Forecasting Spend When Usage Is Spiky
A media company streams live events. Its AWS bill varies from $9,000 in quiet months to $31,000 during major event months, with three or four major events per year. Compute is the largest line, and most of it is on-demand instances that run only during events. A finance analyst proposes buying three-year, all-upfront commitments covering the peak capacity to get the deepest discount. The engineering team warns that event dates shift and that instance families may change as the platform evolves. The CFO wants predictable monthly costs and no large upfront cash outlay this fiscal year.
1. Evaluate the analyst's proposal against the stated constraints.
Read the worked answer
The proposal conflicts with all three constraints. A three-year all-upfront commitment covering peak capacity requires a large cash payment now, which the CFO explicitly rules out. It also assumes peak capacity is stable for three years, but event dates shift and the engineering team expects instance family changes, so the commitment may not match future usage. Committing to peak also means paying for that capacity in quiet months, when actual usage is a third of peak.
The deepest discount is not the right objective here; the right objective is a predictable bill that tracks spiky usage without stranding spend. A commitment sized to the steady baseline, if any, plus on-demand or flexible capacity for event bursts, fits the constraints better.
2. Design a purchasing mix that gives the CFO predictability without a large upfront payment.
Read the worked answer
First, establish the baseline: identify the compute that runs every month regardless of events, and cover only that with a one-year, no-upfront commitment, which lowers the rate without a cash outlay. Second, handle event bursts with on-demand or short-term flexible capacity, accepting a higher unit rate for the small number of hours involved. Third, where the platform supports it, use a savings mechanism that applies automatically to whichever instance family is running, reducing exposure to family changes.
Then build a monthly forecast range rather than a single number, showing baseline plus expected event months, and set budget alerts at the top of the range. This gives the CFO a predictable band, avoids upfront cash, and keeps flexibility for shifting event dates.
3. What monitoring should the company put in place to revisit this decision?
Read the worked answer
Track commitment utilization monthly: what percentage of purchased baseline capacity is actually consumed, and whether any is wasted. Track on-demand spend during events to see whether bursts are growing enough to justify a larger commitment next cycle. Monitor instance family drift so the team knows when a commitment no longer matches the platform. Review event calendar changes against forecast assumptions each quarter. If baseline usage proves stable for two consecutive quarters, consider a modest increase in committed coverage, still avoiding all-upfront terms given the CFO's preference.