Multi-Tenant or Single-Tenant: The Architecture Decision Behind a B2B Product, Explained for Business Owners
November 11, 2026 · 6 min read · Etzal Group Team
A company has built a workforce scheduling system for its own use and now wants to sell it to others. The first prospect, a hospital group, asks: "Will our data be in the same database as other customers?" The second, a small clinic, asks only the price. The founder realises that one decision, made before the first sale, will determine cost per customer, how upgrades happen and which of those two buyers can say yes. That decision is multi-tenant or single-tenant.
Three models
Shared everything
One application, one database. Every record carries a tenant identifier, and the software filters every query by it.
- Cheapest to run. Adding a customer adds rows, not servers.
- Simplest to upgrade. One deployment updates everyone.
- Isolation depends entirely on the code. One missing filter is a data leak.
- One customer's heavy use can slow the others.
Shared application, separate databases
One application, and a database or schema per tenant.
- Stronger separation of data. A query cannot cross tenants by accident as easily.
- Per-tenant backup, restore and deletion are straightforward.
- A tenant can be moved to its own server or region.
- More operational work: changes to the data structure must be applied to every database.
- Cost rises with tenant count, moderately.
Single-tenant
A separate instance of application and database per customer, in your cloud or theirs.
- Strongest separation. The answer enterprise security teams like to hear.
- Customisable per customer, which is also its danger.
- Most expensive to run and maintain. Every customer is a deployment to monitor, patch and upgrade.
- Versions drift unless upgrades are disciplined.
Most successful products run mainly on the first or second model and offer the third at a premium.
What changes with the choice
Cost per customer
With shared infrastructure, a small customer costs little to serve, so a low price can be profitable. With an instance each, there is a floor on cost per customer, which rules out small accounts.
Onboarding speed
Multi-tenant: a new customer is created in seconds, which allows self-service sign-up and trials. Single-tenant: provisioning takes time unless heavily automated.
Upgrades
Multi-tenant: everyone gets the new version together. Fast for you, and customers cannot refuse. Single-tenant: upgrades can be scheduled per customer, which enterprises like and which multiplies your work.
Customisation
Multi-tenant pushes you towards configuration: settings, custom fields, workflows. That is healthy. Single-tenant tempts you into per-customer code, and within two years you maintain twelve products.
Security and compliance conversations
Enterprise questionnaires ask about data segregation, encryption, where data is stored and who can access it. Multi-tenant can pass these with good design and evidence. Some buyers, by policy or regulation, require a dedicated environment or data in a specific country.
Reliability
In a shared system, an outage affects everyone, and so does a fix. In separate instances, failures are contained and operations are heavier.
Making multi-tenant safe
If customers share a system, isolation must be designed, not assumed.
- Tenant context set once, centrally. The system determines the tenant from the authenticated session and applies it automatically to all data access. Developers should not have to remember a filter in each query.
- Enforcement in the database where possible, such as row-level security, as a second layer.
- Never trust a tenant identifier sent by the browser.
- Automated tests that try to cross tenants. For every data endpoint: sign in as tenant A, request tenant B's record, expect refusal.
- Separate files and caches by tenant, not only database rows. Leaks often happen in exports, search indexes and cached pages.
- Background jobs carry tenant context explicitly.
- Staff access is controlled and logged. Support staff see a customer's data only with a reason, and the customer can see that access happened.
- Limits per tenant on requests and heavy operations, so one customer cannot exhaust the system.
- Independent penetration testing that includes tenant isolation.
Most data exposure incidents in shared systems come from an ordinary bug in access control, not from an exotic attack.
What enterprise buyers ask
Prepare honest answers.
- Is our data separated from other customers, and how?
- Where is it stored? Country and provider.
- Is it encrypted in transit and at rest?
- Who at your company can access it, and is that logged?
- Can we get our data out, in what format, and is it deleted at the end?
- Can we have a dedicated instance or a specific region?
- How are upgrades scheduled and announced?
- What happened in your last incident?
A clear architecture description and test evidence win more of these conversations than the word "dedicated".
A way to decide
Answer four questions.
- Who are your customers? Many small and mid-sized organisations point to multi-tenant. A handful of large ones with strict requirements point to separate databases or instances.
- What will they pay? If the price per customer is low, only shared infrastructure works.
- What do regulations or their policies require? Data residency or mandated isolation may decide it for some segments.
- How big is your team? Operating fifty instances needs automation and people. A small team is safer with one system.
A common, sound path:
- Start multi-tenant, with strict isolation and one codebase.
- Design so a tenant can live in its own database when needed.
- Offer a dedicated environment as a priced option for customers who require it, deployed from the same code by automation.
- Never fork the code for a customer. Differences are configuration.
Moving later is expensive
Converting a product built for one customer into a multi-tenant system touches every table, query, file path and job. Splitting a tightly shared system into instances is easier but still real work. Decide deliberately early, and keep the door open with clean tenant boundaries.
Common mistakes
- Copying the system for each new customer because it is quick the first three times.
- Tenant filtering left to each developer's memory.
- No tests for cross-tenant access.
- Forgetting files, search and caches.
- Per-customer code branches.
- Promising a dedicated instance without pricing its operation.
- No per-tenant limits.
- Support staff with unlogged access to all customers.
- No plan for per-tenant export and deletion.
We design and build B2B products with tenant isolation, per-tenant configuration and automated deployment as part of Enterprise Software. If you are turning an internal system into a product, contact us.
Summary
Multi-tenant systems serve many customers from one application and are cheap to run and easy to upgrade, with isolation that must be engineered and tested. Single-tenant instances give the strongest separation at the highest operating cost. Choose by who your customers are, what they pay, what rules they face and how big your team is. Most products do best starting multi-tenant with strict isolation, one codebase and a priced dedicated option.
Frequently asked questions
What does multi-tenant mean?
It means one running system serves many customer organisations, called tenants, with each tenant's data kept separate by the software. Customers share the application and usually the infrastructure. Single-tenant means each customer has its own separate instance of the application and database.
Is multi-tenant less secure than single-tenant?
Not inherently. A well-built multi-tenant system with enforced isolation and tested access control can be very secure, and it is easier to keep patched because there is one system. The risk is that an isolation mistake exposes one customer's data to another. Single-tenant limits that particular risk by separation and adds the risk of many instances falling behind on updates.
Can we start multi-tenant and offer dedicated instances later?
Yes, and many products do. Build so that a tenant's data can be placed in a separate database or environment by configuration, keep one codebase, and automate deployment. Then a dedicated instance is a deployment choice with a price, not a fork of the product.
Related service
Website Development
Fast, SEO-ready websites that turn visitors into customers.
