SaaS

The model it replaced was a purchase: you bought a licence, installed the software on your own machines, and owned that version until you bought the next one. SaaS turns the same software into a rental that never finishes.

The customer gets no servers to maintain, updates that arrive without a project, and access from anywhere. The vendor gets predictable revenue, one codebase to support instead of a dozen shipped versions, and direct visibility into how the product is actually used.

The metrics that come with it

Because revenue arrives monthly rather than at signing, the whole business is measured differently, and these terms show up in every discussion of a SaaS company.

MRR and ARR — monthly and annual recurring revenue, the run rate of what is already subscribed.

Churn — the share of customers or revenue lost each period. It compounds, which is why a few points matter far more than they sound: 5% monthly churn means losing about half your customers in a year.

CAC and LTV — what it costs to acquire a customer against what they are worth over their lifetime. The relationship between those two numbers is the business.

Payback period — how long until a customer has repaid what it cost to win them. Under a year is comfortable; beyond two, growth consumes cash faster than it produces it.

What is changing now

AI has put pressure on one of the model’s assumptions: that per-customer cost is near zero. Serving a language model costs real money per request, so products with unlimited-usage pricing and expensive inference underneath have margins that look nothing like traditional SaaS.

This is why usage-based and hybrid pricing has returned after a decade of flat subscriptions, and why “seat-based pricing” is being questioned for tools where the work is done by an agent rather than a person.