Predictive Analytics and Machine Learning

Data-Driven Decision Making in Audit, Certification and Standards Management

Intact early recognized that—in the fields of audit, certification and standards management — vast amounts of data are collected but at the same time rarely used to drive decision making. While collecting data is relatively easy, drawing insights from it is a completely different story requiring not only excellent skills in data engineering and data science but also a deep understanding of the respective industry itself. A skill set you will hardly find in a single person.

But there is even more to it. In today’s audit, certification and standards management we are facing ever more complex and dynamic systems with an “increasing variety of factors and their multiple interdependencies” (Kleboth et al. 2016, p. 3). This makes it extremely challenging to fully understand the complexity of the systems we are dealing with—even for experts. We have to constantly monitor and adapt to the systems’ changes. Unfortunately, these obstacles render profound data-driven decision making inaccessible for many organizations.

A Framework for Risk-Based Auditing & Decision Making in Complex Systems

To change this and leverage an organization’s data to inform its decision making, Intact Systems initiated and funded a research project at the Wageningen University in the Netherlands. Since 2014, we developed a framework for risk-based auditing and informed decision making in complex systems. First research findings and a general outline of the framework were published in Elsevier’s Trends in Food Science and Technology in October 2016. We already wrote about this in a previous news article.

Using Available Data to Predict Audit Non-Conformities and Risks

Since then, the project made great strides. As a first really practical research output we developed a set of three prediction models with a starting accuracy of more than 80% to predict audit non-conformities. We are now able to classify the amount of expected non-conformities based on data, which is already available. This data can be used to classify the risk of a potential new certification client and supplier, respectively, or to assess the expected risk class for already certified customers and suppliers.

The three prediction models were developed and tested in a pilot project together with one of our customers that had been following a risk-based approach already. We thus had a large set of hard data readily available in ECERT — Intact’s solution for audit, certification and standards management. The results proved the success of both the risk-based approach and the prediction models.

A Rapid Method for Identifying Key Risk Factors – Expertise, Perception, Machine Learning

In addition to using ‘historic’ (hard) data to cope with the risks in complex systems, we can also include and work with our auditor’s expertise and perception to get even better results. The human brain is extremely good at recognizing and judging the context, which makes an auditor’s perception extremely valuable for improving how we deal with data. We thus used qualitative data from expert interviews, which we cross-validated with the ‘hard’ audit data, to narrow down key risk factors.

This approach is very effective yet time-consuming, which left us with the need for a more time-saving and practical way of drawing people’s knowledge. We thus developed a ‘rapid method’ based on machine learning, which uses game design elements and game principles to quickly draw and cross-validate knowledge from auditors (and other experts). Based on this, we will even be able to assist auditors when they are in the field. This is a story for a future article, however.

Reference Source:
Kleboth, J. A., et al., Risk-based integrity audits in the food chain – A framework for complex systems, Trends in Food Science & Technology (2016), http://dx.doi.org/10.1016/j.tifs.2016.07.010