How long custom business software actually takes
The right question isn't when it will be finished, but how soon the first piece stops wasting your time.
«How long will it take?» is the second question I get, right after the price. The honest answer is that it depends, but not in the way people usually mean: it depends less on how big the job is than on how many decisions have already been made.
And above all: it's the wrong question. The right one is how soon the first thing stops wasting your time.
It isn't delivered all at once
A complete system that goes live on day X is the most expensive way to discover, months later, that one screen wasn't needed and another is missing. You start from the piece that hurts most — the one somebody spends hours on every week — put it into real use, with real data, and watch what happens.
The first piece in production has three effects: it solves a problem, it teaches how you actually work, and it turns the rest of the project from a wish list into a priority list.
What actually stretches the timeline
In my experience the code is rarely the bottleneck. These three things are.
Decisions nobody has made. Who sets special prices? What happens on a partial return? Who releases an order blocked on credit limit? These are questions the company has never had to answer explicitly, because until now a person solved them with their head. Software forces them all to the surface at once, and every answer that's late is time standing still.
Dirty data. If the customer records need cleaning, that cleaning goes in the calendar. It isn't the developer's job, but the project doesn't move without it.
Access to third-party systems. Credentials for the existing system, permission from the website provider, access to the invoicing API. They look like details and they're weeks, because they depend on someone who doesn't have your project among their priorities. Ask for them on day one, not when you need them.
What shortens it
One person who decides, available half an hour a week, with the authority to say yes. Real data from day one, not invented. A test environment where the people who'll use the software can touch it while it's being built, and say «no, it doesn't work like that» while changing it is still cheap.
And a rule that always holds: better a few screens that do one thing well than twenty that do three things badly.
The order of magnitude
For reference, not as a promise: the first module in use is measured in weeks, not days and not quarters. A system covering the whole cycle — records, documents, stock, integrations — is measured in months, and keeps changing afterwards, because the company changes.
Anyone who gives you a precise date before seeing your data isn't estimating, they're selling. And anyone who tells you it can all be done in two weeks has misunderstood the problem — or is hoping you have.