
How to build an MVP, and what to change after launch
How to build an MVP: pick the one thing to test, cut the features, choose how to build it, then decide which feedback to act on. With a made-up example.
An MVP (minimum viable product) is the smallest thing you can put in front of real customers to find out whether your fix for their problem works. Many founders treat it as version one of their big plan with the hard parts left off. It works better as a test, built for the people who have already committed to you, with a pass mark you set before anyone uses it.
Here is how to build an MVP in five steps, what to do with the feedback once people use it, and when to change course. A made-up example closes the page: the physio waitlist startup used across this library, from its first rough build to its first changes after launch.
Before you build: what an MVP is for
The idea comes from Eric Ries and his book The Lean Startup. At its center is his build-measure-learn loop: build something small, measure how customers respond, then decide whether to keep going or change course. The MVP is the first turn of that loop. The Lean Startup is on the list of the best books for startup founders, with a note on when to read it.
An MVP comes after validation. If nobody has paid, signed or given you real time yet, start with how to validate a startup idea. Building is the most expensive way to find out that nobody wants something. A rough prototype you show people to win that commitment is part of validating. The MVP is what the people who committed then use.
If you've done that work, you already know the problem and that it hurts. So the MVP only has to be the simplest thing that fixes it. I always use the same picture: your hair's on fire, and someone offers you a brick that will put it out. You grab the brick, and you don't ask for it to be made softer first. People with a painful problem put up with rough edges as long as the fix works. If your first users keep asking for more features before they'll use it, the core probably isn't fixing the problem yet.
How to build an MVP in five steps
- Write down the one thing you're testing, with a pass mark.
- Pick the smallest kind of MVP that can test it.
- Cut the feature list to what the result needs.
- Choose how to build it.
- Decide what you'll measure before anyone uses it.
Step 1: write down the one thing you're testing
Every startup rests on a pile of guesses. Pick the riskiest one: the guess that sinks the rest if it's wrong. Your MVP tests that one and nothing else.
Write it down in three lines. Here they are from the made-up example further down:
- The guess: people on a clinic's waitlist will reply to a text fast enough to take a slot a few hours away.
- The pass mark: at least 4 in 10 late-canceled slots get refilled during the pilot.
- If we miss it: ask the clinics and the patients why before building anything else.
Set the pass mark before launch. Afterwards it's easy to call any result a success.
Step 2: pick the smallest kind of MVP
Not every MVP needs code. Pick the cheapest kind that can test your guess.
| Kind of MVP | What it is | What it can test |
|---|---|---|
| A landing page | A page about the problem and the result you promise, asking for an email or a pre-order | Whether the problem catches attention, and which words work |
| A clickable mock-up | Screens linked together, with nothing working behind them | Whether people understand it and know what to do next |
| Done by hand | You deliver the result yourself, behind a simple form, an email or a spreadsheet | Whether the result is worth paying for, before you automate it |
| A rough version | Software that does the one core job, with manual steps around it | Whether the core job works in daily use |
| A full first version | The core job, automated from end to end | Whether it works without you in the middle |
The first two rows usually come earlier, while you're still validating. They test interest and understanding, and sign-ups are not validation. From "done by hand" down, it's something the people who committed use. Go as far down the table as your guess needs, and no further. The startup landing page covers how to build one as a demand test.
Doing it by hand gets skipped too often. It shows you whether people value the result, and which steps of the job are worth automating first.
Step 3: cut the feature list
Write down every feature you've thought of. Then sort each one into three groups:
- Must have. Without it, the customer doesn't get the result you promised. This group is your MVP.
- Later. It makes the product nicer, faster or easier, but the result still happens without it.
- Maybe never. Ideas you like. Park them in a list and leave them there.
The test for "must have" is whether the customer gets the result. Log-in screens, settings, dashboards, admin tools and a polished design almost always go in "later" for a first version. If you can do a step by hand for now (send the invoice, update the record), it stays manual.
It's easy to put too much in the first group. Sort it again and ask of each item: would the customers who committed still pay without this? If the answer is yes, it moves down.
Step 4: choose how to build it
No route is right for every founder. Pick the one that gets the must-haves in front of your committed customers soonest, and that lets you change it every week. Expect to throw the first version away.
- No-code tools: form builders, spreadsheets, website builders, automation tools. Often the fastest start, until you need deep links into other systems or unusual logic.
- AI coding tools: a founder who can't code can get a working prototype up quickly, and a developer goes faster. Before real customer data goes in, have someone who reads code check how it stores that data and who can get at it. For the rest of your work, what an AI co-founder can and can't do sorts what to hand to AI from what to keep.
- A freelancer or an agency: costs money and puts a step between you and every change. Keep the brief to the must-haves, pay in short stages, and make sure the contract gives your company the code. Intellectual property for startups explains why that isn't automatic.
- A technical co-founder: the best fit when the product itself is the hard part and will keep changing for a long time. How to find a technical co-founder covers where to look and how to check their work.
Set a date, and work to weekly goals rather than a single launch day. If the must-haves don't fit before the date, cut more. Something rough in front of the people who committed beats something polished that nobody has used. Startup mistakes to avoid covers what waiting until it feels ready costs.
Step 5: decide what you'll measure
Before launch, pick two or three numbers that show whether customers get the result and keep coming back. What people do tells you more than what they say.
- Count: how often they use the core feature, whether they return unprompted, whether they keep paying.
- Don't count: downloads, sign-ups, page views, likes. They feel good and decide nothing.
Then talk to the people using it. Ask about the last time they used it and what happened; whether they like it tells you little. Customer interview questions for startups has questions that work after launch as well as before.
After launch: which feedback to act on
Once people use your MVP, requests arrive: new features, changes, complaints. You can't build them all, and you shouldn't try. Say yes to everyone and the product turns into a pile of half-used features.
Sort what comes in this way:
- Repeated beats single. One person asking is a preference. The same problem from several customers is worth a look.
- Find the problem under the request. A feature request is the customer's own design for fixing something that bothers them. Ask what they were in the middle of when they wanted it, and what they did instead. One change often answers three different requests.
- Don't let one customer set the plan. The one who writes most often, or pays most, isn't always typical. Check what the quieter ones do.
- Weigh it against the one job. Does it help customers get the result you promised? If not, it waits, however good it sounds.
- Check behavior against words. If people ask for a feature but keep using the product happily without it, it can wait.
Keep a simple feedback log, one row per request. The row below comes from the made-up example at the end of the page:
| Date | Who | What they asked for | What they needed it for | How many have said this | Decision, and why |
|---|---|---|---|---|---|
| During the pilot | Receptionists at the pilot clinics | Not having to type each slot into the booking system again | Retyping takes as long as phoning round | All three pilot clinics | Build the booking-system link next. Every clinic raised it, and both letters of intent depend on it |
Write down why you said no, or you'll argue about the same request again next month. Tell customers what you decided, the no's included: people who hear back keep telling you things. Try bigger changes on a few customers before everyone gets them.
When to change course
Ries calls a change of course a pivot. Some signs it's time, and some that it isn't (for the full list, with three companies that started as something else, see when should a startup pivot):
- Several sources point the same way. The numbers, the conversations and who keeps paying all tell the same story. One bad review or one lost customer is not a pattern.
- You gave it a fair run. Changing too early is as costly as holding on too long. Use the pass mark from Step 1: if you missed it by a long way and fixes haven't moved the number, change something.
- You change one thing at a time. The solution, the customer, the problem, or the price and how you charge. Then run the test again.
Each time something changes, update your canvas and keep the old version with its date. This Lean Canvas example shows what that looks like.
A made-up example: the physio waitlist startup, from first build to first changes
This example is invented, the company and every number included. It's the startup from the pitch deck outline for a 10-minute pitch, months before that pitch: two founders filling the appointments independent physio clinics lose to late cancellations. Three clinics have paid for a two-month pilot at €100 a month. Two more have signed letters of intent that depend on a link to their booking system.
What they tested. The clinics had already shown the problem was real by paying. The open question was the patients: would people on a waitlist reply to a text fast enough to take a slot a few hours away? Pass mark: at least 4 in 10 late-canceled slots refilled during the pilot. That's a different test from the one on the validation page (at least three clinics pay for the pilot): that one asked whether clinics would pay, this one asks whether the product does the job. They also tracked how long the first yes took, and whether receptionists used it for every late cancellation or only some.
What they built. To make the pilot offer, the founder who used to build booking-system integrations had put together a rough prototype in two weeks. Once three clinics paid, that prototype became the MVP: the thing the clinics used every day during the pilot. The receptionist types the free slot into a simple form. The tool texts that physio's waitlist, and the first person to reply yes gets the slot. Then the receptionist updates the booking system by hand. A ready-made service sends the texts.
What they left out. The link to the booking system, an app for patients, a dashboard, settings and automatic billing. The founders sent the invoices themselves.
What came back during the pilot:
| What they heard or saw | Where | What was behind it | Decision |
|---|---|---|---|
| "Typing the slot into the booking system again takes as long as phoning round." | All three clinics | Entering everything twice | Build the booking-system link next |
| A clinic owner asked for a monthly report with charts | One clinic | "Is this paying for itself?" | No dashboard. A short monthly message: slots refilled, and what they were worth |
| A receptionist asked for an app where patients pick the times they can make | One clinic | Texts going to people who couldn't come at that hour | Not now. Logged, to look at again if more clinics raise it |
| Patients replied yes after the slot had gone, and some were annoyed | Replies at all three clinics | Nobody told them it was taken | Add an automatic "sorry, this slot has gone" reply |
The result. About half of the late-canceled slots were refilled, above the pass mark. All three clinics kept paying after the pilot and became the first customers at launch. How to get your first customers covers how to find the ones after them. The booking-system link was the next thing built, with the two letter-of-intent clinics next in line for it. The price and the customer stayed the same.
Months later, with the link built and the app listed in booking systems' app stores, the same company pitched with 24 paying clinics and 7 in 10 canceled slots refilled.
Ready to Move Faster?
Get actionable feedback and insights on your challenges and goals. 30 minutes focused on helping you move forward.
No sales pitch. No obligations. Just helpful advice.