Usage-Based Pricing

Customers pay based on what they consume -- events tracked, tokens processed, API calls made, audio hours transcribed. No fixed seat count. Revenue scales directly with customer usage, which aligns vendor incentives with customer growth but introduces revenue unpredictability.

Companies Using This Model

AI Infrastructure

• openai -- per-token pricing (e.g., GPT-5.5 at $5/MTok input, $30/MTok output)
• fireworks -- per-token inference, plus per-hour GPU pricing
• replicate -- per-output-unit (images) or per-hardware-second (private models)
• deepgram -- per audio-hour for speech AI
• databricks -- Databricks Units (DBUs), recently consolidated under AgentBricks
• snowflake -- credit-based consumption (credits map to compute used)
• scale-ai -- usage-based data labeling and AI services
• sierra -- outcome-based AI agents, per-conversation pricing (see also outcome-based)

Data & Analytics

• posthog -- per-event across 10+ products, 90% of companies use it free
• amplitude -- per Monthly Tracked User (MTU)
• mixpanel -- per-event ($0.28/1K events after free tier)
• segment -- usage-based customer data platform
• statsig -- usage-based feature flags and experimentation
• hotjar -- usage-based behavior analytics
• fivetran -- usage-based data integration

APIs & Platforms

• twilio -- per-API-call across 20+ products, ConversationRelay at $0.07/min
• plaid -- per-API-call for financial data
• alchemy -- usage-based blockchain infrastructure
• apify -- usage-based web scraping
• customer-io -- usage-based messaging
• botpress -- usage-based chatbot platform
• voiceflow -- usage-based conversational AI

Infrastructure

• mongodb -- usage-based database
• digitalocean -- usage-based cloud infrastructure
• klaviyo -- usage-based email/SMS marketing

Related Models

These companies use usage dimensions as their primary metric but have distinct model names:

• datadog -- per-host pricing ($15-34/host/mo) with usage-based add-ons for logs, APM, etc.
• crowdstrike -- per-device endpoint security
• pendo -- per-MAU product analytics
• zapier -- per-task automation
• framer -- per-site website builder

Patterns & Trends

Generous free tiers drive adoption. PostHog gives away 1M events/mo free, Mixpanel gives 1M events/mo, Amplitude gives 50K MTUs. The bet: developers who start free will grow into paying customers as their products succeed.

Price deflation in AI inference. Fireworks, Deepgram, and Replicate are competing on price. Deepgram cut Voice Agent API rates while tripling concurrency. This mirrors the broader trend of inference costs dropping.

Value metric diversity. Even within "usage-based," the actual unit varies wildly: tokens, events, MTUs, audio-hours, compute-seconds, DBUs, API calls. Each metric reflects what the vendor considers the unit of value they deliver.

No seats, no friction. PostHog and Mixpanel offer unlimited seats on all plans. This removes the "should we add another user?" friction and lets usage (not headcount) be the growth driver.

Usage-based + enterprise floor. Most usage-based companies still have enterprise tiers with committed spend. Pure pay-as-you-go is rare at scale -- Snowflake, Databricks, and OpenAI all negotiate annual commitments with large customers.