Budgeting
How much does an AI MVP cost? A practical breakdown
What really drives the cost of an AI MVP — scope, the AI itself, integrations and running costs — and how to keep your first version affordable.
By the Jerrax team · · 7 min read
The honest answer is that an AI MVP costs whatever its scope costs. The AI part is rarely the expensive bit any more — most products call an existing model through an API. What moves the price is how many things the product has to do, for how many kinds of user, connected to how many other systems.
That’s good news. It means you control the budget by controlling the scope. This guide walks through what drives build cost, what it costs to keep the product running, and how to keep a first version affordable.
The five things that drive build cost
1. The number of workflows
A workflow is a complete journey through the product: a user uploads a contract and gets a summary back, or a client submits a request and tracks its status. Each workflow needs its own screens, data, logic and testing. One well-defined workflow is a small project; five loosely defined ones is a large one.
2. Screens and how polished they need to be
A focused MVP usually needs a handful of screens: sign-in, the main workspace, a result or detail view, a history or dashboard, and a basic admin view. Every extra screen adds design, build and QA time — and so does every extra state (empty, loading, error) you want designed rather than handled simply.
3. What the AI actually does
AI work ranges widely in effort. Roughly from simplest to most involved:
- One call to an existing model API — summarise, classify, extract, draft. This is where most MVPs should start.
- Answers grounded in your own documents — the product searches your content and gives the model the relevant parts. More set-up, more testing.
- Multi-step agents — the AI plans and takes several actions, often across tools. Powerful, but much harder to make reliable.
- Training or fine-tuning your own model — rarely needed for a first version, and the most expensive option by far.
4. Integrations
Each external system — a CRM, payment provider, calendar, accounting tool — adds authentication, data mapping, error handling and testing. One integration is routine. Several, or one with a poorly documented API, can double the work.
5. Users and roles
A product with one type of user (plus a simple admin) is far cheaper than one with several roles and permissions, or a multi-tenant platform where each customer organisation has its own data, users and settings.
Build cost vs running cost
The build is a one-off. Running costs continue every month, and they’re easy to forget when budgeting:
- Hosting and database. Often small for an MVP, growing with usage.
- AI usage. Model providers bill by usage — typically by the amount of text (tokens) sent and received. Your cost depends on the model you choose, how much text each request handles and how many requests users make.
- Other services. Email delivery, file storage, error monitoring and your domain.
Before you build, estimate AI usage with a simple sum: requests per user per day × average text per request × expected users, priced with your provider’s current rates. Then set a spending limit in the provider’s dashboard so a surprise can’t become a large bill.
Where AI MVP budgets go wrong
- Building for every user on day one. Admin panels, teams, roles and billing before anyone has used the core feature.
- Starting with a custom model. Test with an existing model first; most ideas never need more.
- Native mobile apps too early. A responsive web app works on phones and costs a fraction of two native apps.
- Perfect design before evidence. Clean and usable is enough to learn whether people want the product.
- Open-ended scope. Without a written scope, every new idea becomes “just one more thing”.
How Jerrax prices it
We use fixed packages so the scope — and therefore the price — is clear before any work starts:
| Package | Scope | Price |
|---|---|---|
| Starter | One core workflow, up to five screens, one AI API integration, live in 14 days | $699 |
| Growth | Up to three workflows, around eight to ten screens, additional standard integrations | $1,499 |
| Custom | Multiple roles, complex integrations, advanced AI or larger SaaS products | From $2,999 |
Hosting, domains and AI usage are billed by those providers, ideally on accounts you own. See the full pricing details for what each package includes and excludes.
How to keep your first version affordable
- Write the one workflow that proves your idea as a single sentence.
- List the screens that workflow needs — and stop at five.
- Use an existing model API for the AI step.
- Support one type of user, plus basic admin if you need it.
- Estimate monthly running costs before you build.
- Write down what you’re deliberately leaving for version two.
If you want help with step one, read how to scope an AI MVP you can launch in 14 days.
