Cosmo Wolf, Head of Technology at Metronome, joins us to explore why billing isn’t just about finances but trust. We discuss reliability challenges in AI-era billing, architectural decisions like event sourcing, and how precision, observability, and transparency shape customer experience and give you a competitive edge.

Trust at the Core of AI Billing Infrastructure

The classic per-seat cost hardly works for modern tools, which has made companies move towards usage-based billing. To do that shift, you must be able to build trust as a core product principle, especially when you're dealing with AI workloads where billing can get complex and unpredictable. As Cosmo Wolfe points out, if a customer has a bad billing experience, it can immediately sour any of the goodwill the product had earned from being valuable.

AI billing is particularly prone to missed expectations because the cost is less intuitive. “I thought I used X and they thought I used Y and I don't even know which one's right... Now I'm kind of questioning their core ability to... bill me.” One-off errors become high-visibility moments that erode the relationship you have with customers.

In AI-native products where pricing is still evolving, that moment of truth is even sharper. Customers expect systems that set up the product to limit what their spend was, and when those limits fail, it's not just a financial hit, it’s a product trust issue too.

Event-Sourcing as a Foundation for Explainable AI Billing

“My high level answer… is not losing any data or intent,” says Cosmo, highlighting why Metronome uses an event-sourced billing architecture. For AI use cases, where new billing models like token-based or outcome-based pricing are still evolving, trust hinges on explainability.

“We store all the changes that went into it, without ever actually storing the current state.” That means Metronome can compute pricing state at any point in time, whether it's a mid-month upsell, feature migration, or pricing override.

This approach prevents common pitfalls like, “We’ll just bill them with some spreadsheets for the first couple of months” and then finding out that the spreadsheets were wrong. In AI scenarios, where billing can feel magical or opaque, being able to retrace and validate each pricing step builds both external and internal trust.

Observability and Monitoring for AI-Critical Systems

AI billing isn't just about numbers—it's about ensuring the systems computing those numbers are reliable, observable, and verifiable. Cosmo describes how Metronome integrates mature monitoring practices into its infrastructure. “Every team comes with a report of their services. There can't be any spreadsheet math here.”

Prometheus dashboards feed real-time alerts, while long-term logs are stored in S3 and reviewed every two weeks to uphold internal SLOs. But it's not just uptime that matters, performance and correctness of the business logic do too. “Any change that we’re making to that core kernel of logic, we backtest the whole history of various invoices and pricing and configuration from real-world scenarios to make sure there’s not any regressions in both the output and performance.”

In AI scenarios where compute usage varies wildly, regressions could either mischarge customers or crash systems. Monitoring here means deeply validating both the math and the system behavior to maintain trust across the stack.

Reliability as a Differentiator in the AI Tooling Ecosystem

AI startups and platforms are in a race, not just to innovate but to build trust. Cosmo notes that strong cost controls are becoming a reason customers switch providers. “GCP launched much stronger, soft and hard cost controls and people were switching because of that.”

In AI, where workloads are volatile and billing is tricky, reliability isn't just an engineering goal, it’s a product feature. “We want to make it possible for people to compete with their pricing and packaging,” he says. That includes giving teams reliable tools to implement pricing changes and customers the clarity they need to trust what they’re being charged for. In a fast-moving market of agents, LLMs, and unpredictable usage, those with the most stable, transparent systems will earn the deepest loyalty.

‍