MVP development cost, counted in days

The pages that rank for MVP development cost are agencies' estimates: a table of tiers from a few thousand dollars to a few hundred thousand, and timelines from a few weeks to a year. None of them shows an MVP the agency shipped, with the day work began and the day it went live. We publish no price either, for a reason this note comes to at the end, but the other half of the question can be answered with evidence. Below are three builds Easinvy shipped in September and October 2026, timed from their git history and from GitHub's deployment records, with what each one left out.

Three builds, three clocks

Each row is read from the repository, not from memory. A commit shows when work was saved, not when the thinking began, and a deployment record shows when a build first ran in production. Times are India Standard Time.

We executed all three with AI agents, Claude Opus among them, to move fast, and how they fit comes further down. The figure that matters for a budget is not the speed, though. In every build, the days went where the decisions were.

A store live the next day, and the login it lost

The wellness brand's store began as one repository holding the website and its admin. By the evening of its first day it had a ten-section home page, a blog, plans, a checkout with guest checkout, customer login and an admin dashboard, and it was running in production at 18:58. The next day it was on the brand's own domain.

Customer login was built at 16:32 on that first day. Eight days later, on 16 October, it was taken out. Guest checkout already took every order, so the store went on without accounts. The same fortnight, analytics was "skipped for now", reviews were hidden behind a switch instead of being deleted, and the client's final release went out with payments switched off while the checkout still kept each buyer's email.

An ERP on its server in one night, reshaped after the floor

Chandas Perfumers, an incense maker in India, is the one client in this note we can name. The first commit of its ERP went in at 01:09 on 13 October 2026, with the server's deployment settings already in it, and the backend was running on Railway at 01:44. By 03:47 there were 13 commits across the backend and the screens: purchases, receiving, a workstation, production from a per-box recipe, and sales of cartons that mix fragrances.

The first model of finished stock lasted 81 minutes. Stock kept at three levels, loose boxes, loose dozens and cartons, with repacking between them, was replaced at 02:30 by the plant's real flow from purchase to carton. By 03:24 finished stock was kept in boxes, with cartons that mix fragrances packed at the time of sale. The first version took two weeks.

On 30 and 31 October, after we saw how work actually moved on Chandas's floor, 35 commits reshaped it. Every dispatch was tied to a sale order and packing was split from dispatch. The Stock and Catalog pages were retired, and a new Stock page with expected dates and late flags took their place the next day. A view-only role and a "Record made" tab were dropped, and the separate store step was folded into the floor. What that system holds now is in Custom ERP vs off-the-shelf ERP, from one we run.

Four days on a backend we threw away

The auto-parts store holds the most expensive cut in this note, and the decision to make it was ours. Work began on 28 September 2026 with a storefront and a custom commerce backend on Medusa, with an admin of its own. By 1 October the backend had ten commits, a vehicle-fitment module and an admin built phase by phase. On 30 September the GST and invoice phase was parked, because the shop is not GST-registered and has no GST to charge.

At 14:08 on 5 October a new plan moved the catalogue and the checkout to Shopify, read through its Storefront API. Medusa was removed at 01:02 the next morning, and at 03:15 the store was running in production. From the plan to production took 13 hours. An Excel importer written the same night let the shop load its own products, and that afternoon the storefront took its own brand and the shop's real contact details.

Four of the eight days went into a backend that never served a customer. Most of what an MVP costs is that kind of time: not the hours spent building the right thing, but the hours spent on a thing the first version did not need. A hosted checkout, a catalogue the shop edits in Shopify and no tax code to keep up were worth more to this store than owning its backend.

After launch the store kept narrowing. Vehicle fitment went behind a switch on 18 October, a "Pay on WhatsApp" block joined the thank-you page the same day, and on 8 November the catalogue was cut to one category.

What waited, across the three

Put side by side, the cuts look alike. Each was something worth having, and none of them was needed for the first customer to buy or the first order to be packed:

• Customer accounts, built on day one and removed on day eight
• Payments at the store's final release, with the buyer's email kept
• Analytics, skipped for now
• Reviews, hidden behind a switch
• GST and invoices, parked for a shop that charges no GST
• Vehicle fitment, behind a switch
• A custom backend, replaced by Shopify
• In the ERP, a view-only role, a record screen and a separate store step

A switch is the cheapest of these. The feature stays built and tested, and turning it on later is a setting, not a project.

Where the AI agents fit

We executed these builds with AI agents, Claude Opus among them, to move fast. Claude Opus is a co-author on every commit of both stores, 92 of 92 for the wellness store and 75 of 75 across the parts store's two repositories, and on 21 of the ERP backend's 25. The parts store's first admin was built phase by phase by another agent and reviewed after each phase.

What the agents did not do is decide. Taking out customer login, parking GST, moving to Shopify and reshaping the ERP around the floor were judgements about a business, made by people, and the agents carried them out. The same line runs through AI automation for small business: what stays human. The timelines above are timelines with agents, and we make no claim about how long the same builds would take without them.

What decides the cost of an MVP

An MVP's cost follows its scope: how many user roles and core flows it has, whether it takes payments, its integrations, whether it is web, mobile or both, and any AI features. Those are the factors our MVP development page lists, and the three builds show which way each one moves the clock. Payments and accounts were the first things to wait. An integration chosen well, Shopify's checkout, saved more time than any feature cost. The one decision that cost days was building a backend before the first version needed it.

We publish no price, because two MVPs with the same feature list can differ by a whole backend, as the parts store shows. Easinvy sends the cost and the timeline in a written proposal after a free 30-minute call, with what the first version leaves out agreed in writing before building starts. Every time in this note was read from the repositories and GitHub's deployment records on 9 November 2026.