09 Jul 2026
SaaS tenancy mistakes that show up in month four
The first three tenants hide the cracks. The fourth one shares a table they should never see. Tenancy is an architecture choice, not a later patch.
Early SaaS products often store every customer in one place and filter in the query. It works until it does not. Month four is when a larger tenant arrives, a report crosses a boundary, or a support tool shows the wrong row.
By then the schema is load bearing. Changing it is a migration under traffic, not a tidy refactor.
Name the tenant in every path
If a request can run without a tenant id, it will. Put the tenant on the session, the query, the job, and the file. Fail closed when it is missing. Do not rely on a developer remembering a where clause.
Admin tools are the usual leak. They are built last and run with extra privilege. Give them the same boundary as the product.
Decide isolation before the second customer
Shared database with a strict tenant key. Schema per tenant. Store per tenant. Each has a cost. The wrong choice is pretending you have not chosen.
If you sell to teams that care about data residency or audit, say so in the architecture, not in a later appendix.
Test the boundary like an attacker
Write tests that try to read another tenant. Run them on every change. Export, search, webhooks, and background jobs are where the bugs hide.
Month four is late to discover this. Month one is cheap.
