Twelve weeks is a number clients ask for and agencies promise in marketplace app development. It is achievable for a two-sided marketplace, but only when the scope is cut to one path per feature and the finance work is bought rather than built. This is how one such build went for a home-services company: a marketplace where homeowners book vetted providers for cleaning, repairs and yard work.
Situation, goal and constraint
The company had been matching customers to providers by phone and spreadsheet for three years. Volume had reached the point where two coordinators spent their day on scheduling and payouts. The goal was a web product where customers could book and pay, providers could accept jobs and get paid, and staff could see everything in one place.
The manual process was the real spec. A booking meant a call or form from the customer, a round of texts to find a willing provider, a confirmation back to the customer, then a payout calculated by hand at the end of the week. Every step had a person in it. Our job was to remove the people from the routine path and keep them for the exceptions.
The constraint was a trade-show date twelve weeks out, where the company wanted to sign providers on the spot. Missing it meant waiting a year.
A fixed date changes how you scope. Every feature request gets the same question: does the marketplace fail without it on launch day? If the answer is no, it goes on the post-launch list, in writing, with the reason.
What was built
Architecture. One Next.js application with three role-based areas: customer, provider and admin. One PostgreSQL database with row-level security so a provider can only read their own jobs. Background jobs (reminders, payout batches) ran on a small worker, not inside the web process.
The reasoning was simple. One app means one deploy, one set of dependencies and one place to look when something breaks. Row-level security puts the access rule in the database, so a missed check in application code cannot leak another provider's jobs. The separate worker means a slow SMS provider or a payout batch never blocks a customer trying to book.
Stack. Next.js with TypeScript, Postgres on a managed host, Stripe Connect for payments and payouts, Twilio for SMS, and Resend for email. Hosting on a single mid-sized cloud instance behind a CDN. Nothing exotic, which mattered for the handover.
"Nothing exotic" is a hiring decision as much as a technical one. Whoever maintains this after us needs to find documentation, answers and developers without effort. Every tool on that list passes that test.
Team and timeline. Four people: a lead engineer, two full-stack engineers and a designer who also ran QA in the last three weeks. Two-week sprints, six of them, with a working staging link from sprint one.
At the end of each sprint the client got a demo on staging and a short written note: what shipped, what moved, what was decided. Their time cost was about an hour per demo, plus a weekly check-in with the lead engineer for open questions.
- Sprint 1: data model, auth, provider onboarding, the admin list of jobs.
- Sprint 2: customer booking flow with address, time window and price estimate.
- Sprint 3: Stripe Connect onboarding for providers, card payment on booking, authorization held until the job is marked done.
- Sprint 4: job lifecycle (accepted, en route, done, disputed), SMS reminders, reviews.
- Sprint 5: admin tools for disputes and refunds, reporting, the provider mobile view.
- Sprint 6: load testing, accessibility pass, data import from the spreadsheet, launch.
The hard decision behind 12-week marketplace app development
At the end of sprint two the client asked for a native provider app. Providers are on their phones all day, and the argument was reasonable. We said no, for version one, and here is why.
A native app would have added app-store review cycles, a second codebase and roughly five weeks of work. The provider screens had six actions: see the job, accept it, navigate, mark en route, mark done, see earnings. A responsive web view did all six, and a home-screen shortcut made it feel installed. We agreed in writing that a native app would be scoped after launch if provider feedback demanded it. Eight months on, it has not.
The test we apply is whether the app needs something only native code does well: heavy offline use, background location, deep hardware access. A provider tapping six buttons between jobs needs none of those. "Navigate" simply opens the phone's own maps app with the address filled in.
The second decision was Stripe Connect over a custom ledger. Connect handles identity verification for providers, the split of each payment, scheduled payouts and the year-end tax forms. Building the equivalent is at least six weeks of careful finance work plus an ongoing compliance burden. Connect's fees were the price of shipping on time.
In practice, a customer's card is authorized at booking. When the provider marks the job done, the charge is captured, the platform's commission stays with the company, and the rest goes to the provider's connected account for the next scheduled payout. Nobody calculates anything by hand. Check Stripe's current pricing and authorization-hold limits for your own model rather than relying on our numbers.
What we cut, and what it cost
Three features were deliberately left out of version one: provider availability calendars (jobs were offered first-come), in-app chat (SMS covered it) and multi-provider jobs. Each would have been a sprint on its own. The calendar was added in month four, after the company could see which providers actually wanted it.
The rule for cutting was whether a manual workaround existed. If an admin could cover the gap in a few minutes a day, the feature waited. Payments had no safe manual workaround, so they stayed in.
What the cuts cost was a few awkward moments at launch. Customers occasionally booked a time no provider wanted. The admin dispute tool handled those cases, and the data from them shaped the calendar feature.
Results
The marketplace launched the week of the trade show. Within the first quarter the two coordinators were spending their time on provider quality rather than scheduling. We do not publish the company's revenue figures, but the booking volume that once needed two people is now handled by the software and one part-time reviewer.
What we would do differently
- Start the data import in sprint one. Three years of spreadsheet data had inconsistent addresses and duplicate providers. Cleaning it in the final sprint was the most stressful week of the project. Next time, a throwaway import script runs against staging from week two, so the mess surfaces early.
- Put the dispute flow in sprint three, not five. Disputes shape the job lifecycle, and designing them late forced two small schema changes.
- Load test earlier. The first load test found an unindexed query on the job list. Trivial to fix, but it would have been nicer to find in week six than week eleven.
How a project like this starts
Every engagement begins with a written scope: the features in version one, the ones deliberately out, the stack, the team and a fixed price. The client owns the repository, the cloud account and the payment account from the first commit. If you have a marketplace idea and a date, describe it in the project form and an engineer will come back with a scope and a number within a week.
Frequently asked questions
Can a marketplace really be built in twelve weeks?
A focused one, yes: one booking path, one payment path, one dispute path. Add provider calendars, chat and multi-party jobs and you are at five to six months.
Why not build the ledger ourselves and save the Stripe fees?
Because the ledger is the easy part. Identity verification, payout scheduling, chargebacks and tax reporting are where the time and the risk go. Revisit it when volume makes the fees a line item worth engineering around.
Should marketplace app development start with a web app or a native app?
Usually web. It ships faster, updates instantly and lets you learn what users actually do. Go native first only if the core job depends on offline use, background location or device hardware.
Did the client own the code?
From the first commit. The repository, cloud account and Stripe account were created in the client's name, and we were collaborators on them.



