A Pricing Upheaval in the Age of AI
You've built a SaaS product — now how do you price it? A flat monthly subscription? Pay-as-you-go? Free at first, with paid plans kicking in later? This isn't just a pricing decision. It's a choice that determines the entire business structure of your product.
The problem is that we're in the middle of a SaaS pricing upheaval. As AI moves in, the golden formula of SaaS — the seat-based subscription — is starting to crack. As David George of a16z diagnoses it, the first cost customers try to cut when they adopt AI is labor. Seat-based billing sits squarely in the crosshairs.
Schematic, a software monetization platform, mapped out nine pricing models in February 2026. Here we reframe them from the perspective of a Korean founder — as a framework for judging which combination fits your product.
The Nine Pricing Models at a Glance
Nine SaaS Pricing Models
Look at each model up close and it becomes clear where each one fits.
1. Flat-Rate Subscription: The Simplest Model, but Growth Stalls
₩299,000 (about $220) a month, unlimited use. It's the simplest model there is. Easy to sell, easy for customers to understand, easy to forecast.
The problem: revenue doesn't grow when usage grows. Light users and heavy users pay the same. That's a great deal for the heavy users, but it means the company captures no additional revenue from the very customers extracting the most value.
Most companies keep a flat subscription as the base and layer usage or add-ons on top. Going flat-rate alone is increasingly rare.
2. Per-Seat: The Golden Formula of SaaS, Now in Crisis
For the past 15 years, this was the SaaS standard. $25 per user per month. More users, more revenue; fewer users, less. Simple and predictable.
Collaboration tools (Slack, Notion), design tools (Figma), and CRMs (Salesforce) all grew up on per-seat pricing.
But in the AI era, this model is under direct assault. David George puts it precisely: "The first cost customers try to cut when they adopt AI is labor. Seat-based billing becomes the target."
If a tool once used by 100 employees can be run by 50 employees backed by AI agents doing the same work, the seat count gets cut in half. Same customer, half the revenue. SaaS businesses that rely on per-seat pricing alone are likely to be the first casualties of the AI era.
3. Usage-Based: A Candidate for the AI-Era Standard
API calls, events processed, compute time, data volume. Customers pay only for what they actually use.
AWS, Twilio, and Stripe built enormous businesses on this model. Whether headcount rises or falls, revenue grows with activity. Even if AI agents shrink seat counts, the API calls those agents generate keep climbing. It's a natural fit for the AI era.
The downside is just as clear. From the customer's side, the monthly bill is unpredictable. The anxiety of "how much will it be this month?" blocks adoption. That's why usage-based pricing is usually designed as a baseline plus overage.
4. Credit-Based: The Best Fit for AI Products
This model is fast becoming the standard for AI products.
Customers buy credits up front, and each task draws down the balance — a little for light work, a lot for heavy work. ChatGPT's message credits, Midjourney's GPU hours, tokens on the Claude API: all the same structure.
Why do credits suit AI? Because costs vary wildly from task to task. A simple text request and a video generation can differ in cost by a factor of 100. Bill that purely by usage and the invoice swings too violently. With credits, pricing stays predictable — say, 50,000 credits for ₩50,000 (about $36) a month — while each deduction still tracks the real cost of the task.
If you're building AI-powered SaaS in Korea, the credit model deserves a first look. Users can control their own consumption, and the company can manage its volatile infrastructure costs.
5. Tiered Plans: The Most Familiar Packaging
Starter, Pro, Enterprise. The three-tier package is the standard.
The lowest tier protects existing revenue. The middle tier absorbs expansion. The top tier leaves room for negotiation and custom terms.
There's a reason three tiers became the norm: customers can compare at a glance, salespeople can explain them easily, and the price differences are clear. Four or more tiers tend to trigger choice paralysis instead.
That said, this model is rarely used on its own. It's usually designed as tiers plus usage caps plus add-ons.
6. Add-Ons: How to Keep the Core Product Lean
Audit logs, advanced permissions, premium integrations, higher API limits — features that not every customer needs.
Splitting them off as add-ons keeps the core product lean and the pricing simple. "The base plan is $29 a month; audit logs are $10 more." Customers buy only what they need, and the company earns incremental revenue.
The real value of add-ons is reducing the burden of shipping new features. You don't have to redesign the entire pricing structure every time you build something new. You just ship it as another add-on.
7. Freemium: The Most Powerful Model — and the Most Dangerous
Start free, convert to paid past a certain limit. Notion, Slack, and Dropbox all grew this way.
The upside is obvious. The barrier to entry is zero. It spreads by word of mouth. Users get comfortable with the product before they're ever asked to pay.
The downside is just as obvious. Set the limits wrong and you accumulate free users without growing revenue. Infrastructure costs climb because of those free users, and if they never convert, the losses pile up.
Everything hinges on where you set the limit. Too tight, and users leave before they're hooked. Too loose, and they stay free forever. The heart of freemium design is finding the moment right after users feel the product's real value — and nudging them toward paying right there.
8. Marketplace Fees: The Standard for Platform Businesses
Take a fee whenever a transaction happens. Uber, Airbnb, and Coupang (Korea's e-commerce giant) all run on this structure.
It might be 15% per transaction, a flat fee, or a paid listing. You never force a subscription on users, yet revenue follows as transaction activity grows.
There's a key difference from SaaS: a marketplace has to build a two-sided market. It only works once you've gathered both suppliers and buyers. That makes the early build far harder than SaaS, and revenue can be close to zero until you reach critical mass.
9. Outcome-Based: A New Possibility in the AI Era
The most fascinating model — and the hardest. You're not selling access; you're selling results.
An AI sales tool charges ₩50,000 (about $36) per lead captured. An AI fraud-detection tool takes a fee for every fraudulent transaction it blocks. An AI customer-service bot bills per inquiry resolved.
For customers, it's the most attractive model imaginable: no results, no payment. For the company, it ties the product's value to revenue more directly than anything else can.
The problem is measurement. How do you define a "result"? How do you prove the result happened because of your product? If measurement is imprecise or disputes break out, the entire model collapses.
As AI automation products multiply, this model is likely to spread. AI tends to work in domains where clear outcomes — items processed, hours saved — are easy to measure.
The Rise of Hybrid Models
Almost no company picks just one of the nine. Most combine two or three.
A few combinations come up again and again.
Subscription + usage overage. Unlimited use within the base plan's cap, then per-unit billing past it. Claude and ChatGPT Enterprise use this structure. It captures predictability and scalability at the same time.
Subscription + credits. A flat subscription covers the core features, while AI tasks draw down credits. An increasingly common pattern for SaaS with AI features layered in.
Tiers + add-ons. Lay down three tiers, then sell extra capabilities into each tier as add-ons. The shape of standard B2B SaaS.
Self-serve + sales-led. Small customers sign up themselves with a credit card; large customers negotiate custom contracts with the sales team. The same product, sold through two different channels.
The core idea: build predictable revenue with subscriptions, and capture expansion revenue with usage or credits. You need both.
How to Choose a Model
The answer lies in how your customers actually use your product. Look for three patterns.
If customer usage is stable, go with subscriptions or per-seat. Tools people use roughly the same amount every day: email clients, CRMs, collaboration tools.
If customer usage is volatile, go usage- or credit-based. Tools that might get used 100 times one month and 10,000 the next: AI APIs, infrastructure tools, data-processing tools.
If usage varies dramatically from customer to customer, go tiers + add-ons. When small teams and large enterprises use the same product in very different ways. Most B2B SaaS lands here.
Why Pricing Needs a Redesign
Until now, SaaS pricing was a decision you made once and lived with. Choose per-seat, and it held for five years.
The AI era is different. Seats are shrinking. The load AI generates no longer scales with user counts. Automation produces results in ways human labor doesn't. The assumptions behind the old pricing models are breaking down.
To return to David George's diagnosis, there are two paths: use AI-native products to accelerate revenue growth, or redesign the cost structure to lift operating margins. On both paths, redesigning the pricing model is central.
If you run a SaaS business today, you need to test whether your pricing model still works in the AI era. If you depend solely on seats, add usage or credits. If you run flat-rate only, build a structure that captures additional value from heavy users. The era when one model was enough is over; hybrid design is becoming the standard.
If you're starting a SaaS company now, design for hybrid from day one: subscription as the baseline, usage or credits for expansion. Start with both axes in place, and you can respond flexibly as the market shifts.
Price is not just a number. It is the business model itself — the answer to how the value your product creates gets converted into revenue. The companies redesigning that model for the AI era will be the winners of the next five years.
