SaaS (Software-as-a-Service)

Definition

SaaS is a software distribution model in which a vendor hosts an application in the cloud and provides access to customers on a subscription basis. The vendor manages infrastructure, maintenance, updates, and security — customers pay for access, not ownership.


Core Ideas

Business Model Mechanics

Key Metrics

  • MRR (Monthly Recurring Revenue) — baseline monthly subscription revenue
  • ARR (Annual Recurring Revenue) — MRR × 12
  • Churn Rate — percentage of customers/revenue lost per period
  • LTV (Lifetime Value) — average revenue per customer over their entire relationship
  • CAC (Customer Acquisition Cost) — total sales/marketing spend ÷ new customers
  • LTV/CAC Ratio — must be > 3 for a healthy business; > 5 is great

Revenue Formulas

  • MRR = (# customers) × (avg monthly revenue per customer)
  • Net Revenue Retention = (MRR end of period from existing customers) ÷ (MRR start of period)
  • Payback Period = CAC ÷ (MRR per customer × gross margin)

Pricing Models

ModelDescriptionBest For
Per-seatPrice per user/monthTeam collaboration tools
Usage-basedPrice per API call, GB, eventInfrastructure, developer tools
FreemiumFree tier; charge for advanced featuresConsumer products, bottoms-up B2B
Flat-rateSingle price for all featuresSimple products, small markets
TieredFeature-gated tiers (Starter/Pro/Enterprise)Most B2B SaaS

The Pricing Ladder (a worked case)

Pricing rarely starts where it ends. One observed progression, from a solo trading-data product (TradeInsight) into an AI agent product:

  1. Free — first three months, ~300 users. The signal being bought here is do people like it, nothing else.
  2. Paid, priced per period — conversion was poor; the cancellation reason volunteered most often was simply “too expensive.”
  3. Subscription + usage credits — a plan that unlocks the same capability across CLI and web UI, metered by consumption.
  4. Enterprise tier — for the buyers that need contracts rather than credit packs.
  5. Pay-by-output (aspirational) — the customer pays only for output produced.

Two things are worth extracting:

  • Retention data beats conversion data for diagnosing pricing. Across two years: ~1000 free users, 221 people who ever started a trial or paid, and 17 still there. A funnel that leaks 92% of paying-intent users is a product problem wearing a pricing costume — repricing would not have fixed it.
  • Pay-by-output requires confidence and infrastructure. It only becomes viable with faster product iteration, scalable infrastructure, and an enterprise motion — which is why this rung usually stays a long-term goal rather than a launch strategy.

The Vertical Harness Pattern

A specific shape of AI-native vertical SaaS: take a general-purpose agent harness that has proven itself in one domain, and rebuild it for another.

The template case is Claude Code — it stood out among AI agents by being CLI-first and narrowly coding-focused before expanding. The transplant: “Claude Code for coders, X for quant analysts and traders” — the harness, tools, and domain skills wrap the LLM, and the model itself is initially bring-your-own-subscription, later served through the platform.

Why the shape works commercially:

  • The harness, not the model, is the product — so model progress is an input, not a competitor.
  • Vertical depth (domain skills, data access, evaluation) is the moat, matching the vertical-vs-horizontal table above.
  • BYO-model at launch removes inference COGS from a pre-revenue product; bringing models in-platform later converts that cost into a pricing lever.

Go-To-Market

  • Bottom-up (PLG) — product-led growth; users adopt, then companies buy
  • Top-down (SLG) — sales-led; direct outreach to enterprise buyers
  • Community-led — build audience around a problem, then productise
  • Content-led — SEO, blog, YouTube → inbound leads

Vertical vs Horizontal SaaS

VerticalHorizontal
TargetOne industry (e.g. legal, healthcare)Cross-industry (e.g. CRM, HR)
CompetitionLess direct competitionHighly competitive
Pricing powerHigher (specialist expertise)Lower (commoditised)
Market sizeSmallerLarger

AI-Era SaaS Considerations

  • AI as a feature (copilot, summarisation, classification)
  • AI-native products built on LLMs from day one
  • Moats: proprietary data, fine-tuned models, deep workflow integration
  • Cost structure: inference costs as COGS; margins compressed vs traditional SaaS

Relationships