AI Is Delivering More Value Than Ever. So Why Does Nobody Want to Pay for It?

AI delivers labor-level outcomes but remains priced like software within pre-AI budget buckets designed for workflow automation, creating core tension where vendors over-deliver value to protect retention while buyers resist pricing increases. AI credit pricing expanded from 35 to 80 companies among top 500 software vendors in 2025, with inconsistent definitions making cross-vendor comparison impossible and CFO spend explanation difficult. Proving AI ROI requires measuring changed work units like invoices processed per FTE or tickets resolved per agent rather than vague "time saved" claims. CFOs should baseline workflows measuring throughput after AI adoption, demand billing transparency showing team-level usage and cost centers, and establish success metrics upfront before signing deals rather than scrambling for measurement post-contract.

I sat down with CJ Gustafson, CFO and creator of Mostly Metrics, and Kyle Poyar, creator and writer of Growth Unhinged, about the strange moment we're living through in AI pricing. 

The discussion surfaced something that's been nagging at me for months: AI is getting dramatically better, delivery is faster than ever, but pricing pressure from buyers keeps going down.

That's backwards. And it's creating real tension between buyers and sellers that finance leaders need to understand.

I kicked off the conversation by asking AI itself to describe its journey over the past year in finance and procurement. The response was, enlightening, for lack of a better word: 

"Over the past year, I've gone from being a clever parrot perched on your shoulder, good at repeating patterns and sounding confident, to something more like a mycelium network under a forest - connecting signals, sensing what's off, and moving the right nutrients to the right places. For finance and procurement, that means I'm less about flashy outputs and more about invisible leverage."

What even is a mycelium network? As overly complex as that description is, it does capture perfectly how AI has become so embedded, so invisible in how it delivers value, that we've lost track of what it's actually doing for us.

That invisibility is part of the problem.

The Picasso Problem: Why We've Stopped Valuing What AI Actually Does

CJ shared a story that perfectly captures the psychology at play here. Picasso is sitting in a café, sketching on a napkin. Someone recognizes him, asks to buy it. He finishes in a few minutes and quotes a million francs. The buyer freaks out: "A million? That took you five minutes." Picasso's response: "No, it took me a lifetime."

Any expert knows this problem all too well. And that's exactly what's happening with AI.

"We've packed basically all human knowledge into something that answers in a second, and now we're anchoring on the second," CJ explained. "Not the lifetime. Not the infrastructure. Not the fact that it's replacing real work."

Think about how quickly our expectations shifted. Remember the first time you asked ChatGPT to write a poem about something ridiculous? Jaws hit the floor. We couldn't believe what we were seeing.

Now we get frustrated when deep research on public company financials takes two seconds longer than expected.

I call this the "microwave effect." When something becomes fast and easy, we stop appreciating the value. My teenagers stand by the microwave complaining that their entire dinner is taking a minute and a half to heat up. The same phenomenon is commoditizing AI before we've even understood what it's worth.

Why AI Keeps Getting Undervalued: It's Stuck in the Wrong Budget Bucket

Here's what CJ identified as the core tension, and I think he's exactly right:

Most software pricing still sits inside pre-AI budget buckets. IT budget. Finance budget. Ops budget. G&A efficiency tools. Those budgets were designed to automate workflows. They were never meant to replace labor or decision-making.

"AI is delivering labor-like outcomes, but it's being priced like software and being compared to historical budget envelopes," CJ said. "That mismatch is the core tension. Until budgets move from tools to labor outcomes, pricing will stay under pressure."

This creates a bizarre dynamic. Vendors are delivering labor-level outcomes but selling into software budgets with annual uplift caps. So they over-deliver value to protect logos instead of resetting price.

And buyers aren't being irrational either. Procurement teams are trained to defend downside, not underwrite upside. I've never met a procurement professional who gets compensated on underwriting upside. They're the goalie. They get promoted for holding the line and showing they got a great deal, not for paying more because a tool is 3x better.

AI Features Are Becoming Table Stakes - Just Like Single Sign On (SSO) Did

CJ made a comparison that immediately resonated with me. Remember when companies charged extra for single sign-on? It seemed completely normal at the time. Now it's table stakes - bundled in, expected, not a premium feature.

"I think we're going the same direction with bundling in note taking with any app," CJ said. "I have five different apps that can do note taking for me now. That used to be something you'd buy on its own."

AI features are following the same trajectory. What vendors hoped would be expansion revenue is increasingly becoming baseline expectation. The Slack AI situation CJ mentioned is a perfect example: it launched with firm pricing, but over time it's become much more negotiable as Slack realized this is table stakes that customers just expect. 

AI Credit Pricing Is a Mess - And CFOs Are Going to Feel It First

Kyle has been tracking the top 500 software and AI companies. At the end of 2024, only 35 of those 500 had any sort of AI credit model. By 2025, that number jumped from 35 to about 80, and the trend is accelerating.

The problem? Credits look different for every single company. Some say a credit is a token. Some say it's an API call. Some say it's an action in the product where one action might be worth five or ten credits, another might be worth one.

"It's really hard to compare across vendors when all of them say they have credit pricing, but they use a different logic under the hood," Kyle explained.

Kyle made a memorable comparison: "My grandma grew up in the Depression. And if you ever left a light on when you exited a room, you would not hear the end of it. 'Who left the AI on' is going to be the new 'who left the light on.'"

From a CFO standpoint, this is a nightmare. If spend can move materially month to month, I need to see it in real time. Not at renewal. Not in a QBR deck. Now. The vendors that win here will be the ones who make invoicing boring, predictable, transparent, and explainable.

I'll be blunt: if I can't explain the bill, I can't defend the spend.

You Can't Prove AI ROI Without Measuring the Right Work Unit

Here's the uncomfortable truth I shared during our conversation: evaluating how well software is doing at the job we hired it to do already needed bolstering at most companies. We weren't very good at it before AI arrived. We just kept buying another tool, buying another tool - that's what causes SaaS sprawl.

AI makes this harder because it's running invisibly in so many applications. We may be interacting with it constantly and not even know. It's creating value that's genuinely hard for consumers to quantify.

When Kyle works with companies, he asks a simple question: "If we adopt this tool, what changes around your goals, your OKRs, your performance targets?"

If teams have a good answer, great. The tool is adding measurable value. If the answer is more like "this helps us stay a little more compliant" or "we're doing something we weren't doing before" - that's where things get murky.

CJ put it even more directly: "What work unit changed? That's how you have to measure it."

Is it invoices processed per FTE? Tickets resolved per agent? Deals reviewed per lawyer? Forecast iterations per analyst? Days to close? Errors per transaction?

If you can't name the unit of work, you can't measure improvement.

Outcome-Based AI Pricing Sounds Ideal - Here's Where It Gets Complicated

Kyle brought up Intercom's approach to AI pricing as an example worth studying. Instead of charging per seat, they charge 99¢ per resolution when AI successfully handles a customer support ticket. Their resolution rate has gone from 25% to 70%, essentially tripling their monetization as the product gets better.

For customers, there's clear ROI because the alternative - having humans resolve tickets - costs far more than a dollar per resolution.

"If you're paying nothing for tokens or credits and you're only charged when there's a tangible business outcome, you wouldn't need to worry about controlling usage," Kyle noted. "You could use it as much as you want. You're only paying when there's positive value for your company."

I like the concept. Outcome-based pricing aligns incentives. But I worry about the execution. If the vendor decides "outcome achieved" but the customer feels like the question wasn't actually answered, you've got a recipe for constant arguments. The measurement has to be crystal clear before the contract is signed, not scrambled together after the fact.

If AI Handles the Tedious Work, What Should Procurement Actually Be Doing?

Kyle made an observation that should get every procurement professional thinking about their career trajectory. If AI saves your team twenty hours a week of tedious work, what do you do with that time?

"Instead of just handling the back end of a deal, we can get much more involved proactively around helping identify where technology could improve teams' lives, where we have duplicate spend or opportunities to actually invest more, how we're measuring the accountability of the vendors against what they promised," Kyle said.

There's more proactive, strategic work procurement teams can do with saved time. But it requires a real change in behavior and scope. If you're not doing manual, tedious work anymore, what are you doing?

That question is uncomfortable. But it's also exciting because the answer can be much more impactful for the overall company.

The Warning for Finance Leaders and Software Buyers

CJ summed up where we're headed with a warning that should get every finance leader's attention:

"If we don't fix the value perception problem, AI gets priced like infrastructure before it's priced like impact. Buyers push prices down because they can't explain the value, and vendors respond by protecting retention instead of pushing outcomes."

The downside scenario isn't an AI winter. It's normalization, complacency, and underappreciation.

We complain about a 20-minute flight layover and forget we're sitting in a chair in the sky going 500 miles per hour.

Kyle captured what the next 12 months need to look like: "Last year was the year of AI experimentation. Everyone wanted to talk about the AI stuff their teams were doing. They found new budget to fund AI initiatives. They might have had five coding tools in use at their company. This next year is the more exciting year where it's going to be about AI ROI."

There's an interesting retention angle here too. When Gemini 3 came out and people started saying it was better than ChatGPT, Kyle made an observation that applies to all of us:

"ChatGPT has a history of years of conversations. If I asked ChatGPT what it knows about me, it would scare the hell out of me. The outputs are already highly personalized to what I'm looking for, and I have this shorthand back and forth as if it's been operating as my chief of staff. I'm not going to turn off models just because something might be a little bit better."

CJ agreed: "The switching costs - are they higher or lower now? Have you been training something on your company similar to an employee? I'm not sure if another AI will be able to ramp just as fast."

This has interesting implications for both buyers (lock-in risk) and vendors (retention moats built on context, not just features).

The Real Question: What Are You Doing With the Saved Time?

Here's where the conversation got personal. CJ raised something I've been wrestling with:

‍"What are you doing with that saved time? Are you making more sales with it? Can we prove that, or is it hand-wave? Is it like you don't have to work as hard so now you can take your dog for a walk for an extra hour today?"

There's something to be said for overburdened employees having better quality of life. But at some point, these investments need to pay dividends. They help both sides of the P&L - accelerating growth and improving efficiencies.

The problem is we don't always hold the business accountable to achieving those efficiencies. We talk about how this tool is going to save the marketing team time. But then when the marketing team asks for another headcount, we forget we hired the tool to avoid that in the first place.

Three Things CFOs Should Do Before Their Next AI Renewal

After this conversation, here's what I'm telling other CFOs:

1. Stop asking "is AI helping?" and start asking "what work unit changed?" Baseline your workflows. Measure throughput after AI - not feelings, but volume per person, cycle time, error rates. Then make an explicit decision about what happens to the time you freed up. Do you hire less? Redeploy? Increase output with the same team? If you skip that decision, the ROI stays hidden forever. You get a faster organization but not a cheaper one.
2. Demand billing transparency now, before your next renewal. If your vendor can't explain in plain language who used what, for what business process, at what cost, push back. Usage rolled up to teams, workflows, or cost centers with clear time periods and unit economics you can reason about. That's the minimum bar. I've had this problem myself. When I was doing multiyear discount analysis recently, I found myself using AI heavily, churning through iterations. And as CFO, I paused and asked: what is this costing me? Guess what—I had no clue. I'm not the admin of our instance, so there's no transparency to even know as you're working how much it's costing you.‍
3. Get your measurement in place before you sign deals - not after. The best AI deals happen when both sides agree upfront how success will be measured. Sellers anchor pricing to metrics the CFO already tracks. Buyers fund outcomes, not tools. When AI pricing works, it's because both sides agreed what "better" actually means before the contract was signed.