One of the biggest challenges with any conduct surveillance or fraud detection system is to effectively manage false positives. In this article, we cover some essential techniques to dramatically reduce your surveillance false positives with better data processing.
What is a false positive in surveillance?
In short, false positives are mislabelled surveillance alerts that indicate there is suspicious behaviour, when actually, there isn’t. Regulators mandate that every alert generated by trade or communication surveillance systems must be reviewed. So these false alerts present a serious problem for financial organisations.
With teams spending so much time closing erroneous alerts, firms are becoming unstuck. Their automated surveillance systems are in effect adding to their workload, instead of optimising it.
So we know the culprit, but how do we solve the problem? We believe the first step is with better data processing techniques.
Why do false positives occur?
A false positive occurs when an alert is triggered by a rule that, in fact, should not meet that rule, or policy criteria. While this often occurs when a policy has been written too broadly, many false positives can be avoided with better data processing and classification techniques.
Watch out for our next article on the effective management of false positives coming soon.
How to reduce your surveillance false positive rate
You can make real impact on the number of alerts your surveillance system generates by getting right to your data. Clean and accurate data to feed your compliance system with is crucial. Here are four ways to better handle your voice and eComms data, thus preventing false positives from being created in the first place:
- Detect all non-relevant data
- Get control of your voice data
- Search regularly for language spoken
- Use Machine Learning to improve accuracy
1. Detect all non-relevant data
Firms produce on average 1 million hours of audio per year and tens of millions in electronic communications (emails, SMS and chats). However, up to 74% of that data is often classed as ‘non-relevant’ and does not require day-to-day surveillance.
This could be duplicate data from an employee with multiple audio feeds, a group email chain or standard disclaimer information and marketing newsletters. Accurately identifying this information before it arrives in your alert inbox is vital to avoid unwanted false positive alerts.
2. Get control of your voice data
A recurring problem for firms has been how to handle audio data for surveillance purposes. Historically, this has been due to the volume and complexity of transcribing voice into text. However, more and more firms are now trusting advanced voice surveillance systems.
The accuracy of speech-to-text technology has significantly improved over the last few years. Now, global organisations can not only ingest 100% of their communication data (voice and eComms) but also monitor employee conversations as they switch from channel to channel; something that can further aid conduct surveillance.
3. Search regularly for language spoken
When ingesting a communication, you should also regularly search for the language spoken. This technique is particularly useful for global teams where employees are speaking English as a second language or regularly communicating in more than one language.
Accurate language and dialect detection is key. This increases transcription accuracy for calls and helps the system alert you to any instance of language-switching within the same conversation (where an employee hides a message in another language to avoid detection).
4. Use Machine Learning to improve accuracy
When a surveillance system ingests communication information, it starts to identify and classify parts of each communication. This helps users perform searches and build policy criteria. For example, the system can tag a price within an email or understand if the employee is fixing a price or fixing their kitchen sink.Machine Learning understands the context of each communication. Then, it uses the context to correctly classify the information inside the email or call. In effect, the system learns what communication content relates to policy criteria, or not, to reduce the number of false positives.
Why False Positives must be addressed?
Although most false positives don’t pose an immediate threat to the business, they are a huge drain of time and resource. Any time spent investigating erroneous or non-malicious alerts is valuable resource that could be spent on actual cases of market abuse or misconduct.
You will never be able to eliminate false positives entirely, in fact it is important not to refine policies to the point where they cause false negatives. False negatives are far more threatening to an effective surveillance function as these represent hidden risks your compliance team may we unaware of.
However, the more you clean and process the data going into your surveillance systems, the more effective results you will receive out of them.