A new pricing challenge has emerged for SaaS companies leveraging AI: about 1% of users are generating 75% of AI-related costs, forcing firms like Figma, Clay, and Higgsfield to rethink how they capture value and manage expenses.
- 1% of users generate 75% of AI inference costs in high-growth SaaS products.
- Traditional flat pricing fails to capture the vast value spread and spiky costs.
- Solutions include usage-based guardrails and value-aligned pricing tiers.
What happened
Ulrik Lehrskov-Schmidt, CEO of Willingness to Pay and a veteran in B2B pricing, has uncovered a striking pattern in AI-driven SaaS products: a very small segment of users, about 1%, accounts for the vast majority—up to 75%—of AI compute costs. His research and consulting work with companies like Figma, Clay, and Higgsfield revealed that flat pricing models no longer align with the value customers derive or the costs incurred. For example, some users who generate extremely high token usage were paying the same fees as more typical users, leading to unsustainable losses.
This phenomenon is linked to what Lehrskov-Schmidt calls the 'double hockey stick'—a steeply skewed distribution where both user value and costs follow power laws rather than uniform consumption or value models. Traditional plans with fixed monthly fees or limits often fail to differentiate between heavy and light users, resulting in margin erosion, unexpected high spend, and missed revenue opportunities in AI-powered products.
Why it matters
The disconnect between AI usage costs and pricing directly impacts SaaS companies’ profitability and growth trajectories. If the few power users are not priced in accordance with the value they extract or the cost they impose, companies risk losing money on high-demand customers or stalling revenue growth by capping pricing at low flat fees. The stakes are increasingly high as AI model costs fluctuate and customers innovate on usage patterns.
Lehrskov-Schmidt’s analysis highlights the importance of recognizing and pricing for heterogeneous customer value. Whereas demand for a product’s feature may be optional or critical depending on the client, AI compute costs scale steeply for some. This complexity necessitates pricing models that track customer outcomes and usage metrics paired with consumption guardrails to control runaway costs without alienating key accounts.
What to watch next
SaaS companies should closely monitor their AI cost curves relative to user value and adjust pricing accordingly. Strategies could include introducing scaled pricing tiers, usage caps, or overage fees tied to AI consumption metrics. Monitoring the evolving 'dance' of cost and value curves—shaped by AI efficiency improvements, pricing shifts, and customer behaviors—will be critical.
Companies like Higgsfield, Figma, and Clay are testing these approaches, aiming to balance revenue capture without triggering churn among power users. Decision-makers should also consider transparent communication about usage policies and cost drivers to manage customer expectations. The ability to implement flexible, data-driven pricing and conservation guardrails is emerging as a key competitive advantage in the AI SaaS space.