In NZ, retailers lose over $2M per day from retail theft.
But retail crime is a global problem that affects all retailers and law enforcement.
The true cost of crime to society is staggering. In NZ, retailers lose over $2M per day from retail theft. But retail crime is a global problem that affects all retailers and law enforcement.
Eyedentify works with retailers and Police, connecting people with information to prevent crime. By providing a cloud-based software platform that simplifies the crime reporting process, Eyedentify aggregates information across all retailers, providing smart data to our customers to enable actionable intelligence and analytics to prevent crime.
Eyedentify is already operational and working with retailers and Police across New Zealand and is being trialled in Australia. Eyedentify works with major retailers including Countdown, Briscoes, and Rebel Sport in New Zealand, and is engaging in trials with Woolworths, Queensland Police, and the National Retail Association in Australia. Our customers have expressed a strong desire for the information/data collected to be turned into actionable intelligence.
While there have been some attempts in crime prediction previously, these have primarily been based on analysing Police data. In contrast, the Eyedentify dataset is mainly industry-driven – therefore we capture a huge amount of data that would not typically be included in the traditional Police dataset.
The problem faced by Eyedentify, retailers, and Police, is if this significant amount of previously unknown data is collected, how can it be turned into meaningful information and used to prevent crime?
In order to deliver this insight to our customers, we want to build an algorithm that will take a series of data points (including both Eyedentify data and some external data) as inputs and provide a real-time, customised list of the most likely offenders and their estimated likelihood of offending. Our proposed challenge is to apply mathematics and statistics to build such an algorithm, creating a real-time high risk offender rating. This would likely be based on a combination of weighted variables that factor to the unique characteristics of each of our user types, giving stores up to date information on which offenders are most likely to commit an offence in their store at that given time.
The outcome is embodied in the question: WHO IS MOST LIKELY TO OFFEND IN MY STORE NOW?
This question can be broken into three core components. Each of these components will have their own hypotheses that need to be tested by focusing on a variety of data points. These components are summarised in below:
- Who is most likely...
This component requires a focus on Offender behaviour, including their incident history, preferences for particular products, vehicle links, geographic areas of operation, and associations with others.
It is important to note that although we capture personal attributes of a person (e.g. name, gender, ethnicity), these attributes will not be used as inputs into the algorithm. The behaviour-based focus of the algorithm will ensure that the insights are not discriminatory in any way (reducing common racial profiling).
- In my store...
This component ensures that the list of most likely offenders is tailored to a particular store. Therefore the focus will be on building a store profile to understand both its susceptibility to retail crime, as well as its likelihood of being a target.
This will require analysing factors such as store incident history, location (e.g. proximity to other high-theft stores, proximity to motorway on-ramps, etc.), and product types.
This component embodies the real-time requirement for the algorithm. It involves analysing incident patterns based on time of day, day of the week, seasonal patterns and weather, and real-time information such as an offender’s vehicle being in the area (through use of Licence Plate Recognition).
Note that there is some overlap between the three components. For example, a ‘Hot Product’ can be linked to all three components – i.e. offenders with an apparent preference for those products, stores that sell that type of product, and seasonal trends for that product type.
How is our solution innovative or different?
While there have been some attempts in crime prediction previously, these have primarily been based on analysing Police data. In contrast, the Eyedentify dataset is mainly industry-driven – therefore we capture a huge amount of data that would not typically be included in the traditional Police dataset. In addition to the unique nature of our dataset, the industry focus of our platform represents an opportunity to increase collaboration across the community to help solve what is a community problem.