To write a price increase letter customers accept, state the new price and effective date in the first two sentences, name the specific cost change behind it, and offer at least one real choice. Customers judge cost-justified increases as fair and profit-justified ones as unfair. This guide covers timing, structure, real AI cases, and two adaptable templates.
Why a price increase in an AI product is different
An AI product raises prices from a weaker margin position than classic software ever did. AI-native companies run gross margins of 50 to 60%, against the 80 to 90% baseline of traditional SaaS (Bessemer Venture Partners, February 2026). At scaling stage, model inference alone eats roughly 23% of revenue (ICONIQ State of AI survey of ~300 executives, January 2026).
Worse, the costs underneath your price keep moving. In one recent product cycle, Anthropic cut Opus output pricing 67%, from $75 to $25 per million tokens, while OpenAI cut its primary GPT model 20% (PricingSaaS Trends Report, Q1 2026). Costs move up too: newer models spend far more tokens per task on long-horizon work. Your unit economics are tied to model selection, not customer behavior.
So a price increase in an AI product is usually a unit-economics correction, not a revenue grab. The math changed underneath you, and the letter's job is to say so plainly. Repricing has become a routine capability question for any billing platform built for AI products, because the price on the page and the cost underneath it drift apart constantly. The letter is where you close that gap without losing the customer.
When should you raise prices?
Raise prices when the unit economics say so: when your heaviest customers are unprofitable, when a model change moved your costs, or when the product's value has grown well past its price. The signals are the same for any SaaS price increase, but four of them show up earliest in AI products:
- Per-customer margin is negative on your heaviest accounts. Under flat-rate pricing, the customers who use you most pay the same as the ones who barely log in. If you can see margin per customer, the correction is obvious before it becomes urgent.
- You started rate-limiting for cost reasons, not abuse. When Anthropic added weekly caps to Claude Pro and Max plans in July 2025, fewer than 5% of paid subscribers were affected; those few were running agentic coding workloads that were uneconomic at a flat price. A cap added for cost control is a pricing change you have not announced yet.
- A model-cost shift changed your COGS. Pass-throughs run in both directions; the Anthropic and OpenAI cuts above are the downward case, and token-hungry long-horizon models are the upward one.
- Everyone around you is already iterating. PricingSaaS tracked 8,394 pricing events across 498 companies in 2025, and 23.1% of them were price increases. Among AI companies, 37% plan another pricing change within 12 months, per the same ICONIQ survey. A price increase letter is a recurring operational task, not a once-a-decade crisis.
Timing favors the small. The fewer customers you have, the cheaper the change is to make and to communicate; the pricing advantages of early-stage products fade as the customer base grows.
Decide the structure before you write a word
The structure decision comes before the letter, because the structure determines what you can honestly promise. In 2025, price restructuring made up 26.7% of all tracked pricing events, more than plain increases (PricingSaaS Trends Report, Q1 2026). Four structures cover most AI cases:
| Structure | What changes | When it fits | Named example |
|---|---|---|---|
| Simple raise | Same plan, higher price | Value grew, costs are stable | Canva Teams (September 2024) |
| Flat to hybrid | Base fee plus included allowance, usage billed above it | Heaviest users are unprofitable | GitHub Copilot (April 2026) |
| Credit repricing | Same headline price; a credit buys less | Model costs shifted under a credit system | Cursor (June 2025) |
| Phased migration | New customers first; existing customers at renewal | Enterprise contracts, high churn risk | Replit (June 2025) |
Two of these deserve a closer look.
Flat to hybrid with grandfathering. GitHub moved Copilot from flat subscriptions to usage-based billing in April 2026, stating that agentic workflows had made flat pricing financially unsustainable (DevOps.com, April 2026). The structure did the reassuring: annual subscribers kept their pricing until renewal, enterprises got a temporary credit buffer, and a billing preview tool shipped before the switch so users could model their new bill. If you are heading this way, designing hybrid pricing for AI products covers how to size the base fee, the included allowance, and the overage charges above it.
Credit repricing, the silent lever. Credit-based pricing grew 126% year over year across tracked companies in 2025 (PricingSaaS Trends Report, Q1 2026), partly because a credit layer lets you reprice what a credit buys as model costs shift, without changing the headline price. The hazard is silence. Anthropic marketed its Max plan as 20 times more usage than Pro, but after the July 2025 caps the weekly envelope on Sonnet worked out to roughly 6 times. A burn-rate change without a letter is still a price increase; customers just find out from their balance instead of from you.
What goes in the price increase letter
A price increase email to customers lands when it leads with the change, justifies it with the specific cost story, and gives the customer a real option. In Chargebee's 2026 survey of 1,454 consumers in the US and UK, 58% of subscribers who noticed a price increase accepted it when the value was clearly communicated (Chargebee Global Consumer Insights, 2026). The pattern behind that number is old: the dual entitlement research of Kahneman, Knetsch, and Thaler (1986) found people judge cost-justified increases as fair and profit-justified ones as unfair. Forty years later, that is still the whole game.
Seven elements, in order:
- The change and the effective date, in the first two sentences. New price, old price, the date it applies. No warm-up paragraph.
- A real notice period. 30 days minimum for monthly plans; for annual contracts, apply at renewal.
- The specific reason. Name the cost change: "the newer models spend roughly 3x more tokens per task" beats "to continue delivering value." If the honest reason is a cost correction, say that, not a feature list.
- What stays the same. Data, features, support, existing credits.
- At least one real choice. Lock the current price with an annual commitment, keep the legacy plan until renewal, or take a documented downgrade path. A letter with no options is a notification, not a communication.
- A named human to reply to. Replies should reach a person, not a ticket queue.
- A founder signature. At Seed through Series B, the founder signs. It signals the decision was owned, not automated.
Price increase letters from real AI products
Three AI pricing changes from 2024 to 2026 define what works and what fails in a price increase announcement: Cursor's June 2025 rollout (a communication failure despite sound unit economics), Canva's September 2024 Teams increase (value-vague justification that triggered backlash), and GitHub Copilot's April 2026 restructuring (structured sequencing that landed quietly). As of mid-2026, these three cases cover most of what can go right or wrong.
Cursor, June 2025: the canonical communication failure. On June 16, Cursor changed Pro from 500 fast requests per month to a $20 pool of API-rate credits. The underlying decision was economically sound; newer models were spending far more tokens on long-horizon tasks, and Cursor was absorbing the difference. The communication failed: users did not know that "unlimited" applied only to Auto mode, burned through $20 of Claude usage in a few prompts, and hit overage charges with no visible balance and no warning.
On July 4, CEO Michael Truell published an apology (Clarifying our pricing, July 2025) and refunded unexpected charges from June 16 to July 4. The company was past $500M in annual recurring revenue when this happened (TechCrunch, July 2025); scale does not protect you from a bad rollout.
Canva, September 2024: value-vague justification. Canva raised its US Teams plan from $119.99 to $500 per year for five seats, a 317% increase softened by a 40% first-year discount, and pointed to its AI feature set as the reason (TechCrunch, September 2024). Users read "we added AI features" as justification-shopping for a revenue decision, and the communication arrived by private email rather than a public announcement. Backlash followed, then a partial walk-back. Features are not a cost story.
GitHub Copilot, April 2026: the structured contrast. Same underlying pressure as Cursor: model multipliers meant a Claude Sonnet request went from 1x to 9x credit consumption. The difference was sequencing: preview tooling first, grandfathering for annual subscribers, buffers for enterprises, then the switch. Structure did the reassuring before the letter had to.
For a fuller teardown of how one company communicated a structural change, see Clay's pricing rewrite.
Two templates you can adapt. First, the flat-to-hybrid migration letter:
Subject: Your [Product] plan changes on [date]
Hi [first name],
On [date, 30+ days out], [Product] Pro changes from $[X]/month unlimited
to $[X]/month with [N] included [units], then $[rate] per [unit] above that.
Why: our per-request costs are real and variable. The newer models spend
roughly [multiple]x more tokens per task than the ones we launched with.
At a flat price, our heaviest users lose us money, and we would rather
fix the structure than quietly degrade the product.
What stays the same: your data, every feature, and your existing credits.
Your options:
- Do nothing, and the new structure applies on [date].
- Lock your current price for 12 months by switching to annual before [date].
- Preview your last 3 months of usage against the new plan: [link].
Reply to this email and a human reads it. I am that human.
[Founder name]Second, the credit-rate change notice:
Subject: [Product] credit rates change on [date]
Hi [first name],
From [date], one credit buys [new amount] of [model/feature] instead of
[old amount]. Your monthly price does not change; what a credit buys does.
For a typical account, that means about [X]% more spend for the same usage.
Why: [provider]'s pricing for [model] changed on [date], and we are passing
part of that through instead of absorbing it silently and cutting corners
elsewhere.
Your dashboard now shows your live burn rate, so nothing lands as a
surprise: [link]. If the new rates would push your bill above $[threshold]
this month, we cap it there while you adjust.
Questions: reply here.
[Founder name]Which mistakes turn a price increase into churn?
To raise prices without losing customers, avoid the six recurring communication mistakes visible across the Cursor, Canva, and GitHub Copilot cases:
- Burying the change. If the new price appears in paragraph four, the reader finds it on the invoice instead.
- Value-vague justification. "We've added so much" invites the Canva reaction. Name the cost change or the specific value, with numbers.
- Repricing credits without a letter. Changing burn rates in the backend is a price increase delivered by surprise; that was the core of the Cursor episode.
- No notice period. A change effective the day of the email reads as a fait accompli.
- No way to preview the new bill. GitHub shipped a preview tool before switching; Cursor did not, and refunded 19 days of charges.
- Apologizing for the correction. If the math requires the change, an apologetic tone signals you do not believe your own reasoning. Be direct about the why and generous with the options.
Can your billing stack execute the change?
Your billing stack can execute a price increase without an engineering sprint if it supports five capabilities: plan versioning, coexisting price schedules, per-plan margin math, customer-visible burn rates, and segmented migration timing. A price increase letter makes promises the billing system has to keep. "Your plan is unchanged until renewal" requires old and new prices to coexist; "preview your new bill" requires usage data per customer. Before the announcement goes out, verify each:
- Plan versioning without code changes. If repricing needs an engineering sprint, you will change prices less often than the market moves. At the far end of this spectrum, OpenAI's pricing lead can execute a change in Metronome's UI in under an hour, against the quarter-long engineering cycle it once took (Metronome / OpenAI customer story, 2024).
- Coexisting price schedules. Grandfathering is only real if version N and version N+1 can run at the same time, per customer.
- Margin math per plan, before the change ships. You should know the gross margin of the new pricing before customers see it, not two invoices later.
- Customer-visible balance and burn rate. Every failure case above involved customers who could not see what they were spending.
- Segmented migration timing. New customers now, monthly at next cycle, annual at renewal; three cohorts, three dates, one letter each.
Whether billing runs on Stripe Billing, Metronome, Orb, or Credyt, walk this checklist before you hit send. The letter is one afternoon of writing; these five capabilities are what make its promises true.
Next steps
Now that you've read our guide on writing a price increase letter, you're probably wondering how to run the change itself.
The letter is the visible part of a price increase; grandfathering, previews, and margin math are what your billing infrastructure has to execute.
That's Credyt.
Credyt is real-time monetization infrastructure for AI. Here is how it handles a pricing change:
- Billing model versioning: change pricing rules in the dashboard without touching code, applied to new events immediately.
- Product versioning and catalog management: full product version history, with a diff review before price and entitlement changes save.
- Margin-aware pricing: see gross margin per plan in real time as you configure the new prices.
- Hybrid billing: base fee, included allowance, and usage-based overages in a single plan configuration.
- Revenue vs cost correlation: watch per-customer margin before and after the change.
Learn how Credyt handles pricing changes in the Credyt docs.
Related resources
- How to implement consumption based pricing. The migration companion: observability first, billing infrastructure second, pricing last.
- What is usage-based billing?. The mechanics primer for the models referenced throughout this guide.