COGS < CAC, Always

The first principle of SaaS economics is mathematical, not strategic: the cost of serving a customer must be less than the cost of acquiring one. If it isn't, no amount of growth saves you. You are losing money faster the better you sell.

That sentence is obvious. The corollary is not: every other interesting business decision — pricing tier design, infrastructure choice, free-tier scope, hiring plan, sales-led versus PLG motion — is downstream of that one inequality. Most founders treat it as one consideration among many. It isn't. It's the consideration the others have to fit inside of.

The two numbers and what they actually contain

COGS — Cost of Goods Sold — in SaaS, mostly means:

• Hosted infrastructure (compute, storage, egress, third-party APIs you proxy).
• Customer-success and onboarding cost for the median customer.
• Per-seat licensing of anything you embed (Stripe fees, AI inference, observability cost-per-event).

It does not mean R&D, marketing, or sales. Those are above the line. COGS is what it costs to keep an existing customer served, this month, on the plan they're on.

CAC — Customer Acquisition Cost — means:

• Sales salaries (loaded), allocated to closed-won deals.
• Marketing spend, allocated to acquired customers.
• Tooling, content, events — anything whose purpose is to produce a new logo at the top of the funnel.

It does not include onboarding once they're a customer. That's COGS.

The first time a founder writes these two down honestly, two things usually happen: the COGS number is bigger than they thought, and the CAC number is much, much bigger than they thought.

The interesting place is the ratio

A single-period comparison isn't enough. The real question is the ratio across the customer lifetime:

Lifetime Value ÷ Customer Acquisition Cost — the LTV/CAC ratio.

The textbook target is 3. Below 1 and you're lighting money on fire. Between 1 and 3 and you have a small business, not a venture-scale one. Above 3 and you have room to invest in growth.

But the textbook number assumes COGS is well-behaved. It almost never is. When founders dig in, they tend to find:

• A handful of enterprise customers consuming 10x the infra of the median.
• A free tier whose COGS quietly approaches the COGS of a paid tier.
• An AI feature whose inference cost per active user is roughly the entire margin on the plan.

Each of those is a pricing or product decision masquerading as an engineering problem.

What this implies, concretely

Three rules that fall out of the inequality, in order of how often they're violated:

Price the highest-COGS feature explicitly. If 80% of your variable cost comes from one feature — say, an AI inference workflow — that feature should have its own price tier or its own usage meter. Hiding it inside a flat seat price means your worst customers subsidize on your best ones.

Audit COGS quarterly, not annually. Cloud bills don't grow linearly with customer count. They grow stepwise — a new region, a new compliance requirement, a new managed service — and each step changes the math. A quarterly review catches drift before it becomes a pricing emergency.

Refuse the deal whose COGS exceeds the contract. This sounds obvious. It isn't. There is always pressure to take the marquee logo at a discount, or to ship the custom data-residency requirement for the strategic enterprise account. Sometimes that's correct as a strategic bet. It is never correct as a unit-economics bet. Know which one you're making.

The reason this matters more than it should

The companies that get this wrong don't fail dramatically. They fail slowly, and they spend the years between "promising" and "out of money" telling themselves that growth will fix the margin. It won't. Growth without unit economics is a more efficient way to lose money.

The companies that get it right tend to have a single sentence written down somewhere — in a Notion doc, in a board deck, in a founder's head — that says exactly what the gap is between what it costs to serve a customer and what they pay. Once that sentence exists, every other decision gets a little bit easier.