How Bad Could It Be? Inside ShadowIQ's Consequence Scoring
Part 1 of Inside the Risk Model: how ShadowIQ grades the realistic worst outcome of an event — and why the AI is never allowed to be calmer than the official advisory.
A transport strike and a terror attack are both "security events." If your risk tool scores them on one dial, that dial is meaningless.
This is the first post in our Inside the Risk Model series, walking through the four questions ShadowIQ scores for every alert: how bad (consequence), is it still live (likelihood), are your people in its way (exposure), and how solid is the reporting (confidence). Today: consequence.
One question, answered narrowly
Consequence answers exactly one thing: if a traveller were caught in this event, how bad could the outcome realistically be?
Not how likely it is. Not how close your people are. Not how credible the report is. Those are separate questions with separate scores — and keeping them separate is the whole trick. A tool that lets "this is scary" bleed into "this is likely" or "this is nearby" manufactures false alarms, and false alarms are how risk tools lose their audience.
The five-step ladder
ShadowIQ grades consequence on a five-band ladder:
- Negligible — awareness only. No meaningful impact on a traveller.
- Minor — small and reversible. Minor delays, a peaceful protest, petty-crime background.
- Moderate — material disruption worth a precaution. A strike, closures, a serious weather warning.
- Major — serious personal harm is plausible. Violent unrest, a significant earthquake, a disease outbreak.
- Catastrophic — life-threatening. Mass-casualty violence, a tsunami, anything where being present can be fatal.
Five bands, not a 0–100 score. Bands force an honest claim ("this is the serious harm category") instead of false precision ("this is a 73"). They're also how travel risk is actually discussed by security teams, insurers, and the standards bodies — so the output maps onto frameworks your organisation already uses.
Anchored to the official picture
Here's the part we think matters most: the AI is not allowed to quietly out-optimist the official advisory.
Every consequence grade is checked against the government travel advisory picture for the destination. If the official advisory for a region reflects a serious environment, our scoring can't drift below what that supports. If the advisory picture is calm, a routine event can't masquerade as a crisis either.
Why anchor to advisories? Because they're the reference point your duty of care will be judged against. If something goes wrong, the first question in any review is "what did the government advisory say?" A risk score that contradicts the advisory without a very good reason isn't insight — it's liability.
Graded by AI, checked by people
The first consequence grade on every alert is applied by AI, at a speed and scale no analyst team could match. But it doesn't end there: every alert is re-reviewed by a second, independent AI, and every high-severity alert goes to a human analyst — the automation is not allowed to make the serious calls alone. When analysts correct a consequence grade, that correction feeds back into calibration, so the grading improves against real human judgement rather than drifting.
That loop — AI proposes, humans keep it honest — runs across the whole model, and we'll cover it properly later in this series.
Why it matters to you
When you see Major on a ShadowIQ alert, it means one specific thing: a person caught in this event faces plausible serious harm — as graded by a model that is anchored to official advisories and audited by humans. It does not mean "probably going to happen" and it does not mean "your group is in it." Those are the next two posts.
Next in the series: The Event Happened. Is It Still Dangerous? — why "confirmed" and "high risk" are not the same thing.