What is an MVP in software? A founder's guide to the first version

An MVP, or minimum viable product, is the smallest version of a product that real users can get value from, so you can learn what works before building more. Here is what belongs in one, examples, and the mistakes founders make with AI products.

An MVP, or minimum viable product, is the smallest version of a product that real users can actually use and get value from, so you can learn whether the idea works before you spend more building it. It isn't a demo or a mock-up. It does one job well enough that someone would use it, and ideally pay for it.

What is an MVP in business?

In business terms, an MVP is a way to reduce risk. Instead of building everything you can imagine and hoping people want it, you build the one thing you most need to prove, put it in front of real users, and let their behaviour tell you what to do next.

What is an MVP in software development?

In software, an MVP is a working product with:

• One core job, done well, for one specific kind of user.
• The essentials around it: accounts, the data it needs, basic security and a way to get help.
• A way to learn: a measure of whether it is working, such as tasks completed, repeat use or payment.

Everything else waits. The discipline is in deciding what to leave out.

Minimum viable product examples

• A booking tool that lets customers book one type of appointment with one business, before supporting many services, staff and locations.
• A document tool that turns one kind of document into a spreadsheet, and checks the result, before handling every format. Our own product BankStatementPro started from exactly one job: turning a PDF bank statement into a checked spreadsheet.
• A marketplace that matches one type of buyer with one type of seller in one place, run partly by hand at first, before automating the rest.

What makes an AI MVP different

AI products add three questions that a normal MVP doesn't face:

1. What exactly does the AI do? "It uses AI to help" isn't a job. "It reads a bank statement and returns every transaction" is. Name the input and the output.
2. What happens when it's wrong? Every AI makes mistakes. Decide who notices and how they correct it: a person reviews results, or the user can fix them easily.
3. How good is good enough? Collect 20 to 50 real examples with the right answers, and agree the bar the AI must meet before launch.

AI also has a running cost per use. Make sure your pricing covers it before you have many users.

Mistakes founders make with a first version

• Building five features partly instead of one feature well.
• Skipping real users. Build for people you've spoken to, not people you imagine.
• Confusing a prototype with a product. A prototype built quickly with an AI tool proves the idea. The product has to survive real users: logins, payments, data that mustn't leak, and results people can trust.
• Forgetting life after launch. Hosting, fixes and improvements start the day real users arrive.
• No measure of success. Decide the one number you'll watch before you launch.

What does MVP development cost?

It depends on what the first version has to do, who uses it, what it connects to, and whether it includes AI. That is why we don't publish a price: we scope the first version with you and then give you a fixed price. Our post on what drives the cost of custom software explains the factors.

Before you build

Our 15-point AI product checklist helps you check whether your idea is ready to build, and what to fix first. If you'd like it built properly, read how we build AI products or book a free consultation.

How to decide what goes in the first version

A simple method that works for most products:

1. Write the one sentence. "[This user] can [do this job] so that [this outcome]." If you need two sentences, you have two products.
2. List everything you want to build, without filtering.
3. Mark each item: needed for the one sentence to be true, or not.
4. Keep only the needed items, plus the essentials: sign-in, the data, basic security and a way to get help.
5. Write the "not yet" list with everything else. It is not a rejection; it is the order of work after launch.
6. Pick the one number that tells you it is working.

A launch checklist for your first version

• Ten or more real users lined up to try it.
• The core job works end to end, including when things go wrong.
• For AI products: tested on your real examples against the bar you agreed.
• Sign-in, backups and error tracking in place.
• A simple way for users to report problems.
• The one number you'll watch, and a date to review it.

Frequently asked questions

How long should an MVP take to build? Long enough to do one job properly, and no longer. The scope decides the time, which is why cutting scope is the fastest way to launch.

Should an MVP be built with no-code or AI builder tools? They are great for testing an idea quickly. Once real users and real data depend on it, you need something built to hold up.

What comes after the MVP? Watch how people use it, fix what gets in their way, and build the next most valuable feature, one at a time.