Skip to content
EncryptTechBD
Home Industries Our Work Blog Pricing About Contact Discuss Your Project
বাংলা (BN)
Software Engineering 6 min read

Database schema design for SaaS: a buying guide

How to specify, score and pilot a SaaS schema design engagement, including tenancy questions, red flags and exit terms.

E
EncryptTechBD · 0 comments
Developer writing code on a laptop
Photo: Anita Peeples / StockSnap

A schema is cheap to draw and expensive to change. That is why buying schema design is not like buying a website. The deliverable is not a diagram that looks tidy in a meeting; it is a data model that survives tenant growth, reporting pressure, migrations and the departure of the person who designed it.

This guide is for a SaaS founder or product lead buying help with a new schema, a review, or a redesign before scale makes the problem urgent. It covers what to specify, how to compare providers, how to run a paid pilot and what to protect in the contract.

What you are actually buying

Four engagements get sold under the same label. Decide which one you need before you shortlist.

  • Greenfield schema design. A new data model for a product that has not launched or is being rebuilt. The output is a schema, migration path, seed data and documentation.
  • Existing-schema review. A structured audit of an application already in production, with findings ranked by risk and cost to fix.
  • Multi-tenant redesign. A migration from a single-tenant or shared-database model to a deliberate tenancy model, often with isolation, billing and reporting changes.
  • Performance and migration engineering. Partitioning, indexing, archiving and zero-downtime migration work on a schema that is broadly sound but under strain.

If a provider cannot tell you which of these it is selling, the scope will drift.

Write the requirement brief first

A good brief is one or two pages. It gives every vendor the same problem and makes their questions comparable.

  • Tenancy model. Shared tables with a tenant key, schema per tenant, database per tenant, or a hybrid. State your billing, isolation and reporting needs rather than choosing a label.
  • Access patterns. The ten most important reads and writes, with expected frequency and latency tolerance.
  • Reporting. Which reports must run against live data, which can run against a replica or warehouse, and who needs cross-tenant aggregates.
  • Growth assumptions. Rows per tenant, tenants per year, data retention and peak traffic. Use ranges, and label them as assumptions.
  • Migration constraints. Whether the product can pause, whether a parallel run is possible and how much downtime the business can tolerate.
  • Data classification. Personal data, financial records, documents and anything regulated. This affects encryption, retention and access design.
  • Team skills. The database engine and framework your team can operate after handover.
  • Budget envelope. A range for design, build and review, plus a reserve.

Provider categories and trade-offs

Independent data architect or database consultancy. Strong on modelling, review and migration operations. The risk is a single-person dependency or a narrow database-only view; check whether documentation, product modelling and handover are included.

Product engineering studio. Can design and build the application around the schema. Strong when the schema is inseparable from the product work. The risk is that database review becomes a small part of a larger build and receives less scrutiny.

In-house team with external review. The team owns the design and an external reviewer challenges it. This is often the best value when you have capable engineers but no senior data specialist. It requires the team to share drafts early, not present a finished design for approval.

An evaluation scorecard

Adapt the weights to your context. Score each criterion from 1 to 5 and require evidence, not adjectives.

  • Tenancy correctness — 20%. Can they explain how tenant isolation is enforced in the schema, the application and the backup path?
  • Migration safety — 20%. Have they performed a reversible migration on a live system? Ask for the runbook, not a verbal description.
  • Query and index literacy — 15%. Can they read an execution plan and explain why an index will or will not be used?
  • Testing — 10%. What tests prove the schema works: constraints, referential integrity, tenant isolation and migration rollback?
  • Documentation — 10%. Entity relationships, data dictionary, assumptions and decision records. Ask to see a sample.
  • Handover — 10%. Will they transfer migrations, scripts, diagrams and access, and support your team during the first change cycle?
  • Commercial clarity — 10%. Fixed deliverables, change-request process and what is explicitly excluded.
  • References — 5%. A comparable SaaS engagement, with permission to speak to the technical owner.

Questions that separate candidates

  1. How do you represent tenants, and what prevents one tenant from reading another tenant’s rows through an application bug?
  2. Which queries drive this model, and how did you validate the indexes against realistic data?
  3. How do you deploy a schema change to a large table without a long lock?
  4. What happens to reporting queries when transactional volume doubles?
  5. How do you handle soft deletes, audit history and retention without turning every query into a join?
  6. How is the schema versioned, and how does a developer know which migration is safe to run?
  7. What will you hand over, in which format, and who owns it?

Listen for trade-offs. A candidate who claims one model has no downsides is either inexperienced or selling.

Run a paid pilot

Do not buy the whole redesign from a proposal. Buy a small, bounded review of one module: the tenant table, the orders module or the reporting path.

Define the inputs, the deliverables and the acceptance criteria in advance. A useful pilot produces a written model, a migration sketch, a list of risks and a review session with your engineers. It should also show how the provider responds when your team disagrees.

At the end, score the pilot and decide whether to continue, narrow or stop.

Red flags

  • A guarantee that the schema will never need to change.
  • No access to migration history or a refusal to explain existing tables.
  • ORM-only thinking, with no discussion of constraints, indexes or query plans.
  • No load or volume testing before a migration.
  • Documentation promised only at the end, with no drafts during the work.
  • A refusal to hand over scripts or to support a knowledge transfer.

Contract and exit terms

Own the schema documentation, migration scripts and decision records. Name the environment and repository where they will live. Require handover sessions, not just files. Clarify what happens if the provider leaves mid-migration, who can approve a schema change after launch, and how long they will support the first production cycle.

Add a warranty period with a clear definition of a defect versus a change request, and keep an exit reserve.

Decision framework

Choose a greenfield design engagement when the product is pre-launch and tenancy, billing and reporting are still open decisions.

Choose an existing-schema review when the product works but performance, integrity or migration pain is rising.

Choose a multi-tenant redesign when the current model cannot support isolation, pricing or cross-tenant reporting without manual work.

Choose in-house design with external review when your engineers are capable and the gap is senior challenge.

If you need help scoping work, start with Custom Software & Integrations. The pricing page explains how discovery and build are separated. For failure modes to watch in an existing system, read database schema design problems and fixes for SaaS.

Sources

Services that deliver this

Want this implemented properly?

Pricing →

Every engagement starts with a written scope. If a requirement is outside it, we quote it separately before starting.

Responses (0)

No comments yet. Add the first one below.

Leave a response

Your email address will not be published. Required fields are marked *.

Keep reading

More field notes