I looked for this framework in the public conversation about AI pricing for a year. It does not exist. The strategic decision around what to do as inference costs fall under your priced outcomes is one of the most consequential CPO and CFO decisions of the next 24 months, and there is no public playbook. So here it is.
Why the cliff is real
The math is not subtle. Token costs dropped from around $10 per million tokens in 2024 to around $1 per million in 2026, and the industry trajectory suggests around $0.50 per million by 2027. If your outcome price is $0.50 per resolution and inference is 25% of revenue today, in 24 months inference will be 5 to 8% of revenue. Gross margin rises from 60% to 80% or more if you hold price. That is a 20-point margin lift, and it sounds great. Until the competitor with 5% inference cost prices at $0.30 per resolution and offers free trial conversion. Now you are sitting at $0.50, your customers are looking at $0.30, and the cliff hits.
The three scenarios
Scenario A, hold prices, works for a differentiated product with sticky customers, low competitor threat, and high switching costs, such as enterprise products with long contracts. Revenue stays stable and gross margin rises 20-plus points over 24 months. The risk is a 12-month-out competitor undercut you cannot respond to without admitting your old price was inflated.
Scenario B, pass full savings, works for high-volume, price-sensitive products in competitive markets at growth stage. Revenue per outcome drops with cost, margin stays at trough level, and volume grows because customers can afford more. The risk is volume not compensating for the price drop.
Scenario C, pass 50% of savings, is the pragmatic middle and the most common choice. It captures half the margin and halves the competitive exposure.
How to plan for it
Four moves. Build the 24-month model now, not at month 12, using three inputs: current gross margin per outcome, the inference cost trajectory (industry-wide around 50% per year, plus another 10 to 25% if your routing improves), and the competitor pricing trajectory. Track three competitor prices monthly and set a threshold, for example any direct competitor pricing more than 25% below you, that triggers a price review. Decide which scenario you are running, write it down, and get CFO sign-off. Then communicate the decision to sales, because reps making one-off discounts in the field is the number one way the cliff strategy gets undermined.
You will be wrong about competitor pricing, so track actual prices monthly and update the model. The Jevons cliff is dynamic, and "set and forget" is the failure mode. Most teams price once at launch and do not revisit until a customer churns to a competitor eight months later, by which point the cliff already happened. Build the cliff model on day one of pricing strategy this week.