Stale Pretzels and Opaque Pricing: The Case for a Nutrition Label on Your AI Stack

There's a specific kind of despair that hits somewhere around gate C14, when you're staring at a touchscreen kiosk that wants $18.75 for a pretzel that's been under a heat lamp since the red-eye landed. You pay it, because what's the alternative? You're a captive audience, the menu doesn't list ingredients, and there's a strict no-refunds policy.

Now swap the airport for a procurement office and the pretzel for an enterprise AI seat. The transaction feels eerily similar — and a growing number of CFOs have had enough.

The Captive Audience Problem

Talk to procurement leads at mid-size US companies and a pattern emerges fast. The average enterprise now runs somewhere between 100 and 300 SaaS subscriptions, and nobody can fully explain what half of them do. The AI tools are the worst offenders, because the pricing pages read like they were written by lawyers negotiating with themselves.

Seat-based pricing that shifts to usage-based the moment you scale. "Contact sales" buttons where a number should be. Overage rates buried in a PDF addendum that legal signs at 11pm before a quarterly close. One procurement director at a logistics firm in Ohio told me his team discovered a workflow automation platform had quietly tripled per-task pricing — six months after they'd committed. The contract allowed it. He didn't.

"It's the airport thing," he said. "You're already onboarded. Your data's in there. Migration costs more than the increase. So you pay the $19 pretzel and you keep walking."

Why 2026 Changes the Math

Here's what's different this budget season: the AI line item is no longer a rounding error. What started as a $200/month experiment in someone's marketing department has compounded into line items that make CFOs do a double-take during planning cycles.

Several finance leaders I spoke with are treating Q4 as an audit window. Not a security audit — a value audit. The questions they're asking are refreshingly blunt:

• Who owns this tool, and do they still work here?
• What's the actual cost per active user, not per assigned seat?
• If this vanished tomorrow, what breaks — and how fast could we rebuild?
• Are we paying for AI features that a spreadsheet already handles?

That last one stings. Multiple CFOs admitted that during their first real AI spend review, they found overlapping tools doing near-identical jobs — three different summarization products, two dashboarding layers, one abandoned agent platform still billing annually.

What a SaaS Nutrition Label Would Actually Say

The airport pretzel at least has a regulatory framework behind it. Your AI dashboard has a landing page.

So imagine if vendors had to publish a standardized label. Not marketing copy — facts. Something like:

Serving size: 1 seat, billed annually

Active ingredients: LLM inference (third-party model, provider rotates without notice), vector database, integrations layer

Filler: Admin seats you didn't ask for, a mobile app nobody uses, "AI-powered insights" that are a SQL query with better margins

Allergen warnings: Data used for model training (yes/no, and whose data), subprocessors list, retention window

Refund policy: Printed in bold, because currently it's a shrug

The comparison to food labeling isn't just cute — it's structural. Nutrition labels work because they standardize disclosure across an entire industry, which turns shopping from a trust exercise into a comparison exercise. Right now, comparing two AI platforms requires a week of discovery calls and a lawyer on retainer.

Vendors Aren't Villains, Exactly

To be fair to the software companies: some opacity is defensible. Usage-based pricing genuinely is harder to quote upfront when a customer's usage is unpredictable. And the fastest-moving AI startups reprice constantly because underlying model costs are repricing constantly.

But there's a difference between dynamic pricing and a shell game. The vendors getting renewed — the ones procurement leads actually praise — share a few traits: published pricing tiers, a real self-serve option, overage alerts before the invoice rather than after, and export tools that don't require a support ticket to leave.

Transparency, it turns out, is a retention strategy. The companies that make it easy to calculate your true cost are the ones that survive the audit.

What Procurement Teams Can Do Before the Fiscal Year Closes

You don't need Congress to mandate a label. You can build your own, and the teams doing this well follow a similar playbook:

Run the renewals calendar backward. Every auto-renewal in the next two quarters gets a value review 60 days out, not 5. Auto-renewal clauses are the heat lamp keeping your stale pretzel warm.

Track consumption, not spend. Dollars tell you what you paid. Usage tells you what you got. The gap between those two numbers is where the audit lives.

Demand the subprocessor list in writing. If a vendor can't tell you which AI models touch your data, that's an ingredient list you shouldn't eat from.

Price the exit before the entrance. Know what data export costs and how long migration takes before you sign. Leverage is cheapest on day one.

The Bottom Line

Nobody expects enterprise software to become a vending machine. But the direction of travel matters. CFOs entering 2026 planning are no longer asking whether AI tools are worth buying — they're asking whether each one is worth renewing, and they want receipts.

The vendors that figure this out first will eat the market share. The rest are just kiosks at gate C14, waiting for travelers with no better option.

And honestly? Even the pretzel posts its calories.