Chargebee's billing overhaul shows how CodeRabbit, Gorgias and Zapier actually price their AI features

Chargebee, the subscription billing platform that processes recurring revenue for more than 6,500 businesses, spent 2026 rebuilding its pricing engine specifically for AI products, and in July it became reachable directly from Claude's official connector marketplace. The product change matters less for what Chargebee itself charges than for who is now named as running on it: CodeRabbit, Gorgias, Zapier, Lambda and LimeChat are all customers, according to SaaStr's reporting on the rollout, and no two of them bill their AI features the same way.

Five companies, five different pricing units

CodeRabbit is the clearest example: its agent's clock time, not its token spend, sets the price. SaaStr reports the company bills by the minute its code-review agent is actively working, turning an unpredictable token cost into a flat rate a customer can forecast in advance. Gorgias instead splits what it charges between raw ticket volume and the share of conversations its agent actually resolves. LimeChat combines a flat platform fee with a charge per successful conversation, and told SaaStr that automating the dunning process alongside that model lifted on-time collection from 40% to 100%. Zapier takes a different approach entirely, leaning on the platform mainly to test pricing experiments and negotiate enterprise contracts without pulling engineers in each time.

The common thread, per SaaStr's account of Chargebee's pitch, is that AI products tend to launch on prepaid credits and drift toward paying for outcomes once a customer trusts the system with more autonomy. Chargebee's 2026 catalog redesign is built so that drift doesn't force a migration: a vendor can change the unit it sells without moving existing subscribers off their current plan.

The new guardrails exist because usage pricing can blow up a budget

Because billing by usage turns adoption into a live finance risk, Chargebee's release this year also shipped three kinds of guardrails: a webhook-based warning once an account passes a usage line a team sets for itself, a ceiling on overage charges fixed in advance for the whole plan, and a "hold-and-authorize" step that locks in the credits a request needs before it runs — so when several people draw from one shared, nearly-empty credit pool, only whichever request gets there first is guaranteed to go through. Gorgias CFO Kunal Agarwal put the underlying problem in blunt terms, in comments SaaStr attributed to Chargebee's August release:

"You can't price what you can't see."

Gorgias' own numbers, also reported by SaaStr, show why that mattered inside the company: one AI feature was quietly costing more than $4 per interaction before Gorgias started tracking cost by interaction, by margin and by feature, and customers who used more than 70% of what they'd bought churned far less than lighter users.

Billing access now reaches into the AI tools marketers already use

A separate, connectivity-focused change sits alongside the pricing rework. In May, Chargebee put three MCP (Model Context Protocol) servers into beta so assistants such as Claude, Cursor and ChatGPT can read a company's live billing data directly: one answers API questions, one builds a sandbox product catalog out of an existing pricing page, and one retrieves live customer, subscription, invoice and transaction records. Access runs through dedicated API keys or OAuth, built to "ensure the AI's permissions strictly mirror each user's existing Chargebee access level." In July, that server became a one-click install from Claude's official connector marketplace — Chargebee's notes warn that a team is "responsible for actions AI clients perform through MCP" and should review them before they run.

Chargebee's current Flow pricing underscores how literally this billing-for-AI pitch is being taken: the self-serve tier costs either $0 plus 0.80% of monthly invoicing, or $99 a month plus 0.65%, with the two options breaking even at $66,000 in monthly invoicing and both including 100 million usage events a month. A company invoicing $500,000 a month would pay roughly $3,349 on the flat-fee option — about $40,000 a year, a fraction of what a dedicated billing engineering team would cost, though it scales with revenue for as long as the relationship lasts.

Why this is a marketing problem, not just a finance one

None of this is Chargebee's problem to solve alone. Every software marketer selling, or about to sell, an AI feature inherits the same questions Gorgias and CodeRabbit had to answer: what unit are we actually charging for, does the pricing page describe that unit honestly, and can sales and finance point to the same number when a customer disputes an invoice. A pricing page that still says "per seat" while the product quietly caps, throttles or meters AI usage underneath is the exact mismatch Agarwal's comment was warning about.

• Name the exact billing unit for every AI feature in your pricing copy, not just in the contract fine print
• Confirm your CPQ or quoting tool and your billing system agree on what was sold before a renewal exposes the gap
• Track cost per interaction internally even if you don't bill that way yet, so a pricing change isn't a surprise margin hit