AI features can grow revenue while quietly compressing gross margin. Here’s how SaaS finance and pricing leaders can price AI capability against actual delivery cost, not competitive pressure, and protect margin through discounting, contract, and renewal discipline.
Many SaaS companies launched AI features quickly because customers expected them, competitors were advertising them, and investors wanted an AI story. That urgency was understandable. It also meant AI features arrived with variable delivery costs that traditional SaaS pricing was never built to absorb. Token usage, inference requests, GPU workloads, vendor APIs, support burden, and infrastructure scaling can all vary significantly by customer, even within the same pricing tier. SaaS companies need to price AI features carefully to protect gross margins.
When pricing does not reflect those costs, AI features can grow revenue while compressing gross margin at the same time. Two customers can generate identical ARR and look equally healthy on a revenue report, while one consumes a fraction of the AI infrastructure the other does. Left unmanaged, that gap surfaces months later as a margin trend finance cannot immediately explain.
Pricing AI features well is not a matter of matching competitor packaging or including whatever feature the market expects. It has to be grounded in unit-level economics: what each AI feature costs to deliver, which customers use it heavily, how usage scales, and where pricing needs to change before margin erodes.
Why AI Feature Pricing Is Different
Traditional SaaS pricing worked because delivery costs were relatively stable. Per-seat licenses, tiered subscriptions, and flat annual contracts scaled predictably because infrastructure costs grew gradually alongside the customer base, and gross margins stayed in a narrow, attractive range. AI features break that pattern. Token consumption, inference requests, GPU workloads, API consumption, vector database usage, and orchestration complexity all scale with how a customer actually uses the product, not with seat count or contract value. A customer running an AI copilot for occasional summaries consumes a small fraction of the resources of a customer running autonomous agent workflows or generative reporting across an entire team, even when both pay the same monthly fee.
This is also why AI feature costs generally belong in cost of goods sold rather than operating expense: they exist because a customer is using the product, and they scale directly with that usage. Our article on what counts as COGS in SaaS and AI companies covers that classification in more detail, but the pricing implication follows directly from it. If AI delivery cost is variable and usage-driven, pricing built around a flat, uniform cost assumption will eventually misprice a meaningful share of the customer base.
The Pricing Mistake SaaS Companies Keep Making
The most common mistake is pricing AI features based on market pressure rather than profitability visibility. The reasoning usually sounds sensible in the moment: competitors are including AI in their packaging, so the feature gets bundled into existing tiers at little or no additional charge. That approach holds up until usage scales. Once token costs are effectively uncapped and customer behavior is unpredictable, what looked like a competitive advantage becomes a source of margin compression, particularly in enterprise accounts where a small number of power users can drive a disproportionate share of infrastructure cost. Finance teams should price AI features based on actual delivery costs.
Simon-Kucher’s Global Software Study 2025, based on responses from more than 500 software executives, found that many companies have already launched new AI capabilities, yet fewer than one in five report more than 10 percent revenue impact from those features. The gap is rarely the technology. It is usually a commercial model set by feature excitement rather than a clear view of usage, cost, and customer willingness to pay.
Start with Unit-Level Product Profitability
Before you price AI features, finance and product teams need a clear view of what they actually cost to deliver: token consumption by feature and by customer, infrastructure allocation, vendor and API fees, and the support burden the feature generates. Without that visibility, pricing decisions are guesswork dressed up as strategy. With it, teams can answer more useful questions: which AI features generate positive margin at current pricing, which customers consume resources disproportionate to what they pay, and where usage is scaling faster than the contract value meant to cover it.
This is the same discipline behind allocating AI spend to individual accounts, covered in our article on how to calculate cost per customer with AI spend. The allocation logic is not complicated, but it depends on connecting usage data to specific customers and products rather than reporting AI cost as a single aggregate line item.
Why Blended Margins Hide AI Pricing Problems
Most SaaS companies still evaluate profitability at the company or portfolio level. That view works well enough for stable, low-variance cost structures. It breaks down once AI usage enters the picture, because blended averages can make two very different accounts look identical.
| Metric | Customer A | Customer B |
| ARR | $75,000 | $75,000 |
| AI Usage Cost | $4,500 | $31,000 |
| Support Cost | $2,000 | $8,500 |
| Estimated Margin Profile | Strong | Weak |
Both customers generate the same ARR and would appear equally healthy on a standard revenue report. Once AI usage and support cost are allocated to each account, the margin profiles diverge sharply. Customer A is a strong account at its current price. Customer B is being delivered at a cost its contract does not support, and the gap will widen as usage grows. That does not mean the account should be walked away from. It usually means Customer B is a candidate for a usage cap, a different AI tier, or a renewal conversation grounded in actual consumption rather than the prior year’s terms.
Set Discounting Guardrails Before Deals Close
Discounting is not the issue by itself. The issue is discounting without knowing how AI usage, support burden, and infrastructure cost will affect the margin on that specific deal. A discount that looks reasonable against ARR alone can turn an already thin-margin AI tier into a loss-making account once usage ramps up.
The fix is straightforward in concept: tie discounting guardrails to gross margin thresholds, usage forecasts, and support expectations, not just deal size. Sales teams should know, before a deal reaches signature, the point at which a discount takes an account below the margin the business needs from that product line. That is a different question from whether the deal is large enough to be worth closing, and finance needs both questions asked before approval, not after.
Usage-Based Pricing and Contract Design
Unlimited AI usage is rarely sustainable once delivery cost varies meaningfully by customer. Most SaaS companies are moving toward some combination of usage thresholds, prepaid AI credits, consumption-based billing, overage fees, tiered AI access, and premium feature gating for the heaviest workloads. Paddle has argued that AI is pushing SaaS companies toward hybrid pricing models that combine a predictable subscription baseline with usage tiers or outcome-based differentiation, because flat seat-based pricing can undercharge power users who consume heavy compute. The companies protecting margin are generally the ones layering usage-based components on top of a stable subscription base, rather than leaving AI access flat across every account.
Contract language needs to keep pace with these pricing changes. Agreements written for stable, low-variance software delivery rarely account for variable AI consumption, infrastructure surcharges, or usage-based renewal adjustments. Updating standard templates to include usage thresholds, overage terms, and language addressing AI consumption growth gives finance a way to reprice accounts before margin erosion compounds, instead of after.
Renewals, Sales Incentives, and CPQ
AI pricing decisions do not stop at the initial contract. Renewal teams need usage data before they negotiate. If AI consumption grew materially during the term, an account may need a different tier, a usage cap, or a repriced renewal to preserve the margin the business originally underwrote. Treating renewals purely as retention exercises, without checking consumption trends, is how margin erosion compounds quietly across multiple contract cycles.
The same discipline needs to reach sales compensation and quoting. Commission plans built purely around ARR and bookings reward heavy-usage, deeply discounted deals that look identical to profitable ones on a pipeline report. Battery Ventures’ State of AI 2025 research highlights how inference and compute costs can compress margins at the AI application layer. That margin pressure is one reason pricing, discounting, and quote approvals need better cost context than ARR alone can provide. CPQ workflows should reflect that directly, incorporating margin floors, AI usage assumptions, and discounting guardrails so quotes carry the right cost context before they are approved, not after.
The Bottom Line
AI features can be a genuine source of revenue growth, but only when pricing reflects what they cost to deliver. Market pressure, competitor packaging, and feature excitement are not a substitute for unit-level economics. SaaS companies that understand what each AI feature costs, which customers drive disproportionate usage, and where pricing needs to change before margins erode will grow AI revenue without quietly subsidizing it.
Tools like Cogs’z can support this work by connecting AI usage, cost allocation, and profitability data at the customer and product level, but the larger discipline is organizational. Pricing decisions need to reflect the actual cost of delivery, and that requires finance, sales, and product working from the same view of margin rather than separate assumptions about what AI features are supposed to cost.
Brad Perry is the CEO of Cogs’z, a profitability management platform built for B2B SaaS companies. Brad co-founded DealerSocket, an end-to-end platform in the automotive industry, where he experienced firsthand the margin challenges that Cogs’z is designed to solve. Cogs’z automates customer-level cost allocation so finance, CS, and sales teams share a single, accurate, view of who’s profitable and why. Learn more or request a demo at cogsz.com.
References
- Simon-Kucher — Global Software Study 2025
- Paddle — 2025 State of B2B SaaS Pricing
- Battery Ventures — State of AI 2025