the cognistx blog

AI in Supply Chain: Anomaly Detection

September 3, 2019

Supply chain and logistics companies have a high velocity and volume of data. It is not possible to track data quality with human personnel. This makes automation essential.

A typical example

Anomaly detection is all about automating data sets that have high velocity and volume to make better decisions. There is no industry that needs accurate decisions more than the supply chain industry. In the UK alone, supply chain inefficiencies cost $2 billion with more than 100 million hours wasted in procurement alone. Machine learning is just one of the ways that companies in the supply chain industry are confronting these challenges. In an OpexAnalytics case study, a CPG company that used anomaly detection for demand planning saw a better forecast than their current solution 75% of the time. Using an anomaly detection platform for demand planning can help suppliers to ship goods with more efficiency.

Armada sse case

Armada is also a 3PL company located in Pittsburgh, PA. The company distributes goods to restaurants, which includes ordering and delivering goods that are needed. Cognistx met Armada in 2017, and a proof of concept was established in the Summer of 2018.

What we found (Business Problems)  

Armada had an abundance of data: over one million records retrieved per day. Armada used basic statistical methods and visualizations to understand present day activity of their data compared to historical data, using an in-house tool called 24track. Armada wanted Cognistx to be able to find anomalies through the use of AI and rank them from high-risk to low-risk, in terms of damage to operations. They also wanted a way to validate the statistically atypical data points by sending an automated anomaly email to internal personnel as needed. These requirements were achieved in a three-phase process.

What we did

Phase 1: Exploratory Analysis

Cognistx retrieved the data and performed an exploratory statistical analysis. The team had to assess the rules that would be used for analysis, what information was needed from the client (purchase orders, pricing, etc.), and which software would be implemented to visually display the clients’ issues. The team focused especially on data elements such as ship/lead time variations, orders per day, unit of measure variation (product specific), seasonality on metrics, and more.  

Phase 2: Product Conceptualization

The Cognistx team set out to determine how the data analysis to detect anomalies could be productized. The team needed to determine what software needed to be developed to support the ongoing process.

The Outcome

Armada receives over one million records per day. Because of the abundance of data retrieved, the management to process all of these records locally, would be intangible.

The final decision was to move all data onto a cloud-based system, that can be easily accessible by the Armada and Cognistx team. Cognistx’s delivery method of choice was a dashboard. The dashboard included top 10/50/100 etc. anomalies happening each day/week/month. This was achieved by looking at the individual warehouses and comparing the current activity to historical distribution. Besides the rules given by Armada, additional rules were used to find anomalies in real-time. Another feature to the dashboard was the ability to capture feedback (based on a  supervised learning model) and determine which data is considered “good” and “bad”.  

KPI: How Success was Achieved

This is the first time a system like this has been implemented for Armada. Through the use of large-scale data integrity monitoring, users can begin to use the product and look to see how it improves their operations. The system automates the majority of the tasks for them, now allowing the focus to be on processing the system in an efficient manner, reducing the number of bad records detected. The KPI can effectively increase productivity.

Interested in more practical applications of AI in supply chain? Contact us today.

Recent Posts