Understanding Impact Estimates & Confidence
How to read the rand figures, the P10–P90 ranges, the confidence percentage, and the T+30 verification loop — so you trust the right numbers and challenge the rest. About 7 minutes.
Every decision in KeyOne carries an impact estimate — a rand figure that says, roughly, how much this is worth. Estimates are how the queue ranks itself and how you decide where to spend your attention. This guide shows you how to read them like an analyst.
The three numbers on every estimate
Open any work item (see Working a Decision) and the Impact Estimate card shows:
- Estimated upside (green) — the value you stand to gain by acting. Used for decisions and tasks.
- Estimated exposure (red) — the value at risk if you don’t act. Used for risks.
- Confidence (yellow) — a percentage saying how sure the estimate is.
A single item usually leads with one of upside or exposure, depending on its type.
Reading the P10 / point / P90 range
When a range is available, you’ll see three figures — P10, a central point, and P90:
- P10 — the low end. There’s roughly a 10% chance the true value is below this.
- Point — the central, best-single-guess estimate.
- P90 — the high end. There’s roughly a 10% chance the true value is above this.
DiagramRead an estimate as a range, not a single number
So about 80% of the likely outcomes sit between P10 and P90. The width of that band is itself information:
Worked example (illustrative). Two OSA risks both have a point estimate of R 40,000 / week.
- Risk A spans P10 R 35k → P90 R 46k — a tight band. You can plan around the R 40k with confidence.
- Risk B spans P10 R 8k → P90 R 120k — a wide band. The R 40k is a guess in a large range; treat it as “could be small, could be big,” and weigh that uncertainty in how much effort you commit.
Same headline number, very different decisions. Figures are illustrative.
Tip: when two items have similar point estimates, prefer the one with the tighter range and higher confidence — you’re more sure the value is really there.
What “confidence” actually means
The confidence percentage reflects how much KeyOne trusts this particular estimate, based on how much comparable history it had to learn from. A high confidence means the estimate rests on many similar past events; a low confidence means it’s a sparse-data extrapolation.
Low confidence does not mean “ignore this item” — a low-confidence, high-value risk can still be worth investigating. It means “this number is softer; verify before betting big on it.”
“Why this estimate?”
Don’t take the number on faith — open the “Why this estimate?” expander on the item. It shows:
- The methodology — how the figure was calculated.
- Any fallback reasoning — if the primary method lacked data and a simpler one was used, it says so.
- Calibration context — the tier, the number of comparable past events (n_events), and the model’s typical error (e.g. its MAPE, the average percentage it’s off by).
This is your audit trail. If an estimate looks too good or too scary, this is where you find out why.
Estimates are snapshots — and they get checked
Two things to keep front of mind:
- An estimate is a snapshot taken at detection. It reflects what was known when the issue was found. Conditions move; the number doesn’t silently re-write history.
- A T+30 verification loop checks it. Roughly 30 days after action, KeyOne compares the estimate against what actually happened. This is how calibration improves over time — yesterday’s misses sharpen tomorrow’s estimates.
These are estimates, not guarantees. KeyOne is explicit about this on the card itself: impact figures are informed estimates with a verification loop, not promised outcomes. Use them to prioritise, not to forecast the books.
How to use estimates in practice
A sensible reading order for any item:
- Glance at the headline (upside or exposure) to size it.
- Check the range width — tight means plan on it, wide means hedge.
- Check confidence — high means trust it, low means verify.
- Open “Why this estimate?” if the stakes justify a closer look.
- Act, then let the T+30 loop tell you whether the estimate held.
Common pitfalls
- Chasing the biggest point estimate blindly. A huge point with a huge range and low confidence may be worth less, in expectation, than a solid mid-size one.
- Reading the estimate as a promise. It’s a prioritisation tool. The T+30 check exists precisely because estimates aren’t certainties.
- Ignoring low-confidence risks. Low confidence is a flag to verify, not a reason to dismiss.