Chargebee AI定价机制与Gorgias降本提效复盘

Chargebee has been around 14+ years and runs billing for 6,500+ businesses. It makes the list this week because of what it shipped in 2026 and who is running on it.

Every AI company we talk to has changed its pricing at least twice, usually from seats to credits, then to actions, then to outcomes. Each change breaks something downstream: the quote, the entitlement, the invoice, the rev rec schedule. Chargebee spent this year rebuilding around that problem.

What’s new in 2026:

• Credit, action and outcome pricing now run on a single catalog, with margin controls built for agents
• An MCP server went into beta in May, and nearly 400 teams have connected it to their AI tools
• CPQ now sits on the same billing record, so what sales quotes is what finance bills
• Pricing is now one self-serve plan, Flow, at $0 + 0.80% of invoicing or $99 + 0.65%, with 100M usage events a month included
• Gartner named it a Leader in recurring billing for the third straight year in August

#1. The AI customers each price on a different unit

No two of the AI customers Chargebee names charge the same way.

• CodeRabbit: prices its AI code review agent by active agent minutes instead of tokens, so customers get a predictable per-minute rate
• Gorgias: splits AI support pricing between ticket volume and resolved conversations
• LimeChat: a fixed platform fee plus a charge per successful conversation, with on-time collection going from 40% to 100% after automating dunning
• Lambda: hybrid billing for AI infrastructure
• Zapier: uses it to run pricing experiments and enterprise contracts without pulling in engineering

Chargebee’s site also lists HeyGen, DeepL, Writesonic and Messari.

#2. Credits, actions and outcomes sit in one catalog

Chargebee says most AI products launch on credits and move toward outcome pricing as customers trust the agent with more. The product is built so that move does not require a migration. The catalog changes, existing subscriptions stay on the old plan unless you choose to move them, and rev rec follows.

On outcome pricing, escalations and failed attempts get ingested for analytics but are never rated. A support agent that attempts 138,000 conversations and resolves 10,000 bills for the 10,000, and you still have the data on the other 128,000.

Chargebee VP of Product Prittam Bagani walks through how AI companies actually price, from CodeRabbit’s per-minute agents to Lovable’s credits to outcome-based models:

#3. Hard caps and hold-and-authorize on shared credit pools

Heavy usage looks like adoption in the product dashboard and like a loss in the P&L. The 2026 controls go after that gap:

• Threshold alerts fire a webhook when an account crosses a set percentage of its usage
• Overage rates and hard ceilings are set once per plan
• Hold-and-authorize locks the credits the first request needs when several users share a nearly empty pool, and blocks the rest

Gorgias CFO Kunal Agarwal, in Chargebee’s August release: “You can’t price what you can’t see.”

#4. The MCP server went from beta in May to a Claude connector in July

In May, Chargebee shipped MCP servers that let Claude, Cursor and ChatGPT read and act on live account data. July added an official Claude marketplace connector, configurable tool inputs so admins can restrict or default the fields an AI client can send, and coverage for checkout, product and customer lookups.

The buyers are finance and RevOps. A billing lead closing the books can ask why an invoice diverged from the contract and get the answer with account history attached.

Chargebee’s own release notes say you are responsible for what AI clients do through MCP, and tell you to review actions before they run. I’d connect it read-only first.

#5. CPQ is built on the billing engine, and the first 50 quotes are free

AI companies hit enterprise demand earlier than prior generations of software did. The first $300K deal with a ramp, a credit commit and custom limits usually arrives before there is a quoting system.

Reps can quote multi-year ramps, usage pricing and commit tiers across consumption products, all synced to the billing record, and renewal quotes inherit the terms of the existing subscription. CPQ Lite is free for the first 50 quotes for Billing customers. Full CPQ, with multi-product and ramped quotes and approval workflows, is a sales call.

Chargebee CEO Krish Subramanian’s Beelieve ’26 keynote on scaling AI-era business models, including the new CPQ and the Growth experimentation platform:

#6. What it costs: 0.65% to 0.80% of invoicing

Flow is $0 + 0.80% of monthly invoicing, or $99 + 0.65%, and the two options break even at $66K a month. By my math, a company invoicing $500K a month pays about $3,350 a month on the $99 option, roughly $40K a year. Enterprise adds multi-entity, account hierarchy and contract terms, with NetSuite and Salesforce integrations as paid add-ons.

That is a percentage of revenue for as long as you are a customer. The alternative is billing engineers on your own payroll, and $40K a year is a fraction of one.

#7. Where it fits and where it doesn’t

Chargebee fits AI companies running self-serve and sales-led at once, with a finance team that needs rev rec and collections on the same record.

The pushback comes from usage-native competitors. Flexprice, which sells against Chargebee, argues that usage-first products with real-time metering often pair it with, or switch to, a usage-native platform. That is a competitor talking, so run your own event volume through a trial before you sign.

Four top learnings from Gorgias, whatever billing tool you buy

Agarwal spoke at Chargebee’s Beelieve ’26 event in May. Gorgias runs 300 million+ conversations for 17,000 customers, and AI agent usage grew 350% in 12 months.

• $4. One popular feature turned out to cost over $4 per interaction. Gorgias now tracks cost at three levels: LLM cost per interaction, fully loaded gross margin, and cost by feature.
• 70%. Customers using more than 70% of what they bought churned far less. So Gorgias sells a smaller package first and expands.
• 3 days. The attribution window for AI-assisted sales started at seven days. Customers pushed back and it settled at three.
• 90 days. Pricing gets evaluated every 90 days.

On model routing, his line was: “Not everything needs the Porsche of LLM models.”

Come meet the Chargebee team at SaaStr AI Annual 2027.