← FinOpsAid blog
Which of these savings are real? Reading the Savings page and its confidence labels
Potential annual savings, opportunity counts, the four categories and what high, medium and low confidence actually mean on the FinOpsAid Savings page.
Savings is a list of claims about money you could stop spending. Every claim carries a dollar figure and a confidence label, and the confidence label is the more important of the two — a $40,000 low-confidence estimate and a $4,000 high-confidence one are not the same kind of finding, and taking the first to a stakeholder before the second is how a cost programme loses credibility.
Conventions used below are in the grammar behind every number.
The KPI row
| Tile |
What it counts |
Direction |
| Potential Annual Savings |
Sum of every recommendation's estimated annual saving. |
Ambiguous — a big number means a big opportunity, which means a lot is currently being wasted. |
| Opportunities |
Total open recommendations. |
Same ambiguity. |
| High Confidence |
How many of them are labelled high confidence. |
This is the number to lead with. |
The ratio between tiles one and three is the page's real headline. A large total with almost no high-confidence items is a page full of modeling. A modest total that's mostly high confidence is a work plan.
Note that every figure is annualised. It's twelve times a monthly estimate, not a saving you bank this month.
Confidence, and what each level is actually claiming
Confidence is about evidence, not about size. A high-confidence finding names a thing you can point at; a low-confidence one describes a shape you'd have to reorganise around.
The four categories
Recommendations are grouped into cards, each headed by that category's total annual saving as a green chip.
| Category |
What it covers |
Typical confidence |
| License |
Idle and never-used seats. |
High — a seat has a name and a last-active date. |
| Compute |
Runners, Codespaces, machine sizing. |
Mixed — idle Codespaces are high; right-sizing is lower. |
| CI-efficiency |
Workflow triggers, caching, redundant runs, failure waste. |
Medium — the saving needs a change to land. |
| Governance |
Policy and access findings with a cost consequence. |
Varies. |
A category card with no items simply doesn't render, so an absent category means nothing was found there — not that it wasn't checked.
Each recommendation row
Four elements, and they're designed to be read in this order:
- Title — the action, stated as an action.
- Evidence — a one-line factual basis, in smaller grey text. This is the part to read. It's what turns "reclaim 34 seats" into "34 seats with no activity in 90+ days", which is the difference between an assertion and an argument.
- Estimated annual saving — with
/yr attached so it can't be misread as monthly.
- Confidence chip.
If you're going to quote one of these to anyone, quote the evidence line alongside the number. The first question you'll be asked is "how do you know", and the answer is already written there.
Turning the page into a plan
A workable sequence:
- Filter to high confidence. Those are the findings where the only open question is whether to act, not whether the number is right.
- Verify a sample. Click through two or three to the underlying page — Licenses for seats, Codespaces for environments. Confirming a couple by hand is what makes the rest credible.
- Send findings to owners, not to a channel. Seat and Codespaces findings have named owners. People clean up their own things.
- Set the policy that prevents recurrence. Reclaiming today's waste is an afternoon; a retention policy or a machine-type constraint is what stops it coming back next quarter.
- Re-check after the next sync. Recommendations are recomputed, so a finding that disappears was either fixed or is no longer supported by the data.
The categories that resist this treatment are the modeled ones, and the honest position is to present them as options with a stated method rather than as savings. That distinction is the whole argument in savings recommendations you can defend.
What people get wrong on this page
- Quoting Potential Annual Savings as a target. It's a ceiling assembled from estimates of varying quality.
- Reading the confidence chip's colour rather than its label. High confidence is good news.
- Treating annualised figures as monthly. Everything here is × 12.
- Skipping the evidence line. It's the only part that survives a challenge.
- Cleaning up without changing policy. The same list regenerates.
The 60-second read
- High Confidence count against total Opportunities — that ratio is the page's quality score.
- Category chips: where is the money, by kind of work?
- Read the evidence lines on the top three, not the titles.
- Pick the highest-confidence item with a named owner and act on it this week.
- Note which category keeps regenerating; that's where a policy is missing.
Frequently asked questions
What do high, medium and low confidence mean on a savings recommendation?
High confidence rests on an observed fact with a named owner, such as a specific seat idle for 180 days, so the only open question is whether to act. Medium rests on an observed pattern where the saving depends on a change landing. Low rests on a model, where both the size and the achievability are estimates.
Is the potential annual savings figure money I will actually save?
No. It is a ceiling assembled from estimates of varying quality, annualised by multiplying a monthly figure by twelve. The more useful number is the high-confidence count beside it — the ratio between the two is effectively the page's own quality score.
Which part of a recommendation should I quote to a stakeholder?
The evidence line, not the title. It states the factual basis — for example 34 seats with no activity in 90 or more days — and it is the part that answers the first question you will be asked, which is how you know.
Want a defensible savings list for your org? Connect your GitHub org — read-only and free during beta — or explore the demo first. Next: how to read the Security & Governance page.