Do AI-Credit Rollover Windows Cut Downgrade Churn?

AI credits are no longer a minor packaging detail. In 2026 pricing pages and filings, the dominant pattern is still a monthly allocation that resets, often with pooled credits, add-ons, or pay-as-you-go when customers exceed included usage. That construct protects arpu and acv_expansion by forcing sustained excess demand into monetized paths, but it can also strand unused capacity and frustrate customers whose AI usage is lumpy rather than linear.

The benchmark: monthly reset is still the default

Figma is the clearest 30-day no-rollover benchmark in the sample. Its AI Credit Terms say seat credits reset monthly, cannot be shared or transferred, and expire at the end of each month without rollover. For Organization and Enterprise customers, Figma also offers an annual AI Credits Subscription with monthly-refreshing pooled credits, with consumption flowing from seat credits to subscription credits to PAYG charges.

Miro shows a similar monthly cadence. Its AI credits reset on the first calendar day of each month for some plans and on the renewal date for Starter and Business, with paid add-ons available for eligible paid plans when usage exceeds the included balance. Microsoft also describes monthly reset behavior for AI Builder and Copilot Credits assigned to an environment, with usage failing if both currencies are unavailable or exhausted.

GitHub is moving Copilot to usage-based billing with monthly GitHub AI Credit allotments. The notable design choice is not rollover, but pooling. GitHub says pooled included usage across a business helps eliminate stranded capacity, while budget controls can operate at enterprise, cost-center, and user levels.

Why rollover is now a retention lever

The strongest warning signal comes from Figma's Form 10-Q for the quarter ended March 31, 2026. Figma began enforcing AI credit limits in March 2026 and introduced monthly AI credit add-ons plus pay-as-you-go usage. The filing says customers may curtail or stop AI-powered product usage under the AI credit billing model, creating retention risk.

Figma also reported elevated support volume, public and customer dissatisfaction, and reduced usage by certain customers after enforcement. That does not prove that monthly expiration causes downgrade churn. It does show that credit-limit enforcement can create measurable friction, and that this friction matters because Figma also disclosed that cost of revenue rose 253 percent year over year, including a $33.7 million increase in technical infrastructure and hosting costs relating to AI and platform usage.

The implication for pricing teams is practical. If AI usage is volatile, a strict monthly reset can make customers feel penalized for timing rather than value received. But unlimited rollover would weaken the monetization logic of add-ons and PAYG, especially when AI-related hosting costs are rising materially.

The middle ground: capped 60- or 90-day banks

A bounded rollover bank is best treated as a hypothesis, not a proven benchmark. The current sample supports the idea that vendors are trying to absorb volatility through adjacent mechanisms: pooling, promotional included usage, introductory credits, and nonexpiring paid credits. GitHub's Copilot usage-based billing announcement explicitly frames pooling as a way to reduce stranded capacity.

Customer.io offers another useful contrast. Its AI credits documentation gives paid-plan customers a one-time 100,000-credit launch grant that expires after 90 days, while purchased credits cost $10 per 100,000 credits and do not expire. That separates activation support from paid capacity: introductory credits are time-boxed, but purchased credits preserve trust by not expiring.

For B2B SaaS, a 60- or 90-day rollover cap could reduce stranded-capacity frustration without turning included credits into an open-ended liability. The cap matters. Credits could roll only up to a fixed multiple of monthly entitlement, expire after the rollover window, and be consumed before PAYG only within that window. Sustained excess usage would still graduate into add-ons or PAYG, preserving arpu, nrr, and acv_expansion potential.

Conclusion

The evidence does not yet prove that AI-credit rollover cuts downgrade churn. It does show that strict enforcement of monthly AI limits can create customer dissatisfaction, support burden, reduced usage, and potential churn risk, while AI infrastructure costs remain material. The right test is therefore not whether rollover feels more customer-friendly, but whether a capped 60- or 90-day bank improves downgrade churn and activation-to-paid conversion_rate without reducing add-on attach, arpu, nrr, cac_payback, or acv_expansion.