← Notes

27 July 2026 · 7 min read

What bespoke software development actually costs

Nobody publishes real numbers for bespoke software development, which is why every quote feels like it was invented on the spot. Some of that secrecy is self-serving. Most of it is that the honest answer genuinely is "it depends" — so here is what it depends on, in plain terms.

The number is driven by scope, not hours

The single biggest cost driver is not how fast anyone types. It is how many distinct things the software has to do, and how many ways each of those things can go wrong.

A tool that lets five people log a job, see the list, and mark it done is a small build. The same tool with roles and permissions, an audit trail, email notifications, an export, and a customer-facing portal is not slightly bigger — it is several times bigger, because each addition multiplies the states the system can be in.

When you ask for a price, you are really asking someone to guess at that multiplication. That is why a good studio will push hard on scope before quoting, and why a quote that arrives instantly is usually a quote that will change later.

What actually adds cost

In rough order of impact:

• Integrations. Every external system — a payment provider, an accounting package, someone's legacy database — brings its own failure modes. Two integrations are more than twice the work of one.
• User roles. "Admin and everyone else" is cheap. Five overlapping permission levels is a project in itself.
• Data migration. Moving ten years of messy spreadsheet history into a clean structure is frequently the largest single line item, and it is almost always underestimated.
• Compliance. If you handle health, financial or children's data, the requirements are real and they are not optional.
• Undefined edge cases. "What happens if two people edit the same record?" is a question that either gets answered up front for a small cost, or gets discovered in production for a large one.

What barely adds cost

Also worth knowing, because people often spend their budget in the wrong place:

• Design polish, if it is decided before the build rather than during it.
• Extra pages that reuse existing patterns.
• Most cosmetic changes, if the underlying data model is right.

The pattern is consistent: structural decisions are expensive to change, surface decisions are cheap. Spend your thinking time on the former.

Fixed price versus hourly

We quote fixed scope at a fixed fee, and we are not neutral on this, so take it as an argument rather than a fact.

Hourly billing puts the risk of a bad estimate entirely on you, and quietly rewards slow work. Fixed price puts that risk on the studio, which forces the scoping conversation to be rigorous before anyone commits. The trade-off is that changing your mind mid-project has to be handled as a change to the scope, rather than absorbed silently.

Neither model is dishonest. But if you are given an hourly estimate, ask what happens when it is exceeded — the answer tells you who is carrying the risk.

When you should not commission bespoke software

This is the part most studios leave out.

If an off-the-shelf product does 80% of what you need and the remaining 20% is convenience rather than necessity, buy the product. A £40/month subscription that you can cancel beats a five-figure build you have to maintain forever.

Bespoke earns its cost when the software encodes something specific to how your business actually works — a process that is genuinely yours, that the generic tools force you to fight. If you cannot name that thing in a sentence, you probably are not ready to commission a build yet.

How to get a number you can trust

Ask for three things before you accept any quote:

1. A written scope stating what is included and, more importantly, what is explicitly not.
2. A named process for changes — not "we'll be flexible", but what actually happens to price and timeline when you add something.
3. A straight answer on ownership. You should own the code, the accounts and the domains from day one.

If a studio will not put those in writing, the number they quoted is not really a number.