This guide is for SaaS founders, CTOs and engineering leads whose cloud bill is growing faster than revenue, or who are about to price a plan without knowing what a tenant costs to serve. Our SaaS platform engineering guide summarises the cost habits in a few lines; this page turns them into working controls.
In this guide
- Why SaaS cloud cost needs its own controls
- Attribute every cost first
- Cost per tenant: three ways to allocate
- Guardrails that catch spend early
- Levers that lower the bill
- AI cost per tenant
- An operating rhythm
- Alternatives: who runs the controls
- Worked examples: SaaS products Netbase builds and runs
- What delivery record exists, and what does not
- Limits of this guide
- Frequently asked questions
- Next step
Why SaaS cloud cost needs its own controls
In a SaaS business, cloud spend is part of the cost of goods sold: every new customer adds some of it. Three things make it hard to manage with an ordinary IT budget.
- Shared resources hide the cost of a tenant. A pooled database or cluster serves hundreds of customers, and the bill does not say which one used what. Microsoft's multitenant guidance names this as the core difficulty: most services do not break usage down by your definition of a tenant.
- Usage, not headcount, drives the bill. A single customer that imports a large dataset or runs heavy reports can change the month's total.
- AI features add a price per call. Model calls and tokens are billed per use, so one feature can move a plan's margin without any change in infrastructure.
The FinOps Foundation's unit economics capability describes the goal: connect technology spend to business value through metrics such as cost per customer. For a SaaS product that means cost per tenant, per plan and per core transaction.
Attribute every cost first
Controls are only as good as the attribution under them. Before any saving:
- Separate environments by account or project. Production, staging and development each get their own account or project, so a forgotten test cluster cannot hide inside production spend.
- Tag every resource. Environment, service, owner and, for dedicated resources, tenant. Enforce the tags in infrastructure as code rather than by convention.
- Name an owner for untagged and shared costs. Networking, logging, support plans and data transfer rarely carry a tenant tag; allocate them by a documented rule, not by guesswork.
- Keep usage events per tenant. The same events that feed billing feed cost allocation. A SaaS product that emits them from its first release, as recommended in our SaaS MVP architecture guide, can answer cost questions later.
Cost per tenant: three ways to allocate
| Method | How it works | Choose it when |
|---|---|---|
| Direct attribution | Dedicated resources carry a tenant tag, and the bill for that tag is the tenant's cost | Tenants run on their own database or stack |
| Metered share | A usage measure per tenant, such as storage, transactions or requests, splits the pooled bill | Tenants share infrastructure and one measure tracks what drives cost |
| Estimated share | Each tenant is charged a proportion of the shared cost, checked against a periodic baseline | Many small tenants with similar usage and flat pricing |
Most products combine them: direct attribution for large tenants in a dedicated tier, a metered share for the pooled tier, and an estimate for the remaining overhead. Microsoft's guidance recommends checking an indicative measure against a measured baseline from time to time, because a tenant that runs heavy reports can use far more compute than its storage suggests. The AWS SaaS Lens treats cost per tenant as part of its cost optimization pillar for the same reason. How the tenancy model itself changes cost is covered in multi-tenant SaaS architecture.
Guardrails that catch spend early
Attribution tells you where money went; guardrails tell you before the invoice.
- Budgets with alerts per account, service and team. Google Cloud's documentation is explicit that an alerts-only budget does not cap spending, so a budget is a warning, not a brake.
- Anomaly detection. AWS Cost Anomaly Detection, for example, uses machine learning to flag unusual spend by service, account, region or usage type. Route alerts to the owning team's channel, not to a finance inbox nobody reads.
- Tenant limits enforced in the product. Plan entitlements cap storage, seats, API calls and AI usage, so a single tenant cannot consume a plan's margin. This is the only guardrail that acts before the cost is incurred.
- Automatic shutdown outside production. Development and test environments stop outside working hours unless someone opts out.
Levers that lower the bill
| Lever | What it changes | Watch out for |
|---|---|---|
| Rightsizing | Instances, databases and containers sized to measured load | Sizing to averages and missing the peak |
| Autoscaling and scheduling | Capacity follows demand; idle environments stop | Scale-up delays on sudden spikes |
| Commitment discounts | Lower rates for steady baseline load | Committing before the baseline is known |
| Storage lifecycle | Old files and backups move to cheaper tiers or expire | Retention rules the contract or law requires |
| Data transfer and caching | A CDN and regional placement cut transfer charges | Moving data away from where tenants are |
| Logs and monitoring retention | Verbose logs sampled and kept only as long as useful | Losing the logs an incident needs |
Apply levers in that order of risk: remove waste first, then resize, then commit. A discount on capacity you do not need is still waste.
AI cost per tenant
AI features need the same controls at a finer grain. Meter model calls and tokens per tenant and per feature, map them to entitlements with a monthly allowance, choose a smaller or cheaper model where quality allows, cache repeated answers, and alert when a tenant's AI spend leaves its usual range. Netbase works with the major commercial and open-source AI models, chosen per product rather than tied to one vendor, which keeps model choice a cost decision. Each item below states how mature it is at Netbase.
-
Available capability
AI features metered per tenant
Machine learning, NLP and generative AI features with per-tenant usage limits; not yet tied to a published cost case.
-
Available capability
AI-assisted cost anomaly triage
Unusual spend grouped by likely cause for an engineer to confirm; not yet tied to a published cloud cost case.
An operating rhythm
-
Weekly: engineering looks at anomalies
The team that owns a service explains any alert and closes it or opens a fix.
-
Monthly: engineering and finance review unit costs
Cost per tenant, per plan and per core transaction, compared with the previous month and with price.
-
Quarterly: decide the big levers
Commitments, tenancy changes for the heaviest tenants, and plan limits or prices that do not cover their cost.
-
At every release: check the cost impact
A new feature states which cost it adds and how it is metered, before it ships.
Alternatives: who runs the controls
| Route | Strength | Choose it when |
|---|---|---|
| Provider-native tools | No extra licence; budgets, tags and anomaly alerts included | One provider, and a team that can build the reports |
| Third-party cost platform | Cross-provider views and ready tenant allocation | Several providers or a large bill that justifies the subscription |
| In-house FinOps role | Owns the rhythm and the decisions across teams | Spend large enough to fund a dedicated person |
| Managed cloud partner | Reviews, rightsizing and alerts run for you | No platform team, or one that should stay on the product |
Whichever route you choose, the product itself must emit usage per tenant; no external tool can recover it later. Netbase builds these controls through cloud platform engineering and SaaS development, and runs monthly and quarterly cost reviews as part of managed cloud services. Most Netbase projects are delivered on fixed-price contracts agreed after discovery.
Worked examples: SaaS products Netbase builds and runs
- Cloodo Workspace. A multi-tenant digital workplace that Netbase built and operates as a Netbase Business Division, with CRM, HRM, Cloud ERP and AI modules in one product. See the Cloodo record.
- Cloud ERP for a US client (not named). A multi-tenant cloud ERP on AWS that Netbase has developed as offshore development and managing partner since 2020, where every customer business works in its own workspace on shared infrastructure. See the cloud ERP record.
- Printcart. The web-to-print SaaS of the Netbase Business Divisions, serving many merchants from one platform. See the Printcart record.
Netbase works on AWS, Google Cloud, DigitalOcean and Cloudflare; no cloud partner tier is claimed.
Plan the next step with a Netbase consultant
What delivery record exists, and what does not
- What exists. The records above describe multi-tenant products that Netbase builds and runs on shared infrastructure, which is the setting these controls are written for.
- What does not. No Netbase record publishes a cloud bill, a saving, a cost per tenant or a margin, for its own products or a client's. Nothing on this page is offered as a measured cost result.
Limits of this guide
- Provider features are cited from their own documentation on the access date; names, limits and prices change.
- The allocation methods produce estimates, and the right precision depends on your pricing model.
- The security side of cost controls, such as who can create resources and change budgets, follows the same role-based access and MFA rules as any admin access.
Frequently asked questions
Pick a usage measure that tracks what drives cost, such as storage, transactions or requests, split the pooled cost by it, and check the result against a measured baseline from time to time.
Not by themselves. Budgets alert when spend crosses a threshold; stopping spend needs automation you add, or limits enforced in your own product.
After a few months of stable baseline usage, and only for that baseline. Remove waste and rightsize first.
The engineering lead owns the alerts and levers; finance owns the monthly review of unit costs with them.
Next step
Share your current cloud bill structure, your plans and your largest tenants, and we will book a solution review to set up attribution, guardrails and a review rhythm. You can also see cloud platform engineering or more Netbase insights.
Related services and solutions
AI-ready SaaS development from a team that runs its own SaaS
Netbase provides SaaS development services for founders and product teams to launch and scale multi-tenant subscription software with AI features customers will pay for. We build and operate our own SaaS platforms, Printcart and the AI-powered Cloodo workplace, and bring that experience to client products. A typical SaaS MVP takes 8–12 weeks, depending on scope, integrations and review speed.
Learn More
AI-ready cloud platform engineering for SaaS and commerce workloads
Netbase provides cloud platform engineering for SaaS and commerce teams that need a secure, repeatable home for their software and its AI features, on AWS, Google Cloud, DigitalOcean or Cloudflare. We design the architecture, write the environments as code, connect managed and AI services and hand over a platform your team can run, built the way we run ours.
Learn More
SaaS product accelerator: launch an AI-ready SaaS on proven Netbase modules
A SaaS product accelerator is a set of reusable Netbase modules for accounts, billing, roles and integrations that helps founders and product teams launch subscription software faster, with room for in-product AI from the first release. Reusing these modules can cut development time by up to 60%, and the approach is proven on Printcart, Netbase's own web-to-print SaaS.
Learn More
Discuss a project
Netbase JSC helps organizations design, build, modernize, and operate digital products and AI-enabled business systems.+84 937 869 689
91 Nguyen Chi Thanh, Dong Da, Hanoi, Vietnam
Get in touch
Tell us what you want to build, modernize, or operate.