An architecture review cost checklist is filled in by the architect or the team designing a system before the review, and the reviewer ticks it off during the review, at the first design, before go-live and after <90> days of operation; it records what the system is meant to cost and why, which questions per layer are answered, which pricing models were chosen, what an exit would cost and under which conditions the system is approved. The decision it records: approved, approved with conditions or back to design. Why cost belongs in the review and how the providers' Well-Architected programmes ask the question is explained on cost in architecture and procurement reviews; which design decisions move the bill most on architecture choices and cost. This page is the checklist.
The template
Architecture review cost checklist <system>, review on <date>, architect <name>, reviewer <name>, phase <design, before go-live, after 90 days of operation>
Estimate
| Field | Value |
|---|---|
| Cost per month at today's load | <amount>, source <the provider's pricing calculator, link, date> |
| Cost per month at the load in twelve months | <amount>, assumption <units> |
| Cost per unit | <amount per request, customer, GB or job> |
| Cost basis | <on-demand; with commitments from month> |
| Regions and currency | <region>, <invoice currency>, rate <source> |
| Assumptions | <requests per day, data volume, storage growth per month, retention in days> |
| The three largest items | <service, amount>; <service, amount>; <service, amount> |
Questions per layer
| Layer | Question | Answer | Cost effect per month |
|---|---|---|---|
| Compute | Which instance family and generation, and why not the current one or ARM? | <answer> | <amount> |
| Compute | Auto-scaling with minimum and maximum, shutdown outside working hours? | <answer> | <amount> |
| Compute | Serverless against instances computed for this load curve? | <answer> | <amount> |
| Storage | Storage class, lifecycle rule, versioning and snapshot retention fixed? | <answer> | <amount> |
| Storage | Performance tier (IOPS, throughput) from measured need or from the default? | <answer> | <amount> |
| Network | Data transfer between zones, regions and to the internet estimated? | <GB per month, amount> | <amount> |
| Network | NAT gateway volume, endpoints, CDN in front of the origin? | <answer> | <amount> |
| Data | Database engine and edition, high availability only in prod? | <answer> | <amount> |
| Data | Provisioned against on-demand capacity, retention of backups and logs? | <answer> | <amount> |
| Data | Analytics: data scanned per query, partitioning? | <answer> | <amount> |
| AI | Model and rate card, tokens per request, caching and batching, reserved against on-demand throughput? | <answer> | <amount> |
| Observability | Log volume per day, retention, metric cardinality? | <answer> | <amount> |
Pricing model
- Share of the load stable enough for a commitment named:
<percent>from<month> - Spot or preemptible capacity checked for interruptible load
- Licences: brought along or included in the price, with reasoning
- Marketplace contracts named with term and renewal
- Effect on the account's support plan checked
- No dependence on free tier or credits in the estimate
Exit and portability
- Proprietary services named, with the alternative per service
- Data volume and egress cost of an exit estimated:
<TB, amount> - Infrastructure as code in place; data in open formats
- Contract lock-in:
<term, notice period>
Operations
- Tags or labels in the infrastructure code, per the tagging policy
- Budget and alert created for the system, account per the account checklist
- Cost per unit as a figure on the team's dashboard
- Review after
<90>days of operation scheduled: estimate against invoice
Outcome
| Field | Value |
|---|---|
| Decision | <approved, approved with conditions, back to design> |
| Conditions | <condition>, owner <name>, due <date> |
| Variance of estimate against invoice (after 90 days) | <amount, percent>, cause <text> |
| Next review | <date, occasion> |
How to fill it in
Estimate: the provider's pricing calculator with link and date, because list prices change; the cost per unit is the figure the review checks after 90 days (unit economics). The three largest items show where the questions per layer carry weight.
Layers: every row has an answer and an amount, even when the amount reads "negligible, because". Data transfer is forgotten most often; what moves between zones, regions and outside costs is on data transfer and egress. Storage classes and lifecycle rules are explained on storage tiers and lifecycle, shutdown and scaling on scheduling and auto-scaling, the AI providers' price lists on model pricing and rate cards.
Pricing model: the commitment share is a promise to the commitment purchase decision record, not the purchase itself; which programme fits is on choosing between commitment types. Exit: what an exit may cost under the Data Act is explained on Data Act and switching charges. Operations: the account follows the account checklist.
How to adapt it
One team: architect and reviewer are colleagues; the checklist is filled in on the design's pull request, and the outcome is a comment. Several teams: a reviewer from another team or the platform, and the estimate goes to finance as a budget line.
One provider: the questions per layer name that provider's services directly. Three providers: one more question per layer, whether the choice of provider for this service is reasoned, and data transfer between providers as its own row.
Startup: the estimate, the three largest items, data transfer and the free tier point are enough; exit and commitments come with the first annual contract. Finance-led: the outcome carries the budget line and the cost per unit, and the variance after 90 days is reported in the monthly review.
Sources
Read on 2026-10-21. FinOps Foundation, Architecting & Workload Placement (CC BY 4.0), AWS Well-Architected, cost optimization, COST08-BP01 data transfer modelling, AWS Pricing Calculator, Azure Well-Architected, cost optimization checklist, estimate costs with the Azure pricing calculator, Google Cloud, cost optimization pillar, estimate your monthly costs in Google Cloud, Regulation (EU) 2023/2854 (Data Act).