Skip to content
Tech Interview Prep
Technical interview guide

Total Cost of Ownership (TCO) Estimation

Estimating the real cost of a proposed architecture, including the operational costs that don't show up on a vendor's price sheet.

Read
42 min
Practice MCQs
25
Interview Q&A
25
Edition
v4
Editorial status
Reviewed

Scope: FinOps Foundation, AWS, Azure, and Google Cloud cost-management guidance current 2026-08-31.

Curated: · Written: · Reviewed:

25 questions

MCQ-11A platform team spent $400k over 18 months building an in-house job queue. Running it costs $40k/month in infrastructure and one engineer's time. A managed queue costs $25k/month with a one-time $60k migration fee. The team is choosing what to run for the next 12 months. What should it do?
MCQ-2A platform team is deciding whether to keep its self-hosted CI cluster or move to a managed CI service, and is comparing total cost of ownership over the next three years. The analyst's draft for the self-hosted option lists four lines: $180,000 for the servers, bought two years ago and still in use; $120,000 per year for the two engineers who maintain the cluster; $12,000 per year for the vendor support contract; and $25,000 one-time to upgrade the cluster to the version the comparison assumes. The managed service is quoted at $9,000 per month with no migration cost. Which three-year figure for the self-hosted option is the correct comparable cost?
MCQ-1A platform team is choosing between a managed container service at $4,000/month and self-managed Kubernetes on reserved instances at $1,800/month. Self-managing also needs 0.25 FTE of a platform engineer ($220,000 fully loaded) and a $9,000 one-time migration. Over a three-year horizon, what does a TCO comparison of the two options show?
MCQ-7A team is deciding whether to self-host a PostgreSQL cluster or buy a managed database service. Self-hosting requires one DBA at 50% of her time for operations, backups and failover drills, plus two backend engineers at 10% each for query tuning and schema migrations that the managed service's tooling would otherwise handle. Fully loaded annual cost is $280,000 for the DBA and $240,000 for each backend engineer. What people cost should the self-hosted option carry in the annual comparison?
MCQ-12A platform team is choosing between self-managed VMs and a managed service on a three-year commitment. The self-managed quote is $480,000 for infrastructure and excludes 0.5 FTE of operations at $180,000 per year fully loaded and a $60,000 DR site. The managed quote is $720,000 all-in, includes DR, and includes 40,000 GB/month of egress; the workload's measured egress is 90,000 GB/month, billed at $0.09/GB beyond the included allowance. Which conclusion is correct on a like-for-like three-year basis?
MCQ-10Acme is moving 60 VMs from its own data center to a cloud provider. The finance-approved TCO includes $18,000/month for the cloud estate's compute and storage, $240,000 for migration contractors, and $9,000/month for ongoing cloud support after cutover. The migration plan requires a four-month parallel run before cutover: the on-premises estate keeps running at $22,000/month, the cloud estate runs at $18,000/month, and reconciliation labor costs $15,000/month. Which transition cost is missing from the approved TCO?
MCQ-5A team is building a three-year TCO for a managed streaming platform. The vendor quotes $15,000/month list and offers a $180,000 credit applied over the first 12 months; after month 12 the credit ends and list price applies. Finance will use the TCO to approve the migration. How should the credit be represented?
MCQ-17A team is comparing two three-year options for a service. Option A is self-hosted at $50,000 per year, quoted tax-exclusive, and the 10% sales tax on it is not recoverable. Option B is a SaaS contract at €30,000 per year, quoted tax-inclusive at 20% VAT; the company is not VAT-registered in the EU, so it cannot reclaim that VAT. The model converts at a single base-date rate of 1 EUR = 1.25 USD. What three-year cost difference should the model report for Option A minus Option B, in USD?
MCQ-18A team is validating the compute line of a three-year TCO model for a batch analytics cluster. The model prices the cluster at $480,000 over three years, assuming it runs continuously and is billed per node-hour on demand. Job telemetry shows the cluster is actually needed 6 hours a day, 5 days a week. Which figure is the defensible compute line for the model?
MCQ-21A team runs a 300-node batch fleet on on-demand instances for 42,000 USD per month. Moving to spot would cut the instance bill to 15,000 USD per month, but preemption forces a fallback on-demand lane that costs 5,000 USD per month, and the checkpointing code plus interruption triage would consume 1.5 engineers for the year. A fully loaded engineer costs 240,000 USD per year. What does the annual comparison show?
MCQ-22A service's monthly bill is $20,000: $6,000 compute, $2,000 storage, and $12,000 data transfer and NAT processing. The app tier reads 240 TB/month from S3 and every byte passes through the NAT gateway, which bills $0.045/GB processed. Each compute instance holds a 4-minute warm-up cache, so any restart breaks the 2-second p99 SLO. The team must cut 20% of the total bill this quarter. Which action meets the target?
MCQ-3A telemetry service writes one 200-byte object per event to object storage: 500 million events a month, each object retained 30 days and then deleted. Storage bills at $0.02 per GB-month and writes at $0.005 per 1,000 requests; reads are negligible. (1 GB = 10^9 bytes.) The team can ship one change this quarter. Which change cuts the monthly bill the most?
MCQ-24A team built a 3-year TCO model comparing on-prem ($2.10M) with cloud ($1.95M) for a service at 60 TB stored, 400 vCPU, and 12 TB/month egress, using 3-year reserved pricing and cloud egress at $0.09/GB; on-prem egress is internal. Egress has since grown to 60 TB/month and the provider raised cloud egress to $0.12/GB; storage and compute are unchanged. Re-running the same 3-year comparison with 1 TB = 1000 GB, what does it now show?
MCQ-4A payments platform cost $48,000 in March while processing 4.0 million transactions. In April it cost $57,600 while processing 5.4 million transactions. Finance asks whether the platform became less efficient. What do the numbers show?
MCQ-14A payments team can cut $40k/month by moving a workload from its EU region to a cheaper US region. Which finding is a nonfinancial gate that the savings cannot override?
MCQ-8A platform team runs a shared observability stack for three product teams. The monthly invoice is $51,000: $18,000 for the committed retention and control-plane baseline, $30,000 for log ingest, and a $3,000 one-off migration charge requested by Team A. Ingest for the month was Team A 40 TB, Team B 35 TB, Team C 25 TB. The documented allocation rule splits the baseline evenly across the three teams, splits ingest cost in proportion to ingested bytes, and charges one-off items to the requesting team. What does Team C's allocation line show?
MCQ-9A monthly cloud bill of $480,000 reports 82% allocated. The unallocated $86,400 splits into $54,000 on resources still tagged `team=platform-eng`: a team dissolved last quarter, whose resources were reassigned to two successor teams that are in the CMDB, and $32,400 on untagged resources, $28,000 of which is one shared Kafka cluster consumed by four product teams. Which action does this breakdown justify first?
MCQ-19A platform team's cost model projects $48k/month for a service in the March revision and $30.7k/month in June. Finance asks which assumption changed and who changed it. The team currently emails copies named cost_model_v3_final.xlsx. Moving the model into git primarily gives them what?
MCQ-23A streaming service's monthly compute forecast has missed by +38%, +41% and +36% over three months. A central FinOps analyst builds it from a six-month trailing average of billed usage. Each month's overrun traces to the same team's batch concurrency, which that team raises in a release note the analyst never sees; the analyst learns of each change only when the invoice arrives. Which change would have caught the overrun before the invoice?
MCQ-6A payments service peaks at 4,000 QPS today. Over the last six months, month-over-month peak growth has ranged from 2% to 9%, and the team has no basis to narrow that range. Two capacity options are on the table for a 12-month horizon: Option A handles 12,000 QPS, Option B handles 8,000 QPS. Which statement about the two options is correct?
MCQ-16A demand-planning team's model forecasts 12,400 units for Q3. Backtesting that model over the last 20 quarters, the errors (actual minus forecast) have a 10th percentile of −1,900 units and a 90th percentile of +2,600 units; the mean absolute error is 1,000 units. The team's convention is to give the supply committee an 80% prediction interval. Which interval should it present?
MCQ-13A team is choosing between two designs for a checkout service. Both see the same routine incidents: 2 per year at an average $50k each. A correlated regional failure has a 1-in-10-year probability; it would cost $8M on the single-region design and $400k on the cell-isolated design, which costs $200k/year more to run. An average-only model that ignores the regional failure says the single-region design is $200k/year cheaper. Once the regional failure is costed in, which design is cheaper, and by how much per year?
MCQ-15A team's three-year cost model puts architecture A 8% below B on base-case inputs. The inputs (traffic growth, storage price, egress fees, retry rate) each carry a plausible range. Before committing, the lead asks for a sensitivity analysis. Which question should it answer?
MCQ-20A service's monthly compute forecast was $140,000 for 1,000,000 instance-hours: 600,000 hours of family A at $0.10/hour and 400,000 hours of family B at $0.20/hour. Actual usage was 1,100,000 instance-hours: 800,000 hours of family A and 300,000 hours of family B, and family B's rate had risen to $0.25/hour while family A's was unchanged, for $155,000 actual spend. Which decomposition of the $15,000 unfavorable variance is correct?
MCQ-25A platform team is choosing between self-managed Kubernetes on EC2 and a managed container service for a three-year commitment. The model gives self-managed infrastructure at $1.20M and 0.5 SRE FTE; managed infrastructure at $1.35M and 0.15 SRE FTE. The loaded SRE cost is estimated at $220k per engineer-year, with a plausible range of $180k to $260k. Which recommendation is decision-grade?