Vijayanta Gupta

Essay

The SaaS company after SaaS

. 10 min read

The first growth curve of B2B software rewarded one behaviour: ship a workflow, charge per seat, expand inside the account, and let services close the gap between what the product promised and what the customer needed. That logic ran for fifteen years and built the category. It is now strained. The companies entering the second growth curve are not the ones with the longest AI feature list. They are the ones rebuilding the operating model underneath the product so that AI economics, not seat economics, set the shape of the business.

The decision sitting in front of every B2B software CEO is not whether to embed AI. That is settled. The decision is whether the company emerging from this cycle is recognisably the same business with new features, or a different business that happens to share the same logo. Most boards are still funding the first answer while the market has already priced the second.

The old growth logic, briefly

Seat-based pricing was the engine. It tied revenue to headcount, made expansion a sales-led motion, and gave the CFO a clean forecast. Gross margin sat in a comfortable corridor of 70 to 80 percent because the marginal cost of a new seat was effectively zero. Services revenue closed implementation gaps and dressed up product weakness without distorting the recurring line. Net revenue retention above 120 percent meant a company could miss new logo targets and still hit the plan. Multiples followed. At year-end 2024 the median public SaaS company traded at 6.2 times trailing revenue, a band the market had treated as defensible for most of a decade SaaS Capital Index via Aventis Advisors, March 2026.

That logic compounded as long as three conditions held. First, the customer’s workflow was structurally captured by the software. Second, expansion was cheaper than new acquisition. Third, marginal cost stayed near zero.

Where the logic now strains

All three conditions are moving.

Marginal cost is no longer zero. ICONIQ Capital’s January 2026 State of AI report finds inference averages 23 percent of total revenue at scaling-stage AI B2B companies, with surveyed AI product builders expecting average gross margins of about 52 percent in 2026, up from 41 percent in 2024 but still well below SaaS norms SaaS Mag, 2026. Per-token foundation-model prices fell 60 to 75 percent across 2025, and the savings did not flow through, because mature AI features added retrieval, self-critique, and intent-classification calls faster than per-token prices fell SaaS Mag, 2026. Across the public software index, a new corridor of 60 to 70 percent gross margin has established itself for any company shipping meaningful AI capability SaaS Mag, 2026.

Expansion is no longer cheap. Seat-based expansion assumed the next user is incremental headcount inside the customer. When AI compresses the team needed to do the work, the next user is not added; it is replaced. The expansion line that used to look like organic growth begins to look like a slow contraction the company is fighting against, while AI-native peers grow into the same accounts on outcome-based contracts.

Workflow capture is no longer structural. The workflow that the product monetised, the part where the user opened the software to do the thing, is being absorbed by agents that act across systems. Workflow obsolescence risk is the risk that the product boundary is now too narrow, the part of the value chain the software owns is precisely the part agents do best, and the customer no longer needs to enter the application to get the outcome. That risk is not a feature gap. It is a control-point shift.

The market is repricing all three at once. The median public SaaS forward revenue multiple sat at roughly 3.3 to 3.4 times trailing revenue at the end of Q1 2026, down from 6.2 times eighteen months earlier Aventis Advisors, March 2026. True AI-native platforms command 16 to 18 times revenue, with the steepest premiums for businesses where AI is the core product, not a layer on top SaaSRise AI Software Valuation Report 2026.

The premium and the discount are the same signal: the market believes the second growth curve is a different business model, not a feature cycle.

What AI changes in the value chain

The mechanism is straightforward and uncomfortable. AI moves the unit of value from the seat to the outcome, the unit of cost from licence to inference, and the unit of differentiation from feature parity to workflow ownership and proprietary data.

On pricing, the seat is still a defensible unit when the user makes the decision and the AI assists. It is not defensible when the AI makes the decision and the user supervises. Tomasz Tunguz has been explicit on the implication: when AI agents are two-and-a-half to three times as productive as a human, the seat price either rises by that factor or collapses into outcome-based capture Tomasz Tunguz, 2025. Salesforce already prices the AI-augmented seat at roughly 500 dollars against the 175-dollar non-AI seat, a near three-times step that signals where the curve is going for category leaders Tomasz Tunguz, 2025.

On cost-to-serve, inference is the new COGS line that boards will read in MD&A. Several public SaaS issuers in Q1 2026 began disclosing inference-cost ratios separately, typically four to nine percent of revenue at the COGS line SaaS Mag, 2026. The CFO question is no longer how to scale the gross margin from 75 to 80 percent; it is how to engineer the inference call graph, model mix, and caching so that the company holds 65 percent and does not slide further.

On control points, the question is what the customer cannot replicate. A copilot is not a moat. The moat sits in proprietary workflow data, distribution into bought-and-paid-for accounts, switching cost engineering, and the part of the workflow that crosses systems the customer cannot stitch together themselves. Roadmap quantity is not a moat either; investors are now diligencing the moat layer, not the feature layer, and the multiples spread between the two is the loudest signal in the public market.

The downstream consequences

Five consequences follow, and each one breaks a SaaS habit.

Go-to-market compresses. AI-driven research collapses the buyer journey before sales sees it. The pipeline of the future starts further down the funnel because the prospect has already used an AI agent to research, compare, shortlist, and partially evaluate. The seller no longer educates; the seller validates and proves. That changes the cost of customer acquisition, the role of marketing, the design of the demo, and the structure of the sales team. Companies still running the 2022 playbook are paying for motion that no longer compounds.

Pricing changes shape. The seat does not disappear, but it stops being the only line. Outcome pricing, throughput pricing, and consumption pricing arrive together, and the CFO carries three pricing models inside one contract. The 92 percent of AI software companies now running mixed pricing models in 2025 is not an experiment; it is the new operating reality trendingtopics.eu, 2025.

Services stop hiding product weakness. When the product is run by agents, the services hours that used to mask implementation gaps become visible as gross margin drag. The SaaS habit of treating services as a low-margin onboarding cost no longer survives audit. Either services are productised into reusable agent pipelines, or they show up as the line that explains why gross margin is not where the board expected it to be.

Capital allocation changes. Twenty cents on the dollar that used to go into headcount expansion now goes into model orchestration, evaluation infrastructure, and inference cost engineering. Governance is the fastest-growing line item inside enterprise AI budgets in 2026, at 8 to 12 percent, up from 3 to 5 percent two years earlier Presenc AI, 2026. The board that funds AI features without funding the operating spine behind them is buying a story without the scaffolding.

Headcount strategy is the most exposed line of all. The Klarna case is the canonical 2026 cautionary tale: a 40 percent reduction in workforce, from 5,527 to 3,422, with an AI agent claimed to do the work of 700 customer service staff and 40 to 60 million dollars of cost taken out, followed by a public reversal in 2025 when service quality collapsed and the company began rehiring humans CNBC, May 2025; Time, 2025. The pattern is now well documented enough that every CEO sketching an AI workforce plan is asked to explain how their plan avoids the Klarna outcome.

The counterargument, fairly stated

The strongest objection is that SaaS economics will reassert themselves once inference prices fall enough. Per-token costs have already collapsed and will fall further. Models will commoditise. The structural floor on gross margin will lift. This view is held seriously inside parts of the investor community.

It misreads where the gross margin pressure is coming from. The cost is not in the token; it is in the call graph. Inference call complexity has been rising faster than per-token prices have fallen for two years now, and there is no operating reason to expect that to reverse for products that promise quality, reliability, and reasoning SaaS Mag, 2026. Even if per-token economics improve another order of magnitude, the same product features will pull more calls per outcome, not fewer. The corridor of 60 to 70 percent gross margin is structural, not cyclical.

The board agenda

Five questions belong on the board meeting that picks up the second growth curve seriously.

What workflow do we monetise today, and what part of it survives autonomous agents inside the next thirty-six months? If the honest answer is that the agent removes the workflow, the pricing model has to change before the workflow does.

What is the unit economics of one AI-enabled outcome at full inference cost, with the eval and observability layer included? If the company cannot produce that number for the board, the gross margin guidance the company gives the market is provisional.

Where is the moat layer, and how is the company investing in it? Proprietary workflow data, distribution, and switching cost engineering need explicit funding lines. They will not appear by accident on a roadmap.

What does the next twelve months of go-to-market look like if half the early-stage buyer research has already been done by an agent before sales engages? The implication touches headcount, demo design, content investment, and the pricing of the first contract.

What part of the current company does the second growth curve not need? Every transition the category has been through, on-premise to SaaS, perpetual to subscription, has carried a redundant operating layer for two or three years longer than it should have. Naming the layer earlier is the cheapest decision the board will make this year.

The CEOs who will run a second curve business in 2028 are deciding now what to stop funding, what to test, and what to redesign. The decision is not which AI features ship next quarter. It is which company is still here after.

Sources

SaaS, AI-native, pricing, margin, operating-model

Back to writing