Bubble is good at proving an idea. It gets expensive at the point the idea starts working: when a workflow needs logic Bubble will not express, when the database is big enough that pages crawl, when a customer asks a data question you cannot answer, or when the bill starts scaling with your success. Moving off is not a failure of the build. It is what happens to a build that worked.
You do not need all of these. Two is usually enough to make the arithmetic work.
Searches and repeating groups that were instant at a thousand rows are not instant at a hundred thousand, and editor-side tuning has a ceiling.
Pricing rules, scheduling logic, multi-step approvals or anything recursive. You end up with forty workflow steps standing in for twenty lines of code.
Webhook signature checks, long-running jobs, retries, queues, and anything that has to keep a secret well away from the browser.
Where the data lives, who can reach it, what happens in an audit. Enterprise buyers ask, and the answer has to be about your infrastructure.
Workload-based pricing means the better the product does, the more the platform takes. Owned infrastructure does not work that way.
The pool of Bubble developers is small. The pool of React and Node developers is not, and that matters the day you want a team.
A standard, boring, hireable stack, chosen so that the next team after us is not stuck either. Next.js and React with TypeScript on the front, Node on the API, PostgreSQL underneath. Hosting on infrastructure in your own accounts, whether that is Vercel and a managed Postgres or AWS with Terraform. The data comes out of Bubble's database and goes in with its relationships intact, not as a flat export nobody can use.
We go through the app page by page and write down what it actually does. Bubble apps almost always contain features nobody remembers building.
Bubble's data types become a relational schema with real constraints. This is where the hidden work is, and where doing it badly costs you later.
One area at a time, reviewable as it lands, so you can watch progress rather than wait months for a single reveal at the end.
A repeatable migration script, run against a copy first, so the real cutover is not the first time anyone has tried it.
The Bubble app keeps serving customers until the replacement is proven. Then you switch, and we keep the old one available for a while.
The repository, architecture notes, setup instructions, a deployment runbook, and tests on authentication and billing.
Yes. Bubble exposes the database through its Data API and CSV export, and we write a migration that maps Bubble's data types onto a relational schema, including the relationships between them. We run it against a copy of the data first and reconcile the row counts before anything real depends on it.
It can, and often the first version deliberately does, because a migration and a redesign at the same time makes it impossible to tell which one broke something. Once you are off the platform and stable, changing the interface is cheap. We usually recommend doing them in that order.
For a focused app it is usually weeks rather than months. The variables are how many distinct workflows exist, how much of the logic is undocumented, and how much data has to move. You get a real timeline after the audit rather than an optimistic one before it.
One price for the migration, agreed at discovery and written down before anything starts. What moves it is how much app there is: how many screens, how much logic lives in workflows, and how loosely the data was defined. The audit on its own is a much smaller piece of work and can be bought by itself, which is the sensible first step if you are not sure the migration is warranted yet.
Sometimes no, and we will say so. If the app is serving its users, the data volume is modest and nothing in the roadmap needs logic Bubble cannot express, staying is the cheaper answer. The audit exists to give you that verdict honestly, including when it costs us the project.
Each one gets replaced by a library or a direct integration, and most have a well-supported equivalent. Where a plugin is doing something genuinely unusual we flag it during the audit, because that is exactly the kind of thing that turns a clean estimate into a surprise.
Describe what you’re building, what it has to do and when it needs to be live. You’ll get back the architecture we’d use, the trade-offs, and a realistic timeline, before anything is signed.