SaaS pricing strategy used to be simple: pick a per-seat per-month tier structure, add a free plan, ship. In 2026 that's a money-losing strategy for entire categories of products, particularly anything AI-powered, where one power user can rack up $400 in inference costs on a $50/seat plan. The dominant pricing models have fragmented, and the right choice now depends on your unit economics, your buyer's psychology, and how much friction you can absorb.
This guide is the pricing playbook we walk founders through at DevEntia when they're building or repricing a SaaS product.
The four pricing models that win in 2026
| Model | What you charge for | Best fit | Watch out for |
|---|---|---|---|
| Per-seat | Each user with login access | Collaboration tools, B2B with predictable usage | Sandbagged seats, deal compression |
| Pure usage-based | API calls, GB transferred, runs, etc. | Infrastructure, AI inference, transactional services | Bill shock, hard to forecast revenue |
| Tiered (good/better/best) | Feature bundles at fixed prices | Mid-market B2B, products with clear feature ladders | Tier confusion, undercut by competitors |
| Hybrid (seat + usage) | Base seat fee plus usage on specific actions | AI products, platforms, data-heavy SaaS | Complexity in billing and explanation |
The 2026 reality: most successful AI-powered SaaS companies run a hybrid model. Most pure collaboration tools still run per-seat. Pure usage-based is concentrated in infrastructure and developer tools.
The data behind the shift to usage-based
The OpenView SaaS Benchmarks Report has tracked the rise of usage-based pricing for years. Their data shows usage-based companies grow ~30% faster than purely subscription-based peers and have higher net dollar retention. The reason is straightforward: usage-based pricing scales with customer success. When customers use more, they pay more, without renegotiation.
Snowflake, Twilio, AWS, OpenAI, Datadog, and Cloudflare all built nine- or ten-figure businesses on usage-based or hybrid models. The pattern isn't accidental.
Per-seat pricing: still default for collaboration tools
Where it wins
Per-seat works when the value scales linearly with the number of people using the product. Slack, Notion, Figma, Linear, all per-seat-dominant because more team members = more value derived. The buyer's mental model is simple ("$15 per teammate per month"), forecasting is trivial, and the procurement conversation is well-understood.
Where it breaks
- Power users vs casual users. A per-seat plan punishes teams where most users barely log in. Buyers either remove "lurker" seats (revenue loss) or buyers feel they're overpaying.
- Sandbagging. Sophisticated buyers maintain shared logins to avoid seat fees. Common at the SMB level.
- AI cost imbalance. One power user generating 1,000 LLM calls a day costs you $50–$200 in inference. Your $30 seat fee is underwater. This single dynamic is why pure per-seat is broken for AI products.
Best practices if you stay per-seat
- Set per-seat pricing based on who actually does the most valuable work, not headcount
- Differentiate "viewer" vs "editor" tiers
- Use minimum-seat commits at the higher tiers to capture procurement budget
- Build usage caps into seat plans, soft warnings before billing impact
Pure usage-based pricing: scale-friendly but forecast-hostile
Where it wins
- The cost of serving a customer is variable (API calls, compute, storage)
- Your most valuable customers want to start small without seat commitments
- You're selling to developers or technical buyers comfortable with consumption mental models
- You compete with infrastructure where usage-based is the default
Where it breaks
- Bill shock kills retention. Customers churn when they get a bill 3x what they expected
- CFO buyers hate it. Variable bills make budget approval harder
- Revenue forecasting is hard. Your investors will ask for ARR; you'll have to convert usage to "annualized run-rate" with caveats
- Free riders. Generous free tier can be exploited at scale
Best practices if you go pure usage
- Hard spending caps customers can set themselves
- Email alerts at 50%, 75%, 100% of last month's spend
- Monthly commit discounts (commit $X/month, get 20% off vs pay-as-you-go)
- Detailed cost dashboards, customers want to know where the money goes
Hybrid (seat + usage): the dominant pattern for AI SaaS
The pricing model that most modern AI products eventually settle on. The structure:
- Base seat fee covers user access, basic features, and a reasonable usage allowance
- Usage tier applies to expensive operations (AI inference, file processing, high-volume API calls)
- Enterprise tier often includes negotiated commits and volume discounts
Notable examples in 2026: Cursor charges per-seat with usage-based limits on premium models. Linear charges per-seat for the core product but adds AI features as additional usage. ChatGPT Enterprise charges per-seat with generous-but-finite usage allowances.
Why it wins
- The buyer's psychology gets the predictable seat fee they expect
- Your unit economics get protection on power users
- You capture more revenue from customers who derive more value
- You can offer a fixed "all-you-can-eat" enterprise tier at a premium
Where it gets messy
Explanation. Hybrid pricing pages take longer to read and understand. Sales conversations get harder. Smaller customers, in particular, can struggle with two-axis pricing. Mitigation: surface the seat fee prominently, show usage limits on each tier, hide the per-unit overage details unless asked.
The framework: 5 questions to pick a model
- What's your variable cost per customer per month? If it's significant relative to your price, pure per-seat will eventually break.
- How predictable is usage from customer to customer? Wide variance favors usage or hybrid.
- Who's the buyer? CFOs prefer fixed bills, developers prefer usage, ops teams prefer tiered.
- What's your sales motion? Self-serve favors simplicity. Sales-led can support complexity.
- What's your competitor's pricing? Standing alone with weird pricing is harder than blending in then differentiating.
The pricing experiments that consistently move the needle
Three plays that ProfitWell's pricing research consistently shows raise revenue without churn:
- Raise prices on new customers, grandfather existing. Most teams underprice and never test the ceiling. A 15-25% price increase rarely impacts conversion meaningfully.
- Add a higher-tier "enterprise" plan with concierge onboarding and SLAs. Even if few buy it, it anchors your other tiers up.
- Move from monthly to annual default with 15-20% discount. Improves cash flow and retention.
The free tier question
Free tiers acquire users but cost real money. The 2026 wisdom is more nuanced than "always have one":
- Have a free tier if your product has viral or word-of-mouth distribution, your variable costs are low, and you have a clear conversion to paid
- Don't have a free tier if your variable costs are high (most AI products), your buyer is enterprise (they expect to pay), or you have strong outbound distribution
The replacement pattern that's growing: a 14-day free trial of the paid product, no free tier. Forces conversion conversations early. Works particularly well for B2B SaaS in the $40-$200/seat range.
What changed because of AI
The single biggest pricing innovation forced by AI:
"AI credits" or "compute units" as a quasi-currency abstraction over actual model costs, letting customers buy a budget that translates internally to whatever model the product routes their request to. This protects the customer from raw model price volatility and the SaaS vendor from underpricing.
OpenAI, Anthropic, and most AI-product layer companies now use this pattern. Expect it to become standard for any product where end-users trigger expensive backend operations.
The most common pricing mistakes founders make
- Pricing on cost-plus instead of value. Anchor on the customer's outcome, not your cost.
- Three tiers always, even when two work better. Two-tier (free/pro or pro/enterprise) often converts better than the default three.
- Ignoring annual contracts. Annual deals dramatically improve unit economics.
- Underpricing the entry tier to "be competitive." Race-to-the-bottom rarely wins. Differentiation wins.
- Never raising prices. Most SaaS products are 30-50% underpriced 2 years after launch.
What to do this quarter
- Run the unit economics math on your current top-decile customer. Are you profitable?
- Survey 10 customers on willingness to pay (use the Van Westendorp method).
- Test a 15-20% price increase on new signups for 30 days. Measure conversion impact.
- Add an "Enterprise, talk to us" tier even if you have nothing to put in it. Anchor your pricing.
For more on building SaaS products that justify the price you charge, see our SaaS development cost piece, knowing your build cost helps you price intelligently.
Working with DevEntia
We help founders build SaaS products with pricing baked in from day one, not bolted on after launch. Browse our services or tell us about your project.