AI Pricing Is Changing the Software Deal. Are Enterprises Ready?

Tony:
Hello, I’m Tony Mangino from TC2, and this is Staying Connected—where we talk about what really matters to enterprise customers navigating today’s technology and sourcing decisions.

Today I want to talk about something that is becoming a major sourcing issue, and that’s how enterprises actually pay for AI.

I’m not talking about GPUs or data centers. I’m talking about the commercial model inside the software agreements companies are signing every day.

For years, enterprise SaaS was relatively familiar. You had users, you had seats, and you negotiated a price per seat. AI is starting to break that model.

A MarketWatch article recently looked at this issue and described an evolving market that’s experimenting with traditional seats, prepaid AI credits, usage-based charges, per-conversation pricing and even outcome-based models.

I’m joined today by my colleague Julie Garder to try and unpack this.

So, Julie, the first question for a customer is becoming surprisingly basic: What exactly am I buying?

Guest:
Right. Is it a token? A credit? A conversation? An agent action? A successful resolution?

And once you understand that, you get to the harder question: How much of whatever that unit is are we actually going to consume?

Tony:
Which is difficult when a lot of enterprises are still figuring out where AI really fits in their operations.

The Meter Is Changing

Guest:
Exactly. The seat model has plenty of problems. We’ve all seen shelfware. But at least everyone understands the unit.

If you have ten thousand users at a negotiated price, finance can build a reasonably predictable budget around that.

AI is different because two people using the same software product may generate completely different levels of underlying consumption. One person may use the AI capability occasionally. Another group may have agents running workflows all day.

So suppliers are trying to align price more closely with actual use.

Tony:
Which makes sense from their perspective.

Guest:
It does. But it creates a different kind of risk for the customer.

Business Insider recently reported on legal AI provider Legora moving new customers toward usage-based pricing because the actual AI consumption can vary significantly from one user to another. And that’s part of a broader market experiment. Some suppliers want pure consumption. Some want a subscription plus consumption. Others are looking at outcome-based models.

There really isn’t one settled commercial standard yet.

This Is Really About Risk

Tony:
And that’s what makes this more than a pricing discussion. It’s really a conversation about who carries the risk.

Guest:
Absolutely. With a seat model, the enterprise carries the risk that it buys too many licenses. With consumption, it carries more risk that usage exceeds forecast. With outcome pricing, the risk moves into how the outcome is defined. And with a hybrid model, you can end up with both a fixed commitment and variable exposure.

So when a supplier tells you, “Our AI pricing is usage-based,” you still don’t know enough to judge the deal. You need to understand the meter, how quickly consumption can scale and where your financial exposure stops.

Tony:
AWS introduced an interesting capability in Marketplace this month called AI Insights. Among other things, it helps customers understand a seller’s pricing unit, how charges scale with usage and how multiple pricing dimensions interact.

I looked at that and thought, if we need an AI tool to explain how the AI is priced, we’ve probably proven the point.

Guest:
Exactly. Greater transparency is useful. But understanding the formula and knowing whether you have a good commercial deal are still two different things.

How Do You Commit to Something You Haven’t Adopted Yet?

Tony:
Here’s the question I think a lot of sourcing and finance teams are going to struggle with: How do you make a three-year commitment when you don’t know what your usage is going to look like twelve months from now?

Guest:
Very carefully.

Early in an adoption cycle, vendors tend to have optimistic assumptions about how quickly usage will grow. Maybe they’re right. But an ambitious adoption plan is not the same thing as a defensible contractual commitment.

If the enterprise commits based on that optimistic scenario and adoption is slower, you’ve essentially recreated shelfware with a different unit. Instead of unused seats, maybe you have unused credits or committed spend tied to workflows that never get deployed.

But there’s an opposite risk as well. You negotiate a very attractive entry point, adoption takes off and suddenly consumption blows through the budget.

Tony:
So you really have to model both directions.

Guest:
Exactly. I’d want to understand what happens if adoption is slow, what happens if it lands where we expect, and what happens if the technology is wildly successful.

That last scenario matters because commercially, you don’t want success to create a cost problem. If the solution starts delivering real value and the business wants to expand it, the economics should get better at scale—not become progressively more punitive.

Definitions Can Be More Important Than Rates

Tony:
Outcome pricing is particularly interesting because on the surface it sounds almost ideal. Don’t charge me for the software. Charge me when it actually accomplishes something.

Guest:
Conceptually, it’s appealing. MarketWatch discussed examples where AI agents might be priced around successful customer-service resolutions or other measurable activities.

But the minute you do that, the definition becomes the economics.

What counts as a successful resolution? If the AI handles ninety percent of an interaction and a person completes the last step, is that chargeable? If the customer comes back ten minutes later with the same issue, is that a second outcome? If the supplier changes the underlying model, does the measurement methodology change too?

Tony:
So something that looks like an operational definition can end up determining millions of dollars of spend.

Guest:
Exactly. Which is why procurement can’t negotiate the rate in isolation.

The business owner has to understand what is being measured. IT has to understand how the technology behaves. Finance has to understand the exposure. If everyone is negotiating around a different understanding of the meter, the stated unit price doesn’t tell you very much.

Focus on the Commitment Before the Discount

Tony:
Julie, If you were negotiating one of these deals today, where would you spend your time first?

Guest:
I’d spend a lot of time on commitment flexibility, especially if the enterprise hasn’t deployed the technology broadly yet.

Can the commitment ramp over time rather than start at full volume? Can unused credits roll forward? Can the customer move commitment between products or use cases? What happens if the supplier materially changes the product or the measurement model?

Those protections can be more valuable than squeezing another percentage point out of the initial unit rate.

Then I’d look at visibility. The enterprise needs timely consumption data, useful thresholds and alerts. You don’t want the first indication that adoption has accelerated to be the invoice.

Tony:
And you still negotiate the unit economics.

Guest:
Of course. But I’d pay just as much attention to how those economics behave as volume increases.

If consumption doubles or triples, does the unit rate improve? What does overage cost? Are there caps or other protections? You don’t want a structure where the supplier encourages adoption and then captures essentially all of the economic upside once usage takes off.

Be Careful When AI Shows Up Inside a Renewal

Tony:
Here’s another situation I think we’re going to see constantly. The enterprise has a major SaaS renewal coming up, and the supplier says, “We’ve added all these AI capabilities. Here’s the new package.”

What do you do?

Guest:
As much as possible, separate the decisions.

Start with what you’re paying for the existing platform. Then identify the incremental AI capability and the incremental cost. Does everybody need it? How much consumption is actually included? What happens if the enterprise only deploys it to part of the population?

Bundling can make year-one pricing look very attractive while making it difficult to understand the long-term economics.

Tony:
And once the AI capability becomes embedded in the workflow, taking it back out could become difficult.

Guest:
Exactly. That’s why you want flexibility before the dependency develops.

The supplier may be happy to encourage broad adoption today because broad adoption strengthens its position at the next renewal. The enterprise has to think one step further ahead.

Benchmark the Use Case, Not Just the Unit

Tony:
This creates an interesting benchmarking problem too. How do you compare two offers when one supplier uses credits, another charges per interaction and a third prices based on outcomes?

Guest:
You normalize them around the same enterprise use case.

Instead of asking which token or credit is cheaper, ask what the actual workflow costs under each model. What does this customer-service population cost? What does this agent deployment cost? What happens at low, expected and high consumption?

Once you do that, you can start comparing the economics on a common basis.

The unit rate still matters. But increasingly, the construct around that unit may matter even more.

Closing

Tony:
And I think that’s the takeaway.

AI isn’t just changing what software does. It’s changing what enterprise customers are being asked to buy.

Seats were imperfect, but everybody understood them. Tokens, credits, conversations, agent actions and outcomes introduce a different kind of commercial risk.

So, before you negotiate the discount, understand the meter. Before you make the commitment, model the consumption and understand how the economics behave if adoption turns out very differently from what everyone expects today.

To our listeners, if you would like to discuss AI or software pricing, or if you’d like to discuss other technology strategy, sourcing and cost reduction needs with Julie, me, or any of our TC2 and LB3 colleagues, please give us a call or shoot us an email.

You can also stay current by subscribing to Staying Connected, by checking out our website, and by following us on LinkedIn.