The AI Pricing Model Question for SaaS

Every SaaS company with AI features eventually faces the same question: include it in the base plan, put it behind a premium tier, or charge for consumption?

There's no universal right answer. But the wrong answer is expensive, and more companies have discovered that the hard way after shipping and pricing AI features.

This article is for founders and revenue leaders who are actively working through this decision. Not a vendor survey. A framework for reasoning through the tradeoffs given your specific product, market, and cost structure.

The three pricing models for AI in SaaS

The patterns that have emerged in the market break down into three distinct models, each with a different core logic.

Model 1: Bundled into existing tiers. AI is included with the base plan or standard tier. Users don't pay more to access it. The bet is that AI drives engagement and retention, which protects revenue through lower churn even if it doesn't grow average revenue per user (ARPU) directly.

Model 2: Premium tier or add-on. AI is available at a higher price point, either as a separate add-on or as a differentiating feature of a higher tier. The bet is that AI delivers enough demonstrated value that users will pay more for it, or that AI unlocks a new buyer persona willing to pay at a different price point.

Model 3: Usage-based or consumption pricing. AI access is priced by use, whether measured in API calls, tokens generated, queries executed, or outputs produced. The bet is that AI value correlates with usage, so customers who get more value pay more.

Each model is internally consistent. Each has situations where it's clearly the right choice and situations where it backfires.

Key Facts: AI Pricing Models in SaaS

• 68% of SaaS vendors restricted AI features to premium tiers in 2025, while 37% planned pricing adjustments within 12 months as competitive pressure built toward bundling (Getmonetizely, 2025)
• By 2025, 85% of SaaS leaders had adopted usage-based or hybrid pricing models, with 61% using hybrid pricing that combines a base subscription with usage-based AI components (Flexera, 2025)
• 78% of IT leaders experienced unexpected charges on a SaaS bill due to consumption-based or AI pricing models, highlighting the forecasting problem with flat-price AI bundles (Zylo, 2025)

The 4-Model AI Pricing Decision

The 4-Model AI Pricing Decision is a sequential evaluation framework that maps each SaaS company's AI feature to one of four pricing structures. Bundled: AI is included at all paid tiers; optimizes for adoption and retention rather than ARPU. Add-on: AI is a separately priced module; appropriate when AI delivers a demonstrably different capability from the base product. Usage-based: AI is priced by consumption (tokens, queries, outputs); best for developer tools and API products where value correlates with usage volume. AI-tier: a new pricing tier defined by AI capability ceiling rather than seat count; defensible when the AI tier enables measurably different outcomes, not just faster execution. The decision sequence runs: adoption viability, then retention impact, then cost structure, then competitive context.

Bundling AI: the retention argument

The most important thing to understand about bundling AI is that it's primarily a retention decision, not a revenue decision.

When AI is bundled into the base tier, two things happen. First, every user encounters the AI feature in their normal workflow. Adoption is high by default because there's no friction, no upgrade decision, no separate onboarding loop. Second, as users build habits around the AI feature, the product becomes stickier. Churn decreases because switching away from your product means giving up an AI workflow that's embedded in how they work.

Notion's approach illustrates this. When Notion AI launched, it was priced as a separate add-on at $8 per user per month. Adoption was moderate. In 2024, Notion moved to include AI in all paid plans. Adoption increased sharply. More importantly, users who use Notion AI as part of their daily writing workflow are meaningfully less likely to churn. The AI became a retention asset, not just a revenue line.

Figma took a similar path. AI capabilities were woven into the product experience rather than gated. The result is that AI isn't something Figma users think about purchasing. It's just part of using Figma.

The case for bundling is strongest when: your AI feature is embedded in a high-frequency workflow, your competitors are moving toward bundling (making premium gating a competitive disadvantage), and your primary risk is churn rather than ARPU. AI features as product: where to add them explains how to identify the high-frequency insertion points that make bundling defensible.

The risk of bundling is that LLM API costs are real and scale with usage. If your AI feature gets high adoption and your cost-per-active-user goes up by $5 per month but your ARPU doesn't change, you've compressed your margin. The bundling decision requires a careful cost model before committing. How much headroom do you have before the cost math breaks?

Premium tier: the revenue argument

Premium AI pricing is defensible when the AI feature delivers demonstrably different outcomes, not just faster execution of the same workflow.

GitHub Copilot is the clearest example. The individual tier at $10 per user per month is the standard entry point. GitHub Copilot Enterprise, at $39 per user per month, adds features like custom model fine-tuning on your codebase, policy controls for enterprises, and deeper integration with enterprise GitHub features. The higher price is justified by a different buyer persona (enterprise with security requirements) and a demonstrably different capability set (codebase-specific context, not just general code completion).

That's the model that works for premium AI pricing. There's a capability cliff between tiers, not just a label.

The premium model fails when it's used to gate features that should be in the base plan. If your AI feature is genuinely a workflow accelerator for daily tasks, putting it behind a premium forces a decision that most users won't make. They don't upgrade. They just work without the AI, and the habit never forms. When they're evaluating renewal, the AI feature was never part of their daily experience, so it doesn't register as a reason to stay.

HubSpot learned a version of this lesson. Earlier iterations of HubSpot AI features were gated behind higher Enterprise tiers. The adoption data showed that users who never encountered the features were less likely to expand. More recent HubSpot product decisions have moved toward making AI foundational across tiers, with advanced AI for more complex use cases at higher tiers. The tiering logic shifted from "pay to access AI" to "pay for more sophisticated AI."

Salesforce Einstein Copilot is priced at $50 per user per month on top of existing Salesforce licenses. That's a significant additional cost for large enterprise users. Salesforce can hold that price because: enterprise buyers are accustomed to high Salesforce spend, the features are genuinely differentiated from standard Einstein Analytics, and the buyer persona (enterprise sales operations) has clear ROI metrics to point to.

Premium AI pricing works when you can answer "what outcome does the AI tier enable that the base tier can't, and what's the dollar value of that outcome?" If you can't answer that clearly, the pricing tier will struggle. 5 Dimensions of AI ROI provides the framework for quantifying what an AI tier actually delivers in measurable business outcomes.

Usage-based: the value alignment argument

Usage-based AI pricing aligns the price with the value delivered, at least in theory.

Stripe Sigma charges for query execution. OpenAI's API pricing charges per token. Salesforce Einstein features have usage-based components for AI predictions and generations. The logic is clean: customers who run more queries, generate more outputs, or make more AI-assisted decisions presumably get more value, so they pay more.

The practical challenges are real.

First, usage is hard to predict. Enterprise buyers in particular dislike variable costs that are difficult to budget. A fixed annual commitment is easier to approve than a monthly invoice that depends on how much their team uses the AI. Usage-based pricing can slow enterprise deal cycles and increase the frequency of conversations about cost management.

Second, the correlation between usage and value isn't always tight. A team that runs fifty AI queries per month and makes one high-quality decision from the output might get more value than a team that runs five hundred queries and treats the outputs as noise. Usage doesn't measure outcomes.

Third, usage-based pricing creates a behavioral dynamic where users think before using the AI, which is the opposite of what you want for habit formation. The cognitive overhead of "is this query worth running" reduces adoption at the margin.

Usage-based pricing works best for developer tools and API products where the buyer is technical, comfortable with variable billing, and has a clear consumption model to work with. It's harder for horizontal SaaS products where end users aren't thinking in terms of API calls. a16z's analysis of AI pricing models finds exactly this split: AI-native API products lean toward usage-based, while human-facing SaaS products tend to retain subscription or bundled structures because usage-based billing creates cognitive friction that suppresses adoption.

Competitive dynamics

Your pricing isn't set in isolation. It's set in a market.

If your top three competitors have bundled AI into their base plans, you can't effectively premium-gate AI without losing trials. A prospect evaluating four CRM options where three include AI and yours costs $X more per user per month for AI will consistently choose one of the three. Not because your AI is worse. Because the mental accounting of "extra cost for something the competition includes" creates friction at the comparison stage.

Conversely, if nobody in your market has bundled AI yet and customers are accustomed to thinking of AI as an add-on, early premium pricing can work. You're capturing revenue from early adopters who place high value on the feature before the market norm shifts to bundling.

The competitive dynamic that most SaaS companies are underestimating right now: the shift from AI as a premium feature to AI as a baseline expectation is happening faster than pricing teams are adjusting. What justified a premium tier in 2023 is a bundled expectation in 2026. The window for premium AI pricing in most horizontal SaaS categories is narrowing. OpenView's research on usage-based pricing shows that 38% of SaaS companies now use some form of usage-based pricing, up from 27% in 2023, and that public usage-based companies outperform the broader SaaS index on net revenue retention (NRR), which suggests competitive pressure on pricing structures is increasing across the category. The AI arms race in SaaS documents how competitive pressure is compressing these pricing windows.

The cost structure problem

LLM API costs are real and they're not fixed.

A typical GPT-4 class API call costs roughly $0.01 to $0.05 depending on input/output length and the specific model. If your AI feature serves 10,000 active users and each user makes 20 AI-assisted actions per month, you're running 200,000 API calls per month. At $0.02 average, that's $4,000 per month in LLM infrastructure cost, or roughly $0.40 per active user per month.

For most SaaS products, that's absorbable. But it scales with active usage, not with seats. If you sell 10,000 seats but only 2,000 are active, the 2,000 active users drive your costs. If adoption improves to 8,000 active users, your costs quadruple, but your revenue might not change if AI is bundled.

Before committing to a bundled AI pricing model, you need a realistic cost projection:

• What's the estimated AI API cost per active user per month at current usage patterns?
• What's the projected cost at 2x adoption and 5x adoption?
• Does bundled AI pricing still work at those adoption levels?

The companies that get this wrong are the ones that launch bundled AI at low adoption levels where costs are negligible, then find themselves compressing margin six months later when the feature takes off.

Some SaaS companies address this with soft usage caps: "AI features included, reasonable use, enterprise-level usage on request." This is pragmatic but creates ambiguity that customers notice.

The cannibalization question

Some SaaS companies fear that their AI features will automate away value they currently charge per-seat for.

This fear is most acute in products where the value proposition is partly "give each user their own workspace." If AI can do the work of five users, why are you paying for five seats?

The honest answer is that this cannibalization risk is real in some products and minimal in others. For products where the primary value is collaboration and shared context across humans, AI augments the workflow rather than replacing the humans. For products where the primary value is individual task execution, the risk is higher.

The defensive move isn't to avoid building AI. It's to make sure your AI features strengthen the collaboration use case rather than enabling individual users to do more with fewer seats. Features that surface insights across the team, support handoffs, and improve coordination are both more defensible strategically and more difficult to replace with standalone AI tools.

"Premium AI pricing is defensible when you can answer the question: what outcome does the AI tier enable that the base tier cannot, and what is the dollar value of that outcome? If you can't answer that clearly, the pricing tier will struggle. Bundling is defensible when AI retention impact is real and competitors are moving toward inclusion." (Rework Analysis, 2025)

"The behavioral economics of usage-based AI pricing create a dynamic where users think before using the AI, which is the opposite of what you want for habit formation. Flat pricing removes the cognitive overhead. Usage pricing adds it back. For horizontal SaaS with human end users, that overhead suppresses adoption at the margin." (Rework Analysis, based on a16z AI pricing research, 2025)

AI Pricing Model Comparison

Sources: Bessemer Venture Partners AI Monetization Playbook 2025, a16z AI Pricing Models Research 2025, Getmonetizely Pricing Guide 2026

Rework Analysis: The window for premium-gating AI in mid-market SaaS is narrowing. What justified a separate AI add-on in 2023 is a bundled expectation in 2026. Teams pricing AI as an add-on in categories where the top three competitors have bundled should model the trial-to-paid conversion gap against the ARPU gain from add-on conversions. If bundling reduces churn by 5 percentage points, the NRR math typically beats add-on pricing unless add-on conversion exceeds 35%. Most horizontal SaaS add-on adoption falls well below that threshold.

What the 2025-2026 market tells us

Looking at the major SaaS platforms, a pattern is emerging.

Linear includes AI features in all paid plans. No separate AI tier. The bet is that AI-assisted issue creation and summarization are core to the daily developer workflow.

Notion moved from add-on to bundled. Usage data drove the decision.

GitHub Copilot maintains a tiered model with a clear capability differentiation between Individual and Enterprise. The tiering is justified by demonstrated outcome differences.

HubSpot is moving AI deeper into the product across tiers, with more sophisticated AI reserved for higher tiers, but basic AI broadly available.

Zendesk includes AI features at all tiers with usage caps, premium AI agent volume available at higher tiers.

Salesforce maintains premium AI pricing at the Enterprise level where the buyer has high willingness-to-pay and clear ROI metrics.

Rework bundles AI capabilities as part of the product tiers rather than gating them separately, keeping the pricing straightforward around the Starter and Standard packages. This fits the team-oriented use case where AI compound value comes from shared context across users.

The pattern: AI is moving toward a baseline expectation across mid-market SaaS. Premium AI pricing is holding primarily at enterprise level where the capability differentiation is genuine and the buyer is accustomed to add-on pricing.

The decision framework

There's no universal AI pricing answer. But here's the analytical sequence:

Start with adoption. If AI adoption is low, the question isn't pricing. It's insertion point. Low-adoption AI features don't justify premium pricing and don't benefit retention regardless of how you price them.

Then retention impact. Does the AI feature, when used, correlate with lower churn? If yes, bundling protects that value. If the correlation is weak, premium pricing is more defensible because you're not leaving a retention lever on the table.

Then cost structure. At projected AI adoption levels, what's the cost per active user per month? Can you absorb that in the current plan margins, or does the math require either usage-based pricing or a premium tier to stay viable?

Then competitive context. What is the market expectation? If competitors are bundling, you need a strong argument for why customers will pay extra for yours.

Work through those four questions honestly and the pricing model usually becomes clear. The companies that skip straight to "what will customers pay for this" often end up with a pricing structure that works in the short term and creates problems when competitive dynamics shift. a16z notes that AI is now driving a shift toward outcome-based pricing as AI-native companies like Decagon start pricing per-resolution rather than per-seat, which suggests a fourth pricing model emerging beyond the three covered here, one that will eventually pressure traditional SaaS tiers more directly.

---

Learn More:

• 5 Dimensions of AI ROI: quantifying AI tier value in terms buyers and boards will accept
• AI Features as Product: Where to Add Them: finding the high-frequency insertion points that justify bundling
• The AI Arms Race in SaaS: Speed to Ship: how competitive pressure compresses AI pricing windows
• Buy vs. Build for SaaS AI Features: the build/buy decision affects cost structure and pricing options
• How AI Reshapes the SaaS Operating Model: broader context for AI's impact on SaaS business models