A monthly cloud cost review turns an undifferentiated SaaS bill into decisions: which products, teams, tenants, and workflows consume resources, what each unit costs, and which savings actions deserve attention next. This workflow provides a provider-neutral allocation model, repeatable formulas, a spreadsheet layout, and a review cadence you can reuse as usage, pricing, and architecture change.
Overview
Cloud FinOps for SaaS is not simply a search for the lowest invoice. The objective is to connect infrastructure spend with business activity while preserving the reliability, security, and performance your product requires. A useful monthly review should answer four questions:
- What did we spend, and how does it compare with the approved budget?
- Which products, environments, teams, and customers drove the spend?
- What is the current cost per tenant, transaction, API request, or other meaningful unit?
- Which specific actions can reduce waste or improve unit economics without creating unacceptable operational risk?
Start by separating costs into three categories. Direct costs can be assigned to a product, environment, or customer with reasonable confidence, such as a dedicated database or tenant-specific workload. Shared costs support several products or customers, such as a shared Kubernetes cluster, observability platform, network gateway, or managed cache. Unallocated costs are costs whose ownership is not yet known. The goal is to reduce this last category over time rather than distribute it arbitrarily.
Allocation is most useful when it supports decisions. A precise-looking allocation based on weak assumptions can be less valuable than a transparent estimate marked as shared or provisional. Record the method and confidence level for every important allocation.
How to estimate
Use a billing export or invoice-level report as the financial source of truth, then join it with operational usage data. Provider billing dimensions differ, so map the available fields into a common internal structure instead of assuming every provider exposes the same tags, projects, accounts, or service names.
A practical cost-allocation spreadsheet can use these columns:
- Billing period: the month being reviewed.
- Provider and account: the cloud provider, billing account, subscription, or project.
- Service and resource: the billed service and, where available, resource identifier.
- Environment: production, staging, development, testing, or shared.
- Product or team: the accountable owner.
- Customer scope: a tenant, customer segment, or shared pool.
- Allocation method: direct, usage-based, capacity-based, or fixed shared allocation.
- Amount: the billed cost after applying your chosen treatment of credits, taxes, discounts, and refunds.
- Confidence and notes: high, medium, or low, with the assumption documented.
Then calculate total cost and unit cost. The basic formulas are:
Total SaaS cloud cost = direct costs + allocated shared costs + approved overhead
Cost per unit = total cost assigned to a scope ÷ number of units produced in that scope
A unit might be an active tenant, processed transaction, customer order, API request, gigabyte-month stored, inference request, or another measure that tracks value. Choose one primary unit for management reporting and retain secondary measures for diagnosis. For example, cost per active tenant may be useful for pricing decisions, while cost per transaction helps explain changes caused by usage intensity.
For shared services, choose an allocation driver that reflects consumption. A shared database might use database storage, connection time, query volume, or a documented blend. A shared Kubernetes platform might use requested CPU and memory, observed usage, namespace ownership, or a reserved platform fee. Do not change the driver every month merely to make a product look better; change it when the architecture or business question changes, and record the effective date.
For an operating view, compare the current period with the previous period and with a budget or forecast:
- Month-over-month variance = current cost − previous cost.
- Budget variance = actual cost − budgeted cost.
- Unit-cost change = current cost per unit − previous cost per unit.
- Cost intensity = cloud cost ÷ selected business output, where that ratio is meaningful for your product.
Investigate both total-cost increases and unit-cost increases. A higher bill may be healthy if revenue-producing usage grew faster. Conversely, a flat bill can hide deteriorating efficiency if customer activity declined.
Inputs and assumptions
Before allocating costs, establish a small governance checklist:
- Account structure: decide whether environments, products, or teams should have separate accounts, projects, subscriptions, or billing groups. Separation improves ownership, but excessive fragmentation can add operational overhead.
- Metadata standard: define required tags or labels such as owner, product, environment, cost center, and data classification. Document allowed values and what happens when metadata is missing.
- Ownership map: assign a person or team to each product and shared platform. A tag without an accountable owner rarely creates action.
- Billing treatment: document whether the report includes credits, taxes, refunds, support charges, commitment discounts, and marketplace items. Use the same treatment each month.
- Usage sources: identify the system that provides active tenants, transactions, requests, storage, compute time, or inference volume. Check that its reporting period matches the billing period.
- Shared-cost policy: define which costs remain centrally funded and which are allocated to products. Include a confidence note when a proxy is required.
Keep pricing assumptions separate from observed usage. A rate may change because of a provider price update, a discount plan, a commitment, a region, an instance family, or a billing rule. Record the relevant billing dimension and the date checked rather than hard-coding a rate without context. This is especially important for storage, data transfer, managed databases, Kubernetes platforms, and GPU-based AI workloads.
Use a savings register alongside the allocation sheet. Each row should include the finding, estimated monthly opportunity, implementation effort, risk, owner, due date, status, and validation method. Label estimates as potential savings until the next bill confirms them.
Worked examples
Example 1: Allocating a shared SaaS platform
Assume a SaaS platform has direct product costs of 4,800 units of currency for the month and shared platform costs of 2,000. The team chooses to allocate the shared amount using each product's share of measured compute and database usage. Product A accounts for 60% of the driver and Product B for 40%.
Product A receives 4,800 direct cost plus 1,200 shared cost, for an allocated total of 6,000. Product B receives 0 direct cost in this simplified example plus 800 shared cost. The shared-cost method and the driver should appear in the sheet so that product owners can challenge the assumption constructively.
Example 2: Cost per active tenant
Suppose the full monthly platform cost is 6,800 and the product recorded 850 active tenants. The estimated cost per active tenant is:
6,800 ÷ 850 = 8 cost units per active tenant
If the next month costs 7,200 and active tenants rise to 1,000, cost per tenant becomes 7.20. Total spend increased, but unit economics improved. Review the result with other measures, such as support load, storage consumed, and high-usage tenants, before treating it as a complete efficiency signal.
Example 3: Turning a variance into an action
A monthly review finds that storage costs rose faster than tenant count. The action is not simply “reduce storage.” First, compare new data growth, retention settings, backup copies, snapshots, and idle volumes. A safe action might be to confirm retention requirements, identify unattached volumes, and test a lifecycle policy in a non-production scope. The savings register should capture the owner, expected opportunity, risk to recovery objectives, and the date when the result will be measured.
For related architecture decisions, review the guidance on multi-tenant SaaS cloud architecture, managed PostgreSQL for SaaS, and the Kubernetes cost-optimization playbook. These choices affect which allocation drivers are available and which savings actions are safe.
When to recalculate
Run the full review monthly, but do not wait for month-end when a material change needs an earlier response. Recalculate unit costs after a pricing or discount-plan change, a major architecture migration, a new region, a substantial customer launch, a change in retention policy, or a shift between shared and dedicated infrastructure.
At each monthly review:
- Close the billing period and reconcile the export with the invoice or financial record.
- Check missing or invalid metadata and assign every material unallocated item.
- Compare total cost, budget variance, service mix, and unit costs with the prior period.
- Investigate unusual changes in compute, storage, databases, network transfer, observability, caches, and AI or GPU workloads.
- Review idle resources, oversized capacity, stale snapshots, unused environments, and unexpected data growth.
- Evaluate commitment or discount options only after confirming stable usage and the relevant billing terms.
- Assign a small number of savings actions with owners, dates, risk notes, and validation criteria.
- Update the spreadsheet assumptions and record what changed.
Refresh the workflow whenever provider billing dimensions, pricing terminology, export formats, tagging limitations, or discount rules change. The exact labels and rates in cloud billing consoles are not permanent; verify them against the current provider documentation and your account configuration before making a commitment.
The most valuable outcome is a reliable operating habit: every month, reconcile the bill, explain the movement, update cost per business unit, and convert the clearest finding into an owned action. Over time, this makes cloud cost optimization part of SaaS planning rather than a reactive exercise after an unexpected invoice.