EXECUTIVE SUMMARY
For a decade, B2B pricing expert Ulrik Lehrskov-Schmidt has lived in the weeds of monetization strategy. As the founder and CEO of advisory firm Willingness to Pay, Lehrskov-Schmidt and his team have steered hundreds of pricing projects and authored several seminal texts on software-as-a-service (SaaS) financial models. Yet, until recently, he deliberately withheld publishing a dedicated framework for artificial intelligence pricing.
The hesitation wasn’t born of a lack of interest; it stemmed from a missing puzzle piece. Across the market, traditional software giants were pivoting to generative AI, agentic workflows, and Model Context Protocols (MCPs). Simultaneously, they were watching their historically enviable gross margins—long comfortable at 85% to 90%—plummet into the mid-50s. Traditional pricing models could not account for this structural bleeding.
By April, however, Lehrskov-Schmidt crystallized a phenomenon he calls the "double hockey stick." In a comprehensive breakdown alongside SaaStr, he used real-world case studies from industry pacesetters like Figma, Clay, and Higgsfield to decode how AI has fundamentally severed the historically tidy relationship between software costs and customer value.
The core revelation is stark: AI has introduced extreme divergence. A microscopic fraction of users can rack up astronomical inference costs, while value creation itself follows a ruthless power law. Companies that fail to decouple cost management from value capture risk sliding into unprofitability. Conversely, those that master the dynamic interplay between cost spikes and value distribution can unlock unprecedented enterprise value—even accelerating top-line growth past the $1 billion ARR threshold.
DETAILED CHRONOLOGY: THE EVOLUTION OF SOFTWARE MONETIZATION
To understand why modern AI pricing is in a state of chaotic flux, one must trace the evolutionary path of software economics over the past twenty years.
The Era of Flat Costs and Uncaptured Value
In pre-AI B2B SaaS, the financial equation was remarkably forgiving. Companies routinely relied on flat pricing tiers—charging $5,000, $50,000, or $500,000 annually per tier—because the marginal cost of serving an additional user was practically negligible. Compute and infrastructure costs scaled predictably, and customer success overhead was broadly uniform.
This uniformity created a hidden structural flaw: companies routinely left millions of dollars on the table. Every product delivers vastly differing value depending on the buyer. Just as a bottle of water is virtually worthless at airport security but priceless to a lost desert traveler, enterprise software carries wildly asymmetric utility.
Lehrskov-Schmidt’s analytics consistently reveal that software value distribution follows a power law. At a typical mid-market company boasting a few hundred million in ARR and 10,000 to 20,000 customers, a single enterprise account can single-handedly account for 30% of total user volume or infrastructure utilization. Furthermore, the top 20% of customers typically capture 80% of the software’s value, with the top 20% of that elite cohort capturing 80% of the remainder.
Historically, because traditional B2B service costs remained flat, this massive dispersion of customer value manifested merely as an invisible opportunity cost. It never explicitly scarred the Profit and Loss (P&L) statement, granting founders the luxury of ignoring it.
The AI Disruption: Introducing the Second Curve
The advent of generative AI shattered this comfortable status quo by introducing a second, volatile curve: spiky costs.
Unlike traditional SaaS, where an extra database query costs fractions of a cent, AI inference requires heavy, resource-intensive compute cycles. Across dozens of high-growth AI pricing audits, Lehrskov-Schmidt’s team observed a striking rule of thumb: roughly 1% of users frequently generate 50% to 75% of total AI inference bills.
In a platform with one million active users, that statistical reality means a mere 1,000 individuals can absorb three-quarters of the underlying operational expenditures. One early-stage client learned this lesson the hard way, discovering that their least profitable customer was consuming 380 times more in backend token costs than the total revenue they generated for the platform. Operating under a flat-rate pricing scheme, the company was actively losing money on every single API call this power user executed.
The Four Cost-Value Combinations
To systematically navigate these dual dynamics, Lehrskov-Schmidt maps market strategies across a four-quadrant matrix defined by the intersection of cost behavior and value distribution:
- Spiky Value / Flat Cost (Classic B2B / Clay): Characterized by companies where a few power users extract massive business value, but underlying compute costs remain relatively manageable. The primary mandate here is better value pricing and credit structures to capture high-end utility.
- Spiky Value / Spiky Cost (Advanced AI Platforms / Figma): The most complex operational environment. Companies must implement sophisticated value-capture mechanics while simultaneously deploying strict cost guardrails.
- Flat Value / Flat Cost (Consumer Subscription / Netflix): Traditional media or basic utility models where every subscriber extracts roughly equivalent value and incurs uniform distribution costs. (Note: Should Netflix pivot to generating personalized, custom-rendered video streams on demand for every user, its cost curve would instantly transition to spiky).
- Flat Value / Spiky Cost (High-Risk Prosumer Tools / Higgsfield): The danger zone. Platforms that charge flat consumer fees while offering uncapped access to expensive underlying AI architectures.
SUPPORTING CONTEXT & METRICS: DECONSTRUCTING THE CASE STUDIES
Higgsfield: The Prosumer Trap of "Unlimited AI"
A glaring example of the "Flat Value / Spiky Cost" trap can be seen in prosumer creation tools like Higgsfield. Ahead of recent product overhauls, Higgsfield’s pricing architecture featured neat, predictable tiers at $19, $49, and $99 per month—with several advanced generation features explicitly labeled as "unlimited."
This design creates an existential economic mismatch: revenue per user is strictly capped by the subscription tier, while cost per user has no upper ceiling. Because most everyday users extract moderate, uniform value from the platform while a microscopic fraction of heavy power users execute continuous, resource-heavy generations, the platform inevitably absorbs punishing margins.
Lehrskov-Schmidt stresses that companies caught in this dynamic must implement strict fair-usage limits and throttling mechanisms. Without cost guardrails, flat pricing structures guarantee that heavy users will disproportionately bleed platform profitability.
Figma: Managing Margin Compression While Accelerating Beyond $1B ARR
Few companies have navigated the AI cost-value transition as publicly or successfully as Figma. Faced with the integration of powerful design automation features, Figma made the strategic decision to roll out AI capabilities broadly across its user base.
Predictably, initial usage was intense. Figma’s gross margins experienced a notable contraction, slipping from an elite 92% down to 82% as backend inference expenses spiked. However, rather than retreating, leadership rolled out a phased monetization strategy: introducing credit counting, structural enforcement, and finally, direct monetization.
The financial outcome validated the strategy. While gross margins dipped temporarily, revenue growth significantly outpaced underlying compute costs. Furthermore, the revenue distribution became healthier and more heavily concentrated at the top. Lehrskov-Schmidt argues that Figma always possessed a massive, uncaptured value curve; traditional per-seat pricing simply failed to reflect the true utility of design workflows. The sudden shock of AI costs provided the necessary catalyst to overhaul their monetization architecture, ultimately accelerating top-line growth well past the $1 billion ARR milestone.
Clay: Layering Action Credits Over Data Credits
B2B go-to-market data platform Clay faced a different balancing act. Characterized by spiky value and relatively flat infrastructure costs, Clay originally built its business model around selling data credits for lead enrichment and sales workflows.
Over time, however, management recognized that small accounts were extracting vastly more enterprise value than their baseline subscription prices implied, while enterprise accounts exhibited an even wider utility gap. By introducing action credits as a secondary credit pool alongside traditional data credits, Clay leveraged the market’s collective awareness of AI’s value to overhaul its pricing tiers. Today, Clay’s power users pay several orders of magnitude more than legacy rates, successfully aligning revenue generation with the underlying value curve.
OFFICIAL STRATEGIES: IMPLEMENTING THE "BRAKES AND KILL POINTS" FRAMEWORK
For companies suffering from margin compression due to runaway AI costs, Lehrskov-Schmidt’s advisory team rejects the simplistic advice that "costs should just be passed through to the consumer."
Cost and value are fundamentally independent variables. If a software platform turns $1.00 worth of compute tokens into a mere $0.10 of real-world customer value, no amount of creative pricing engineering will fix the underlying product-market mismatch. Forcing a price increase on a mismatched customer base results in immediate, catastrophic churn.
Instead, businesses must surgically manage their cost exposure using a structured, step-by-step operational playbook:
- Segment Users Granularly: Distinguish between onboarding cohorts (who may require temporary headroom to fully experience product-market fit before hitting constraints), flagged fraudulent actors, and standard production users.
- Establish a Hard Kill Point: Define the exact technical threshold at which compute-heavy tasks automatically halt—such as a specific ceiling on concurrent actions per second.
- Deploy Operational Brakes: Introduce intermediate friction points that gracefully degrade utility rather than enacting hard blocks. Examples include halving execution speed after a user crosses 1,000 automated edits per minute, capping concurrent sessions, or introducing mandatory cooldown wait times between heavy jobs.
- Communicate Transparently: Notify users of threshold breaches via targeted in-product messaging, email notifications, or system integrations.
- Slowly "Boil the Frog": Deploy the initial brake precisely at the aggressive kill point so it initially impacts only a microscopic fraction (e.g., 0.01%) of users. Gradually tighten the threshold toward 0.1%, and monitor how affected users respond. Typically, users fall into four distinct behavioral buckets: they keep executing tasks, voluntarily throttle their usage, churn, or upgrade their subscription tier.
Once user retention and upgrade mix stabilize at an acceptable threshold—affecting roughly 4% of the total user base—a company can typically resolve 60% to 80% of its overarching cost leakage. At that stage, a secondary, stronger brake can be layered behind the first until healthy gross margin targets are fully restored. Using this exact methodology, Lehrskov-Schmidt’s team has helped historically profitable companies that nosedived post-AI deployment recover up to 22 points of gross margin within weeks, entirely without raising headline prices.
FUTURE OUTLOOK: AGENTS, MCPS, AND THE NEXT FRONTIER OF MONETIZATION
As the software industry transitions from traditional chat-based interfaces to autonomous AI agents and Model Context Protocols (MCPs), pricing models must continue to adapt at breakneck speed.
The Death of Per-Agent Seat Pricing
Early attempts by software vendors to price autonomous AI agents identically to human employee seats—charging a flat monthly fee per agent—survived roughly five minutes in the wild.
Human productivity operates within a natural, predictable bell curve. While top-performing enterprise sales representatives routinely outperform peers, no individual human employee is a trillion times more productive than another. Autonomous agents, by contrast, operate with zero such constraints. A single well-architected agent can execute millions of complex enterprise transactions in the time it takes another to complete one. Consequently, counting the mere number of deployed agents tells leadership nothing about the economic value being generated. Future-proof pricing must anchor directly to functional output and verified business outcomes.
The MCP Evolution: From Read-Only to Write Access
As technical architectures shift toward MCPs—with infrastructure tools evolving rapidly from passive, read-only data fetchers to active, write-enabled executors—founders face immediate anxiety that direct protocol access will cannibalize legacy software add-ons.
Industry leaders advise treating early MCP rollouts as a deliberate, company-funded research phase. By following Figma’s iterative sequence—launching functionality openly, tracking granular usage telemetry, introducing precise credit meters, and eventually enforcing monetization structures—companies can safely discover where enterprise value truly resides before locking in permanent pricing tiers.
Conclusion: The Roadmap Ahead
The era of lazy, one-size-fits-all software pricing is officially over. As market leaders confront the realities of the double hockey stick, survival requires a sophisticated, dual-track mastery of both value capture and cost containment. For founders navigating this transition, the coming months will demand operational discipline, aggressive segmentation, and a willingness to tie revenue directly to the asymmetric reality of AI utility.
