What a six-week AI MVP actually contains
What does a six-week AI MVP actually contain?
Six weeks is enough to launch a real AI product if scope is locked on day five. Here is the week-by-week plan, exactly what ships, what waits for v1.1, and what it costs in INR and USD.
Our Launch 6 programme promises a production MVP with AI in the core in six weeks at a fixed price. People reasonably ask what that includes, how six weeks is possible, and what is left out. The short version: everything a first cohort of real users needs, and nothing they do not. The longer version follows, with the week-by-week plan, the scope-lock template we use, the cost, and a case study pointer.
In scope, every time
- Authentication, roles and an admin view
- One complete user workflow from start to finish
- One production AI capability: a copilot, an extraction pipeline, retrieval over documents, an agent step or a recommendation
- Analytics instrumentation so week-seven decisions are based on usage
- Cloud deployment with CI/CD, monitoring, evals and rollback
- Handover documentation and thirty days of post-launch support
Week by week
| Week | What happens | What you sign off |
|---|---|---|
| 1 | Kickoff, access, data sample, user segment and workflow chosen, AI capability chosen, architecture sketched, design frames | Scope lock on day five |
| 2 | Foundations: auth, roles, data model, CI/CD, environments; AI capability baseline measured | Friday demo on staging |
| 3 | Core workflow end to end; AI capability iterated against the eval set | Friday demo |
| 4 | Admin, analytics, edge cases, integrations on the critical path | Friday demo; feature freeze |
| 5 | Hardening: load test, security checks, eval regression, rollback rehearsal, accessibility | Go/no-go for launch |
| 6 | Production launch to the first cohort, monitoring live, handover pack, v1.1 backlog priced | Launch on day 42 |
How six weeks is possible
Week one ends with a scope lock. AI handles specification drafts, scaffolding, test generation and boilerplate, which is most of the typing. Senior engineers spend their time on architecture, business logic, security and the parts of UX that decide whether people come back. Fridays are demos, and the demo on day 42 is on production. None of that is a shortcut; it is the removal of the two things that make MVPs take six months, undecided scope and repetitive coding. We describe our approach in How we use AI to build software.
The scope-lock template
Scope lock is a one-page document signed on day five. It contains:
- Segment: the one type of user the MVP serves, named
- Workflow: the one job they complete, from first screen to done, in numbered steps
- AI capability: what it does, what data it uses, and the eval threshold it must clear
- Integrations on the critical path: named systems and what is read or written
- Explicitly out: the list of things that go to v1.1
- Success metric: the number that tells us on day 60 whether the MVP worked
Anything not in the document after day five goes to a priced v1.1 backlog. That discipline is what makes the price and the date fixed.
What waits for v1.1
A second workflow, integrations that are not on the critical path, a native mobile app, and polish that does not change the core decision. All of it goes on a backlog with its own price, so the MVP invoice stays fixed and the roadmap after launch is already written.
What it costs
Launch 6 is fixed-price between $26,500 and $45,500 (₹17.6–30.4 lakh), the range depending on the AI capability and integrations locked in week one. A ten-day Sprint Zero at $3,250 (₹2 lakh) is credited in full if the MVP starts within 60 days. After launch, a Care Plan or a feature retainer continues the work. All prices are on the pricing page.
Who it is for, and who it is not
- For: startups with a validated idea and a launch date; enterprises spinning out a new digital product; teams that want a fixed price and a fixed date.
- Not for: products that need three or more workflows at launch; scope that cannot be locked in week one; teams without a decision-maker available on Fridays.
What good looks like on day 42
Real users on production. Authentication, roles and admin working. The core workflow completing end to end. The AI capability with an eval score you can quote. Analytics recording the events that matter. A codebase with CI/CD, infrastructure as code and documentation that your team or ours can extend. And a v1.1 backlog, priced. For a sense of what the AI capability can be, the in-app copilot case study shows a copilot that went from Sprint Zero to beta in nine weeks and to general availability at twelve.
Common questions we get in week one
"Can we add one more thing?" Yes, to v1.1. "Which stack?" React or Next.js, Node.js, MongoDB or Postgres, React Native for mobile, and whichever models win the benchmark for your task. "Do we own it?" Yes: code, infrastructure, prompts and documentation transfer on final payment. "What if the AI capability fails the eval?" We find out in week three, not week six, and the scope lock names the fallback. The programme page has the full detail: AI-Accelerated MVP.
A worked example: a compliance-document MVP
A compliance startup needed a product that let mid-size companies upload policies, ask questions and generate audit-ready summaries. Scope lock on day five: segment, compliance managers at companies with 200–2,000 staff; workflow, upload, ask, export a summary; AI capability, retrieval with citations over the uploaded documents, threshold set on a golden set of eighty questions. Weeks two to four built auth, roles, the upload and ask flow, retrieval with permissions and an export. Week five hardened: load test, security scan, eval regression, rollback rehearsal. Day 42, twelve pilot customers on production. The v1.1 backlog, priced, held integrations with document-management systems and a second workflow for policy drafting.
The AI capability options and what each needs
| Capability | Data you bring | What we evaluate |
|---|---|---|
| Copilot inside the product | API surface, sample accounts | Actions executed correctly; answer accuracy on real asks |
| Document extraction | Sample documents including the ugly ones | Field-level accuracy by type; exception rate |
| Retrieval over documents | Corpus and real questions | Recall, precision, groundedness |
| Agent step in a workflow | Historical cases and tool access | Completion rate; escalation quality |
| Recommendation | Event and catalogue data | Offline metrics, then an A/B design |
Why fixed price and fixed date are possible
Because scope is locked, hardening has its own week, and AI removes the repetitive work. The programme is not a discount; it is a discipline. The most useful external reference on the underlying idea is Eric Ries's minimum viable product, which Launch 6 applies to AI products with the addition of evaluation as the definition of "viable". Programme details and the credit from Sprint Zero are on the AI-Accelerated MVP page.
Team and timeline
Launch 6 is a squad of four to five: a product-minded principal, two engineers, an AI engineer and a part-time designer, with a delivery lead keeping Fridays honest. Your side provides a decision-maker every Friday, access to the systems on the critical path, accounts with your cloud and model providers, and real users ready to try the product in week six. The calendar holds when those four are in place and slips when the Friday decision-maker is not.
Before you start: a checklist
- Confirm the decision-maker for Friday demos
- Pick the one user segment and the one workflow
- Choose the AI capability and gather its data
- Grant access to the systems on the critical path
- Open cloud and model provider accounts
- Line up first users for day 42
- Prepare to send everything else to v1.1
What happens in week seven
Week seven is where MVPs are won or lost, and it is why analytics are in the fixed scope. With real usage data you can see which step of the workflow users complete, where they drop, whether the AI capability is used or bypassed, and which v1.1 items actually matter. We run a review at day 60 against the success metric named in the scope lock, and the roadmap that follows is based on that evidence rather than on the original assumptions. Clients who continue do so under a Care Plan or a monthly product squad, and the squad that built the MVP is the squad that continues, so nothing is lost in handover.
Related reading
Related reading: Multi-tenant SaaS architecture for the foundation every MVP includes, How much does AI development cost? for where the fixed price sits, and Why AI copilots inside SaaS beat chatbots for the most common AI capability chosen.
Six weeks is not a stunt; it is what happens when scope is decided, hardening is scheduled and the repetitive work is automated. If your idea can be locked to one segment, one workflow and one AI capability, day 42 is a realistic date, and the fixed price is the proof that we believe it.
Frequently asked questions
Can the MVP have more than one AI feature?
▾
One in the fixed scope; a second is usually the first v1.1 item and often reuses the same infrastructure.
What if our data is not ready?
▾
Sprint Zero finds that out in ten days; the MVP is scoped around what exists, and data work becomes a named backlog item.
Do you continue after launch?
▾
Yes, through a Care Plan or a monthly product squad; most Launch 6 clients stay for six to twelve months.