Archive | Upgrade to Premium | Sponsor

Hi {{FIRST_NAME|readers}},

Quick ask before we get into it this week.

The internet is being flooded with AI-generated slop content faster than most people realize. Procurement analysis is no exception:

We spend a lot of time making sure our content isn't that.

Practitioner-written articles. Buyer-protective analysis. Actual opinions, formed by actual humans, sourced from actual projects and conversations with actual procurement leaders.

If that's why you're here, can you take 15 seconds to list us as a trusted source in your Google preferences? It’s literally 1 check box.

It's a small signal, and it helps our work show up ahead of the slop in search results and Discover feeds.

Thank you,

Now onto the good stuff.

Onwards!

📰 In this week’s edition:

Note: Some of the content listed above is only available in the email version of this newsletter. Don’t miss out! Sign up for free to get the next edition.

Three years ago, cloud storage looked cheap…

Then the invoices got weird. Extra fees for moving your own data. Charges for regions, transfers, and services that were never in the sales pitch. Analysts estimate these hidden costs quietly add 10 to 15% to what enterprises spend on the cloud.

That wasn't a pricing accident. It was the plan.

We're watching the sequel now. Same director. New actors. Bigger numbers.

If you're leading procurement and you've been letting your IT team handle AI vendor conversations because "it's a tech thing," this is your warning shot. In the next few months, AI cost governance is going to land squarely on your desk... Your IT category manager is going to be working overtime. And the tools you'd normally use to manage this kind of spend don't exist yet.

Let's walk through why.

Before we go further, three terms worth pinning down. If you already know them, skip ahead. If you don't, they're the vocabulary you'll need for every AI vendor conversation from here on out.

Got it? Good. Now here's why the difference matters.

Two very different companies. Two very different situations. Same story. Same quarter.

That is not bad luck. That is the pricing model working exactly as designed.

Every major AI provider is running the same commercial strategy. Nothing about it is surprising if you've watched previous technology cycles. It goes like this:

Step 1: Subsidize adoption. Offer flat, predictable, seat-based pricing so companies can say yes fast. Low friction. Easy to budget. Easy to pitch to a CFO who wants a fixed line item.

Step 2: Become essential. Once the tool is woven into daily workflows (drafting emails, coding, doing research, answering customer questions), removing it becomes organizationally painful. Adoption quietly becomes dependency.

Even better for the vendor if customers have reduced their workforce because of these new capabilities. That changes the math entirely. Once the roles are gone, the org chart is redrawn, and the institutional memory has walked out the door, the vendor can raise the price to just under the customer's opportunity cost and still stay in place. Meaning: right below the cost of rehiring all the people who got let go, retraining them, and rebuilding the process the AI replaced.

That's not a ceiling most CFOs want to test.

Step 3: Absorb the real cost, short term. Running these models is expensive. (Have you seen frontier AI capex spending the last few years? Hundreds of billions of dollars, funneled into data centers, chips, and power infrastructure, with more announced every quarter.) Every query costs the provider real money in electricity, hardware, and infrastructure. But providers are willing to take losses today to win the adoption race, because they're playing for the position they'll hold when the subsidies come off.

Step 4: Shift to usage-based pricing once switching costs are high. This is where we are now. This is where the surprise invoices start showing up in enterprise finance departments, and where the stories about companies "accidentally" spending fortunes on AI in a single month come from.

There's nothing sinister about any of this. Someone has to pay for the compute eventually. "Eventually" happens to be arriving now.

But it means something specific for anyone with a procurement lens: the era of cheap, predictable frontier AI is ending.

Not gone. Ending. If your organization is comfortable using smaller AI providers further down the market (there are dozens of them, and they're competitive on price for many use cases), cheap access might stick around for a while. But if you've standardized on one of the big-name frontier models, the pricing conversation is about to get harder.

Here's the discipline forming in real time that doesn't have a name in most org charts yet: LLM token budgeting and governance.

(And they said AI was going to obliterate jobs 😅)

Let's break down what that actually involves, because it isn't obvious yet.

On the procurement side, you're going to be responsible for asking questions that don't have standard vendor answers today:

On the IT side, your colleagues will need to build the operational controls:

Neither side has the tooling for any of this yet. Most enterprise software vendors, including the ones selling into procurement, are only just beginning to build the guardrails buyers need.

If your organization has standardized on any AI tool with usage-based billing tucked underneath a friendly flat-rate sticker, this is the moment to ask exactly what happens the month your teams actually use it the way the demo suggested they should.

Now let's turn the camera on the software vendors selling into procurement.

Every provider on the market right now is racing to ship some version of an "AI Workflow Studio." Different names, same idea: a place where customers can build their own agents, automate their own workflows, and configure their own LLM-powered copilots.

Most of this functionality is still in the honeymoon phase. There's no explicit line-item cost yet. Vendors are giving it away as part of the base subscription because they need to prove the value before they can charge for it.

But here's the trap sitting under every one of those studios.

Agentic workflows (meaning AI that acts autonomously across multiple steps) don't sip tokens. They gulp them. A single chat query might consume a few thousand tokens. A multi-step agent that reads a contract, checks it against policy, drafts an approval, and routes it to three stakeholders can burn through hundreds of thousands. Every step, every retrieved document, every tool call gets metered.

That cost is currently being absorbed by the ProcureTech vendor via their API bill from OpenAI, Anthropic, or Google. But no software company can absorb that indefinitely. The economics don't work.

Eventually, it gets passed through to the customer. It always does.

We haven't seen a single S2P platform offer transparent token budgeting or governance functionality yet. Pricing pages are silent on it. Contract language sidesteps it. That silence has a shelf life, and the shelf is shorter than most people think.

You’re about to start seeing this error message pop up in a number of your enterprise applications:

That's the future this section is about. Here's how we think we get there.

Three developments we expect to see, in roughly this order:

None of this exists in a mature form in any ProcureTech platform yet. Which is exactly why it's an opening.

The first ProcureTech vendors to ship real token budgeting and governance (a working admin control, not a marketing slide) gets to walk into renewal conversations with something none of their competitors have.

Not "we support AI." That dog has been shipped to the barn out in the country already...😅 It has become the least useful sentence on any website.

"We help you control what AI actually costs you."

That is the sentence procurement buyers will remember when the invoice from somewhere else in the tech stack lands on their desk with a number they didn't budget for.

Here's the buyer-protective takeaway, and it's worth repeating because we don't think it's landed for most procurement teams yet.

The golden age of cheap frontier AI was never permanent. It was a customer acquisition strategy wearing a "we're so excited to democratize AI" costume. Every playbook in tech history says the same thing about this phase: it ends, and it ends when the vendor decides you can't leave.

We aren't saying don't use these tools. We use them. Most of your competitors use them. Refusing to engage is not a strategy.

We're saying: your business is going to need procurement professionals to act like it for this new spend category in short order. Ask the questions your CFO will wish you had asked. Build a governance model now, while the numbers are still small enough to be embarrassing rather than career-ending.

Because the bill was always going to come.

The only question left is who's ready when it does.