The SaaS Model Didn't Die, It Learned to Prove Its Worth
There is a precise moment in the cycle of any business model where the collective narrative stops describing reality and starts producing it. The SaaS sector reached that moment more than a year ago, and the industry is still processing what it means. It is not the collapse that some anticipated with the term "SaaS-pocalypse," but neither is it a frictionless return to the growth of 2021. What is happening is more uncomfortable and more useful than either of those two versions: buyers, investors, and capital markets are demanding proof that the software they contract actually changes something measurable in the people who use it.
That seems obvious. It wasn't.
During the period of abundant liquidity that followed the pandemic, the recurring revenue model functioned as a sufficient argument in itself. Having a reasonable renewal rate and an upward-sloping growth curve was enough to sustain valuation multiples that today seem difficult to justify. Multiples on ARR—annual recurring revenue—reached levels that discounted years of future growth as if that growth were a structural certainty. According to data from SaaS Capital, those multiples are now at lows not seen in over a decade, which reflects not only an interest rate adjustment, but a shift in what kind of evidence is considered convincing when allocating capital.
The sector is not collapsing. The projection toward 2030 still points to more than 900 billion dollars in global market value, with a compound annual growth rate of around 18%. For 2024, spending on software as a service was estimated at close to 232 billion dollars according to Gartner. The absolute figures do not contradict the expansion narrative. What changed is the quality of evidence required for those figures to translate into favorable valuation. And that change has very concrete operational implications for those who build or finance these businesses.
Artificial Intelligence Doesn't Replace Software; It Pressures Its Justification
Part of the narrative panic around the "end of SaaS" comes from a hasty reading of the role that artificial intelligence is playing in the sector. The simplified argument goes: if AI can generate code on demand, build autonomous workflows, and replicate functionalities that previously required annual contracts, then per-seat subscription models lose their reason for being. There is something true in that pressure. There is also much that exaggerates the speed of change and underestimates the real friction involved in adopting it.
What AI is effectively doing is lowering the marginal cost of producing generic software. That puts pressure on the layers of the market that competed primarily on standard functionality at an acceptable price. A basic CRM, a task management tool, a forms platform: these categories face real price compression because the barrier to replication has been reduced. But software that operates at the intersection of specific industrial processes, proprietary data flows, and deep business logic cannot be replicated with a prompt. Complexity has not disappeared; it has been redistributed.
Companies that buy specialized software for industries such as fashion, manufacturing, or logistics are not acquiring isolated functionalities. They are buying the accumulation of operational knowledge that took years to codify, plus the infrastructure for integration with legacy systems, plus built-in regulatory compliance. None of those things are generated in real time with an assisted coding tool. What AI can do in this context is accelerate the detection of anomalies in those processes, automate repetitive decisions within already-designed workflows, or connect data sources that previously required costly manual integration. That does not destroy the model: it forces it to demonstrate with greater precision than before exactly where it is generating incremental value.
The true effect of AI on the SaaS sector is not one of substitution but of demand. It forces providers to be more specific about which part of the customer's problem they are solving, with what measurable outcome, and under what conditions of adoption. That demand existed before; AI makes it impossible to avoid.
When Pricing Stopped Being Per Seat and Started Being Per Outcome
The shift in pricing models that is accelerating across the sector has implications that go far beyond contractual mechanics. The move from per-user licenses toward consumption-based or outcome-based pricing fundamentally alters how risk is distributed between provider and customer, and what kind of operational relationship is needed for the model to function.
In the per-seat model, the provider charges for access and the customer assumes the risk of adoption. If users do not use the tool, the contract renews anyway until someone in procurement reviews the invoice. That decoupling between usage and payment was for years a source of comfortable margins for providers and a source of silent waste for buyers. The data from Zylo on "shadow AI" and unpredictable software costs is not an anomaly: it is the contemporary expression of a structural problem that existed long before AI entered the picture.
The outcome-based model closes that decoupling by force. If the contract specifies that the provider charges in proportion to the reduction in productive cycle time, or to the percentage increase in conversion rate, or to the decrease in errors in an operational process, then the relationship becomes verifiable. That is good for customers. For providers, it implies that they need to instrument their products with enough depth to measure those results reliably, and they need to have the conviction, backed by data from previous customers, that the product actually produces them.
Here a trap emerges that deserves attention. Several providers who are adopting the language of "value-based pricing" or "outcome-oriented models" are not building the measurement infrastructure that model requires. They are using the vocabulary of outcomes without the verification apparatus that would make them credible. That is not a model transition; it is a cosmetic reframing of the previous contract. The difference between an outcomes model and an access model with outcomes-based marketing lies in whether the provider accepts that its compensation will vary when results do not materialize. Very few actually accept that in practice.
The move toward consumption-based pricing being adopted by infrastructure platforms and some application layers is a more honest version of this transition. The customer pays for what they use, the provider has an incentive for usage to be high because it reflects real adoption, and both parties have visibility into the relationship between activity and cost. The problem is that this model introduces budget volatility that finance teams at large organizations handle poorly, which generates internal resistance in the purchasing process even when the product is superior.
Vertical SaaS Has an Advantage That Horizontal Cannot Copy Quickly
The distinction between horizontal and vertical software has always existed, but the current market pressure is making it more strategically relevant. Horizontal software competes on adoption scale and functional breadth. Vertical software competes on depth of domain understanding and on the switching costs that depth generates. In an environment where AI is lowering the cost of producing generic functionality, the horizontal advantage erodes faster than the vertical one.
A provider that has spent ten years building software for the textile production chain has embedded in its product a knowledge of traceability standards, material waste logic, integration with specific machinery, and regional regulatory compliance that cannot be replicated by copying its interface. That knowledge took years to translate into software logic because it required real conversations with plant operators, with production directors, with quality auditors. AI can accelerate the next iteration of that product. It cannot compress the ten years of accumulated learning embedded in the current one.
The metric that best predicts the durability of a vertical SaaS is not the growth rate of new contracts, but net revenue retention, which measures whether existing customers are expanding their usage and spending over time. According to available data, investors and lenders are using this metric, alongside gross revenue retention, as the most reliable indicator that a product has genuine stickiness. Net retention above 110% indicates that expansion within the installed base is compensating for customer churn, making the model self-sufficient in terms of growth without relying exclusively on new customer acquisition.
That pattern is harder to build in horizontal software because it requires customers to find reasons to expand usage within the same platform, and those reasons compete with the offerings of dozens of alternatives that do the same thing with minor variations. In the vertical, expansion occurs more naturally because the provider has visibility into other friction points within the same operational process it already knows well. The next problem to solve is right next to the one it already solved.
The trap for vertical providers is confusing depth of domain knowledge with strategic comfort. Knowing an industry well is an initial advantage, not a permanent guarantee. If that knowledge is not updated with changes in the customer's processes, with new regulations, with the evolution of available technology, it becomes technical debt disguised as specialization. The providers that maintain high net retention over time are those that continue learning from the customer with the same intensity after year three as they did in year one.
Durability Is Not a Narrative, It Is a Revenue Architecture With Real Friction
The SaaS sector arrives at 2026 having passed through a stress test that was part valuation correction, part macroeconomic adjustment, and part genuine pressure from new technologies. What emerges from that process is not a broken model in search of narrative redemption. It is a more legible model, where the metrics that always should have mattered—retention, expansion within the installed base, customer acquisition cost relative to lifetime value—are now receiving the attention they deserved long before now.
The term "SaaS-pocalypse" accurately described the fear, not the reality. What is happening is a selection process within the sector: providers that built their growth on real adoption, on demonstrated retention, and on organic expansion within their customer base are passing through the filter with their models intact. Those that grew on the ease of selling in a technology spending environment without scrutiny are facing the cost of having deferred that conversation.
For those who build or finance these businesses, the adjustment does not require a philosophical change but rather a clarity that the previous context made optional: demonstrating, with their own verifiable data, that the product produces the result that justifies the contract. That is not a new standard imposed by the market. It is the standard that always defined whether a business model was sound. The market has simply stopped accepting the option to defer it.
