AI-powered software is breaking the traditional SaaS billing model, and CFOs are scrambling to respond, as PYMNTS reports. The per-seat subscription — the pricing structure that carried Salesforce, Slack, Workday and an entire generation of B2B software — was not designed for products whose marginal cost now moves with every customer interaction.

That is the core of the problem, and it deserves a precise framing before any product or finance team acts on it.

Why does AI break per-seat pricing?

Classical SaaS pricing rests on an economic assumption: once the software is built, serving one more user costs the vendor almost nothing. That assumption justified flat per-seat fees, predictable annual contracts and the clean deferred-revenue lines CFOs built their forecasting models around.

Generative AI inverts the assumption. Every query, inference call or generated token triggers compute costs — often from third-party infrastructure priced by usage. A customer with ten licenses can generate costs that vary by an order of magnitude depending on how intensively they use the product. The seat count stops being a proxy for either value delivered or cost incurred.

For product managers, this means the pricing page is no longer a packaging exercise. It is a unit-economics problem that sits at the intersection of gross margin, packaging and customer behavior.

What does the scramble look like on the finance side?

According to PYMNTS, CFOs are the roles under the most immediate pressure, because the billing model change hits three of their core instruments at once:

• Revenue forecasting. Usage-based revenue is inherently less predictable than contracted seats, which complicates quarterly guidance and valuation multiples that investors anchor to net revenue retention.
• Margin visibility. When cost of goods sold moves with customer usage, gross margin becomes a variable the customer partly controls — a scenario most SaaS finance models were never built to model.
• Contract and billing infrastructure. Metering, rating and invoicing for consumption require systems that flat-seat billing never needed, and retrofitting them touches billing, rev rec and compliance.

The reported dynamic is a scramble precisely because these are not independent problems. Fix metering without fixing forecasting and you get accurate invoices for revenue you cannot predict.

What pricing options are vendors actually weighing?

The reporting points to a shift away from traditional subscription billing, and in practice product teams evaluating the move face a limited menu, each with real tradeoffs:

• Pure usage-based pricing. Aligns revenue with cost and value, but makes revenue volatile, punishes adoption-heavy customers, and complicates enterprise procurement, where buyers demand budget certainty.
• Hybrid models — platform fee plus usage. A subscription floor preserves predictability; overage or metered tiers capture heavy usage. This is the direction much of the market is drifting, but it doubles the pricing surface a PM has to design, test and defend.
• Outcome-based or credit-based pricing. Pre-purchased credits or fees tied to delivered results buffer vendor margin risk, but credits add cognitive load at renewal and outcome pricing requires a measurable, trusted definition of the outcome — a bar many AI features cannot yet clear.

Each option fails in identifiable conditions. Usage-based pricing fails when buyers require predictable budgets — most large enterprises do. Hybrid pricing fails when the usage tiers are set without cost data, which happens when product and finance teams price off competitors rather than their own inference bills. Outcome pricing fails when the outcome metric is gameable or disputed.

What should product managers do with this?

Treat pricing as a live product surface, not an annual revision. Concretely:

• Instrument usage per customer segment before changing the price, not after. You cannot set usage tiers without knowing your cost curve.
• Model gross margin per account, not per product. AI cost concentration means a small number of power users can flip a nominally profitable tier into a loss.
• Bring finance into pricing discovery early. The PYMNTS reporting makes clear the constraint is organizational as much as technical — the CFO's forecasting model is a stakeholder in your pricing page.

Where is this heading?

The likely end state is not the death of subscriptions but their restructuring: a predictable base fee for access, plus metered layers where AI compute drives cost. Software pricing converged on per-seat simplicity for two decades; AI's variable cost structure is now forcing the divergence, and the teams that instrument usage early will set the terms everyone else has to renegotiate against.