The world has been running on the same model for decades. A company builds a team of analysts — smart people, sitting at desks, reading news feeds and government alerts and social media, and logging incidents one by one. It works. The problem is the maths. Twenty analysts have to cover 168 hours a week across 195 countries. So they don't. They cover the countries clients are paying for, during business hours, in the time zones that matter to the paying customers. On a Saturday night, there are one or two people watching the planet. That's not a staffing problem. It's a structural one — and hiring more people doesn't fix it, it just doubles the bill.

So the market tried a shortcut. Build platforms that pull in data from dozens of sources, run it through AI, and publish it automatically. Fast, cheap, scalable. Except the data isn't original — it's re-packaged from feeds the platform licences from other providers. The AI can't distinguish a genuine report from a recycled video that's three years old and thousands of miles from the claimed location. The platforms that do this well spend tens of millions a year on feed licences and need their AI models retrained every single day to keep up. The ones cutting corners on that are quietly publishing misinformation as intelligence.

The next attempt was more industrious. Build pipelines that hunt raw government data — 911 calls, police logs, city clerk PDFs, API endpoints buried in municipal websites. Standardise it. Then use AI to model down to street level, inferring what you can't source from demographic and geographic proxies.

It works — in cities where governments publish reliable, consistent data. But in Venezuela, the government stopped releasing official crime statistics years ago. In Niamey, there is no 911 system. And here's the problem no model can solve: every city is unique. Comparing Cape Town to Paris with the same algorithm is like reading a map drawn by someone who's never left the office. The coverage bar looks full. The confidence is borrowed.

The crowd model starts from a different premise. Instead of watching the world from a desk, you put people in it. Local reporters — residents, not tourists — who file from where events actually happen. No shift pattern. No collection plan based on client spend. No recycled content laundered through an AI. The event happens, the person who lives there files it, a human verifier checks it, and it's published. That's it.

In an age of misinformation, only a real human who was there can tell you what actually happened. Crowd Survey adds another layer: the same reporters answer weekly questions about security, sentiment, and economic conditions. That's how we capture the pre-escalation signal — the deteriorating mood, the rising tension, the changed behaviour — before any incident has occurred. You have to be there first.