What actually counts as an MVP
An MVP is the smallest version of your product that lets a real customer complete one genuinely valuable journey, well enough that you can measure whether they come back and do it again. The operative word is 'one'. Not the core three journeys you eventually need — one. The discipline of an MVP is entirely in what you leave out, not what you include.
It is not a prototype. A prototype (a Figma click-through, a no-code mockup) tests whether people understand the concept. An MVP is real, running software that a customer actually uses to get something done, with real data and real consequences if it breaks. If nobody can lose or gain anything real by using it, it's a prototype, not an MVP, and you should call it that.
It is also not a stripped-down version of your five-year vision. Founders frequently build an MVP that includes three user roles, an admin panel and a billing system because 'we'll need all of it eventually'. Eventually is not now. The test for whether something belongs in the MVP is not 'will we need this' — you will need almost everything eventually — it's 'does the single journey break without it'. If the journey works without it, cut it.
Choosing the one journey: the exercise that saves six weeks
Write down every journey your product could plausibly support. For a carpooling app that's: post a trip, book a seat, message a co-rider, rate a trip, manage recurring trips, handle payments, report an issue, and more. Most founders try to build six of these in the first release. Pick one.
The right one is usually the journey that, if it worked perfectly and nothing else existed, would still make someone use the product again next week. For a marketplace it's almost always the core match — one side posting, the other side booking or buying — not the surrounding features like ratings or messaging, which matter enormously later but don't test whether the core exchange has value.
A useful forcing test: if you had to launch tomorrow with only one screen, which one would it be? Whatever you answer is the journey. Everything else — settings, profile editing, notifications preferences, admin tooling — gets a line in the cut list, dated, so it can be revisited once the core journey is proven, not before.
No-code vs AI-assisted code vs agency: choosing the right build route
- —No-code (Bubble, Softr, Glide) suits founders testing a concept for near-zero cost with simple, mostly-standard logic — a form, a list, basic workflows. It breaks down fast once you need custom business logic, non-trivial permissions, or anything that needs to scale past a few hundred users; migrating off it later is expensive and often means a full rebuild.
- —AI-assisted code (a fractional product lead or senior engineer working with tools like Cursor, Claude Code or similar) is the middle path: production-grade code from day one, at a fraction of traditional build time because the tooling handles boilerplate, and someone experienced is still making the judgement calls about what to build. This is the route that produces a real six-week, £12k-£25k MVP.
- —Agencies suit a well-funded team with a broad, already-validated spec and no internal technical leadership — you're buying a team and a process, not just code. The cost is usually 3-5x an AI-assisted fractional build and the timeline is 3-6 months including the sign-off cycles agencies build in to manage their own risk, which is often more process than a single-journey MVP actually needs.
- —The wrong choice in either direction is common: using no-code for something with genuinely custom logic wastes months hitting walls, and hiring a full agency for a single-journey MVP means paying for a process built for much bigger scope than you have.
What to instrument before you launch, not after
- —The one activation event that proves the core journey completed — a trip posted, a booking confirmed, an account upgraded. Wire this up in week 4, not as a retrofit once you notice you have no data.
- —A 7-day and 30-day return metric tied to a specific user, not just aggregate traffic. Vanity metrics like page views tell you nothing about whether the MVP is working; return usage tells you almost everything.
- —Drop-off points inside the core journey itself — where in the flow people abandon, not just whether they finished. A funnel with three steps and the drop-off at each is worth more than a single conversion percentage.
- —Session recordings (Hotjar, PostHog or similar) for at least the first 50 users, watched personally, not just reviewed as aggregate heatmaps. The specific moment someone hesitates tells you more than any dashboard number.
- —A simple feedback capture inside the product itself — one question, asked at the moment of use, not a separate survey emailed a week later when the context is gone.
The launch checklist
- —The core journey has been completed successfully by at least ten people outside your immediate team and network.
- —Every error state in the core journey shows a message a real user could understand and act on — no raw error codes or blank screens.
- —Payments, if applicable, have been tested with a real card in production, not just Stripe test mode.
- —The activation event and at least one retention event are firing correctly and visible on a dashboard you can check without asking an engineer.
- —There's a way for a user to report a problem that reaches a human within a working day.
- —You have a rollback plan — you can disable or fix a broken core journey within the hour, not the week.
- —You know the one number you're checking every day for the first two weeks post-launch, and who's checking it.
What an MVP costs, and why the range is so wide
A focused, single-journey MVP built with a fractional product lead working alongside AI-assisted engineering typically costs £12,000 to £25,000, delivered inside six to eight weeks. The variation within that range is driven almost entirely by scope discipline, not by team seniority or tooling — a tightly scoped single journey with off-the-shelf auth and payments sits at the bottom; a second necessary surface (an admin view, a second role, a native wrapper) moves it toward the top.
The comparison points matter. DIY with no-code tools costs close to nothing in cash but consumes 100+ hours of founder time and frequently stalls on custom logic, meaning the true cost is time and a real risk of never shipping. An agency quote for the equivalent single-journey MVP typically runs £40,000-£100,000+, because agency pricing is built around a team structure and process designed for much larger scopes, and includes sign-off cycles that a six-week build doesn't need.
The number that should worry you more than the build cost is the cost of building the wrong thing. A £15,000 MVP that tests the right journey and tells you clearly it doesn't work has done its job. A £15,000 MVP built without a discovery step first, that turns out to solve a problem nobody has, has cost you the build fee plus the six weeks you can't get back — which is why a discovery sprint before the build, at £4,000, is often the cheapest insurance available.