Book Appointment

FinOps Questions Healthcare Software Teams Keep Asking

Healthcare Software

Teams that maintain software for orthopaedic and joint-care administration, think bookings, referrals letters, equipment inventories, often find the cloud invoice creeping onto every meeting agenda, and the same practical queries follow. What follows is a set of frequently heard FinOps questions with short, direct replies. None of it concerns clinical content; it is about how software spending is read and managed. Skim the headings and jump to whatever you are stuck on.

What does FinOps actually mean for a software team?

It means the people who build and run the software take part in decisions about what it costs, alongside finance and the people who own the product. The Ops Atlas page on FinOps describes it as a shared practice, not a department, and says the name comes from joining finance with development-and-operations culture. For a small team the plain version is: look at the numbers regularly, know who can change what, and write down the decision.

Is it just a way to lower the bill?

No. The goal is value for money, and at times the sensible move is to pay more, for example on a faster service that cuts the time staff wait for a screen to load. Judge the bill per unit of work: cost per booking or per referral processed is more telling than the total.

Where do we start if nothing is set up?

Switch on the detailed billing export for each cloud account first; one major provider can reconstruct only the present and preceding month, so late starters lose history. Next, decide who owns each application and make sure that owner is attached to the resources, using tags where the provider supports them. On AWS the Atlas notes that a cost allocation tag has to be activated before it shows in reports, with a delay of around a day at each stage. Then schedule a recurring weekly review. Only after that should you shop for extra tooling.

Do we need a special tool?

Not at first. Every large provider ships its own console for cost exploration, budgets and alerts. Among open-source helpers, the Atlas mentions Infracost, which shows an estimate of the cost impact of infrastructure changes during code review and also policy-as-code approaches that mark a resource first and act on it later. Treat them as aids. The Atlas is explicit that owning software does not amount to practising FinOps.

What is the difference between showback and chargeback?

Showback reports each team’s cost without moving money. Chargeback transfers the cost into the team’s own budget. Showback is always needed, while chargeback is a matter of company policy, and neither is the more advanced of the two. If your organisation has not agreed on allocation rules yet, stay with showback.

How do discounts for committing to usage work?

Suppliers offer lower prices if you commit to a level of usage for one or three years. The risk is committing too high. The Atlas suggests sizing to the lowest hourly level your eligible usage reaches, then deducting any planned change, and buying in several steps. Because the discount is applied hour by hour, an unused hour is lost, so daily averages mislead.

The Atlas supplies a worked lab: demand of 100 an hour during business hours, 40 an hour at other times, and a 30% discount. Committing to 28 hourly units came out ideal, with ESR at 21.4% and coverage at 71.3%. Push the commitment to 42 and coverage reaches 80.9%, a figure that looks fine in a slide, while ESR slides down to 6.0%.

Why not just chase a high coverage number?

Because coverage tells you how much usage is touched by a commitment, not whether you are better off. The lab above shows coverage rising while the real savings fell. Prefer the effective savings rate: the money you truly kept, measured against a pay-as-you-go price for identical consumption.

How frequently should the numbers be reviewed?

A layered rhythm is the Atlas suggestion. Check anomalies daily, meet each team weekly, run showback and forecasting monthly, revisit commitments and indicators quarterly, and settle budgets and renewals yearly. A month’s final invoice can take anywhere from three days to twelve to turn up, yet the daily feed is enough for steering.

What should we watch for with temporary or interruptible capacity?

Interruptible machines are cheaper because the provider can reclaim them. The Atlas notes AWS gives roughly a two-minute warning, and another provider’s equivalent is 120 seconds if you configure it. It also flags a quirk on a third cloud: the default eviction behaviour, deallocate, keeps the attached disks, and those continue to be billed. Read the eviction settings before relying on a saving.

Our containers look cheap on paper. Why is the bill still high?

Allocation models for Kubernetes often charge the larger of what a workload requested and what it used, and treat the remainder of the machine as idle. If requests are padded far above real usage, you pay for the padding. The OpenCost specification, per the Atlas, works this way.

The bill jumped over the weekend. What do we do?

Treat it like an incident with a short runbook. The Atlas includes an illustrative one for a weekend logging spike, and the broad shape is sensible for any surprise. First, confirm the jump is real and note which account and service it sits in. Second, find the owner and ask what changed: a new release, a looping job, a debug setting left on. Third, stop the bleeding if it is safe to do so, keeping service levels intact. Fourth, record what happened and what check would have caught it earlier. A daily anomaly glance exists so Monday morning is not the first time anyone notices.

Remember that built-in alerting can lag. The Atlas says one provider’s anomaly detection may be up to a day behind and needs ten days of history before it works, so do not treat it as a live alarm.

Which unit measures should a healthcare software team pick?

Choose one measure the business understands, such as cost per booking, and one that engineers can act on, such as cost per gigabyte stored. The Atlas separates these as business metrics and resource-efficiency metrics. Follow each across months for a single product, not across unrelated products, since a trend in one system is a fairer test than a league table.

How do we build a forecast people trust?

Begin with the history, then adjust it for things you know are coming: a new module going live, a campaign that will send traffic to the booking pages, a legacy system being switched off. A straight trend line ignores all of those. The Atlas describes a ladder, from forecasts that only extend history, to rolling trends that factor in commitments, and finally to models driven by business events. Moving one rung up each quarter is a realistic ambition. Keep a small holdback in the budget sized to how wrong your forecasts have been before, since no model is exact.

Who should attend the weekly review?

Keep it small: the application owner, one engineer who knows the system, and a finance contact who joins when there is something to explain. Fifteen minutes is enough if the chart is ready beforehand. End every session by naming one action, one owner and one date.

Where can we read more?

Within this site, the guide to cloud cost management through a FinOps training provider offers a beginner perspective, and a rethink of how IT operations are put together gives wider context.

Keep this ordering in your head above all else: make the data visible, give each cost an owner, review on a calendar, and only then buy tools or commitments. Teams that reverse that order usually end up with an expensive dashboard and the same arguments as before.