Pricing AI: A Practical Guide for a Hype-Filled World

This is the first of a series of blogs that may eventually become a book on Pricing AI. Why? Because AI is exploding into products everywhere, and nobody seems to know how to price it.

Let’s start with two truths:

• Buyers trade money for value.

• Companies exist to create value for customers.

These are the foundations of business and pricing. Whether you’re selling seats on a SaaS platform or tokens to an AI model, the fundamentals haven’t changed. Not even a little.

AI is new. The technology is evolving fast. The use cases are expanding. The hype is real. But pricing? Pricing is still about value.

It’s still about how much a buyer is willing to pay for a result they care about.

It’s still about choosing the right pricing metric.

It’s still about segmentation, versioning, and positioning.

It’s still about the buyer’s perception of value, not your cost or effort.

And yet AI creates confusion. It breaks some of our habits, especially in the SaaS industry. We’re used to charging per user. AI flips that model upside down.

It also creates pressure. Traditional SaaS has near-zero marginal cost. AI doesn’t. Every query, every prompt, every token has a real, measurable cost. That means your pricing model needs to scale with usage (and hopefully value) or you’re bleeding margin.

In the coming weeks, I’ll publish a new post each week unpacking a layer of this challenge. Eventually, I may combine them to create a book on Pricing AI. I’ll keep the tone practical, the structure clear, and the focus where it belongs: on delivering and capturing value.

In this first post, let’s elaborate on one of the biggest problems we face in pricing AI. In a future post, we’ll talk about how to address it.

What Are You Charging For?

In pricing terms, this is the pricing metric—and AI is forcing companies to rethink it. In traditional SaaS, the most common metric has been per user or per seat. But AI automates human work. That means we can serve more people with fewer users.

That’s a problem if your whole business is built around charging per seat.

Here’s a simple example. Let’s say you sell customer support software, and a client has 20 agents. They pay $50 per user, so $1,000/month. You add an AI-powered assistant to make their support agents more efficient. Suddenly, they only need 10 support agents to do the same job. Their work product is the same or better. But now, they only pay you for 10 users, which is $500/month. They’re getting more value. You’re making less money.

Unless you change the metric.

And we’re starting to see that shift in the wild.

Examples of AI Pricing in the Real World

1. OpenAI: A Weirdly High Top Tier

Sam Altman made headlines when OpenAI launched a $200/month plan for power users of ChatGPT. That was a huge jump from the existing $20/month tier, and both tiers are still available. Many called the price arbitrary.

But here’s the thing: it was probably intentional. Not in the sense that it was precisely calculated, but in the sense that it was meant to anchor perceptions of value. It’s the Which One problem: if you’re deciding between free, $20, and $200, the $20 feels like a deal and the $200 signals serious power.

There’s no “per user” logic here. This is versioning based on access, capacity, and maybe even ego. Some people just want the best. And at $200/month, the margin works even if usage is high.

2. OpenAI API: Classic Token-Based Pricing

When developers access GPT models via API, OpenAI charges per token—fragments of words that models read and generate.

This is the most common form of token-based pricing. It’s a direct usage model. You use more tokens, you pay more. Of course, this is a version of cost-plus pricing, so it’s not tied to the value buyers receive. There are advantages, though:

• It matches cost to usage, so sellers know their costs are covered.

• It works for developers and businesses doing high-volume integration with uncertain demand.

• It creates predictability for OpenAI, and pressure for its customers to optimize.

But token pricing isn’t always intuitive for buyers. That’s why many companies build on top of OpenAI and repackage it using simpler models, like per message, per document, or per output.

Token-based pricing works great when your buyer understands and cares about scale, has uncertain demand, and receives varying amounts of value per use. It works less well when you’re selling to end users who don’t want to think about tokens.

3. FinAI: 99 Cents per Resolved Call

FinAI sells an AI-based support assistant to financial institutions. Their pricing? 99 cents per completed support call. That’s a brilliant move. Why?

• It scales with value delivered (calls resolved).

• It aligns with outcomes (the thing the buyer cares about).

• It gives FinAI predictable revenue tied to the success of its own product.

This is outcome-based pricing, not tied to usage or inputs. And that’s what makes it powerful.

4. Microsoft Co-Pilot: The Old Model Still Works… Sometimes

Microsoft charges $30/user/month for Co-Pilot access inside Word, Excel, Outlook, and other apps.

It’s familiar. It’s simple. And it fits Microsoft’s world. CIOs and IT buyers are used to user-based pricing. Microsoft sells in bulk. This model doesn’t perfectly match usage or value, but it matches how buyers buy.

And sometimes that’s more important than elegance.

5. The Companies That Don’t Charge Yet

Many companies are offering AI for free. That’s fine, for now. However, it’s not sustainable if the AI delivers real value or drives significant costs.

If you’re paying to run an LLM behind the scenes and you’re not monetizing it, you’re hoping AI is a differentiator, not a revenue driver. That’s a short-term play.

The real question is: If you removed the AI tomorrow, would your customers miss it enough to pay for it?

If the answer is yes, it’s time to think about how you charge for it.

The Takeaway: The Fundamentals Still Apply

AI may have dramatically changed how you offer your product or services, but it hasn’t changed pricing fundamentals. Instead, it has probably forced you to think about pricing differently than you have in the past. You may have to rethink concepts that you knew intuitively about your business before AI. You may have to:

• Know your buyer

• Understand what they value

• Choose a metric that scales with that value

• Use segmentation and versioning differently to capture different levels of willingness to pay

Because AI changes:

• Who does the work

• What gets measured

• Where value is perceived

• And how buyers compare alternatives

When you price AI like traditional SaaS, you often miss the value and misalign incentives.

So if you’re adding AI to your product, or building something new altogether, start by asking:

• What value is the AI delivering?

• Who cares about that value?

• How can I price in a way that grows with impact?

Those are pricing questions. Not AI questions. And the answers are grounded in the same fundamentals we’ve always used.

AI is just a new technology. We’ve done this before with semiconductors, with SaaS, with hybrid cars. Every time an innovation hits the market, we have to figure out how to package it, position it, and price it. But the underlying pricing concepts don’t change. It’s still about creating value for customers and capturing a fair share of that value in return.

Share your comments below.

Now, go make an impact!