Seat-based pricing built many of the biggest SaaS successes of the past 10 years. Notion, Slack, Zoom, Jira… the list goes on.

It was a great innovation because it let you bet on your customers and through that, the tech market at large: As your customers’ business expanded, their team grew, and so did their software spend.

You’d participate in the upside of the growth of a company without investing.

Customers were also fine with this because most of these tools are collaborative. But that dynamic is crumbling for two main reasons:

1. Teams are smaller (and grow less)

2. AI products have real cost

That’s why usage-based or blended models are taking over.

The past 10 years (up to 2022-23ish) were a golden age for tech. Money was cheap, so it was easy to raise. The millions raised flowed into growing teams, and it seemed like every company was always hiring (on every team).

As teams and companies everywhere grew their head count, they needed new tools to collaborate, which is why collaboration tools were some of the biggest successes of the pre-AI SaaS era.

Take Figma (which Adobe was willing to buy for $20B), whose entire initial pitch was about problems that only arise when teams grow.

Invision and Sketch did much of what Figma initially did, but Figma’s killer feature was real-time collaboration, which obviated versioning (and with it, much miscommunication and parallel work).

Collaboration is easy when you’re tiny, but gets harder as teams grow. Many other 2010-2023 SaaS successes follow the same pattern:

• Slack did text-based communication, but email gets too messy for big teams.

• Notion did more or less what Google Workspace does, but juggling docs get too messy for big teams.

• Asana did more or less what Excel spreadsheets and emails do, but those get messy for big teams.

(all of these products are also superior to the incumbents in other ways, but the first versions especially were often “just” collaboration-ready alternatives)

Pricing should follow how you create value. And if you create value by orchestrating collaboration, charging per team member makes sense. Each team member makes collaboration messier, so collaborative tools get more valuable with each user.

But this also means that charging by the seat means your growth is dependent on customers’ teams growing.

Today, many teams are shrinking. Companies are more hesitant about hiring. New companies hire less because AI increases worker productivity. So the fundamental assumption of seat-based pricing makes growth harder.

One of the premises of pre-AI seat-based tools is that you got more or less unlimited usage (with rate limits so high you’d rarely bump into them). That’s because each individual bit of usage is practically free for the providers.

Slack’s pro plan offers unlimited messages.

It’s probably not even worth calculating the cost of storing each individual message. Their aggregate AWS bill may be giant, but each individual message is basically free.

That’s why software companies have margins other industries could only dream of (often 75%+).

This is no longer true the minute you build with AI.

LLM providers typically charge per million tokens. That means every bit of usage has a direct cost. It also means a single power user can bankrupt you.

Two other factors are subscription fatigue (companies are tired of recurring costs) and, looking into the future, AI agents, who will only access products through API.

All of this indicates that seat-based pricing won’t be as prevalent as before. But how will pricing and monetization change?

keeps talking about the shift towards usage-based pricing. One part of that is that many companies are removing seat caps (example from Webflow):

You may not get unlimited seats like you do on PostHog (below), but seats are getting more flexible—an admission that they’re no longer the core value driver.

Many companies are switching to usage-based pricing. When your hear usage-based pricing, you might think of pure-play usage-based like Twilio:

But usage-based pricing exists on a gradient. There are many hybrid models:

• Credits: Prepay for some amount of usage and use it up over time.

• Subscription with overages: Pay a subscription that includes some usage, pay per unit if you use more.

• Subscription and usage: Pay a subscription and pay for usage on top of it.

Kyle Poyar has studied this intensely and found that usage-based pricing (in all of its forms) is becoming more and more popular:

The report for 2024 isn’t out yet, but I can’t see that number going down.

That doesn’t mean usage-based is right for every company. Returning to our previous examples, nobody wants to pay by the Slack message or by the Figma frame.

But as usage-based pricing is taking hold, another pricing model is only emerging.

Usage-based pricing has the downside that usage doesn’t always correlate with value (e.g. Notion doesn’t strictly get more valuable the more pages you have). Seat-based subscriptions have the downside that expenses don’t always correlate with usage.

Outcome-based pricing remedies both of these. It’s what it sounds like: You pay whenever the AI tool achieves an outcome. Intercom is a pioneer of this. It’s AI support agent Fin charges per resolution, i.e. per support chat that didn’t result in a support ticket being submitted.

This is great because it aligns the incentives: Intercom is incentivized to keep chats as short and effective as possible (they pay OpenAI for usage, so their margins increase the less they use). Customers don’t pay for hallucinations.

This is what Sarah Tavel describes as selling work, not software: Instead of selling software that augments a human worker (like Notion, Slack etc. do), AI can complete the work itself—and is compensated like a human (commission-