A marketplace is two products plus the operations between them. What belongs in the first version — one complete deal, trust between strangers, an admin console and a data model you won't have to rewrite — and what can safely wait.
- Published
- Reading
- 13 min
- Author
- Yaroslav Tsarenko

Most marketplace ideas sound simple: connect people who need something with people who offer it. So the MVP sounds simple too — a catalog, profiles, a "request" button. In practice, the first version of a marketplace is harder than the first version of an online store, because you're building two products at once, one for each side, plus a third one nobody sees: the operations between them.
This piece is for founders planning a marketplace or a B2B platform with several roles who want to know what belongs in version one and what would be a waste of the first few months. It's based on two marketplaces I built: AllShip AI, a US freight marketplace, and AddSale, a B2B marketplace from Latvia. If there's only one seller on the platform, you, and the buyers are your regular wholesale customers, that's not a marketplace but a B2B ordering portal, and its first version looks different.
How a marketplace MVP differs from a regular MVP
A regular product has one side: either the user gets value or they don't. A marketplace only creates value when two sides find each other and agree on something. You can have a thousand sign-ups and zero deals — a failure, even though every dashboard looks green.
So the metric is different too. In a16z's 13 metrics for marketplace companies, one of the key ones is match rate: how successfully the two sides find each other. The same piece suggests tracking "zeros" — searches and requests that led nowhere. For an MVP, this is the most honest number you have: how many requests turned into a completed deal.
That leads to three differences that shape the build:
- two funnels instead of one — sign-up, profile and first action for each side separately;
- trust as a feature — people are making a deal with a stranger, and the product has to explain why that's safe;
- operations — moderation, disputes, manual matching. Early on, your team does this work, and does it through the admin console.
The unit of an MVP is one complete deal
The most useful rule I know for marketplaces: the first version is one complete deal path, from the buyer's first step to completion, with no gaps. Not ten half-finished sections — one flow that real people actually walk through.
For a freight marketplace, that path looks like this: a shipper posts a load → carriers bid → the shipper picks one → a driver picks up → delivery is confirmed → both sides see the deal closed. For a B2B marketplace like AddSale: a business submits an application → providers see it with all the requirements → take it on → move it through the stages to integration.
Run your feature list through one question: "Will the deal fail without this?" If the answer is "no, it'll just be less convenient," the feature goes into the next version.
What to build in version one and what to postpone
Here's how I usually sort marketplace features at the start. It isn't dogma: in your niche, one of the "later" features may be critical. But the logic holds — the deal and trust come first.
| Feature | Version one | Later | Why |
|---|---|---|---|
| Sign-up for both roles | yes | — | no roles, no market |
| Listing or request | yes, structured | templates, imports | matching only works on comparable data |
| Search and filters | basic | smart search, recommendations | with few listings, fancy search finds nothing |
| Response from the other side | yes | automated matching | early on, matching can be manual |
| Deal statuses and notifications | yes | — | without them both sides email support |
| Participant verification | yes, basic | manual → automated checks | trust is the main reason deals don't happen |
| Admin console | yes | analytics, reports | moderation and disputes start on day one |
| Chat between the sides | often | files, templates | depends on the deal |
| Ratings and reviews | simple | weighted ratings, badges | with few deals, a rating says nothing |
| Native mobile app | no | after demand is proven | a responsive web app covers the first months |
| Loyalty, referrals | no | once there's something to retain | first make sure deals happen at all |
A word on chat. It looks like a small feature, but it pulls in moderation, notifications, file uploads and the question of what happens when the two sides take the deal off-platform. If the deal is standard and the terms fit into form fields, start without chat. If the terms differ every time, you need it from day one.
The data model: what you can't "add later"
Most of a marketplace's interface can be rebuilt in a week. The data model can't. It decides whether the product survives its first six months without a rewrite. Here's what I write down before the first line of code:
- Roles and accounts. Can one person be both a buyer and a seller? Are there companies with several team members? At AllShip, a carrier and a driver are separate roles with separate permissions, and one carrier company can have several drivers. Bake in "one user, one role" and three months later you're migrating the whole database.
- Deal statuses. A deal is a state machine: "created → has offers → selected → in progress → completed", plus "cancelled" and "disputed" branches. Who can move a deal to the next status, and when, must be checked on the server — not by which button the UI happens to show.
- Change history. Who changed what in a deal, and when. Without it, every dispute is one person's word against another's.
- Money as records, not a field. Even if payment happens off-platform for now, store amounts, fees and settlement statuses as separate records. A number in a
pricecolumn can't be reconciled with anything later.
The classic mistake in AI-generated prototypes is a data model that describes screens instead of a market. I wrote about how to spot that moment and what to do about it in vibe coding an MVP: 10 signs you need a developer.
Trust beats features: verification, honest matching and bots
A marketplace doesn't sell features — it sells confidence that the other side is a real person or company who will hold up their end. So trust isn't an "About us" page; it's concrete mechanisms in the product:
- participant verification — documents, licenses, public registries. At AllShip, carriers are verified against FMCSA data (the US freight carrier regulator) before they can bid;
- proof of delivery — photos at pickup, a signature on delivery, live tracking;
- honest matching — no generated bids, no "active" sellers who don't exist;
- bot protection — email verification, rate limits, an invisible CAPTCHA on forms.
Honest matching sounds obvious, but it's exactly where marketplaces fool themselves. When I started on AllShip, the database held 34,921 bot accounts and generated bids. From the outside, the marketplace looked busy. The real accounts I migrated into the new system came to 16, plus 5 carriers.
Fake activity hurts twice. A real user posts a request, gets "bids" from carriers that don't exist and never comes back. Meanwhile, the team stares at sign-up numbers and can't figure out why there are no deals. Sixteen honest accounts and a clear picture beat 35,000 and an illusion.
Chicken and egg: what code solves and what it doesn't
The classic marketplace problem: buyers won't come until there are sellers, and sellers won't come until there are buyers. NFX collected 19 tactics for the chicken-or-egg problem, and most of them aren't about code. Code helps in three ways.
1. A narrow niche. The smaller the market at launch, the faster deals appear in it. One city instead of a country, one category instead of ten, one customer type. Technically, that means the MVP doesn't need complex search, multiple currencies or ten languages.
2. Value for one side even without the other. If the product works as a tool for sellers — managing requests, a profile, statuses — they'll show up before the crowd of buyers does. At AddSale, providers don't just get contact details; they get structured applications with every requirement spelled out and a shared seven-stage pipeline. That's value their own CRMs didn't give them.
3. Manual work that the admin console supports. In Do Things That Don't Scale, Paul Graham describes Airbnb's founders going door to door in New York, helping hosts improve their listings. At launch, your team connects the two sides by hand, makes calls, sends reminders. The product should support that: manually assigning a provider, notes on a deal, manual status changes recorded in the history.
Demand can be built with code too. For AddSale we built 50 pages in English and Latvian aimed at the searches businesses make when they're looking for a provider. Pages like that keep working when there's no ad budget. How to get them cited in AI answers from Google and ChatGPT as well is covered in a website that gets leads when AI answers.
The admin console is half the product
Founders often push the admin console to "later" because users never see it. In a marketplace, that's a mistake: for the first months, the admin console is how your team does everything the product doesn't automate yet.
Users see
- Listings and profilescards, photos, prices, terms
- Search and responsesfilters, bids, offers
- Deal statuswhat's happening and whose move it is
Your team needs
- Moderation and verificationwho passed checks, who is blocked
- Disputesdeal history, evidence, decisions
- Manual matchingconnect the sides before there's an algorithm
A minimal admin console for a marketplace MVP can:
- list every participant with their verification status and block bad actors;
- open any deal with its full change history;
- change a status or assign a provider by hand, recording who did it;
- handle a dispute with messages, photos and documents in one place;
- show simple numbers: requests, deals, and "zeros" — requests that got no response at all.
Permissions deserve their own line. The admin console shows data from every side, so authorization has to be checked at every entry point, not just at the login page. At AllShip I audited authorization across 109 entry points — routine work once a product has several roles.
Regulations that shape the architecture
If your marketplace sells to consumers in the EU, a few requirements land not on your lawyer's desk but directly in your data model and admin console. This isn't legal advice — check the specific obligations for your business with a lawyer. But it pays to know about them before development starts.
- Digital Services Act, Article 30. An online marketplace must obtain from each trader their name, address, phone and email, an identification document, payment account details, a trade register number (where registered) and a self-certification that they'll only offer products and services that comply with EU law — before the trader starts selling. Micro and small platform businesses are exempt, but if you plan to grow, the fields and the review workflow are cheaper to build in from the start. The full text is on EUR-Lex.
- DAC7. Since January 1, 2023, platform operators report every year to tax authorities on their EU-resident sellers: identity details and the amounts paid to each seller. The report for a year is due by January 31 of the following year. For the product, that means per-seller totals have to be stored in a way you can export, not pieced together from bank statements.
The US, the UK and other markets have their own rules, but the logic is the same: model participants and money as if a tax authority or a bank will ask to see them tomorrow.
How long a marketplace MVP takes
The honest answer: it depends less on the number of pages than on three things — how many roles there are, how complex the deal is, and how many external systems you need to connect. A marketplace with one seller role and a standard purchase is weeks of work. A platform with verification, bidding, delivery tracking and disputes is months.
A reference point from a real project: AddSale — a two-sided B2B marketplace with two account types, nine-tab provider profiles, a seven-stage lead pipeline and 50 SEO pages in two languages — launched in four months with a team of two: me and a designer.
What that kind of scope costs in 2026, and how to compare quotes from different teams, is in a separate guide to MVP development cost.
- 1–2 weeksModel the marketsides, the deal, statuses, what can go wrong
- next 4–6 weeksOne complete dealrequest to completion, on a preview link
- in parallelTrust and adminverification, moderation, disputes, notifications
- launchFirst usersnarrow niche, manual matching, weekly fixes
- afterScaleautomate what keeps repeating
The order that saves the most time:
- Model the market on paper. The sides, the deal, statuses, roles, what can go wrong between the sides. A week or two of conversations is cheaper than a month of rework.
- One complete deal. From request to completion, on a preview link you can show your first users. Each week adds another piece of the flow.
- Trust and admin in parallel. Verification, moderation, notifications, manual matching. Without them, your launch ends up living in the founder's inbox.
- Launch in a narrow niche. Real deals, manual work, weekly fixes.
- Automate what repeats. Only after you've done it by hand several times and know how it should work.
Off-the-shelf platform or custom build
Not every marketplace needs custom development at the start. If the deal is standard — a product or service, payment, a review — marketplace builders let you test the idea in a few weeks. A custom build pays off when:
- the deal has several stages, documents or proof of completion;
- participants must be checked against external registries;
- matching is complex: bidding, criteria, manual qualification of requests;
- there are several roles with different permissions inside one company;
- the marketplace is part of a bigger product, with integrations and its own logic.
If you already have a prototype on a builder or one generated with AI, that time isn't wasted: it shows which flows your users actually need. Carry over the flows and the data, not the code.
That's how it went with AllShip: the product lived in three separate codebases, each solving part of the problem its own way. Instead of porting code, I pulled out the flows people actually used and built one Next.js app with a clean data model and dashboards for shippers, carriers, drivers and admins. Real data moved to the new system; the bots stayed behind in the old one.
A useful rule for the move: port one complete deal first and test it with real users, then everything around it. That way the old system keeps running until the new one proves it does the job better.
A checklist before you hire a developer
Answers to these questions shrink the estimate and remove half the surprises along the way.
| Question | Why it matters |
|---|---|
| Who are the sides, and which one is harder to attract? | decides where to start |
| What does one complete deal look like, step by step? | that's the MVP scope |
| Who moves a deal to its next status? | permissions and server-side logic |
| What can go wrong between the sides? | disputes, cancellations, evidence |
| How will you verify participants? | verification from day one |
| How does the marketplace make money? | commission, subscription and pay-per-lead are different data models |
| What's the launch niche? | one city, category or customer type |
| Who on your team does the manual work in the first months? | what tools the admin console needs |
If half of these don't have an answer yet, that's normal. That's where good development starts: the market model first, code second.
The bottom line
A marketplace MVP isn't the minimum set of pages. It's the minimum that lets strangers complete a deal safely: one complete deal path, trust, an admin console and a data model that won't fall apart in three months. Everything else — ratings, native apps, smart recommendations — gets added once deals are actually happening.
If you're planning a marketplace or a multi-role B2B platform, start with the checklist above. And if you'd like to walk through your market model with an engineer who builds these products himself, see how I approach marketplace development, or get in touch.
Was this useful?
– likes · – views
Questions people ask
What features does a marketplace MVP need?
Only what one complete deal requires: sign-up for both sides, a listing or request, search or matching, a response from the other side, agreeing on terms, deal statuses, notifications and a simple admin console for moderation and disputes. Ratings, loyalty programs and native apps can wait.
How long does it take to build a marketplace MVP?
It depends on the number of roles, integrations and trust requirements, not on the number of pages. As a reference point, AddSale, a two-sided B2B marketplace with provider profiles, a seven-stage lead pipeline and 50 SEO pages in two languages, was launched in four months by a team of two.
Which side of a marketplace should you build for first?
Usually the side that is harder to attract, in a very narrow niche: one city, one category, one type of customer. The product should be useful to those first users even before the other side shows up, for example as a tool for their own work.
Should I use a marketplace platform or build a custom one?
For a standard deal (a product or service, payment, review), off-the-shelf marketplace builders are a fast way to validate the idea. Custom development pays off when the deal is not standard: multiple stages, documents, participant verification, bidding, proof of delivery or complex matching.
Which EU rules affect how a marketplace is built?
For online marketplaces selling to consumers in the EU, the Digital Services Act requires collecting and checking trader details before they can sell, and DAC7 requires reporting seller income to tax authorities every year. Both become fields in the data model and workflows in the admin console, so plan for them from the first version and confirm the details with a lawyer.