Usage-based pricing has moved from cloud infrastructure niche to the dominant SaaS billing model. Here's how leading companies actually implement it — the metrics they chose, the billing infrastructure they use, and the results.
| Company | Pricing Model | Value Metric | Billing Type |
|---|---|---|---|
| Snowflake | Pure consumption | Compute credits | Prepaid credits + on-demand |
| OpenAI | Per-token + subscription | Tokens (input/output) | Prepaid credits + API metering |
| Twilio | Per-message/call | Messages, minutes, calls | Pay-as-you-go + volume commits |
| AWS | Per-resource-hour | Compute hours, storage GB, requests | On-demand + reserved + savings plans |
| Datadog | Per-host + ingestion | Hosts monitored, logs ingested | Committed + on-demand overage |
| Stripe | Per-transaction % | Payment volume | % of volume processed |
| HubSpot | Hybrid (seats + contacts) | Marketing contacts, seats | Tiered subscription + overage |
| Anthropic | Per-token credits | Tokens (input/output) | Prepaid credits + enterprise commits |
| MongoDB Atlas | Consumption | Compute, storage, data transfer | Pay-as-you-go + committed |
| Vercel | Hybrid (plan + usage) | Function invocations, bandwidth | Free tier + plan + overage |
| Salesforce (Agentforce) | Per-conversation | AI conversations | $2 per conversation |
| Confluent | Consumption | Confluent Units | Committed + on-demand |
| Elastic | Consumption | Elastic Consumption Units | Prepaid + on-demand |
Snowflake's consumption model drove record net revenue retention (158% NRR at peak) by perfectly aligning price with value. Customers buy credit packages upfront or pay on-demand. The lesson: pure consumption creates powerful expansion but requires guardrails when customers learn to optimize.
📖 Snowflake's Consumption Model: Lessons for Every SaaS Company
AI API providers charge per input and output token, often with different rates per model. Credits abstract the underlying token cost, giving providers room to adjust model pricing without breaking customer commitments. The challenge: tokens are a terrible value metric for business users.
📖 Why Tokens Are a Terrible Pricing Metric
Salesforce's $2/conversation model is the highest-profile attempt at pricing AI by outcome rather than input. It's outcome-adjacent — charging per conversation regardless of complexity — which creates margin risk on complex interactions but simplicity for buyers.
📖 Salesforce's Agentforce Pricing Experiment
HubSpot layers marketing contact overage charges on top of tiered subscriptions. The subscription handles feature access; the contact count drives expansion. This hybrid model is the template most B2B SaaS companies are adopting.
📖 HubSpot's Hybrid Pricing Playbook
Microsoft keeps per-seat pricing for Office 365 and Teams as the anchor, then layers Copilot consumption on top. The seat price provides stability; AI usage provides expansion. This transitional model is the most common path for companies moving from seats to usage.
📖 Microsoft Raised the Price of Predictability by 65%
Related reading:
Snowflake (compute credits), Twilio (per-message/call), AWS/GCP/Azure (per-resource-hour), OpenAI (per-token credits), Datadog (per-host + ingestion), Stripe (per-transaction percentage), and many AI startups. Usage-based pricing is the fastest-growing SaaS billing model.
It depends on the category. Infrastructure: compute hours or data processed. Communications: messages or minutes. AI: tokens or credits. Payments: transaction percentage. The best metric maps to customer-perceived value, not your internal cost structure.
Yes — usage-based companies typically see 120-140% net dollar retention because revenue expands as customers use more. Snowflake reported 158% NRR at peak. The risk is contraction: customers can also optimize usage and shrink spend.
The most common pattern is a platform fee or committed spend floor plus per-unit overage. The subscription provides base revenue predictability; the usage component captures expansion. HubSpot, Slack, and most enterprise SaaS use this hybrid approach.