Replicate launches prepaid credits in 2 weeks, significantly reducing fraud with Metronome

Replicate makes it simple for developers to run and deploy machine learning models in the cloud. Thousands of businesses build their AI products on Replicate—from individual creators experimenting with cutting-edge generative models to large enterprises running production-scale workloads. 

Replicate supports some of the most diverse and demanding use cases in AI with a platform that abstracts infrastructure complexity. They handle the unpredictable, cost-variable nature of inference behind the scenes, so their customers can focus on building. 

As their customer base and usage scaled, Replicate’s original billing-in-arrears model created mounting challenges. With the launch of Google’s text-to-video model with audio, Veo 3, (initially priced at $6 per 8-second video generated), Replicate’s risk profile changed overnight, attracting more fraud than expected. The continued maintenance and updates of their piecemeal, homegrown anti-fraud measures were a Herculean effort, consuming hundreds of engineering hours quarterly. 

Rather than continue layering on defenses, the team moved to a new billing model: prepaid credits with auto-recharge. The approach aligned with market precedent, preserved their free trial offering, and let legitimate users get started quickly—all while blocking bad actors before they could make a move.

Key outcomes from the partnership include:

• 14 days from first line of code to first customers using prepaid credits
• 90% of users migrated to prepaid credits model
• Fraud reduced significantly with rollout of credits

Fraud spike with high-value models

Veo 3’s premium outputs increased the ease and impact of abuse, making the continued use of arrears billing too risky.

“Moving to prepaid credits essentially solved the immediate fraud problem.” — Ming Lu 

Homegrown controls became costly to maintain
Early-charge thresholds and internal monitoring systems for suspicious activity worked, but the system had grown more complex and harder to grok as time went on.

Preserve conversion
The team worried that a prepaid requirement might add enough friction to put a dent in conversion. The goal: reduce fraud and closely monitor impact on signups or paid conversion.

“We launched credits in two weeks and kept conversion steady—while shutting down legacy fraud code,” said Ming Lu, Product Lead at Replicate.

Replicate’s business depends on supporting a wide range of AI models with unpredictable compute costs. They need billing infrastructure that can keep pace with constant experimentation, product launches, and evolving monetization strategies. Metronome has been that foundation since the initial partnership with Replicate in 2022.

Built for usage-based billing
Replicate’s workloads are nondeterministic and span different hardware and models. Metronome handles event metering  and credit balances in real-time, so Replicate ensures that every customer is charged accurately the moment usage occurs.

Speed and autonomy for engineering
Clear APIs and documentation helped Replicate’s team implement prepaid credits with auto-recharge in just 14 days, without waiting on vendor customizations or slow support cycles. Engineers could configure pricing, events, and credit logic directly, keeping momentum in their own hands.

Partnership beyond the API
The shift to prepaid credits touched both payments and UX, and required tight coordination across teams. Metronome’s responsiveness and collaboration ensured Replicate could introduce an entirely new billing model smoothly, with minimal disruption to end users.

“The Metronome team was very responsive and willing to closely work with us to navigate any challenges we ran into,” said Ming. 

Source of truth for usage and credits
Replicate relies on Metronome data during support investigations and as an operational “single source of truth” for credit balances and usage. Real-time dashboards and controls keep everyone—internally and externally—informed on current usage and spend. 

Real-time alerting at scale 

With prepaid credits, real-time alerting is essential. It gives customers visibility to manage spend and recharge before their balance runs out, preventing product interruptions. At the same time, it safeguards Replicate’s business by ensuring usage never exceeds available credit balances, cutting off a common path for fraud.

A fast, collaborative rollout
With Metronome, Replicate replaced its arrears-based model with prepaid credits for self-serve users while continuing their invoicing model for enterprise accounts. The change was live in just 14 days, from first line of code to first customers, driven by Metronome’s clear APIs and collaborative approach to implementation.

Fraud cut at the source
By requiring credit purchase before usage, the prepaid system immediately and drastically reduced fraud, shutting down abuse before it could happen. What had once required a patchwork of fraud thresholds and custom defenses became a win–win for customers and the Replicate team with a simple, reliable credit balance.

Seamless for existing and new users
The rollout of prepaid credits was smooth and nondisruptive for users, ensuring retention of current customers and maintaining key customer metrics. Despite concerns around added friction, signups and paid conversion remained stable. Preserving a free trail softened the transition and kept the developer experience intact.

Operational simplicity
The collections and support teams now spend far less time chasing down payments and handling billing tickets. And with reclaimed attention and energy, engineering can now focus on product improvements instead of maintaining an ever-growing, brittle patchwork of fraud defenses.

Contact us here to learn Metronome helps AI companies balance risk and growth.