The features went out as a toggle on the existing plan. No standalone tier, no usage metric, no commercial logic. The question of how to charge for AI has been deferred three roadmap cycles in a row, and the inference bill keeps growing.
For Product teams · $20M–$75M ARR · 150–400 employees
We have three tiers but customers keep asking for features from the wrong tier.
We price per seat, but our best customers barely add seats — they just do more with each one.
We've got the usage data and the AI, but no framework connecting the two to revenue.
20% of your users open the AI daily and generate 10× the sessions of everyone else, on the same plan, at the same price as customers who've never touched it. The packaging was built for a product you no longer ship.
Infrastructure got built; the commercial design didn't. What costs how many credits, what a credit is worth, how limits map to tiers — all deferred. Buyers find it confusing and default to the cheaper flat tier.
LLM inference cost per power user is material and isn't reflected in price. A generous bundle on a compute-variable product is a margin hole that widens as usage scales.
Per-seat is easy to forecast but structurally capped when value comes from doing more per seat. NRR plateaus at 110% when correct metric design could reach 130%+.
A competitor launched outcome-based AI pricing and the CEO wants a response, but there's no framework to evaluate or design one, and no model of what automation does to seat demand.
Poor packaging logic doesn't stay contained — it distorts the whole system:
Free-to-paid conversion stuck at 1–3% when correct tier logic reaches 6–9%
Power users on the wrong tier either over-consume and feel throttled or under-consume and feel over-priced — both churn
Roadmap gets distorted by churn signals that are actually pricing signals — engineering ships the wrong fixes
Competitors who adopt usage metrics capture the upside from your power users and enterprise workflows
Every AI feature sits on a value chain from inputs to actions to outcomes. The further right you price, the more revenue tracks the value delivered. Most packaging is stuck at the input end.
Map credit cost to the value a job delivers rather than document length or raw compute, with tier limits that create natural, legible upgrade triggers.
An AI cost-to-serve analysis and pricing floor so AI features are priced to a margin target instead of eroding it unnoticed.
Use the usage data you already have, plus WTP research, to decide which AI features sit behind the paywall and which drive conversion.
Engagements that started where you are — usage data everywhere, commercial logic nowhere.
45,000 freemium users converted at 1.8% — AI features in the free tier were indistinguishable from paid. We moved the three features that correlated with conversion behind the paywall and added a 14-day full trial. Conversion more than doubled to 4.3% in three months.
Credits were priced at cost-plus with no link to value, and buyers couldn't understand them. We mapped credit cost to review complexity, built a spend-modelling tool and added pre-committed enterprise tiers with rollover rights. Conversion and ARPA both rose; NPS improved.
Three tiers with overlapping features buyers couldn't tell apart, and ASP flat for 18 months. We restructured packaging around three features power users valued, accelerated time-to-value and introduced targeted increases for new customers — a 17% ASP uplift with no material churn.
Ayon quickly established business relationships and created business models to support growth initiatives.
12–16 weeks · $20M–$75M ARR · 150–400 employees · NDA on request
Bring your current tier structure and a read on AI usage by cohort. You'll leave with a view of where the packaging is leaking value and the highest-impact change to your AI pricing architecture.
Not you? Executive view · Growth view · Finance view · AI-native view
