A small invented parts desk makes this slippery topic much easier to hold in your head. Picture a small team that runs the online catalogue for an aircraft parts reseller. They are not cloud experts. They have a storefront, a search service and a nightly stock sync, all renting servers by the hour. One quarter, the owner asks a fair question: are we paying too much for those servers? The tale below is the kind that FinOps, the habit of managing fluctuating technology spend as a group, tends to produce. Every name and number in the tale is illustrative, apart from the worked figures I borrow from a published exercise and credit below.
Chapter one: the tempting discount
The team lead, whom we will leave unnamed, finds that the hosting provider offers a lower hourly rate if you promise to keep paying for a fixed amount of capacity for a year. The brochure says savings of around a third are possible. The lead’s first thought is the one most of us would have: look at last month’s busiest hour, promise that much, and enjoy the discount.
This is where a good reference earns its keep. The Ops Atlas page on FinOps includes a small worked exercise that shows why the busiest hour is the wrong anchor. In its model, demand sits at 100 units an hour while the office is open and falls to 40 units outside those hours; committed capacity earns a 30 percent discount. Anyone who has run a parts counter knows the shape: crowded in the working day, nearly silent overnight.
Chapter two: the numbers that surprised everyone
According to the exercise, the sweet spot is a promise of 28 units per hour. That choice gives an effective savings rate of 21.4% while covering roughly 71.3% of consumption. Push the promise up to 42 units per hour and coverage rises to 80.9 percent, which looks wonderful on a slide, yet the savings rate drops to 6.0 percent. At 56 units per hour the savings rate goes negative, minus 9.4 percent, meaning the team would pay more than if it had never signed anything.
The lead reads this twice. A bigger promise feels more serious and more thrifty, and the arithmetic says the opposite. The reason is simple once you see it. A commitment is billed every hour whether or not the work shows up, so capacity promised above the quiet overnight floor is paid for during hours when nobody is browsing the catalogue.
Chapter three: what the team actually did
The lead follows the playbook the Atlas outlines, in roughly this order.
- She finds the floor of steady use, the amount needed even at the quietest hour, and bases the promise on that.
- She subtracts anything planned to change, since the warehouse team intends to retire the old search service within the year.
- She prefers flexible plans over resource-specific ones where possible, so the discount survives a change of machine type.
- She buys in two steps rather than one, leaving room to learn.
- For the first weeks she checks how much of the promised capacity is really used each day.
Notice that none of these steps needs a special product. They need a spreadsheet, a calendar and a willingness to say “let’s wait a month”.
Chapter four: the mistake they nearly made
The team also considered shrinking their servers. That is a sensible idea, called rightsizing, but the order matters. If they had bought a resource-specific promise first and shrunk the machines afterwards, the promise would have kept charging for capacity that no longer existed. The Atlas lists this among its recognised traps. Rightsizing comes first, observed across a full business cycle with a clear rule to roll back if the storefront slows down, and the purchase follows.
Parts people know a similar trap. Ordering a batch of spares before you have finished the inspection findings means you may stock items the aircraft no longer needs. Our own guides on avoiding errors when buying control-system parts and on common slips with electrical components for aircraft carry the same lesson: buy after the facts are in, not before.
Chapter five: the quiet meeting
Three months later, the team runs a thirty-minute review every week. They look at spend per thousand catalogue searches, not the raw invoice. They glance at a short list of unexplained lines. Every month, they compare the forecast with what happened and write down why the two differed. The lead describes the habit as “boring in the best way”. The Atlas describes the same rhythm as daily checks for spikes, weekly team reviews, a monthly close and a quarterly look at commitments.
Perhaps the most telling moment came when the owner asked whether the discount plan “worked”. The lead did not answer with the invoice total, because the catalogue had grown and the total was higher. She answered with the unit cost per thousand searches, which was lower, and with the savings rate on the commitment, which was positive and close to the model.
One side effect pleased the warehouse manager most. Because the weekly review showed which line items belonged to which service, a forgotten staging copy of the catalogue was spotted and removed within a fortnight. Nobody had been negligent; the copy simply had no owner and no reminder attached, which is exactly what labelling and a regular look are meant to prevent.
What this fictional story is not
It is not a prediction that your numbers will match the exercise. The 28-unit sweet spot belongs to that model workload and says nothing about your own usage. It is also not an argument for always buying commitments. Teams with unstable workloads, or a migration in progress, may be wiser to wait. The whole point of FinOps is that people with different jobs look at the same data and make a decision together, rather than one person guessing alone.
The takeaway for a parts business
If you run an online catalogue, an inventory feed or a customer portal, start with the quietest hour of your week, not the loudest. Check the real savings rate, not only the coverage percentage. Rightsize before you commit. And keep a weekly habit of reading the numbers with the people who change them. Your servers will not complain, but your margins will notice. If your team is tiny, even a one-page note that records the floor you chose, the date you chose it and the reason will save a long argument when the renewal comes around and nobody remembers why the number looked right.

