The shift from per-seat SaaS to usage-based and AI-powered pricing has outpaced the billing infrastructure most finance teams inherited. Vayu's 2026 CFO Signal Report, produced with PwC and The SaaS CFO, surveyed nearly 100 finance leaders and surfaced an uncomfortable picture: 71% report measurable revenue leakage, only 11.9% have the automation to support their current pricing models, and 39% still depend on engineering to ship billing logic.\In this exclusive interview, we sit down with Erez Agmon, co-founder of Vayu, to unpack what the data reveals about the state of revenue infrastructure, why "automation" has become a paradox for finance teams, and what it actually takes to run AI-era pricing without breaking the close cycle.\📊 Read the full 2026 CFO Signal Report here.\Ishan Pandey: Hi Erez, welcome to our "Behind the Startup" series. Take us back to the origin of Vayu, your background and what convinced you that revenue infrastructure was a problem worth building Vayu around?\Erez Agmon: I've been working with finance teams for about 10 years, and I've kept seeing the same frustration time and again. Every time companies changed how they price, the systems just couldn't keep up.\In the past, it was less common and easier to fix because subscription models were simple, making it relatively easy to fix. However, over the last few years, AI has spurred a rapid shift to usage-based models, making pricing far more dynamic and complicated. Suddenly, finance teams were expected to operate in real time to be able to ride the AI wave, yet they lacked the systems to do that.\During the subscription era, and far too often today, finance teams relied on spreadsheets, engineering teams for basic changes, and manually gathering their own data. Even then, it created the feeling of being responsible for revenue, but not having control, and now that feeling has become a full-blown five-alarm fire for finance teams.\I started Vayu together with Shenhav Avidar and Shay Gross. All of us come from fintech, from companies like PayPal and Melio, and we were looking for a problem worth solving. This one turned out to be inescapable and with surprisingly few solutions for what amounts to a problem that could easily turn into an extinction-level event for companies.\Ishan Pandey: The report opens with a striking number, only 11.9% of finance teams have the automation to support their current pricing models, even though 54% have already moved to hybrid or usage-based pricing. Why is there such a wide gap between what GTM teams are selling and what finance systems can actually process?\Erez Agmon: I don't think the gap is surprising. Pricing changed very fast, but infrastructure didn't.\GTM teams are under pressure to capture value, so they moved to usage, hybrid models, and anything that better reflects how customers actually consume the product. But finance systems remained built for fixed subscriptions.\The result is that companies layer new pricing atop old infrastructure. It's financial scaffolding that kind of works, but only with a herculean manual effort behind the scenes.\That's why you see automation's capabilities so far outpacing adoption. It's not that teams don't invest in tools; it's that the underlying model those tools are built on doesn't match how revenue actually behaves today.\And over time, that gap just compounds. The more flexible your pricing becomes, the more fragile your operations get.\Ishan Pandey: You introduce the idea of an "Automation Paradox," where 82% of finance teams still rely on manual spreadsheets despite investing in tools, and the month-end close stays stuck at three to seven days. From a systems design perspective, why do point solutions consistently fail to compress the close cycle, and what does true data unification look like in practice?\Erez Agmon: What we see in most companies is that they didn't really automate the system, they just automated parts of it.\A billing tool is added, maybe something for revenue recognition, maybe a data pipeline. Each one does its job, but they don't really talk to each other without manual handoffs, which results in a fragmented system that requires tons of upkeep and is highly likely to break down.\That's where the time goes. Not in generating the invoice, but in reconciling everything around it. Making sure usage matches contracts, contracts match billing, and billing matches what actually happened.\So the close doesn't get shorter. It just moves the work around.\Real speed comes when all of that lives in one flow. When your usage data, pricing logic, and billing are connected from the start is when companies stop reconciling and start trusting the system.\Ishan Pandey: One of the most alarming findings is that 71% of surveyed companies report measurable revenue leakage, with over 30% losing more than 5% of revenue annually. Walk us through where that leakage actually originates in a usage-based contract lifecycle, and why it tends to scale with company size rather than shrink with maturity.\Erez Agmon: Leakage usually doesn't come from one big mistake. It's a lot of small gaps across the lifecycle.\In most companies, usage lives in one place, contracts in another, and billing somewhere else. And the connection between them is never perfect.\So you might capture usage, but it's not fully aligned with the contract, or the contract has edge cases like credits or custom terms that don't make it cleanly into billing. And by the time you generate the invoice, you're already slightly off.\Some of it gets caught later, but not all of it, and as companies grow, this only gets worse. You have more pricing variations, more contracts, more exceptions. That's why leakage scales with the company. It isn't an execution issue; it's just that the system itself was never designed to keep everything in sync.\Ishan Pandey: The report frames usage-based pricing as a forecasting upgrade rather than a source of unpredictability, with 65% of usage-based teams reporting confidence in their forecasts versus 43% of flat-rate teams. This challenges the conventional CFO instinct that recurring subscriptions are inherently more predictable. What is the data infrastructure, insights, and real-world experience that flips this assumption?\Erez Agmon: I think the instinct makes sense. Variable pricing should feel less predictable.\The reality is that subscription models offer the illusion of predictability because it's a fixed number every month, but it's a black box where you really have no idea what's happening and what kind of value you're getting.\With usage, you're forced to build a much better data layer. You start tracking real customer behavior, how usage grows, where it slows down, and how it changes over time. That leads to real-time data on usage that can be optimized continuously instead of relying on a once-a-year seat-based subscription check-in. Businesses are forced to build better and enjoy better forecasting if they pull it off.\Ishan Pandey: 39% of finance teams depend on engineering to execute pricing and billing logic, and complexity consumes up to 60 hours of engineering time per month. How does this dependency change the power dynamic between finance, product, and engineering, and what does a "finance-native" architecture actually mean at the code and data model level?\Erez Agmon: It creates an unhealthy, unsustainable, and costly dynamic.\Finance is accountable for revenue, but they don't actually control how revenue is defined or executed. Every change goes through engineering, whether it's pricing, packaging, or even fixing billing logic. When that happens, it creates waiting times for an in-demand resource, and engineers' valuable time usually ends up getting wasted on maintaining logic and routine maintenance rather than building new, innovative tools.\GTM can't experiment, finance can't adapt, and engineering becomes a bottleneck without really wanting to be one.\Finance-native architecture gives back control to finance. All of a sudden, pricing logic, usage metering, and billing are defined in a way that finance can actually own and operate, without writing code or opening tickets.\And once that happens, the whole dynamic shifts. Finance stops being dependent and starts operating as part of the growth engine instead of just reporting on it.\Ishan Pandey: 60% of finance leaders prioritize AI, but 38% are blocked by data quality, and only 20% have moved AI workflows into production for billing. There is a lot of marketing noise around "AI for finance" right now. What use cases are real today versus aspirational, and how should a CFO evaluate whether their data foundation can actually support agentic or predictive workflows?\Erez Agmon: There's definitely a lot of noise right now.\The real use cases we see working today are often quite simple. Teams want to ask basic questions and get answers they can trust.\Things like what revenue hasn't been billed yet, which customers are about to go over their commitments, and why a specific account dropped this month. TAI dreams are great, and sometimes they work out, but at the heart of this is the day-to-day work of finance and RevOps.\Most companies struggle to answer simple finance questions without pulling data from three or four different systems and trying to piece it together.\AI's real value, therefore, comes when it sits on top of a clean, connected data layer and just lets users explore and understand what's happening. Ask a question, get a clear answer, and do so without the manual lift of traditional approaches.\Where it's still early is execution. Actually trusting the system to take actions, adjust billing, or automate decisions. Most teams are not there yet because the underlying data isn't reliable enough.\So the way we think about it is in two layers. First, AI is a way to understand your revenue. Then, over time, AI can actually operate parts of it.\But you can't skip the first step. If you can't explain your numbers today, AI won't fix that. It will just surface the gaps faster.\Ishan Pandey: Looking ahead, AI-native products are pushing toward outcome-based and agent-based pricing, where a single transaction might involve pooled credits, model costs, and variable margins. What does the next generation of revenue infrastructure need to look like to support pricing models that do not exist yet, and where do you see the industry being two years from now?\Erez Agmon: We're already starting to see where things are going.\Pricing is becoming much more tied to outcomes. Not just how much you use, but what you actually get. And in AI products, a single "transaction" can involve a lot of moving parts, models, credits, and diverse cost structures. Current infrastructure isn't built for that level of flexibility.\That means the next generation of revenue systems for this new usage-based AI era must be dynamic with the ability to define, change, and experiment quickly, without rebuilding the system every time. It also should be real-time and, perhaps most importantly, everything has to live in one place. Usage, contracts, and billing need to be connected.\If I look two years ahead, I think the gap we see today only gets bigger. The companies that solve this infrastructure layer will move much faster. The ones that don't will keep adding complexity on top of systems that were never designed for it.\Don’t forget to like and share the story! \