Per Seat vs Per Token AI Pricing: Why It’s the Wrong Question (and What to Ask Instead)

Key takeaways

• Per seat suits predictable, person-based use; per token suits flexible, experimental use
• Neither pricing model guarantees a return. The buyer carries the risk of using it well
• Outcome pricing (paying per resolved ticket or paid bill) works best for simple, closed tasks
• The real gap is know-how: knowing which work to give AI and how to judge the answer
• Everyone who uses AI at work is now a manager of AI, deciding where it goes

Should you pay for AI tools per user or per use? It is one of the most common questions we hear from small business owners, and it is the question big companies are stuck on too.

In a recent video, AI strategist Nate B Jones argues that enterprise AI deals keep stalling on exactly this point, and that per seat vs per token AI pricing is the wrong thing to focus on. His view: AI is a new kind of business input, alongside people and money, and the price tag matters far less than whether you know how to put it to work.

Here is what he said, what we think it means for smaller businesses, and two practical guides to help you act on it.

Why is per seat vs per token the wrong question?

Because both are ways of metering, not ways of measuring value. A seat licence counts how many people can log in. A token counts how much text a model reads and writes. Neither tells you whether a single customer was helped or a single sale was made.

Nate’s point is that tokens are useful for budgeting, but a token does not contain a fixed amount of useful thinking. Whether it is useful depends on the person using it. Standard contracts leave that risk entirely with the buyer: you bought it, now make it pay.

That is why so many companies cap their AI spend. They have not yet worked out how to get a reliable return at scale, so they limit the downside. Understandable, says Nate, but it often means they are under-invested.

How do per seat, per token and per outcome pricing work?

Here are the three main ways AI tools are sold today, in plain terms:

What does outcome-based AI look like in practice?

Nate opens with Cedar, a US healthcare payments company that uses an AI voice agent to handle patient billing calls. A patient can ring at 2am, have a confusing bill explained, check their insurance and pay, all in one call, with no hold music.

The point he draws out: the hospital does not care how many tokens that call used. It cares that the bill got paid. That is the logic behind outcome pricing.

He links this to Sierra, the AI customer service company, which he says is actively exploring per-outcome pricing deals. He also mentions Fin, the AI support agent, as another example of pricing per resolved conversation.

Are most businesses under-invested in AI?

Nate thinks so. His argument is that companies set a ceiling on AI spend to avoid risk, when the better question is what to invest in. That is a question of ambition and agency, not procurement.

It also cannot be handed to a chatbot. Deciding which bets to make, and lining staff up behind them, is a leadership job. Companies that answer it well are the ones he expects to raise their AI investment and get the most from it.

Why does buying AI not guarantee a return?

Because general-purpose AI has to be pointed at the right job, and most of that discovery happens after you buy. You can purchase the capability before you know which work suits it, how to change the processes around it, or how to tell a good answer from a bad one.

Nate’s worked example is a hypothetical equipment company (based, he says, on real cases). An engineer notices a recurring fault. An AI agent pulls together maintenance records and past cases in minutes. But only the engineer knows which records are unreliable, which conditions matter, and when two faults are being described in the same words.

At some point, generating more AI explanations stops helping. The engineer has to decide when to dig deeper with the AI and when to pick up the phone to a customer. Repeat that across enough cases and the team may spot a new paid service, such as advice that prevents breakdowns.

That journey, from tool to new product, is what good AI use looks like. And it is the part no price list covers.

What are forward deployed engineers, and why do they matter?

Forward deployed engineers are technical staff who work inside, or right alongside, a customer’s business to turn AI into working systems. They help choose where AI should be used, then design, build, test and launch it using the customer’s own data and processes.

Nate points to OpenAI’s deployment business and to Anthropic’s joint venture with Blackstone, Hellman & Friedman and Goldman Sachs, which he says is aimed at firms such as mid-size manufacturers and regional healthcare systems. He also mentions OpenAI taking a stake in Thrive Holdings and placing research, product and engineering teams inside its companies, starting with accounting and IT services.

His caution: keep it purely as a service forever and you learn too little; turn it into a product too fast and you also learn too little. The balance is where the value is.

When does outcome pricing work, and when does it not?

Outcome pricing works when the task has a clear, closed finish. A billing call that ends in a payment, or a support ticket that is resolved, can be counted and charged for.

It works far less cleanly for complex work, such as redesigning a customer’s whole maintenance programme. There is no single, simple “done” to put a price on. Outcome-based deals are also hard to measure, which is why Nate does not expect everyone to switch to them tomorrow.

Even so, he argues that the first genuinely new business input in centuries justifies some experimenting with how we buy it. If you are offered per-result pricing for an AI agent, our practical guide below shows how to judge whether it is worth it.

Is anyone actually getting a return on AI spend?

Yes, and the public examples come from companies that are very fluent with AI. Nate cites Shopify, whose chief operating officer has described a healthy return on AI spending. According to Nate, Shopify’s headcount has stayed broadly flat for about two years while revenue grew 34%.

Shopify’s own figures support the revenue part: its Q2 2026 results show 34% year-on-year revenue growth. Nate also says Block reported 25% gross profit growth after slimming its team and putting AI at the centre of the business, while noting we will have to see how that continues.

Importantly, Nate says per token pricing is here to stay. If your business already knows how to use AI, paying for usage gives you room to experiment and keep the gains. Different firms will need different deals depending on how fluent they are.

What does it mean to be a manager of AI?

It means deciding where AI effort goes, what it should finish, and when a human needs to step in, whether or not you have “manager” in your job title.

Nate’s example: you can ask an AI agent to finish a regular report faster, or you can ask it to dig into the customer problem behind the report and start fixing it. The second choice is what moves you from user of AI to manager of AI.

He argues this is a real career opening. Many companies are not yet fluent with AI and need champions who can show results that reach the top and bottom line. For leaders, his advice is to put the company’s ability to create value on the table alongside the price, and to think about growth, not just cost cutting.

What should small businesses do next?

Start with the result, not the pricing page. Here is a simple order of work:

1. Pick one job with a clear finish, such as answering pre-sale questions, chasing unpaid invoices or drafting product descriptions
2. Measure today’s baseline: time, cost and quality before AI
3. Choose the pricing model that matches the job: seats for daily assistants, usage for automations, outcomes for closed tasks
4. Name an owner who decides what the AI does and checks its work
5. Review after 30 days and keep, change or cancel based on the numbers

Frequently asked questions

The bottom line

Per seat or per token is a budgeting choice. The bigger question is whether you know which work to hand to AI and how you will measure what it delivers. Your next step: pick one task this week, write down how long it takes now, and run it through an AI tool with a named owner checking the results.

Related posts

• How to Work Out the Real Cost of an AI Task Before You Switch Models
• How to Draft With a Cheap AI Model and Refine With a Premium One
• How to Spot Reward Hacking When an AI Agent Says the Job Is Done

Garry Knight

I'm Garry Knight, the person behind Prodify Digital. I write about email list building, email marketing, SEO, AI search and the tools that connect them. My aim is to make online marketing easier to understand, so creators and small business owners can make informed decisions about building an audience and keeping people engaged. Here you'll find straightforward guides and product reviews that explain what something does, where it fits and which limitations matter. The focus is on clear explanations and useful next steps—not hype, shortcuts or promises of easy earnings.

See Full Bio