Why 92% of your new customers may never use your AI features—and how to fix it
Only 8% of new customers adopt AI features priced as add-ons.
That's the finding from SBI's 2025 State of SaaS Pricing Report—and it should fundamentally change how you think about AI monetization. Their research found that when AI is sold as a separate add-on, only 20% of net-new customers purchase it, and only 38% of those buyers actually use the AI features. Meanwhile, Growth Unhinged and PricingSaaS tracked over 1,800 pricing changes across 500 SaaS companies in 2025 alone—an average of 3.6 changes per company. The message is clear: everyone is scrambling to figure out AI pricing, but most are getting it wrong.
The core problem? Teams are treating AI like a premium upsell instead of a habit to be formed. This post breaks down what's actually working—and provides a practical framework for pricing AI features and agents without destroying adoption or eroding margins.
The AI Pricing Trilemma
Pricing AI features is uniquely challenging because you're navigating three competing forces simultaneously:
Variable costs: Unlike traditional SaaS where marginal costs approach zero, AI has real per-use costs—inference, retrieval, training, and tool calls all consume compute.
Non-linear value: AI value can be unpredictable. A single AI-generated insight might save hours of work—or produce nothing useful. Customers struggle to predict what they'll get.
Autonomous work: Agents operate without a human actively using the UI, breaking the fundamental logic of seat-based pricing.
The result: many vendors are moving toward credits and tokens, rebundling AI into core plans over time, and adopting "adoption-first" packaging that gets customers using AI early—then monetizes power usage and advanced autonomy later.
Credits as a Bridge, Not a Destination
Credit-based pricing has exploded. According to the PricingSaaS 500 Index (reported by Growth Unhinged), 79 companies now offer a credit model, up from 35 at the end of 2024—a 126% year-over-year increase. Household names like Figma, HubSpot, and Salesforce have all adopted credits.
As Metronome notes, "credits are a bridge, not a destination"—useful when costs are known but value isn't. But getting credit design right requires careful decisions:
What Credits Should Buy
Map credits to an "AI event" (summarize a document, generate a draft, run an agent workflow) or to a cost-driver proxy (tokens, seconds, tool calls). The key is keeping the mapping understandable and auditable—customers must be able to reconcile their usage to their bill.
How Credits Should Be Allocated
A common adoption-friendly pattern is per-user monthly credits plus a shared organizational pool. Box AI pioneered this model: each user receives 20 credits per month, with an org-level shared pool of 2,000 additional credits for power users, and options to purchase additional blocks. This reduces internal friction ("who gets to use AI?") while still controlling costs.
Why Fungibility Matters
If customers aren't sure which AI features or agents will be valuable, transferable credits let them experiment without renegotiating packages. Fungible credits can span multiple agents ("agent credits") so customers can shift spend to what's working. For vendors, prepaid credits improve revenue predictability, pull cash forward, and create levers to steer behavior through incentives and bundles.
Key insight: As Metronome's research emphasizes, "customers don't know what a credit does"—which is why dashboards, cost previews, rollovers, alerts, and overage caps are essential to prevent trust erosion. Credits work as a transitional architecture, but durable pricing strategies eventually anchor to value drivers customers can understand and forecast.
The Rebundling Playbook: Add-On → Included → Segmented
A common lifecycle is emerging for AI monetization:
Phase A: Add-on or Credit Pack (Early Market)
Price AI separately to learn willingness-to-pay, manage compute risk, and avoid repricing your whole product too early. Credits are especially useful here because they're flexible and fit uncertain value. However, this is where that 8% adoption problem bites hardest.
Phase B: Include AI in a Premium Tier
As AI becomes table stakes, vendors often simplify: include AI in a higher plan to reduce friction and boost usage. Box made AI features free for Enterprise Plus subscribers after previously charging consumption fees. Notion followed suit in May 2025, discontinuing its $8-10/user AI add-on and bundling unlimited AI exclusively into Business ($20/user/month) and Enterprise tiers.
Phase C: Re-segment by Intensity and Sophistication
Once adoption is high, you can reintroduce differentiation without killing usage. Keep "everyday AI" included (summaries, basic drafting) and monetize higher volume (more credits), advanced models and tools, governance and compliance features, and agent autonomy levels.
Pricing AI Agents: The Layer Cake Framework
Seat-based pricing breaks when agents create value independently of human users. Agentic AI pushes pricing toward consumption and value metrics. Ibbaka developed a practical framework called the "Agentic AI Pricing Layer Cake" with four components:
1. Role: What job does the agent do? This defines the value proposition and anchors customer understanding.
2. Access: A retainer or platform fee to ensure availability and cover baseline costs. This provides revenue predictability.
3. Usage: Charges based on runs, tasks, tool calls, tokens, or minutes. This aligns cost with consumption.
4. Outcomes/Performance: Success-based premiums when results are measurable and controllable. Intercom's Fin AI agent demonstrates this model—charging $0.99 per successful resolution, with the company reporting an average 60% resolution rate across customer deployments.
This framework lets you start with simpler usage pricing, then evolve toward outcome components where you have credible measurement.
Adoption-First Packaging: Monetize After the Habit Is Formed
"Adoption-first" means intentionally designing packaging so customers use AI early and often—because AI value is learned in-workflow, not in demos or sales decks.
Tactics that work:
Include a baseline allowance of AI usage in core plans—enough for real workflows, not just a demo.
Add a shared pool so teams don't fight over who gets AI access.
Make top-ups simple (blocks of credits) rather than forcing an upgrade for every spike.
Keep packaging aligned to the metric: "Start with the pricing metric, then design packaging to support it." Misalignment suppresses both adoption and revenue.
The Trust Imperative: Why Pricing Without Trust Fails
Here's what many companies overlook: complex pricing models create customer anxiety. Research from Forrester indicates that 62% of B2B buyers report that unclear pricing is a major factor in abandoning a purchase process entirely. Meanwhile, around 45% of B2B buyers say unclear pricing is their biggest frustration (per Sopro's B2B buyer research). Pricing anxiety—fear of unpredictable costs, bill shock, or being taken advantage of—suppresses adoption regardless of how much value your AI delivers.
Building pricing trust requires investment in three areas:
Transparency: Real-time usage dashboards, detailed billing breakdowns, and "value receipts" that link charges to outcomes. Customers should never be surprised by their bill.
Predictability: Committed-use pricing options, usage caps, and proactive alerts. OpenAI's soft and hard usage limits, and Hugging Face's usage cap tools, demonstrate this pattern in practice.
Control: Self-service usage controls, budget alerts, and clear upgrade/downgrade paths. Customers need to feel they can manage costs without calling support.
A Framework for Action
Here's a practical sequence for AI monetization decisions:
Step 1—Classify your AI capabilities: Map each AI feature on two dimensions: Autonomy (how independently does the AI operate?) and Attribution (how clearly can you connect outcomes to the AI's work?). This determines which pricing model fits.
Step 2—Quantify value: Identify which value drivers your AI delivers: revenue enhancement, cost reduction, productivity increases, risk reduction, working capital efficiency, or capital investment deferral. Translate these into dollarized benefits.
Step 3—Design your pricing architecture: Determine which layers of the pricing stack to implement based on your capabilities, customer segments, and competitive position.
Step 4—Build trust infrastructure: Implement transparency, predictability, and control mechanisms before launching new pricing.
Step 5—Execute with care: Phased rollouts, "no worse off" guarantees for existing customers, and 60+ days notice for pricing changes. Trust, once broken, is extremely difficult to rebuild.
The Bottom Line
The question isn't whether to monetize AI—it's whether your pricing accelerates adoption or kills it.
The companies winning at AI monetization share common traits: they treat credits as a bridge (not a destination), follow a deliberate rebundling lifecycle, price agents on work performed rather than seats occupied, and build trust infrastructure before rolling out complex pricing.
Most importantly, they prioritize adoption first. Because an AI feature that customers don't use—no matter how cleverly priced—generates zero value and zero revenue.
Sources
SBI/Price Intelligently, 2025 State of SaaS Pricing Report (Part 1)
Growth Unhinged / PricingSaaS 500 Index (Kyle Poyar & Rob Litterst, February 2026)
Metronome, "The Rise of AI Credits" and "AI Pricing in Practice: 2025 Field Report"
Ibbaka, "Agentic AI Pricing Layer Cake" framework (April 2025)
Intercom Fin AI Agent pricing documentation
Box AI pricing and Enterprise Plus announcements (TechCrunch, TechTarget)
Notion 2025 pricing changes documentation
Forrester and Sopro B2B buyer research on pricing clarity
