"SaaS is dead." That was Satya Nadella's verdict on the BG2 podcast in December 2024. A structural observation from the CEO of the largest seat-based software business in the world. Few had more to lose from that statement being true.
My first instinct was that this was a technology statement. I was thinking databases, infrastructure, the way software gets built and deployed.
On reflection, that is not it at all. SaaS is not a technology. It is a commercial model, a way of packaging and selling software. Subscriptions, seats, recurring licences tied to a human user. That is what AI is breaking. And once you see it that way, the implication for every B2B FinTech business is immediate. AI has changed the cost structure of these products. What you can afford to give away, what you need to charge for, and what your investors will start asking about are all shifting. The old pricing model was not built for this.
What followed made the point concrete. Microsoft committed to converting every per-user product it owns into a hybrid model.
The pricing levers that work for AI products
In traditional SaaS, an extra free user costs almost nothing to serve. In an AI product, every interaction consumes compute. That changes how you should price. Three levers work for AI products in financial services, but which one you can use depends on what you can actually measure.
Charge for consumption, not for AI model maturity
Price on how much users run through the product: reconciliations completed, reports produced, data requests processed. Most users will never notice a difference between model versions. They will always notice when they hit a limit. This sounds like the easy option. It is a number. But most products are not logging task-level activity in a way that is clean enough to bill against. That is a product decision before it is a pricing decision.
Price the heavy lifting differently
Some AI tasks cost more to run. A standard query is cheap. A 50-step agent that pulls data from six systems, reconciles it, and produces a board-ready output is not. The cost basis is transparent, and clients understand it intuitively.
Charge for outcomes
This is the hardest lever, and in financial services, the most valuable. Intercom charges $0.99 per resolved support ticket, nothing for failed attempts. The client pays for the result, not the effort, but defining that result is harder than it sounds.
What is the value that you're actually selling?
The first instinct is that the outputs are the business outcomes: documents produced, reports generated, queries run. These look like outcomes but are a proxy for value, not value itself.
In B2B FinTech the outcome is the compliant distribution, the reduced operational cost, the faster fund launch. In every investment case it can be reduced down to three things: risk and compliance, revenue, and cost reduction.
Risk and compliance
Compliance offers the cleanest measurement but carries the most risk. Most vendors will not price like this as it transfers regulatory liability onto their contract. The vendor position has to be: we are responsible for the quality and accuracy of what the system produces; the client is accountable for the regulatory decision.
Revenue
Pricing on revenue contribution is commercially appealing but almost impossible to prove. Too many variables sit between your product and the number on the P&L.
Cost reduction
This is where most AI vendors in financial services play: automate the manual work, streamline the process, reduce the operational overhead. It is the most common value story in the market, and the one where measurement is most likely to let you down.
Compliance carries too much liability to price against. Revenue contribution is too hard to prove. Cost reduction is what remains, and it can be measured. The challenge is measuring it correctly.
The cost reduction challenge
Cost reduction in practice is not about removing whole people. It is about reducing effort in parts of a process: making specific steps faster, removing handoffs, eliminating the stop-start pattern where work sits waiting before it moves forward.
The capacity created is real. You deliver software that demonstrably reduces time and effort. The right framing is capacity, not headcount. This task used to take 47 minutes and now takes 6. That process used to involve four approval steps and now involves one. That reporting cycle used to sit waiting for data for two days and now completes in four hours. Measurable, defensible, and yours to own.
But you can only make that argument if the measurement exists.
To price on outcomes, or even to defend the value of consumption pricing, you need to know what is happening inside the product. Product analytics captures this directly: task completion time, waiting time, stop-start frequency, error rates, rework. Data that does not require clients to self-report improvement. These are the evidence base for your commercial relationship. Getting this measurement built requires product decisions: what to log, how to surface it, what to expose to clients, how to make it legible without a separate data team.
The cost problem nobody is pricing for
In traditional SaaS, more customers means more margin. In AI, it can mean the opposite. The CEO of one of the world's largest AI companies admitted publicly that he had personally set the price for his own product, and got it badly wrong. Users consumed far more compute than he anticipated. For any B2B vendor building AI into their product, this is not a cautionary tale about one company. It is the structural reality: every interaction has a real cost, and if your pricing model does not recover those costs, growth accelerates losses rather than revenue.
Traditional SaaS had predictable costs to serve and predictable revenue from seats. AI breaks both ends simultaneously. Vendor input costs vary with compute, model pricing, and scale. Vendor revenue varies with consumption. That is a new risk profile that neither vendors nor buyers have fully priced in.
The buyer problem is just as real. Procurement is built on annual budget approval, fixed monthly commitments, and clean invoices. Consumption pricing breaks all three. When the bill can vary by 40% month to month, the CFO cannot approve the contract. The pricing conversation has focused so far on how to structure pricing as a seller. The buyer's challenge is different and deserves naming.
One practical mechanism the three levers above leave out: caps. Implement usage limits for each plan as a margin protection tool, then adjust them as you learn actual usage patterns. It is not sophisticated. But it prevents the growth paradox from becoming your problem before you have the data to solve it properly.
The valuation problem this creates
None of this stays inside the product team. It becomes a board problem the moment your pricing model changes. The metrics your investors use to evaluate your company were built for per-seat, not consumption. ARR is a clean metric for a per-seat business. It becomes unreliable for hybrid seat-plus-consumption models. NRR becomes more volatile as consumption ebbs and flows with client activity rather than contract renewal cycles. The Rule of 40 and public market multiples get harder to calculate and harder to compare. The benchmarks used to evaluate SaaS businesses for the past decade are becoming an unreliable guide.
Most pricing conversations start in finance and end in a spreadsheet. The companies getting this right have moved pricing into the product function and treat it the way they treat a roadmap: iterating fast, measuring behaviour, owning the outcome.
