Scoping a SaaS MVP: what to build, what to fake with Zapier and AI, what to skip

Build one workflow in custom code: the one a customer pays for. Buy sign-in, billing, email and analytics from established vendors, run the supporting steps behind the scenes with Zapier, an AI step or a person, and leave the rest off the list until paying users ask for it. The result fits on one page a developer can estimate.

Start with the one core workflow test

Write one sentence in this form: "A [type of user] arrives with [input] and leaves with [result] they would pay for." If you need the word "and" to join two results, you have two products. Pick the one your validation interviews support and park the other. A filled-in version reads: "A bookkeeper arrives with a folder of client receipts and leaves with a categorized expense report they would pay for."

That sentence is your core workflow. Each feature on your list now gets one question: can a user finish the core workflow without it? If yes, it goes in the buy, automate or skip column below. If no, you build it.

Sort your feature list into four columns

Place each item on your feature list in one column. The table shows where typical items land.

Give the data model more care than the screens. A developer can redesign a screen without touching customer records, while a change to stored data means migrating each one.

Buying billing still leaves a build task. Stripe Checkout hosts the payment page, Stripe Billing runs recurring invoices and retries failed payments, and the customer portal lets customers update a card or cancel. Your developer writes the code that listens for Stripe's events and switches account access on or off.

Write a short reason next to each skipped item. My recommendation for most first versions is one admin role and one member role, with a responsive web app in place of native mobile apps. A query tool or a spreadsheet export can stand in for a custom admin dashboard.

What to fake behind the scenes

In a concierge step, the customer knows a person does the work: you set up each account by hand on a call. In a Wizard of Oz step, the product looks automated while a person or a simple automation works in the background.

Both let you test demand before you pay for code, on one condition. Do not tell customers a step is automated when it is manual, and do not promise turnaround times a person cannot meet.

Candidates for a first version:

• Onboarding. A form feeds a Zap that creates the account record, sends the welcome email and assigns you a setup task.
• Reports. You or a scheduled automation builds the weekly report from a database export and emails it.
• Notifications. A Zap watches for new records and sends email or Slack messages.
• CRM sync. New signups and plan changes flow into your CRM through an automation.

I recommend writing each faked step into the scope as a named procedure with an owner.

Where an AI step helps and where it hurts

A large language model is a reasonable shortcut for three kinds of work: drafting text a person will review, classifying incoming items into categories, and extracting fields from documents or emails. In each case, have a person check the output before it reaches the customer or the database, at least at the start. I would keep that human review in place until the output has been checked on real customer data for a while.

An AI step is a risk in three situations.

Guaranteed accuracy. Models can produce wrong answers that read as confident. Keep them away from billing amounts, legal or medical statements and anything a customer will treat as final without review.

Cost at scale. Providers charge by usage, so a step that costs little in testing can become a large monthly bill at volume. Ask your developer to estimate cost per core workflow run from the provider's current pricing page.

Customer data. Sending customer data to a third-party API needs a privacy review before launch. Read the provider's data retention and training terms, and name the provider in your privacy policy if the rules that cover your users require it. Those rules depend on where your users live: UK GDPR in the UK, PIPEDA and provincial laws in Canada, and a growing set of state privacy laws in the US. This is not legal advice; a privacy lawyer should answer those questions for your case.

When Zapier stops being enough

Replace an automation with code once you hit one of these limits:

• Volume. Runs climb from dozens a day to thousands.
• Error handling. A failed step leaves a customer half onboarded, and you learn about it from their email. In code, your developer can add retries and alerts.
• Latency. Instant triggers fire when the source app sends a webhook. Polling triggers check for new data at an interval that depends on your plan, so a user can wait minutes for a result they expect in seconds.
• Per-run cost. Zapier counts a task for each successful action step, while triggers and filters use none. Make counts credits for the module operations a scenario runs. A five-step automation uses several units per run, so compare the monthly bill with the cost of building the step once.
• Logic complexity. Branches and exceptions pile up until only one person understands the automation.

A Zapier MVP is a sound start if you plan the exit. Keep your own database as the source of truth, and give each automation one job so your developer can swap a single step for code.

The engineering basics an MVP still needs

Scope is the place to cut, and engineering practice is the place to hold firm. I would remove features and keep these six items in any quote:

1. Version control. All code lives in a Git repository.
2. Backups. The database backs up on a schedule, and your developer restores one backup as a test before launch.
3. Basic tests on the core workflow. Automated tests cover the path a paying customer takes.
4. Staging. Changes reach a staging environment before production.
5. Error monitoring. A tool reports errors to your developer before customers do.
6. Ownership. You own the repositories, the cloud accounts, the domain and the Stripe account. Your agency gets collaborator access.

What a one-page MVP scope contains

Write these seven parts on one page, in this order, and a developer has enough to estimate:

• The user and the problem, in two sentences, with your evidence that the problem is real.
• The core workflow, as numbered steps from the user's first action to the result.
• The data, as a plain list of the things the product stores, such as accounts and invoices, and how they relate.
• The four columns, with each feature placed in build, buy, automate or skip.
• Integrations, naming each vendor you have chosen or want advice on.
• Constraints: budget range, target date, user regions and compliance needs.
• Done, defined as one observable event, for example "ten pilot customers complete the core workflow and one pays through Stripe."

When you should not build software yet

Hold off on code in three cases. First, you have not yet heard people with the problem describe it in their own words. Second, you can deliver the result by hand, with a spreadsheet and a form, to your first few customers. Third, an existing product solves the problem with some configuration, and your advantage sits in the service around it.

What devinsta does here

We at devinsta scope and build first versions for founders in the US, Canada and UK, so I have a commercial interest in the build route. Our MVP SaaS development service covers the build column in two-week sprints with weekly demos, and you own the code and the accounts. For products that outgrow a first version, we build custom web applications.

For the automate column, we set up Zapier automation and AI workflow automation with human review, plus CRM integration. Our startups page covers how we work with early-stage teams.

If you have a feature list and no scope, book a free 30-minute strategy call and we will sort it into the four columns with you. If you have the one-page scope, send it through the project brief. If the manual route looks like the better next step, I will say so on the call.

Common questions

How many features should a SaaS MVP have?

Count workflows instead of features. A first version needs one core workflow a customer would pay for, plus sign-in and billing bought from vendors. Each other item goes into the automate or skip column until users ask for it.

Can I build a SaaS MVP on Zapier alone?

You can run the supporting steps on Zapier or Make, and some founders test demand with forms, a spreadsheet and automations before writing code. Once customers need their own accounts and stored data, you will want a real database and a small custom app for the core workflow.

Is it safe to use AI features in an MVP?

Yes for drafting, classification and extraction where a person reviews the output. Avoid AI steps where a wrong answer has a cost you cannot absorb, estimate the usage cost per run, and complete a privacy review before you send customer data to a third-party API.

What should I own when an agency builds my MVP?

Own the code repositories, the cloud hosting account, the domain, the database backups and the Stripe account. Create them under your company's email and grant the agency access, so you keep control if the engagement ends.