All insights
Engineering··9 min read

AI-Built SaaS Launch Checklist: 15 Things to Fix Before Real Users Arrive

Your Lovable, Cursor or Bolt app works in the demo. Here are 15 launch fixes for payments, email, backups, monitoring and costs before real users arrive.

AI-Built SaaS Launch Checklist: 15 Things to Fix Before Real Users Arrive

An AI-built SaaS can look finished and still break in its first week with real users. The demo only proves the happy path works. Launch tests everything else: a card that fails, an email that lands in spam, a database that gets wiped, a bill that triples overnight.

We see this pattern in most Lovable, Cursor and Bolt projects that come to us through Finish Line. The core feature works. The parts around it were never built, because nobody prompted for them.

Security is the first gap, and we covered it in Lovable and Vibe Coding Security: 8 Risks to Check Before Launch. This checklist covers the rest: the 15 things that decide whether your launch week is calm or chaotic.

The checklist at a glance

#CheckWhat breaks if you skip it
1Payment webhooks verified and idempotentUsers pay but never get access, or get charged twice
2Every subscription state handledFailed cards and cancellations keep full access
3One real live-mode payment and refundYour first customer finds the bug for you
4Email sent from your domain with SPF, DKIM, DMARCSignup and reset emails land in spam
5Auth flows tested end to endUsers get locked out and email support
6Automated daily backupsOne bad query wipes customer data for good
7A restore you have actually testedBackups exist but nobody can use them
8Separate dev and productionTest data and test keys leak into production
9Error trackingUsers hit errors you never see
10Uptime monitoring with alertsYour app is down and a customer tells you
11Indexes and pagination on big tablesPages slow to a crawl at a few thousand rows
12Spend caps and rate limits on AI and APIsOne user or bot runs up a huge bill overnight
13Slow work moved to background jobsRequests time out and actions run half-way
14Privacy policy, terms and account deletionPayment providers and app stores reject you
15Analytics, support channel and domain basicsYou launch blind and lose early feedback

Payments and billing

Payments are where AI-built apps fail most expensively, because the demo only ever tests a successful card in test mode.

1. Payment webhooks are verified and idempotent

Your app should grant access when Stripe (or Paddle, or Lemon Squeezy) sends a webhook, not when the browser lands on a success page. Users close tabs, and redirects fail.

Check two things. First, the webhook verifies the provider's signature, so nobody can fake a "payment succeeded" call. Second, the same event processed twice does nothing the second time. Providers retry webhooks, and without this you get double credits or duplicate accounts.

2. Every subscription state is handled

Generated code usually handles "active" and nothing else. Real subscriptions also go past due, get cancelled at period end, get upgraded mid-cycle and get refunded.

Write down what a user can do in each state, then test each one. The most common bug we find: a cancelled or failed-payment user keeps full access forever.

3. One real live-mode payment and refund

Test mode hides real-world problems: tax settings, currency, card authentication and live webhook URLs. Before launch, buy your own product with a real card, confirm access, then refund it and confirm access ends. It costs a few dollars and removes the most embarrassing launch-day bug.

Email and sign-in

If users cannot sign in, nothing else you built matters.

4. Email comes from your domain, with SPF, DKIM and DMARC

Many prototypes send email through the platform's default sender. It works in testing because you are emailing yourself. With real users, signup confirmations and password resets start landing in spam or not arriving at all.

Send transactional email through a proper provider such as Postmark, Resend or Amazon SES, from your own domain. Set up SPF, DKIM and DMARC records, then send a test to a Gmail and an Outlook address and check that it reaches the inbox.

5. Auth flows are tested end to end

Test the whole journey, not just the login screen: sign up, verify email, log out, reset password, change email and sign in on a second device. Then try the awkward cases, like an expired reset link or someone signing up twice with the same address.

If you offer Google sign-in, set the production redirect URLs and publish the consent screen. A Google login stuck in testing mode only works for the people you added by hand.

Data safety

Once real customers put real data into your app, losing it is the one mistake you cannot apologise your way out of.

6. Automated daily backups

Check what your database plan actually includes. Some free and starter tiers keep only a day of backups, or none you can download. For a paid product, you want automated daily backups kept for at least 7 days, with point-in-time recovery if your provider offers it.

7. A restore you have actually tested

A backup you have never restored is a hope, not a plan. Before launch, restore last night's backup into a separate database and check that the data is complete and the app runs against it. Write down the steps and how long it took, so a stressful day later is just a checklist.

8. Separate development and production

Many AI-built apps run on one database and one set of API keys for everything. That means a test migration can drop a production table, and test users show up in your real metrics.

Run two environments with separate databases and separate keys. Use live payment keys only in production, and never point a local copy of the app at production data.

Monitoring

Most users do not report bugs. They leave. Monitoring is how you find out before they do.

9. Error tracking on frontend and backend

Add an error tracker such as Sentry to both the browser and the server. Every crash then arrives with the page, the user and the stack trace, instead of a vague "it didn't work" email a week later. Set it to alert you on new errors, not every repeat of an old one.

10. Uptime monitoring with alerts

Use a simple uptime checker such as Better Stack or UptimeRobot to hit your homepage, your login page and one API endpoint every few minutes. Send alerts to the phone of whoever can fix the problem. If a customer is the first to tell you the app is down, this check was missing.

Performance and cost limits

Your demo had ten rows of data and one user. These checks are about what happens at ten thousand rows and a hundred users.

11. Indexes and pagination on large tables

Generated code often loads whole tables and filters them in the browser. That is fine with test data, then pages take seconds once a real customer has a few thousand records.

Add database indexes on the columns you filter and sort by, such as user ID, status and created date. Paginate every list that can grow, and load only the fields the screen actually shows.

12. Spend caps and rate limits on AI and paid APIs

If your product calls OpenAI, Anthropic or any paid API, every request costs you money. One user stuck in a loop, or one bot hitting your signup form, can run up a bill that wipes out a month of revenue.

Set a monthly spend limit in each provider's dashboard. Add per-user limits in your app, such as a daily number of AI generations. Rate-limit public endpoints like signup, login and contact forms.

13. Slow work moved to background jobs

Long AI generations, file processing, imports and bulk emails should not run inside a normal web request. On most hosting, those requests time out after 10 to 60 seconds, and the work stops half-way with no record of what finished.

Move slow tasks into a background job queue, show the user a "processing" state, and retry failed jobs automatically.

Legal, trust and launch day

14. Privacy policy, terms and account deletion

Payment providers, app stores and Google sign-in all expect a privacy policy and terms of service at public URLs. Users also expect a way to delete their account and data, and in many regions the law requires it.

Publish both pages, link them in your footer and signup form, and build a working "delete my account" option. If you use non-essential cookies or tracking for EU visitors, add a consent banner. These are not legal advice, so have a lawyer review them once you have revenue.

15. Analytics, a support channel and domain basics

On launch day you want to know where users drop off and have a way for them to reach you. Add product analytics such as PostHog or Plausible, and track at least signup, activation and payment. Give users one clear support route, like an email address or a chat widget, that someone actually reads.

Then check the basics: your domain redirects to one version (with or without www), SSL is active everywhere, each page has its own title and meta description, and a sitemap is submitted to Google Search Console.

Short on time? Fix these five first

If launch is days away, start with the checks that can cost you money or data. The rest can follow in your first month.

  1. Payment webhooks verified and idempotent (#1)
  2. Automated backups plus one tested restore (#6 and #7)
  3. Spend caps and rate limits on AI and APIs (#12)
  4. Email sent from your domain with SPF, DKIM and DMARC (#4)
  5. Error tracking (#9)

These five take a few days for an experienced engineer, not weeks. None of them need a rewrite. They sit around the code you already have.

FAQ

Can I fix these myself with Cursor or Lovable?

Some of them, yes. Backups, uptime monitoring and spend caps are mostly dashboard settings. Webhooks, subscription states and background jobs are harder, because the AI will write code that looks right without knowing which edge cases matter for your product.

Do I need to rebuild my AI-built app before launch?

Usually not. Most of these checks add missing pieces around working code. A rebuild only makes sense when the core data model or architecture is wrong, and a review will tell you that quickly. For more on that decision, see When an AI-Built App Needs a Real Engineering Team.

How long does it take to make an AI-built app launch-ready?

For a typical MVP, one to three weeks, depending on how many of these 15 checks are missing and how many integrations the app has.

Not sure which of these your app is missing?

That is what our free Finish Line Scan is for. Send us your AI-built app, and we will review it against this checklist and our security checklist. You get a short report on what is blocking launch, what can wait, and a fixed price if you want us to fix it.

AI product development

Get your free Finish Line Scan