Four questions, kept apart on purpose
A risk score is only useful if you know what it is actually saying. The common failure is to mash everything into one number — so a confirmed earthquake that finished yesterday looks the same as a wildfire heading your way, and a rock-solid government advisory looks the same as an anonymous social post.
ShadowIQ keeps four questions separate and scores each one on its own. Only then do they come together into the picture you see.
Consequence — how bad could it be?
The first question is the simplest to grasp: if this hazard reached one of your travellers, how bad could it realistically be? A transport strike is an inconvenience. Civil unrest with violence is not. A major natural disaster is in a different category again.
ShadowIQ places every event on a plain severity ladder — from negligible, through minor, moderate, and major, up to catastrophic — anchored to the kind of language government travel advisories already use. This keeps “how bad” consistent from one event to the next, instead of depending on which analyst happened to read it.
Likelihood — is it still dangerous?
This is the question people most often get wrong, so it is worth being precise. Likelihood is not“did this happen?” Almost everything we ingest genuinely happened. The question that matters for a traveller is: is there still a hazard ahead of me?
An earthquake that struck yesterday and left no ongoing danger is a historical fact, not a live risk. A wildfire that is still spreading is very much a live risk. Both are real events; only one should raise a traveller’s score today.
Worked example — all four events really happened
| The event | Still ahead of you? | Likelihood |
|---|---|---|
| Earthquake yesterday, no ongoing damage or aftershock warning | Over | Low |
| Protest that formed this morning and has since dispersed | Over | Low |
| Wildfire actively spreading toward the area | Ongoing | High |
| Civil unrest escalating night after night | Escalating | High |
Every row is confirmed. Likelihood asks what is still ahead of you, which is why ShadowIQ does not keep crying wolf over events that are genuinely finished.
To describe likelihood we use a defined vocabulary — from “almost no chance” through “roughly even chance” up to “almost certain” — rather than vague words that mean different things to different readers. That ladder comes straight from ICD 203, the analytic standard used by the intelligence community.
Exposure — are your people in its way?
A dangerous, live hazard on the other side of the world is not your problem. The same hazard where your group is standing very much is. Exposure is the question that connects a global feed of events to your specific trips.
It is important to be clear about what this is and is not. ShadowIQ does nottrack your travellers’ phones or watch their live location. It matches events to the itineraries your organisation has entered — the places your trips visit and the dates they are there. An event only counts as exposure if a trip is scheduled in the affected place, during the affected time window.
Worked example — same feed, different itineraries
| The event | Your itinerary | In its way? |
|---|---|---|
| Flooding in a city your group left last week | Right place, wrong time — already departed | No |
| Transport strike in Marseille | Group is in Paris this week | No |
| Protest in the same district during your stay | Present, and inside the time window | Yes |
Place and time both have to line up. A serious event your people have already left behind, or are nowhere near, should not light up their trip — and it doesn’t.
Confidence — how sure are we?
The last question is about the reporting, not the world. How much should you trust this particular claim? A government advisory and an anonymous social post are not equally reliable, and pretending otherwise would be dishonest.
ShadowIQ grades every signal with the NATO Admiralty system: the reliability of the source (from consistently reliable down to cannot-be-assessed) and how well the specific claim is corroborated by others. Crucially, confidence is kept separate from risk. A shaky report of something serious is routed to a human analyst to verify— it is not quietly inflated into a scarier score, and it is not quietly buried. That separation of “how likely” from “how sure” is a rule we borrow from ICD 203.
How it comes together into a trip level
Individual alerts are scored on the four questions above. A trip’s overall level is then set by the most serious hazard that is currently live and actually relevant to that itinerary — a high-water mark, not an average that quietly dilutes one severe threat among many mild ones.
And because likelihood eases as hazards resolve, a trip’s level comes back down on its own once the danger has genuinely passed. You are looking at the risk that is live now, not a pile of everything that was ever reported near your route.
AI and humans keep it honest
Watching the volume of open-source signal that matters is beyond any human team, so AI does the first pass — reading every source and scoring the four questions consistently, around the clock. But “AI-powered” is easy to say and hard to trust, so the reasonable question is: when the AI is wrong, what catches it?
The answer is people, by design. Every serious alert is reviewed by a human — the automation is not allowed to close those on its own — and a continuous, blind-sampled slice of ordinary alerts is put in front of analysts too, so the automation is always being spot-checked rather than simply trusted. Those expert decisions are the reference standard the scoring is tuned against, and the agreement between AI and human judgement is measured, not assumed. We go deeper on this in How AI and Humans Keep the Whole Thing Honest.
Built on standards, not invented from scratch
None of this is a proprietary black box. The bones of the model come from references your auditors, insurers, and security leads will already recognise:
The international travel risk management standard — likelihood × consequence, assessed for the individual traveller on a specific itinerary.
The intelligence community's analytic standard — a defined probability vocabulary, and a firm rule never to blend likelihood with confidence.
A decades-old method for grading intelligence — source reliability and claim corroboration, scored independently.
Standards are a floor, not a ceiling. The value we add is applying them at machine scale, continuously, to open-source signal — and calibrating that automation against human analyst judgement. If you want the reasoning behind each choice, our standards deep-dive and the full Inside the Risk Model series walk through the whole model, one element at a time.
Where our job ends
ShadowIQ is decision-support, and we are deliberate about its edges. It helps you see and understand risk so the right people in your organisation can act with clear, defensible evidence. It is not a live-location tracker, and it does not run the response — we do not coordinate evacuations, manage the incident, or make the duty-of-care decision on your behalf. Those calls stay where the accountability sits: with you.