Pranav Lal is an enterprise technology leader who has architected and scaled GTM systems for many high-growth pre-IPO to IPO SaaS companies.gettyEvery product leader I have spoken with in the last year is wrestling with the same question: We are shipping AI features that cost real money to run. How do we charge for them?The conversation almost always starts in the wrong place. Product and marketing start a working session, debate per-seat vs. usage-based vs. outcome-based pricing and arrive at a decision. The pricing page gets updated. Finance updates the forecast. Six months later, the same companies are quietly walking back their AI pricing, granting custom exceptions and absorbing margin erosion they did not forecast.The pattern is now visible in the data. In Revenera's 2026 Monetization Monitor, a survey of 501 product leaders, 70% of those offering AI features said delivery costs were eroding profitability, and more than a third cited uncertainty around pricing AI features as a direct blocker to aligning price and value. The postmortem usually blames the market, but in my experience having worked with multiple SaaS companies during their growth and public market transitions, the market is rarely the problem. Pricing was treated as a marketing decision when it was actually a systems decision, and the systems were never built to support what the pricing page promised.​When a company moves from per-seat to consumption-based pricing for an AI feature, it is not just changing a SKU. It is committing to a metering infrastructure that did not exist before. Every AI invocation now needs to be counted, attributed to a customer, aggregated in near-real time, surfaced to the customer in a usage dashboard and reconciled against an invoice. None of this is hard in isolation. All of it is hard at the same time, especially when the underlying CRM, billing system and product telemetry were architected when the only billable event was a user license renewal.The customer-side evidence of the mismatch is now hard to ignore: In Zylo's 2026 SaaS Management Index, 78% of IT leaders reported unexpected charges tied to consumption-based or AI pricing models in the past year. Those surprises do not happen because vendors are dishonest. They happen because the metering, contract logic and dashboarding were not architected to deliver on the pricing model the vendor sold. Customers ask why their bill does not match their dashboard. Account executives cannot explain consumption spikes. Renewals get harder. Profit margins get worse.A more honest way to think about AI monetization is to treat it as a systems architecture problem first and a pricing problem second. There are three architectural prerequisites that, in my experience, separate companies that get this right from those that do not.The first prerequisite is metering as a first-class capability. If you are going to charge for AI consumption, you need a metering pipeline that is engineered to the same reliability standard as your billing system. That means defined service-level objectives for event capture, deduplication, attribution accuracy and latency. It means the metering data is the source of truth for both the customer-facing dashboard and the invoice, with no reconciliation lag. Most companies underestimate this and look at metering as a side project owned by analytics. That works until it does not, and when it stops working, the failure is visible to every customer simultaneously.The second prerequisite is contract-aware pricing logic. Enterprise customers will negotiate every consumption-based model you offer. They will ask for floors, ceilings, rollover provisions, true-ups, multiyear ramps and product-mix flexibility. Each of these is a configuration in your CPQ system and a calculation in your billing engine. If those systems were built around per-seat assumptions, they will fight you on every nonstandard term. The companies that want to handle this well must invest in pricing logic that is decoupled from product packaging, so their sales team can negotiate a complex enterprise agreement without three months of engineering work to support it. The companies that handle this poorly may end up with custom contracts that finance reconciles in a spreadsheet, which is the kind of arrangement that creates audit findings under SOX.The third prerequisite is unit economics observability. Charging for AI features means caring about the cost of serving them in a way that most SaaS companies have never had to think about. A single power user running aggressive AI workloads can flip an account from profitable to unprofitable in a quarter, and you may not know it for two more quarters if your cost-of-goods-sold data is not connected to your customer data. The architectural fix is to wire model inference costs, infrastructure costs and support costs into the same systems that hold revenue and consumption data at the customer level. This is not a finance project. It is a data architecture project that finance benefits from.The implication for technology leaders is that AI monetization conversations need to start one room earlier than they currently do. Before product debates pricing models, the systems team should be asked a different question: What monetization models can our current architecture credibly support, and what would we need to build to support the others? That answer constrains the pricing decision in a productive way and surfaces the real cost of it, which is rarely the price itself, but rather the engineering investment required to make it operational.The urgency is not abstract. According to Gartner, 40% of applications will leverage AI agents by the end of this year, up from less than 5% in 2025. Every one of those integrations is a monetization question waiting to be answered.The pricing page looks like marketing copy. It is not. It is a contract, and the systems behind it determine whether your company can keep that contract at scale. The leaders who internalize this in the next 12 months may set the monetization terms in their categories. The ones who do not may spend that same year explaining to their boards why AI revenue is growing slower than AI costs.​​​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?