DevOpsInterviewPrep logo
Platform Engineering & FinOps / 08
mediumNewStripeUberNetflix

Leadership wants proof cloud spend is efficient. Why report cost per transaction instead of percentage savings?

Percentage savings make every quarter look like a win and tell you almost nothing. The candidates who get hired bring a denominator tied to revenue and defend it under pressure.

Updated Sep 2026 · Grounded in researched DevOps, SRE and platform engineering interview loops, written to a senior-engineer editorial bar, and never padded to hit a word count.

TL;DR: Percentages hide scale, growth and direction. A unit cost such as cents per transaction ties spend to the business and answers the question leadership actually has: are we getting cheaper per unit of work as we grow? Report absolute dollars and per-unit trend together, and assess them alongside reliability, quality and the budget.

How to approach it

Concede immediately that percentages are not worthless, they are incomplete, then establish what any credible ratio needs: an honestly allocated numerator and a denominator leadership recognises as the business. Everything after that is mechanics.

A strong answer

Unit cost is allocated spend divided by business volume. Say 150 million checkout calls a month carry 45,000 dollars of allocated compute, database and network cost: about 0.03 cents per call. If compute is the entire cost and volume is unchanged, a 35% reduction yields 0.0195 cents per call. If compute is only 40% of allocated cost, the total reduction is 14% and the unit cost becomes 0.0258 cents. Then traffic doubles next quarter and the bill grows 60 percent. Percentage reporting screams blowout; unit cost fell 20 percent, which is the true statement: every transaction got cheaper while volume drove the total. That inversion is the entire argument.

Percentages mislead in three specific ways. Base effects: 30 percent of a two-million-dollar line and 30 percent of a two-hundred-thousand-dollar line make identical headlines and very different money, so effort drifts toward prestige items instead of dollar mass. Growth masking: a flat percentage alongside doubled traffic is a large real win, invisible in isolation. Baseline gaming: compare against an unusually wasteful month and any quarter looks virtuous. Unit cost helps expose these effects, but its denominator can also be gamed through retries, low-value work or rejected requests. Track successful business outcomes and reliability alongside it.

Allocation honesty decides whether the number survives scrutiny. Include a defensible share of shared platform, observability and network cost distributed by a measurable driver such as request share, and publish the method. A unit cost computed from a team's own compute alone is a flattering fiction, and a sharp interviewer finds it in one question.

The operational payoff is alerting. Raw-spend alerts fire on every legitimate seasonal peak; unit-cost alerts fire when something regressed relative to work done. A unit cost jumping 15 percent week over week at flat traffic is a real signal, usually a retry storm, a hot loop or configuration drift, caught days before the invoice lands.

LensAnswersFails on
Total spendBudget trackingSays nothing about efficiency
Percentage savedQuarterly theatreScale, growth, baseline games
Unit costCheaper per unit of business or notNeeds honest allocation

Reversal condition: very early products lack stable denominators, so track absolute burn and introduce unit metrics once volume is meaningful. And pick the denominator leadership uses: orders rather than requests if that is what the board sees, because a metric nobody acts on is decoration.

What interviewers probe next

"Shared costs make your unit cost noisy." Allocate by drivers, publish the methodology, accept roughly ten percent noise in exchange for honesty; direction beats precision here.

"Seasonality distorts the denominator." Compare against trailing-twelve-month bands or same period last year, never adjacent months.

"Which denominator do you choose?" The one closest to revenue that teams can influence; per-successful-transaction beats per-request, which rewards failing cheaply.

Common mistakes

Quoting one blended unit cost across products with wildly different economics, which averages away every insight.

Celebrating percentage cuts on line items too small to matter in absolute dollars.

Per-request accounting that counts failures, rewarding teams for rejecting work early instead of serving it well.

No published allocation method, so the number dies the first time a director asks what is included.

That one was free, and so are 10 answers per topic without an account. Signing in doubles that to 20, keeps your bookmarks, and tracks which topics you keep getting wrong.one Google click · no card · nothing to cancel
HOW DID IT GO?
0
UP NEXT ON YOUR JOURNEY
DISCUSSION · 0

Nothing here yet. Say how you would answer it.