--- title: "Five Credit-Based Pricing Patterns, Pulled From Our Data" description: "Pricing strategists Steven Forth and Michael Mansard ran the Tanso AI Pricing Finder through their design framework. Across 73 companies, five execution patterns emerged, and one clear anti-pattern." date: 2026-06-17 author: Kat Laszlo canonical: https://tansohq.com/blog/credit-pricing-patterns --- # Five Credit-Based Pricing Patterns, Pulled From Our Data By [Kat Laszlo](https://www.linkedin.com/in/katrinalaszlo/) ยท June 17, 2026 We built the [AI Pricing Finder](/ai-pricing-finder) to track how companies actually price AI products. This month, pricing strategists Steven Forth and Michael Mansard ran the full dataset through their own design framework and [published the results](https://open.substack.com/pub/pricinginnovation/p/credit-based-pricing-execution-patterns). It is the closest thing we have to an independent audit of the data, so I want to share what they found and what it means. --- ## What they looked at The dataset covers 73 companies across 18 categories, captured in June 2026. It is AI-heavy by design: 47 of the 73 (64%) sit in one of seven AI-specific categories (AI API, AI Coding, AI Design, AI Productivity, AI Video, AI Voice, AI Search). For each company we record the real pricing choices, not the marketing language: the revenue model, the value metric, what happens when a customer runs out of allowance, whether unused credits roll over, and whether credits are pooled across a team or locked to a seat. The revenue mix alone tells you the market is still in transition: | Revenue model | Companies | Share | | --- | --- | --- | | Subscription | 39 | 53% | | Consumption / usage | 22 | 30% | | Seat expansion | 10 | 14% | | Transaction fees | 2 | 3% | More than half the companies are still subscription-first. Only about a third have moved to consumption revenue, which is where credit-based design tends to be most fully realized. --- ## The five patterns Forth and Mansard sorted the 73 companies into five execution patterns, scored against a five-dimension index of how credit-native each design is. These archetypes are their reading of the data, not labels we ship in the dataset, but they map cleanly onto the choices we track. | Pattern | Share | Signature | | --- | --- | --- | | Consumption-Native | 30% | Usage revenue, soft caps, pay-as-you-go, pooled. The most credit-native group (OpenAI, Anthropic, Snowflake, Databricks). | | Credit-Subscription Hybrid | 14% | Subscription base with abstract credits and generous lifecycle policies (ElevenLabs, Replit, Windsurf, Clay). The vanguard. | | Subscription-Seat | 36% | Traditional SaaS applied to AI: per-seat, hard cap, no rollover. The largest group, and the one most exposed. | | Seat-Led | 14% | Legacy B2B seat expansion (Salesforce, HubSpot), now layering AI agent credits on top of a seat-priced core. | | Subscription-Pooled | 4% | Subscription revenue with pooled credits but no full consumption pricing (Clay, n8n). The smallest and most transitional. | The headline they drew from this: only 23 of the 73 companies (31%) score in the top band of their credit-native index, and those are overwhelmingly API and infrastructure vendors. Most application-layer AI products are still in transitional or "credits in name only" designs. --- ## The anti-pattern: hard caps The clearest finding is also the most actionable. In our data, 45% of companies (33 of 73) stop usage cold when credits run out. No top-up prompt, no overage, no soft landing. Forth and Mansard call this the worst possible experience for an agentic workflow, because an agent can be halfway through a task when the meter hits zero. Hard caps cluster with subscription revenue and per-seat allocation. They are not random mistakes, they are the rational default of the seat-based model. The fix is not a single setting, it is a move toward hybrid design: pair the cap with auto-top-up and an alert, and treat the hard stop as the exception for compliance or abuse cases rather than the norm. --- ## Credits are not the same as credit-native This is the finding I think is most useful for anyone redesigning their pricing right now. Calling your unit a "credit" does almost nothing on its own. In the dataset, the companies using abstract credits as their metric score across the entire range, from fully credit-native to no better than a seat quota. | Company | Metric | Overage | Rollover | | --- | --- | --- | --- | | Replit | Credits | Rollover | Yes | | Windsurf | Credits | Rollover | Yes | | v0 (Vercel) | Credits | Hard cap | No | | Jasper | Credits | Hard cap | No | All four price in credits. Replit pairs credits with rollover, pooling, and pay-as-you-go, and reads as fully credit-native. Jasper layers credits on a per-seat, hard-cap, no-rollover skeleton and behaves exactly like an old feature quota with a new label. The lesson: never ship abstract credits without redesigning the lifecycle around them. Unit design and lifecycle policy are one decision, not two. --- ## The rollover gap Only 10 of 73 companies (14%) let unused credits roll over. It clusters almost entirely in the more credit-native designs and is essentially absent from the subscription-seat group. The barrier is not design philosophy, it is revenue recognition: finance teams resist rollover because it defers when credit can be recognized as revenue. The buyer-friendly middle path is capped rollover, for example up to one period of allowance expiring within twelve months, sold internally as committed credit pool ARR rather than defaulting to "no rollover" to keep accounting simple. --- ## Why this matters to us I checked their numbers against our source data before writing this. The aggregates line up: 73 companies, the revenue split, the 45% hard-cap rate, the 14% rollover rate, the category counts, and the individual company designs they call out. The dataset is updated continuously, so their analysis reflects the June 2026 snapshot, and a few counts will shift as we add companies. That is the whole point of publishing the AI Pricing Finder. The data is only useful if it is good enough that someone else can build a framework on top of it and reach conclusions we did not hand them. The analysis layer, the five archetypes, the credit-native scoring, and the design recommendations, is Forth and Mansard's, grounded in their Credit-Based Pricing design framework. It is a genuinely useful read, and it lands on the same conclusion we keep arriving at from the data: hard caps and per-seat quotas are a poor fit for AI products, and the companies moving to pooled credits with soft caps and bounded rollover are setting the template for agent-era pricing. You can explore the same dataset yourself in the [AI Pricing Finder](/ai-pricing-finder), filter by category, revenue model, or overage strategy, and see where your own pricing sits. *[Kat Laszlo](https://www.linkedin.com/in/katrinalaszlo/) is co-founder of Tanso. The analysis discussed here was published by Steven Forth and Michael Mansard on [Steven's Substack](https://open.substack.com/pub/pricinginnovation/p/credit-based-pricing-execution-patterns).* ### Want to price with confidence? Book a quick call and we'll walk you through it. More for agents: https://tansohq.com/llms.txt