Is building software still a good business?

Software has economics nothing else on this list has: near-zero cost to serve one more customer, and revenue that recurs. It also has a failure mode the others do not — you can grow revenue every month and still be dying.

Part of the common ways of making money.

Software has one property nothing else on that list has: serving the thousandth customer costs almost nothing more than serving the first.

That single fact produces both the extraordinary upside and the specific way these businesses fail.

The economics

Costs are almost entirely upfront. Building the thing takes months or years and is paid for before anyone pays you. After that, an additional customer costs a fraction of a penny in servers.

Revenue recurs. A customer who paid last month usually pays this month. Revenue compounds rather than resetting — unlike an agency, which starts every month at zero.

Gross margins are very high, commonly 70–90%. Compare with ecommerce, where the product costs real money every time.

That combination is why software companies command higher valuations than almost any other kind. It is not fashion; it is the arithmetic.

The failure mode nobody sees coming

Here is what the economics obscure: you can grow revenue every single month and be dying.

Because revenue recurs, so does its absence. Every customer who leaves takes not one payment but all their future payments. That is churn, and it is the number that decides these businesses.

The arithmetic is unforgiving. At 5% monthly churn, you lose about half your customers in a year. To grow at all you must replace half your base and then add more. Growth becomes running to stand still, and the treadmill accelerates as you get larger — 5% of a bigger number is a bigger number.

This is why a business can post record revenue and be structurally failing. The top line looks fine while the tank drains.

Retention is the business. Acquisition is what founders talk about and retention is what determines whether the thing exists in five years.

Where the money comes from

A customer pays because the software saves them money or time, or earns them more. Straightforward — but two consequences follow that shape everything.

Business customers pay far more than consumers for the same underlying work, because a business can compute what it is worth: an hour of staff time saved has a price. Consumers compare against free.

The price you can charge is set by the value delivered, not by what it cost you to build. This is the reverse of most businesses and it is why software pricing feels arbitrary. It is not arbitrary; it is anchored to the customer’s savings, and you often cannot see that number.

What AI changed

Both directions at once, and being honest about it means acknowledging both.

Building got dramatically cheaper. A small team now ships what recently required a large one. Lower barriers to building.

Which means lower barriers for everyone. If your advantage was that the thing was hard to build, that advantage is smaller each year. Competitors arrive faster and copies are cheaper.

So the durable advantages have shifted away from the code itself toward things a competitor cannot replicate by building the same features: the data you accumulate, the workflows customers embed into, the integrations, the trust, and the cost of switching.

Which is the same lesson as everywhere else on this site. The code is the activity. The defensibility is the structure.

What the evidence says

Reasonable for funded companies, poor for the rest.

Venture-backed startups are studied because investors track them. The well-known finding is that most fail and returns concentrate in very few — a power law again, and a severe one.

Bootstrapped software businesses are much less visible. Many small, profitable ones exist quietly and never appear in any dataset because nobody is required to report. This is one place where the visible sample genuinely understates the base rate of modest success, in contrast to creator income where it overstates it.

What is reliably documented is the shape of the metrics: churn, acquisition cost, and the ratio of customer lifetime value to what it cost to acquire them. Those are the numbers operators actually manage, and they are worth understanding before starting.

Which channel this is

Genuinely ownership, and one of the few unambiguous cases on the list.

The business produces revenue whether or not you worked today. It can be sold, and it sells for a multiple of recurring revenue precisely because the buyer is purchasing a stream rather than a job. It compounds.

The cost is that the early period is severe: months or years of building with no revenue, funded by savings, a job, or investors. That is the entry price, and it is why so many people stop at freelancing — which pays immediately and never compounds.

If you are considering it

Sell before you build, if you possibly can. Charging someone before the thing exists is the only real evidence of demand. Everything else is opinion.

Solve a problem someone already pays to solve. An existing budget means the value is proven. Creating a new category is far harder than people expect.

Measure churn from your first ten customers. It is the number that decides the outcome and the one founders look at last.

Prefer business customers, unless you have a specific reason not to. They pay more, churn less, and can articulate what the thing is worth.

Assume building is no longer the moat. Ask what you will have in three years that a competent competitor cannot assemble in three months.