LEARN. PRACTICE. GROW.Retain the agreed population
The fictional contract includes every eligible request, even when rejected at admission. There are 10500 eligible requests, 10000 admitted, and 9500 correct outcomes. The contractual ratio is 9500/10500, approximately 90.48%; using admitted requests alone gives 95%. Neither denominator is universal: the exercise explicitly defines which applies. Do not change population to turn an indicator green after introducing load limits. Also measure rejections and user impact. Admission control can protect an overloaded dependency while still representing failure against the agreed objective. That consequence must remain visible in the decision and report.
Update forecasts when capacity changes
The new model starts with 600 items, receives 20 per minute, and processes 80 per minute for five minutes. Three hundred remain. Over the next five minutes, a dependency limits service to 20 per minute; the queue stays flat because arrivals continue. When service returns to 80 per minute, another five are needed to drain. Adding five for validation produces twenty minutes to the final milestone, beyond the eighteen-minute window. The model measured no actual infrastructure. Use the sequence to reassess assumptions during recovery, communicate a new forecast, and evaluate alternatives. Doubling workers does not guarantee more throughput when another dependency limits service.
Count work automation does not remove
A fictional task occurs twelve times daily and takes four minutes, totaling 48. The proposal retains ten review minutes and three four-minute exceptions: 22 remain and estimated saving is 26 daily. With 780 build minutes and no other model costs, simple payback occurs after thirty comparable usage days. No automation was implemented and no actual saving observed. Different maintenance, frequency, or exceptions can change the calculation. Measure before and after using the same unit, including rework and quality. More closed tickets or more code lines do not by themselves demonstrate lower operational effort.
Compare cost per correct outcome
Variant A costs 300 in cloud and 16 operating hours at 40 per hour, giving a fictional total of 940. It produces 9400 correct outcomes at unit cost 0.10. Variant B costs 400 in cloud and eight hours at 40, totaling 720 for 8000 correct outcomes, or 0.09 per outcome. Within this comparable scope, the ratio improves despite the higher cloud bill. The calculation does not decide migration alone: check required volume, deadline, quality, resilience, and risk. These figures are fictional and do not represent vendor prices. They are original figures for practicing the link between cost components and useful service outcomes.
Withdraw resources without losing infrequent cycles
FinOps finds a component with no traffic for seven days. Inventory still identifies a month-end consumer running every thirty days. The model does not authorize withdrawal: observation does not cover the cycle and owner validation is missing. Do not delete the consumer from inventory to align documentation with recent silence. Confirm the dependency, replacement destination, data, and reconciliation of the remaining cycle. A daily test may not exercise monthly rules. Define owners, withdrawal sequence, recovery conditions, and completion evidence. Planned saving should be presented as realized only when the resource and corresponding costs have actually been addressed within agreed scope.
Turn analysis into a revised commitment
Conclude with a note enabling an accountable decision: objective, assumption, evidence, recommended option, risk, and review condition. If forecast changes from fifteen to twenty minutes, communicate the change before the window ends and explain what caused it. If automation promises 26 daily minutes, confirm saving and whether recovered time enables useful project work blocks before changing commitments. Preserve limits and stop criteria in automated actions; a write timeout can leave an unknown outcome. Finally, distinguish the eleven executed local groups from actual-system practice and human review, which remain outstanding in this course.
manual_minutes = 12 * 4 # 48/day
residual_minutes = 10 + 3 * 4 # 22/day
net_saving = manual_minutes - residual_minutes # 26/day
build_minutes = 780
simple_payback_days = build_minutes / net_saving # 30
# Fictional estimates; no automation deployed or staffing saving established.
old_unit_cost = (300 + 16 * 40) / 9400 # 0.10
new_unit_cost = (400 + 8 * 40) / 8000 # 0.09Cloud spend rises from 300 to 400 while included operating cost falls. The decision compares full scope per correct outcome and retains capacity and deadline requirements.
Common pitfalls
Excluding rejected requests despite the contract, projecting initial capacity across recovery, or presenting all manual time as automatic saving.
Related topics: Observability · Problem Management · IT Project Management
An improvement must benefit the service within defined scope. Measure correct outcomes, residual effort, dependencies, and risk across the cycle.
Reference: Eliminating Toil · BigSavant APS professional curriculum 2026-09; vendor-neutral operational guidance reviewed 2026-09-30