM provider bills you on when they're not using a seat. 2:59 It's roughly three-quarters of an English word counted on the way into the LLM as you're asking your questions and prompting the LLM and on the way out when you get a response. 3:10 That's the meter running behind everything you're about to hear. 3:14 So you've already heard how this hit Uber and Microsoft in the opening of the show, the two of the most sophisticated tech operations on the planet, same quarter, same mistake, nobody had set an AI usage cap. 3:27 That's not bad luck. That's the pricing model working exactly as designed. Now, sure, these are coders. These are tech companies. They're burning tokens on coding tools. 3:38 It's easy to file that under, "It's not in my world." But the same dynamic is coming for your enterprise platforms the minute that they start leaning on AI LLMs under the hood. 3:49 So if you start using AI agents in your source-to-pay suites or your intake and orchestration platform, they're selling into procurement and racing to ship an AI workflow studio, for example. 4:02 Well, the pricing issues that we're watching unfold with Uber and Microsoft is exactly the issues that they're gonna be having, and they're gonna be passing on to you, uh, as enterprise software vendors, and it's on a timeline that's much, much shorter than you think. 4:18 So here's the playbook so you're well aware. It isn't new. We've watched this movie before, just with different vendor logos on the screen. Step one, subsidize adoption. 4:28 Flat, predictable, seat-based pricing so that the yeses can come faster and the CFO gets a clean item on the, uh, P&L. Step two, become essential. 4:38 Once the tool is woven into daily workflows, uh, in organizations, removing it becomes organizationally painful, so adoption quietly becomes dependency. Step three, absorb the real cost short term. 4:52 So running these frontier LLM models is notoriously super expensive, and providers are willing to eat that cost today because they're playing for the pricing power that they'll hold once they have your habits and once the subsidies come off. 5:07 Step four, shift to usage-based pricing once the switching costs are high enough that leaving isn't really an option for people using the services anymore. 5:18 So that's where Uber and Microsoft's invoices came from, and that's where we are right now. To me, the era of cheap, predictable frontier LLMs are ending. 5:28 Doesn't mean that it's gone, and if you're comfortable going further down the LLM market, there's dozens of competitive providers that can still compete on cost, 5:37 uh, and give you the same reasonable amount of functionality that an LLM is, is giving you today. So that's why I think bring your own LLM is gonna be a thing, by the way, but that's a conversation for another time. 5:49 So if you're in a standardized enterprise LLM today with one of the big names, Claude, Gemini, ChatGPT, and that's the strategy you've adopted, you're gonna have a pretty hard, uh, pricing conversation over the next eighteen to twenty-four months, I think. 6:05 So what does this actually mean for you as a procurement professional starting Monday? 6:10 Well, the next enterprise software contracts that your organization renews or signs, you need to be aware of how LLM usage is currently handled. If it's only seat-based, 6:23 maybe the providers, uh, hasn't realized this yet and is gonna have to come with you with a surprise later on. So figure out how, what their maturity is of understanding this. 6:32 Number two, how is it gonna be handled once the pricing starts to change to usage-based? Because it will. 6:38 And number three, what happens when a team burns through the allotment on your side of AI credits or AI usage within the platform? 6:47 Because on the back end, they're calling the LLM provider, they're gonna get charged, right? 6:51 And so they're gonna need to do something with those costs if you as a customer is using a lot more AI usage or tokens than, uh, than others. 7:01 And then you wanna figure out if, if you can negotiate a hard cap on, uh, into the contract itself, or whether there's different things in the applications that can be done to, to limit usage or distribute usage or budget for usage, uh, so that, you know, you don't have a rogue automation script that doesn't, uh, that, that becomes a half-million-dollar mistake. 7:22 So if you're evaluating procure tech vendors and impressed with the new AI workflow studio functionality, you should be. It's very powerful. 7:29 However, you should know that agentic workflows don't sip tokens, they gulp them. 7:34 And a single multi-step agent reading a contract, checking agai- checking it against policy and routing it through approvals can burn hundreds of thousands of tokens in one run, and that cost is currently sitting on the vendor's own API bill, 7:49 and I don't think it's gonna stay there 'cause the, the, our providers, our enterprise software providers are also using the same playbook. 7:57 Get you to adopt it, get you to use it, get you to see that it's useful, get it in your organizational habits and, and procedures and processes, and then have the real pricing conversation with you. 8:08 Nobody, as far as I know, has shipped real token budgeting and governance functionality yet, at least not in the procure tech space. 8:15 I do know it's coming based on discussions with certain vendors, uh, but I haven't seen anything in terms of website pricing pages, press releases, or contract lang-language that explicitly addresses this, which means the sentence that you should be waiting for in your next renewal conversations or when you're signing the next software deal isn't, "Do you support AI? 8:37 We support AI, yes." Everybody does. The sentence you're looking for is, "We help you control what AI will actually cost you." So what's the takeaway here? 8:47 I think the golden age of cheap frontier AI was never gonna be permanent. It was a customer acquisition strategy. 8:54 Build your governance model now if you're a procurement professional while the numbers tied to these governance holes are still small enough to be embarrassing instead of career-ending. 9:03 I went deeper on where I think token budgeting shows up first inside procure tech platforms in last week's newsletter, and the link is gonna be in the show notes if you wanna dig deeper into this topic. 9:14 If you're not already getting the newsletter, you can fix that at pureprocurement.ca, and I'll be back next week to help you make smarter procurement technology decisions. See you then. 9:24 [upbeat music] That's a wrap on this episode of ProcureTech Unpacked. If this one resonated, subscribe wherever you get your podcasts and sign up for the Pure Procurement Newsletter for weekly insights between episodes. 9:36 If something we covered sparked a question or an idea, we'd love to hear from you. All the links for the reports we discussed are in the show notes. We'll see you next time. [upbeat music] Resonate.
