For years, SaaS pricing was simple: pay per user per month. AI has complicated that. AI features cost real money every time they run, and they can let one person do the work of several, which undermines per-seat pricing. In 2025 and 2026, many software companies moved to credits, usage-based pricing or hybrids.
Whether you're building software or buying it, here's how the main models work.
The main pricing models
| Model | How it works | Good for |
|---|---|---|
| Flat rate | One price, everything included | Simple products with predictable costs |
| Per seat | Price per user | Collaboration tools where value grows with users |
| Tiered | Plans with different features and limits | Most SaaS products |
| Usage-based | Pay for what you use (API calls, messages, GB) | Infrastructure, APIs, AI-heavy products |
| Credits | Monthly credit allowance spent on actions | AI features with variable costs |
| Outcome-based | Pay per result (resolved ticket, qualified lead) | Products that can measure outcomes reliably |
| Hybrid | Base subscription plus usage or credits | Increasingly common |
Per-seat pricing
Pros: easy to understand, predictable revenue, grows as customers grow. Cons: encourages customers to share logins; doesn't reflect heavy AI usage; can undercharge when AI replaces manual work.
Usage-based pricing
Pros: fair (pay for value), low entry price, scales naturally. Cons: unpredictable bills scare buyers; revenue fluctuates; usage must be measured accurately.
Make it workable with:
- Clear units customers understand.
- Usage dashboards and alerts.
- Optional spending caps.
- Committed-use discounts.
Credits
Credits have become the default way to price AI features: each plan includes a monthly allowance, actions cost credits, and customers can buy more.
| Strength | Weakness |
|---|---|
| Covers variable AI costs | Abstract units can confuse buyers |
| Predictable monthly bill | Customers may feel "nickel-and-dimed" |
| Lets you price different actions differently | Credit costs need careful design |
Make credits understandable by translating them into outcomes: "500 credits ≈ 100 meeting summaries or 50 long reports". Our own Zelvorm Intelligence uses free monthly credits for AI features for exactly this reason.
Outcome-based pricing
Charging per resolved support ticket, per qualified lead or per completed task aligns price with value. It works only when outcomes are clearly measurable and largely within your product's control. AI support agents are one area where it's growing.
Choosing a model for your product
- Understand your costs. What does it cost to serve a light user, a typical user and a heavy user, including AI and infrastructure?
- Find your value metric. What grows as customers get more value: users, projects, documents, transactions, outcomes?
- Match price to value metric. Price should rise with value, not with something customers don't care about.
- Keep it simple to buy. Buyers should estimate their bill in under a minute.
- Protect your margins from heavy users with limits, fair use policies or usage charges.
- Test and iterate. Pricing is never finished.
Common hybrid example
| Plan | Price | Includes |
|---|---|---|
| Starter | £19/month | 1 user, core features, 200 AI credits |
| Team | £49/month | 5 users, integrations, 1,000 AI credits |
| Business | £149/month | 20 users, advanced controls, 5,000 AI credits |
| Extra credits | £10 per 500 | Top-ups any time |
Pricing mistakes AI products often make
- Unlimited AI on a flat plan. A handful of heavy users can wipe out the margin from everyone else. Add fair limits or credits.
- Credits nobody can translate. "2,000 credits" means nothing until you say what it buys.
- Pricing on cost alone. If your AI saves a customer five hours a week, price against that value, not just your model costs.
- No visibility of usage. Customers who can't see what they've used feel ambushed by limits or overage charges.
- Hard stops at the worst moment. Blocking a feature mid-task frustrates users; warn early and offer a top-up.
- Never revisiting prices as model costs fall or features improve.
Example: working out the cost of a credit
Say an AI report costs you about 4p in model and infrastructure costs on average, and you want healthy gross margins on usage.
| Item | Amount |
|---|---|
| Average cost per report | £0.04 |
| Credits charged per report | 5 |
| Price of 500 extra credits | £10 (2p per credit) |
| Revenue per report from top-ups | £0.10 |
| Gross margin on top-up usage | 60% |
Then test whether customers understand and accept it, and adjust. Track your real cost per action monthly, because AI model prices change often. (Figures are illustrative.)
Presenting pricing
How you show pricing matters as much as the model. Read pricing page design for layouts that convert, and how to price products online for the maths behind price points.
For buyers: questions to ask
- What exactly counts as a user, credit or unit?
- What happens when I hit a limit: blocked, charged or throttled?
- Can I set spending caps and get alerts?
- Do unused credits roll over?
- How have prices changed in the last two years?
Building a SaaS product?
Pricing affects architecture: you need usage metering, plan limits, billing integration and dashboards from the start. Plan it early. Read what to know before commissioning a SaaS MVP and how much custom software costs.
The short version
SaaS pricing has shifted from simple per-seat plans towards credits, usage-based and hybrid models because AI features have real per-use costs. Choose a value metric that grows with customer value, protect margins from heavy users, translate credits into real outcomes and keep pricing simple enough to estimate quickly.
Building a SaaS product? Our software development team designs billing and metering into the build, and our consulting team helps with pricing strategy.




