The SaaS Downfall — and the Three Archetypes That Survive

SaaS is not dying. Undifferentiated SaaS is. The decade-long subscription party — built on 0% rates, 80% gross margins, and the assumption that any well-designed CRUD app could compound at 140% net dollar retention — is over. What replaces it is not a crash. It is a sorting. The companies that survive the next decade will look less like 2018 startups and more like specialty banks: fewer, deeper, with moats that compound rather than vapourise.

I’ve spent the last decade shipping fintech and AI platforms for companies whose pitch decks all shared the same vocabulary: ARR, net dollar retention, Rule of 40. It was a good vocabulary. It described a world. That world ended, and the clearest way to see what comes next is to stop arguing about the sector and start sorting it by archetype.

The diagnosis (briefly)

Four shifts arrived between 2022 and 2026. They compounded.

• The cost of capital stopped being zero. A revenue multiple that made sense at $r=0.01$ looks obscene at $r=0.05$. The same business — same customers, same product — became worth roughly half. Nothing on the income statement changed. The denominator did.
• AI compressed the product moat. A single engineer with a frontier model now stands up in a weekend what a 2018-era company raised a $15M Series A to deliver. The buyer’s rebuild-vs-buy calculation tilted, hard.
• Budgets consolidated, then contracted. Mid- market and enterprise buyers learned, under duress, they were running 180 SaaS tools and could mostly get by on 60. Procurement became a defence rather than a rubber stamp. Net dollar retention fell across the sector.
• Distribution got harder, not easier. Outbound, paid acquisition, and SEO landing pages are saturated with auto-generated content. Buyers evolved real immunity. CAC rose, conversion fell, both sides of the LTV/CAC ratio worsened in different rooms at the same time.

That is the diagnosis. The interesting question is what the survivors look like — and why their economics are durable in a way that horizontal SaaS no longer is.

The three survivor archetypes

1. The regulated-vertical SaaS

The first archetype owns a workflow that is illegal or punishing to do badly. Veeva (life-sciences, GxP-validated). Toast (restaurant point-of- sale, payments, payroll). Procore (construction project management with regulatory document trails). Guidewire (insurance policy administration). Each of these companies has a customer set whose alternative is not “build it in a weekend with Claude” — it is “fail an audit, get sued, lose a license.”

The math:

• NDR is durable above 110% because switching means re-validating against a regulator. The cost of switching is paid in compliance overhead, not in feature parity.
• CAC is high — vertical sales motions are slow — but amortised over 10–15 year customer lifetimes, vs 3–5 for horizontal SaaS.
• Gross margins compress slightly (60–70%, not 80%) because of compliance and integration overhead. The moat is not gross margin. The moat is the regulatory wall.

2. The services-as-a-software archetype

The second archetype takes work previously done by a consulting firm, an agency, or an in-house team — and replaces the humans with AI agents while keeping the outcome contract. Cresta for sales coaching. Harvey for legal research. Sierra and Decagon for customer support. Klarna’s in-house AI agent, which displaced a $40M/year customer-service contract. An entire new generation of “we don’t sell software, we sell resolved tickets.”

This is not a SaaS business in the 2018 sense. It is a services business with software margins. The economics:

• Pricing is per-outcome, not per-seat. Resolved ticket, qualified lead, drafted contract, processed claim. Per- seat is a tax on your own efficiency the moment software does the work humans used to do.
• Gross margins land at 50–70% — better than agency (10–20%), worse than horizontal SaaS (80%). Inference is a real cost-of-revenue line, not a footnote.
• The moat is not the model. Everyone has GPT-5 and Claude Opus. The moat is the workflow integration: data plumbing, human-in-the-loop fallback, the feedback loop that makes the agent better at this customer’s job quarter by quarter.

The trap in this archetype: many “AI agent” companies are wrappers with no workflow ownership. They die when the customer realises the work is one prompt and a junior engineer away from being done in-house. Workflow ownership — the data, the integrations, the institutional memory — is what nobody rebuilds in a weekend.

3. The distribution-owner archetype

The third archetype made the product cheap and the distribution channel expensive. Shopify owns the merchant relationship; the apps that run on Shopify are commodities. Klaviyo owns the email channel for those merchants; the message itself is commodity. HubSpot owns the small-business CRM gravity well; the individual features are increasingly commoditised. Atlassian owns the developer-tools attention.

In a world where AI made building software cheap, distribution is the part that didn’t get cheaper. Owning where buyers find their software is now worth more than owning the software itself.

• NDR comes from selling new categories into the existing distribution relationship — not from price increases. Shopify Payments. Shopify Capital. Shopify Fulfillment. Each one is a new SaaS attached to the same merchant.
• Margins look healthy because the marginal cost of adding a category is near-zero once the channel is built. The expensive thing was acquiring the merchant in 2017.
• The moat is rebuilt every time a new vendor pays to acquire a customer that already lives on your platform. Distribution compounds.

The new Rule of 40 — one per archetype

The Rule of 40 — growth + profit margin > 40% — was a horizontal-SaaS heuristic. Each surviving archetype has its own equivalent, and grading any of them by the original Rule of 40 will mislead you.

• Regulated-vertical: NDR > 115%, sustained through one full budget-tightening cycle (which we have now had). Growth is secondary; retention is the moat.
• Services-as-a-software: Outcome-payback < 6 months. The cost of acquiring and onboarding the customer must be recovered before the customer notices they could have hired their own engineer to build it.
• Distribution-owner: Cross-sell attach rate (additional categories sold per customer per year) > 1.0. A platform that isn’t expanding into new categories is a platform whose distribution is decaying.

What does not survive

The undifferentiated horizontal CRUD app — the 2018 prototype of SaaS — is the casualty. Per-seat pricing, 80% gross margins, general-purpose workflow, no regulatory or distribution moat. The pitch deck used to read “Notion, but for X.” That category is dead, because Notion itself is one prompt away from being rebuilt for X by anyone who needs it.

The companies in this category will not all fail. Many will be acquired by distribution-owners (archetype 3) at multiples of revenue, not multiples of ARR. Others will compress to bond-like cash-flow yields and trade like utilities — perfectly fine businesses, just nothing like the dream of 2018. The graveyard is the venture-backed ones priced for compounding that never compounded.

What to do, if you’re building

Pick your archetype before you pick your product. Each one has a different go-to-market, a different gross-margin target, a different definition of what “good” looks like. Building a regulated-vertical product, with services-as-a- software pricing, sold through a distribution-owner channel, is the dream. Building a horizontal CRUD app and hoping for an exit is the trap.

• If you’re regulated-vertical: spend the first eighteen months earning the certifications (SOC 2 is the floor; HIPAA, GxP, FedRAMP are the moat). Customer acquisition is slow on purpose. The moat compounds from year three.
• If you’re services-as-a-software: price per outcome from day one. Resist the per-seat instinct even when investors push back. Per-seat caps your upside at the headcount of the customer; per-outcome caps it at their spend on the underlying problem.
• If you’re distribution-owner: the product is the channel. Optimise for category-attach rate, not feature parity with point solutions. Features are easy to copy; channel is not.
• If you can’t name your archetype: you are an undifferentiated horizontal CRUD app. Stop, and pick one. Hoping the market sorts itself out for you is the most expensive form of optimism in software.

The tide went out three years ago. The 2026 cohort of founders is being built differently — narrower, deeper, more expensive to switch off. The companies still pitching “Notion but for X” are the last generation of a pattern that already ended; the companies pitching “the regulated workflow for X” or “resolved-ticket pricing for X” or “the channel for X-shaped buyers” are the next.

That is not a downfall of software. It is a downfall of the 2018 theory about software. The companies being built right now, quietly and unfashionably and profitably, will be very glad they understood the difference early.