How can you anticipate…violent crime

Crime is one of the most studied topics in social science. Entire academic fields have grown around it, producing decades of research on its causes and correlations. We know more about what drives violence between individuals than we do about why people commit terrorist attacks, take part in violent protests, or overthrow governments.

Yet that depth of knowledge is both a strength and a challenge when it comes to anticipating violent crime. The research base is vast, but even with the power of AI to analyse it, modelling violent crime in a way that is genuinely useful for decision-makers means knowing where to draw the line. We can’t build a model that captures every possible factor; doing so would create confusion rather than clarity, and add little real value for our customers.

For a platform like ours, the goal isn’t to track or predict every aspect of criminal behaviour. It’s to identify where the environment for violent crime is changing, and the conditions under which robbery, assault, or homicide become more likely. That means being selective about what we include, grounding it in evidence, and focusing on the forms of violence that matter most to organisations and their people.

What we mean by violent crime

When we talk about anticipating violent crime, we’re not talking about every possible act of violence. Our focus is on the types that directly impact an organisation, its people, or its property. We break that down into three buckets:

1. Violent crime targeted against staff (whether locals or travellers).

2. Violent crime targeting assets or operations.

3. Collateral exposure to group violence.

Is it exhaustive? Probably not, but it provides a solid foundation for modelling the the threats that matter most to the private sector.

Why violent crime is hard to anticipate

Even with a clear scope, violent crime presents unique challenges. It’s influenced by a huge number of factors, many of which are individual and beyond the reach of a platform like ours, and, frankly, probably in breach of Article 5 of the EU AI Act:

“It is prohibited to place on the market, put into service, or use an AI system to make risk assessments of natural persons in order to assess or predict the risk of a natural person committing a criminal offence, based solely on profiling or personality traits.”

In simple terms: we’re not aiming to predict who will commit a crime. 

Another challenge is that violent crime is also hyperlocal. While future versions of our platform will extend from the strategic and operational to the tactical, we’re not there yet.

So when we talk about anticipating violent crime, our goal isn’t to predict individuals or incidents, but to help organisations understand when and where the environment for violent crime is changing. So that intelligence and security teams can take the steps needed to keep their people and assets safe.

How can we do that?

OK, with the caveat that we’re not trying to recreate Minority Report and wire up some precogs to predict the future, let’s look at how we can identify where violent crime is more likely to happen.

There are some key factors we monitor. Many are intuitive, but we still ensure they’re backed up by research before including them:

Crime data: While we usually push back against the idea of simply mapping where bad things have happened and assuming they’ll happen again, the evidence in crime research shows that this approach is a pretty good one. Crime tends to cluster. But we can do better. Past data tells us where incidents occurred, not necessarily where they’ll happen next. Crime moves when its drivers shift: policing increases, lighting improves, or targets become harder to find.

Income inequality: High levels of social inequality are a robust predictor of violent crime, and increases in inequality are associated with increases in homicide and robbery.

Unemployment: violent crimes like robberies and carjackings increase during periods of high unemployment.

Urbanisation: an increase in the degree of urbanisation is associated with a rise in robberies.

Corruption and weak governance: high levels of corruption and weak democratic institutions create an environment where crime and illicit markets can flourish.

Deterrence: the strength of the police and judicial systems matters, and as a result, how people view the chances of being caught if they commit a crime. The evidence about the effectiveness of capital punishment on reducing homicide rates is inconclusive, with little to show whether it increases, decreases, or has no effect relative to long prison sentences

Environmental: the relationship between crime and weather is a field in it’s own right, but to simplify; homicides and other crimes tend to increase as temperatures do. Unsurprisingly, even bad actors don’t like to work while it’s raining, and so rain and wind tend to reduce rates of street robberies.

If you’re thinking, “How can I possibly track all of this? I have a day job and I’m not a data nerd,” that’s exactly why we exist. We harness the power of this research, data and statistical models, to give you their essential insights, without having to spend your days tangled in spreadsheets.

Short-term triggers and how they differ from unrest

We also track immediate or situational triggers that can cause a short-term surge in violent crime. But unlike coups or unrest, violent crime rarely hinges on a single spark. The link between an event and a rise in incidents is more diffuse, shaped by deeper social and economic pressures.

Still, certain conditions are worth monitoring: extended power cuts, national celebrations or festivals (crowds, alcohol, easy opportunities), and heatwaves. These don’t cause violent crime, but they can create short-term openings in already high-risk environments.

So in our platform, you’ll see fewer day-to-day fluctuations in violent crime likelihood scores, but you can be confident that we’re tracking how the broader environment for violence is evolving. That gives you the foresight to adjust your risk mitigation measures accordingly.

Finding the level that adds value

Violent crime has been uniquely challenging to model and track. Part of the challenge has been deciding what is both useful to customers and possible to model responsibly. Our current focus is not predicting incidents at street level but identifying the level of analysis that is genuinely valuable and empirically sound.

Is it helpful to know that crime in a country is “high” or “low”? Maybe if you just need to tick a box somewhere, but it doesn’t really tell you much you can actually use. We believe the balance we’ve struck - breaking violent crime down into some key components (targeted against staff, against assets or operations, and collateral exposure to group violence) and allowing users to track key scenarios within those categories, such as armed robbery at a company site, armed robbery of a company vehicle, or kidnapping of an employee - is the most practical and meaningful approach.