Design Capacity Planning

Problem Statement: Enterprise Capacity Planning

Design a capacity planning platform that turns noisy utilization history, launch calendars, service ownership, and cloud pricing into approved capacity decisions. The key interview move is to show that the platform is not a prettier spreadsheet: it is a control system that protects SLOs while keeping spend accountable.

Aha Moment

Capacity planning answers a different question from autoscaling. Autoscaling asks what to do in the next few minutes. Capacity planning asks whether the organization has enough regional inventory, commitments, quotas, and procurement lead time for the next quarter. If you say this early, the rest of the design has a clear spine.

Whiteboard Note

Draw the loop as metrics -> forecast -> simulation -> approved plan -> procurement/export -> actuals review. Put SRE, service owner, and FinOps on the diagram so the interviewer sees both engineering correctness and business accountability.

Mistakes to Avoid

• Do not average global usage and call it capacity; regional peak and AZ-loss math drive the purchase.
• Do not treat headroom as waste; for tier-0 services it is part of the availability design.
• Do not let approved plans depend on live metrics that can drift after signoff; store assumptions and versions.

Interviewer-Love Note

A strong answer connects three numbers: forecast error, headroom policy, and dollar impact. For example, if checkout forecast MAPE is 9%, a 25-35% buffer is defensible; if MAPE jumps to 20%, the platform should flag the plan as risky before FinOps approves spend.

How to open this one

The framing that signals depth on capacity planning is demand forecasting against headroom and lead time: you project growth, compare it to current capacity and the time it takes to provision more, and pre-position resources before the wall, not after. Lead with the distinction between organic growth and spike events (a launch, a sale), and the failure story that proves it: a forecast misses a step-change in demand and you hit a quota or a hardware lead-time wall mid-incident. That shows you understand capacity planning is about buying time ahead of demand, not reacting to saturation.