From Seats to Tokens: Licensing AI Features the Right Way

Luke Hinchliffe

Head of Marketing at 10Duke.
Connect with me on LinkedIn.

AI features are no longer a differentiator, in fact they're fast becoming the baseline expectation for B2B software. For Independent Software Vendors (ISVs), the immediate challenge isn't building these features; it's figuring out how to charge for them and prevent the nightmare of revenue leakage.

The problem is that traditional licensing models weren't designed with AI in mind. A per-seat license made sense when the primary variable was who was using your software... but AI now changes that question to how much is being used, and the answer can fluctuate wildly from one user to the next.

This blog walks through the core licensing approaches available to ISVs embedding AI features into their products, and what to consider when choosing between them.

Why Traditional Seat Licensing Breaks Down for AI

Consider a design software ISV with 500 seats licensed to a mid-size engineering firm. Under a standard seat model, each user pays the same regardless of how they use the product. But now that the ISV has launched an AI-powered design assistant, usage looks very different across those 500 seats:

• A senior designer might run 200 AI generation requests per day
• A project manager might run 3

Both hold the same seat license. The ISV is effectively subsidising heavy AI consumption at the expense of margin, and has no visibility into which accounts are the heaviest consumers!

At scale, this becomes a serious revenue leakage problem, and it's one that seat licensing alone can't solve.

The Four Main Licensing Models for AI Features

Pricing models for AI features are still crystallising - Bessemer Venture Partners' analysis of AI monetization identifies several competing approaches in play simultaneously across the market. Here are the four that are most relevant to B2B ISVs.

1. Consumption-Based (Pay-Per-Use)

The most direct and simple approach: users are charged based on how much they consume. For AI features, that typically means per API call, per query, per generation, or per token processed.

Works best when: Your AI feature delivers discrete, measurable outputs - a document summary, an image generated, a data analysis run etc.

Watch out for: Unpredictable costs for end customers, which can create friction or abandonment. Heavy users may churn to competitors with flat pricing if your consumption rates feel punishing.

The fix: Consider including a base allocation with each seat, then charging for consumption above that threshold. This gives customers cost predictability while still capturing value from power users.

2. Feature-Based Licensing (AI as a Premium Tier)

Here, AI features are gated behind a higher product tier or add-on module. Customers on a base plan don't have access; those on a Pro or Enterprise plan do.

Works best when: Your AI feature is substantial enough to justify a meaningful price step-up, and you want to preserve a clear good/better/best packaging structure.

Watch out for: If AI features become ubiquitous in your category, keeping them gated may damage competitiveness. Customers may resist upgrading if the perceived value isn't clear.

The fix: Use license usage data to identify which base-tier customers are hitting limits or requesting AI functionality - those are your natural upsell signals.

3. Bundled AI Credits

A number of ISVs are moving to a credit-based system: customers purchase a bundle of credits (monthly, annually, or as a top-up), and AI features draw from that pool. It's a fast-growing approach - AI credit models grew 126% year-on-year in 2025, with a third of software companies planning to introduce them within 12 months. Credits can map to different actions at different rates: a simple query costs 1 credit, a complex generation could cost 10.

Works best when: You offer multiple AI features with different computational costs and want a unified currency for customers to manage.

Watch out for: Credits add cognitive overhead for buyers. If the mapping between credits and actions is confusing, it becomes a barrier rather than a value driver.

The fix: Keep the credit-to-action ratio simple and transparent. Publish a clear rate card. Make remaining credit balances visible in the product interface.

4. Usage-Capped Subscription

A flat subscription fee that includes a defined volume of AI usage per month. Usage beyond the cap either stops (hard cap) or triggers overage charges (soft cap).

Works best when: You want subscription revenue predictability while still accounting for AI cost variability. Works especially well for SMB buyers who prefer flat billing.

Watch out for: Hard caps that cut off access mid-workflow create very bad user experiences. Soft cap overages that surprise customers at invoice time could create churn.

The fix: Build in early warning notifications at 75% and 90% of cap usage, and give customers easy options to upgrade their tier or purchase additional allocation.

Mixing Models with a Hybrid Approach

Most mature ISVs don't land on a single model, they combine them. For example:

• Base seat license covers standard product access
• AI module is an opt-in add-on at the account level (feature-based)
• Within the AI module, a monthly credit allocation is included with overages available to purchase (usage-capped + consumption hybrid)

This gives you multiple revenue capture points, a clear upsell pathway, and cost predictability for customers - while still linking AI revenue to actual AI consumption. It's increasingly the norm: 43% of software companies already use some form of hybrid model, with that figure projected to reach 61% by the end of 2026.

Considerations for Your Licensing Infrastructure

Licensing AI features effectively requires capabilities that many in-house licensing systems simply don't have:

Real-time consumption tracking. You need to measure AI usage at the user or account level, in real time, and surface that data both to your ops team and to the customer. Batch reporting won't cut it.

Flexible entitlement structures. Your licensing system needs to support mixed models: seat counts, feature gates, and consumption quotas - ideally within a single account.

Automated enforcement. When a customer hits their AI usage cap, the system should enforce the limit or trigger an upsell flow without requiring manual intervention.

Usage visibility for customers. Customers who can see their own usage make better purchasing decisions and are less likely to be surprised by bills. Self-service usage dashboards reduce support overhead and improve trust.

If your current licensing system was built for perpetual or simple subscription models, it will struggle to support this level of complexity. That's the practical argument for a modern, cloud-based licensing platform that separates entitlement logic from your core product code.

Legacy Licensing Systems Weren't Built for AI 

ISVs who can't answer basic questions about their AI licensing are likely losing deals. Enterprise procurement teams now routinely ask questions like: "Do costs spiral if our team uses this heavily?" and "Can admins see usage before the invoice is received?" These are standard due diligence questions and not edge cases.

In-house licensing systems typically can't answer these questions. Built to count seats and enforce expirations, they simply weren't designed for real-time consumption tracking or self-service usage visibility. Retrofitting that capability can be very expensive and slow.

The ISVs gaining ground on AI are the ones who've made the value exchange legible, and that requires licensing infrastructure built for the complexity of AI, rather than patched to approximate it.

10Duke helps ISVs build flexible, cloud-based licensing models that support consumption tracking, feature gating, and real-time entitlement management - without rebuilding your product. Book a demo to see how it works.