Nine Steps to Build a FinOps Review Loop From Scratch

If you have ever set up hazard reporting in a safety management system, you already know the hard part is not the form. It is the loop: somebody spots something, someone owns it, a decision gets made, and the result is checked later. Technology spending needs the same loop, and the practice that supplies it is called FinOps. This how-to walks through building one in nine steps, written for a training organisation with a modest cloud footprint and no spare staff.

Where I lean on a reference, it is the Ops Atlas, whose long entry is the Ops Atlas page that documents FinOps practice. Read it later; you do not need it to begin.

Step 1: Decide what you are protecting

In safety work you start by naming what must never be compromised. For cost work the equivalent is service quality. The Atlas leaves no room for doubt here: a bill never justifies a slide in service levels. Write one sentence at the top of your plan, for example, “Student scheduling stays available during training hours; savings come second.” Every later step answers to it.

Step 2: Collect one month of billing data

Export the last month’s charges from each provider. Do not wait for the invoice, which may arrive days after month end; daily data is adequate for decisions, the Atlas says. If any provider only back-fills the current and previous month when you enable an export, switch it on today so you are not missing history later.

Step 3: Make a simple owner list

Create a table with three columns: what the charge is, which team or programme it serves and who answers questions about it. Where you cannot fill the owner cell, mark it orphaned. The orphans are your first safety reports. Some charges, such as taxes and support plans, will never carry a label, and that is normal. Agree a splitting rule, perhaps in proportion to each team’s labelled spend, and record it.

Step 4: Pick one unit and compute a cost per unit

Choose something the academy values, such as cost per enrolled student per month or cost per scheduled lesson. Divide. The number itself is less important than its trend over time, and the Atlas recommends comparing a product only against its own history, not against unlike products.

Step 5: Split every change into usage and price

When this month differs from last month, ask which of two things moved: how much was used, or what each unit cost. The Atlas calls the broader version a price-and-volume variance, adding a third label, scope, for items added or dropped. A short note per month, three lines long, is enough. This is your investigation habit; you would not close a safety report without asking what changed.

Step 6: Set a rhythm and book it

  • Daily, five minutes: look for any sudden jump.
  • Weekly, thirty minutes: each team looks at its own slice.
  • Monthly, one hour: statement of cost per team, a forecast and a variance note.
  • Quarterly: review any long-term purchase commitments and the measures you track.
  • Yearly: budget and renewal planning.

These cadences are the Atlas’s proposal rather than a law, so scale them down. A two-person team can fold the weekly and daily checks into one standing slot.

Step 7: Write a one-page runbook for the first surprise

Prepare for a weekend spike before it happens. The runbook can be five lines: confirm the spike is real; find which service caused it; apply the agreed guardrail, such as a cap on log volume; notify the owner; note the cause in the monthly variance record. The Atlas uses a logging spike over a weekend as an illustrative case, which is a good reminder that the usual culprit is mundane, not exotic.

Step 8: Add light automation, not heavy tooling

Two kinds of helper suit a small team. A policy engine can scan for untagged or idle resources and mark them first, then act after the owner has had time to respond; this mark-then-act approach respects the people involved. An estimator wired into code review can print the likely price tag of an infrastructure change before anyone approves it. The Atlas names both kinds, along with its caution that owning software is a different thing from running the practice. Start with the manual loop, then automate what is already working.

Step 9: Change things carefully, one at a time

When the first review turns up an oversized resource, treat it like an aircraft modification. Look at utilisation including the busy peaks. Agree the service guardrails beforehand. Adjust a single setting per pass, wait through one whole operating cycle, and reverse course the moment a service measure slips. If the resource is part of a purchase commitment, inform the person who bought it before you shrink anything, since a pledge tied to a particular machine type becomes dead weight if you shrink that machine afterwards.

A miniature example, with invented numbers

Suppose a small academy’s scheduling and records system cost 800 last month and 960 this month, and it handled 400 student lessons versus 600 before. Cost per lesson fell from 2.00 to 1.60 even though the bill rose by a fifth. The variance note would read: volume up, rate unchanged, scope unchanged. The owner list shows the records database as the biggest single line, assigned to the registrar’s technical lead. The decision at the monthly review: leave everything alone, and ask the lead to check storage growth before the next intake. Those figures are made up purely for illustration, and a real month will be messier, but the shape of the note is the point: three lines, one decision, one follow-up.

Who does what in a team of three

One person acts as the coordinator, keeps the calendar and writes the monthly note. One person, whoever runs the systems, takes the actions on usage. One person from the business side, perhaps the chief instructor or the administrator, sets the priorities and confirms that nothing important is being starved. The Atlas lists six core personas for larger organisations, but the logic compresses: someone who runs the process, someone who builds, someone who decides what matters.

How you know it is working

When one loop has turned, you should be able to answer, in a sentence apiece: what the unit cost was and which way it moved; how much of the bill has a named owner; what changed in usage and what changed in price; what you decided last time and whether it worked. The Atlas treats maturity as crawl, walk, run, assessed per capability. At the walk stage its sample goal is allocation coverage of at least 85 percent, and at the run stage more than 90 percent, but treat those as sample goals, not as an inspection standard.

Common stumbles

  1. Running it as a savings crusade. The Atlas calls the goal value, and says that sometimes means paying more.
  2. Leaving it to one person. Engineers make most usage changes; a lone coordinator cannot.
  3. Skipping the review because nothing seems wrong. A quiet month is a good time to test the routine.
  4. Forecasting from history only. Add known events such as a new intake or a software launch.

The nine steps fit on one sheet of paper. Print it, name an owner for each step and run the loop once this month. The first pass will be rough, and that is acceptable; a rough loop that runs beats a perfect one that never starts.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top