Research10 min read
MVP Features: What Went Into 43 First Versions We Built
Abstract
We studied the first versions of 43 apps we built and coded their headline MVP features into 19 building blocks. A typical MVP had 5 of them. The most common were an operator dashboard (53%), payments (49%), notifications (42%), booking (40%) and search (37%). Connecting to an outside system was the block most consistently linked to longer builds, at about one extra week. Payments added no time.
Reviewed by Alpit Patel, Founder and CEO · Updated November 14, 2026
Part of the guide: Software development
On this page 12 sections
- What features does an MVP actually need?
- Why we did this
- How we did it
- Which MVP features show up most often?
- Do marketplace MVPs need different features?
- Which features add build time?
- Does the number of features decide the timeline?
- What we left out, and what we added late
- What this means if you are planning an MVP
- Limits of this study
- Data and downloads
- References
Key takeaways
- A typical first version in our data had 5 building blocks (range 1 to 10) across 43 builds.
- The blocks we built most often: an operator dashboard (23 of 43), payments (21), notifications (18), booking or scheduling (17) and search (16).
- Marketplaces are their own shape. 12 of our 16 marketplace MVPs took payments, 8 verified their providers and 5 held money in escrow. Outside marketplaces, verification showed up in 2 of 27 builds.
- Apps that connected to an outside system (an airline fare system, email sync, a supplier network) took 7.7 weeks on average against 6.8 without. Of the blocks that added time, it was the clearest.
- Taking payments added no time (7.1 weeks with, 7.2 without). Payment platforms are a solved problem; other people's systems are not.
- Only 4 of 43 named an AI feature, and only 3 named direct messaging between users.
Research paper
Read it here, or download the paper and the full dataset. Free to use with a link back to this page.
What features does an MVP actually need?
Across the 43 first versions we studied, a typical MVP had 5 building blocks. The same few came up again and again: a dashboard for whoever runs the product, payments, notifications, booking and search. The choice most consistently linked to a longer build was connecting to a system someone else controls.
Source: our coding of the published feature lists of 43 builds, November 14, 2026. Time figures are for the 41 apps. Data file below.
Why we did this
The question we hear most from founders is what has to be in version one. Most answers online are opinion. We had something better sitting in our own case studies: 44 products we built, each with its feature list, its build time and its results written down.
So we went back through them. We wanted to know three things. What do first versions really contain? Do different kinds of products need different things? And which features make a build take longer?
How we did it
The data. Our 44 published builds, all done by our in-house team. One of them (DAIN) has no feature list on its case study, so we left it out. That leaves 43 builds and 247 headline features. Each case study lists 5 or 6 of them; that cap is our own writing format, not a property of the products.
The coding. We sorted every feature into building blocks: the standard parts a software product is made from. We ended up with 19. One feature can belong to several. GoHangars lists one feature as “Payments run through Stripe Connect with automatic operator payouts and platform fees. Insurance certificates are uploaded and verified before each booking is confirmed, operators get an occupancy dashboard, and in-platform messaging keeps enquiries attached to the listing.” That counts as payments, verification, an outside connection, an operator dashboard and messaging. We coded only what the feature list says, not what we remember about the build.
| Building block | What counts |
|---|---|
| Operator dashboard | An admin, operator or owner view with numbers or controls |
| Payments | The product takes money: checkout, deposits, donations, payouts |
| Held payment | Money held until a job or event is done (escrow) |
| Verification | Checking identity, licences, insurance or student status |
| Outside connection | A live link to a system someone else runs |
| Documents | Contracts, e-signature, consent forms, document requests |
| Direct messaging | Users message each other inside the product (comments and group threads count as social) |
| The rest | Search, matching, booking, reviews, notifications, maps, AI, social, approvals, export, offline, self-managed content |
The analysis. We counted how often each block appears and split marketplaces (16) from everything else (27). For build time we used the 41 apps only, because the 2 websites in the set are both short builds and would skew the results. For each of the 8 most common blocks we compared average build time with and without it, and ran a permutation test: shuffle the labels 20,000 times and count how often chance alone produces a gap that big.
Which MVP features show up most often?
The surprise for us was the dashboard. Founders picture the customer screens, but more than half of our first versions named a view for the person running the business as a headline feature: a market operator seeing which stalls are posting, an EMS manager seeing staffing cost against contract value before the event. A product nobody can operate rarely survives its first month, so this belongs in v1.
The bottom of the chart matters just as much. Direct messaging, offline mode and AI are each named in 4 builds or fewer. When a founder tells us chat is essential, we ask what job it does that an email notification does not. Often the honest answer is “none yet”.
Do marketplace MVPs need different features?
Yes. The 16 marketplaces in our data look very different from the other 27 builds.
| Building block | Marketplaces (16) | Everything else (27) |
|---|---|---|
| Payments | 12 (75%) | 9 (33%) |
| Search and filters | 10 (63%) | 6 (22%) |
| Booking or calendar | 9 (56%) | 8 (30%) |
| Verification | 8 (50%) | 2 (7%) |
| Held payment | 5 (31%) | 0 |
| Reviews | 5 (31%) | 1 (4%) |
| Approvals or roles | 1 (6%) | 10 (37%) |
| Outside connection | 4 (25%) | 11 (41%) |
A marketplace has to make strangers trust each other on day one, so it needs payments, checks on the supply side and often money held until the job is done. Hago, a home services marketplace we built, holds payment until the customer confirms the work is complete. Wingman blocks any pilot whose medical or currency has expired from accepting a flight.
Internal tools and B2B platforms need something else: approval steps, roles and links into the systems the team already uses. Spirit Airlines’ group desk needed an approval chain from agent to supervisor to revenue management, tied to the airline’s fare system.
Which features add build time?
For each of the 8 most common blocks, we compared the average build time of the apps that had it with the ones that did not.
| Building block | Apps with it | Weeks with | Weeks without | Difference | How often chance gives a gap this big |
|---|---|---|---|---|---|
| Outside connection | 15 | 7.7 | 6.8 | +0.9 | 0.8% |
| Documents | 12 | 7.7 | 6.9 | +0.7 | 5% |
| Matching | 10 | 7.5 | 7.0 | +0.5 | 24% |
| Operator dashboard | 23 | 7.3 | 7.0 | +0.3 | 47% |
| Map or location | 12 | 7.3 | 7.1 | +0.3 | 52% |
| Payments | 21 | 7.1 | 7.2 | -0.1 | 77% |
| Search and filters | 16 | 6.6 | 7.5 | -1.0 | 0.5% |
| Self-managed content | 8 | 6.4 | 7.3 | -1.0 | 2% |
41 apps, websites excluded. Last column: permutation test, 20,000 shuffles.
Outside connections are the slow part. Spirit Airlines (connected to Sabre) took 10 weeks, our longest build. Homer and the Miami Dolphins suite platform took 9; both connected to outside parties and both handled documents (Homer’s milestone document requests, the Dolphins’ e-signed contracts). When a build depends on someone else’s system, we cannot control its documentation, its test accounts or how fast its team answers. The 7 apps with both payments and an outside connection averaged 8.1 weeks.
Payments are not. Taking money sounds hard, but payment platforms are mature, and 21 of these 43 builds used one. Payments added nothing to build time.
Search builds were faster. We think this is because search usually means discovery apps over the product’s own data (venues, markets, stylists), which are well-understood builds. It is not that adding search saves time.
The 17 apps that had neither an outside connection nor documents had a median of 6 weeks.
Does the number of features decide the timeline?
No. Across the 41 apps, the number of building blocks and build time had a correlation of about zero (-0.02; 0.17 if you add the two websites). HirePrep, an interview training platform, maps to just one standard block and still took 8 weeks. Its hard part was its own logic: adaptive learning paths, mock interviews that branch on the answer given, a modelling test that runs in the browser. No category list captures that.
So feature count is not the main thing we estimate from. We start by asking which parts of the build depend on someone else, and which parts are logic nobody has written before.
What we left out, and what we added late
A few decisions from these builds shaped how we scope every v1 now.
- Left out on purpose. Alendil, a rental marketplace, has no fee line items at checkout, and the system has no way to add them. That was a product rule, not a missing feature.
- Kept to one screen. StackStack’s daily logging is one screen of taps, under 15 seconds. The client had already learned that any app needing more than a few seconds a day gets dropped in week two.
- Added because users needed it. Offline cash logging was not in the original scope for the Coo fundraising platform. We added it because it let organisations keep their existing habits while moving online.
- Added late, used most. Audio recitation came late in Papri’s scope and is now one of the app’s most used features.
What we take from these: write the scope down, then check it against real users as early as you can, and be ready to change it.
What this means if you are planning an MVP
Our view, after building these:
- Start from the job, then pick the blocks. Write down the one thing a user must be able to finish. Then pick the blocks that job needs. Five is normal.
- Put the operator view in v1. Somebody has to run the product on Monday morning.
- Charge from day one if you plan to charge. It costs almost no build time and it is the clearest signal that the product works.
- Treat every outside connection as a risk. Budget about an extra week when the build depends on one, and ask for test access before the build starts.
- Marketplaces: build trust first. Payments, supply-side checks and, for big jobs, held payment. Reviews can be simple.
- Leave chat, AI extras and export for later unless the product is about them.
This fits wider evidence for keeping version one small. Pendo’s 2019 feature adoption study of 615 software subscriptions found that 80% of features in the average product are rarely or never used [2]. CB Insights’ 2026 study of 431 VC-backed startup shutdowns since 2023 found poor product-market fit behind 43% of the 385 with a known cause [3]. A short v1 that reaches real users fast is one of the cheapest ways to avoid both. That is also what Eric Ries meant by an MVP: the version that teaches you the most about customers for the least effort [1].
We go through this process in more detail in our guide on how to scope an MVP. If you want to see how long builds like these take, our timeline study of 44 builds has the numbers.
Limits of this study
- Our builds only. This is one team’s work, not a market survey. Every build in it launched; projects that paused or are under NDA are not here, which makes the set look tidier than real life.
- Headline features, not full specs. We coded what each case study lists. A build can have a feature that its write-up does not name, so a low count means “rarely a headline feature”, not “never built”.
- One coder. One person on our team did the coding. The data file below has every decision, so you can check or recode it.
- Small numbers. 43 builds is enough to see big patterns, not small ones. We ran 8 time comparisons, so treat anything near the 5% line (documents) as a hint, not a finding.
- Build time as reported. Times are whole weeks, so 7 weeks could be 6.5 or 7.4. Most case studies count from kickoff; a few count from the first call or after a separate scoping week.
Data and downloads
The full coded dataset is free to use: one row per build, one column per building block, with a link to each case study.
Every build in the study has its own case study on our work page.
References
- Ries, E. (2009). Minimum Viable Product: a guide. Startup Lessons Learned.
- Pendo (2019). The 2019 Feature Adoption Report. Pendo.
- CB Insights (March 5, 2026). The top 9 reasons startups fail. CB Insights Research.