The Death of Seat-Based Pricing

Let’s admit it: seat-based pricing is dead.

It didn’t die quietly. It died a public, bloody death on the NASDAQ floor.

Salesforce. HubSpot. Workday. ADP. Some of the most iconic SaaS companies of the last 20 years are all feeling the pressure—despite shipping wave after wave of AI features.

So… whodunnit?

Everyone keeps pointing fingers at AI startups, like we’re the ones holding the knife.

But it’s not us. And it’s not as simple as “companies are cutting headcount.”

First, what’s the motive?

Like any good murder mystery, the obvious answer is only part of the story.

Yes, companies like Block cutting ~40% of their workforce (https://www.nytimes.com/2026/02/26/technology/block-square-job-cuts-ai.html) (partly citing AI) matters. Fewer people → fewer seats.

(https://substackcdn.com/image/fetch/$s_!sRej!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff1151e06-07c0-41a7-90db-59600794e71e_299x168.png)

But that’s not the root cause. That’s just the scene of the crime.

The real motive goes deeper. For the last decade, SaaS companies sold into a future that never showed up. Seat-based pricing was built on one core assumption: 1) companies grow then 2) headcount grows then 3) seats grow and so 4) revenue grows

And great sales teams leaned into that. They sold ahead of the curve. “You’ll grow into this.”

That was working, but lately? Not as much.

Now we’re seeing the correction. Companies are flattening, not expanding, and fewer people are doing more work (thanks to AI). As a result, those “future seats” never materialized

So it looks like demand is shrinking. But what’s actually happening is simpler.

The pricing model was misaligned with reality.

Seat-based pricing was never really tied to value—it was just a convenient proxy that is easy to sell, easy to understand, easy to model for investors.

But it only worked in a world where headcount always went up.

That world is gone.

And AI didn’t kill seat-based pricing. It exposed it as a temporary stage.

So where’s the murder weapon

So if AI didn’t pull the trigger… what actually did?

A. Value Is No Longer Linear

In traditional SaaS, value scaled somewhat predictably with seats.

In AI, it doesn’t.

You can have:

A power user who uses the product constantly and gets massive value
A casual user who barely touches it and gets almost none

But more interestingly, you can also have:

A few high-leverage prompts (low usage, high value)
Lots of low-quality iteration (high usage, low value)

Now try pricing that per seat. I’ve tried. It’s almost impossible, and I wind up with questions like:

“Is it worth $200/month for my CRO to ask 3 questions?”

And my gut response is:

“If those 3 questions change your strategy, it’s worth way more than $200.”

But regardless, that uncertainty kills adoption.

B. Costs Are Now Variable

At the same time, the cost structure underneath AI is fundamentally different. It’s based on tokens (or rather, compute), which is both hard to really understand and can swing a ton.

We’ve seen companies like Snowflake successfully price based on consumption (or variable compute that gets used). It works for them because it aligns cost with actual usage and it removes that seat friction. So both the customer and sales team are incentivized to encourage adoption.

But warehouse costs are different than AI costs, and it comes with some problems to wrestle with.

The unpredictability is constant in both cases. Finance comes to reluctantly accept it, at least in some cases. However, non-determinism with LLMs means that the same input (or roughly the same input) might not provide the same output, AND might not give the same output. Importantly, it might not cost the same.

It leads to a psychological barrier: “Is this question worth $20?”

And here’s the kicker: sometimes the expensive answer is worse than the cheap one.

There’s no clean mapping between cost and outcome.

This is where the story gets messy.

We didn’t replace one clean model with another. We broke one… and replaced it with something equally imperfect.

On one side, you have seat-based pricing. It’s predictable, easy to budget, and easy to explain.

But it artificially limits adoption. It forces every user—and every question—into a fixed cost, regardless of the value created.

On the other side, you have usage-based pricing. It’s flexible, scales with actual consumption, and feels more “native” to AI.

But it’s chaotic. Costs are unpredictable. Outcomes are uncertain. And every prompt starts to feel like a small bet:

“Is this going to be worth the $10 of usage?”

Neither model cleanly maps to how AI actually creates value and doesn’t point towards a clear culprit.

Our list of suspects

Right now, everyone is experimenting in real time. There’s no clear culprit who will replace seat-based pricing—just a set of patterns emerging as companies iterate.

Hybrid Models

The most common approach is a blend:

A base platform fee (for access, security, infrastructure)
Plus usage-based pricing on top

This gives companies some baseline predictability while still capturing upside as usage grows. It’s not elegant—but it works.

You’ll see this as common amongst AI startups targeting finance firms, using the need for security & governance to justify the platform fee.

Token Margins

A simple solution is to just price entirely on usage and mark up the token cost, but there’s a key problem.

Customers are getting smarter. They know what tokens cost. They know what model providers charge. So naturally, they ask: “Why am I paying 4x markup for this?”

That forces a hard question for every AI company, including us. Where does your margin actually come from?

You either: 1) build something proprietary (models, infrastructure, data advantages) or 2) prove that your product creates value beyond raw compute (workflows, UX, outcomes)

If you can’t do either, elevated token pricing (i.e. over the latest model from OpenAI or Anthropic) gets very hard to justify.

Services Disguised as Software

Some companies sidestep the problem entirely. They don’t expose usage to the customer because they absorb it internally.

Take Crosby Legal (https://crosby.ai/). They are an AI startup masquerading as a legal firm. They offer legal services for contract negotiations, and internally, they use AI to reduce their costs.

They use AI to reduce the cost of delivering work, but they still charge clients based on outcomes or hours. The customer never sees tokens, models, or compute. They just see redlines.

This works especially well in industries where human labor is expensive and value is already well understood.

Law is a perfect example. Crosby Legal and others like them are very well positioned to grow and succeed regardless of where the rest of the market shakes out with regards to pricing.

Success-Based Pricing

Then there’s the most extreme version of alignment: success-based pricing.

Take Spendly (https://www.getspendly.com/). They analyze vendor spend, renegotiate contracts, and take a percentage of the savings they generate.

They only get paid if they save you money. It’s incredibly compelling. There’s no upfront cost, no guessing, and no risk for the customer. It has to be the easiest sell ever.

But it only works if you have:

High confidence in outcomes
Tight control over which customers you take on

Otherwise, your cost structure can spiral.

It’s risky, so it only works in a limited set of markets. But if you can crack it, it’s “money-printing” model.

Commit-Based Models

The other pattern comes straight from cloud infrastructure. Instead of pure usage, companies sell annual spend commitments with prepaid credits.

It functions much like a “bar minimum.” Customers commit to a baseline amount, then use it however they want: cheap models (“beer”), expensive models (“wine”), pre-set workflows (“well drinks”), or services (“signature cocktails”).

This creates a middle ground of predictability for the vendor and flexibility for the customer. But it requires a much higher level of transparency to monitor & report usage at every level. And the cloud providers can offer bulk discounts because they own the infrastructure — startups cannot.

The hidden door: education

Underneath all of this is something more fundamental. Most people don’t actually understand what they’re paying for.

Tokens aren’t intuitive. Model behavior isn’t predictable. And the relationship between cost and output is… fuzzy at best.

You can ask the same question twice and get two different answers — and two different costs.

Sometimes the better answer is cheaper. Sometimes the worse answer is more expensive. So users are left guessing.

Every interaction becomes a small gamble: “I think this is worth it… but I’m not sure.”

That uncertainty slows adoption, and more importantly, it erodes trust.

The wake of the murder

This doesn’t just affect customers. It breaks internal systems too.

Finance

Finance teams are used to predictability. Seat-based pricing gave them that. Usage-based pricing takes it away.

Now they’re dealing with:

Hard-to-forecast spend
Unclear ROI
Budgets that move after the fact, not before

And the hardest part of all: you can spend a lot… without clearly knowing if you got value.

Sales

Sales teams lose their simplest tool: “It’s $X per seat.”

Instead, they have to explain tokens, usage, variable cost, and non-deterministic outcomes to buyers who just wrapped their head around the fact that ChatGPT can find recipes for dinner.

Every deal becomes more consultative, which slows everything down.

Users / Operators

And then there are the people actually using the product.

They’re making constant micro-decisions:

“Should I run this?”
“Should I iterate again?”
“Is this worth the cost?”

In some workflows—like underwriting, analysis, or agent iteration—those decisions can add up quickly. You might spend hundreds or thousands of dollars just getting to a version you trust.

And you don’t know that upfront.

Whodunnit

So where’s the big Sherlock-style reveal where I say whodunnit?

Honestly, I can’t give it. But for the next few years, it won’t be a single model.

More likely, we converge on a mix:

Services as software
Commit-based systems
Success-based pricing

Not because they’re perfect. But because they’re the easiest to explain.

Over time, the winners will be the ones that balance three things well:

Alignment with outcomes
Transparency (or at least perceived fairness)
Flexibility without overwhelming the user

That’s a hard needle to thread.

What am I doing?

Stepping back, this feels bigger than just pricing. This is a full reset of the SaaS playbook.

And right now? No one has it figured out. Which makes it tempting to try. I’ve spent so many hours fretting and obsessing over pricing, thinking I could reinvent it.

But honestly, that’s probably the wrong move for most startups. There are too many variables changing at once: model costs, user behavior, and product capabilities.

If I try to innovate on pricing right now, I’m layering uncertainty on top of uncertainty for my startup in a way that I don’t need to.

Better to focus on what I can control. Innovate building something people actually want to use. Let the market innovate on pricing.

Which means, for now, I’m operating in the gray area. And if this is a murder mystery, we still haven’t found the real killer.