Understanding the uniqueness of Engine B’s anomaly detection technology
Anomaly detection has been present in some aspects of audit for several years, but it is not often in a way which allows it to be used within a high-quality methodology. Engine B’s technology is now addressing this by providing auditors with a context-rich view of anomalies using machine learning and Knowledge Graph technology.
This article explores how the key features of Engine B’s anomaly detection software are helping audit firms to gain a laser-like focus on fraud risk.
A new approach to anomaly detection
For auditors, being able to identify hidden anomalies in data and make sense of these is extremely important, whether it’s for fraud detection, to detect data abnormalities or intrusion detection.
Engine B is taking a new approach to anomaly detection combining two key features – a ‘smart’, Knowledge Graph based approach and building in a commitment to explainability from the start.
Leveraging intelligent Knowledge Graph technology
The ‘smart’, Knowledge Graph based approach is the use of the agreement weighting formula and combining it with the recursive use of Knowledge Graphs.
Previous approaches have used agreement weighting before, but they have often used it simply to rank risks as low, medium, or high. Engine B’s approach is qualitative instead, asking what does it tell us that these specific algorithms agree or disagree, and how does this inform our view of risk or anomalous behaviour? By feeding this back repeatedly into the Knowledge Graph, the graph is enriched and the auditor can be presented with a fully contextualised view of each anomaly.
Explainability from the start
Explainability is also key to Engine B’s approach. Instead of presenting just an output, instead we make it clear what algorithm(s) is responsible for identifying an anomaly, what features of the item have been flagged as anomalous, what the context of it is, and how it relates to other anomalies. One of the benefits of using Knowledge Graphs is that anomalies are highly visible: the links in the graph which make something anomalous are visible within the graph, letting the auditor see exactly what looks suspicious and how it relates to its context.
The combination of these gives the auditor a really rich view of unusual behaviour within their audited entity and the context of that behaviour, in a way which makes it easy to see what’s relevant and to then apply judgement about risk.
Anomaly detection for every stage of the audit
Anomaly detection tools can be used at every stage of the audit, from identifying patterns in company behaviour using only external data as part of the pre-engagement due diligence process all the way to identifying transactions which breach internal controls in highly suspicious ways.
With the vast reams of data now available within and around organisations, finding what to focus on has become like looking for a needle in an ever-expanding haystack. As auditors we know what kind of market trends and management incentives create a risk of fraud, but it’s much harder to know where to look for specific vulnerabilities or evidence.
Machine learning and Knowledge Graphs for anomaly detection
Using machine learning and Knowledge Graphs for anomaly detection makes this easier again. Normal business processes and transactional behaviour recede, letting you know that light-touch substantive testing will be sufficient. What comes to the front – literally with Engine B’s Knowledge Graph tools – is the relationships, behaviour patterns, and data points which are not normal, either for the audited entity or for its market peers.This saves you time and effort in the audit, making it much easier to adopt a robust ISA 315 processes, clearly documenting your risks and planned responses, focussing your controls and substantive testing where it’s most needed. It makes it easier to make judgements on risk, because all of the evidence is synthesised and presented together with explanations for what’s normal and what’s not. And it means you’re much less likely to be caught out by a fraud or error hidden deep in the detail of millions of complex transactions that look to the human eye like normal business behaviour.
Need to know more? Download our guide for a technical deep dive into Engine B’s anomaly detection technology.