ERP
HEAD
TO
HEAD™
Compare 11 leading ERP products at 1 event
ERP HEAD TO HEAD™

4 ways Machine Learning could work with data from your ERP

Machine Learning (ML) is the subset of Artificial Intelligence (AI) which gives computers the ability to learn without being explicitly programmed. Whilst subject to inflated expectations, there are many practical applications of ML that you will be familiar with today – spam filters on your e-mail, product recommendations from Amazon and Netflix, Google prioritising search results and payment transactions being verified by your bank to name but a few.

In this blog, using a B2C setting as an example, I suggest four ways that ML algorithms could be applied to data that you would reasonably expect to find within your ERP.

Before continuing further, it is important to state that ML needs sufficient data to train the model – a rule of thumb being ten times as many examples than the number of independent variables in your model.  It’s also useful to know that some ML models also require the data to be normalised in advance of training.

Gartner[1] predicts that 70% of organisations will be using AI by 2021. These ideas could be a way of introducing AI into your organisation and allowing you to learn first hand how to get the best from this toolset. Alternatively, if you’re currently undertaking an ERP selection project, you might consider these as functional requirements. This would potentially allow you to use AI capabilities as a differentiator between ERP systems. 

 

1. Customer Segmentation

Customer segmentation allows the customer base to be partitioned into groups of individuals that have similar characteristics. This may then be used to personalise the user experience for each of the members of these segments.

Manually performing this task is resource intensive; ML provides a practical alternative. It is also an unsupervised (i.e. without bias) way of systematically doing this.

For those who are interested in the quantifiable benefits of segmentation, Mailchimp[2] has metrics based from around 11,000 segmented campaigns.

 

2. Customer Churn

Using historical customer data, ML methods make it possible to forecast which customers are at risk of “churn”. Once identified, and perhaps in conjunction with an estimate of customer lifetime value, it becomes possible for the business to decide how best to respond. 

A recent article by McKinsey & Company[3] suggested that using ML in this way could reduce churn by as much as 15% for telecom companies.

 

3. Recommendations

ML provides various ways in which recommendations can be made. A content-based system will try to show recommendations for products that share an attribute with those which are already being purchased. A collaborative filtering approach looks at what is popular with similar customers.

In either case, the user is provided with further products that may be of interest with the opportunity for upselling.

Youtube’s[4] implementation of their Recommendation System has been successful in terms of their stated goals, with recommendations accounting for around 60% of video clicks from the homepage.
 

4. Fraudulent Transactions

Rather than manually defining rules about which transactions should be held pending further investigation by a fraud detection agent, ML can provide an automated way to take historical data and predict in real time which orders are likely to be fraudulent.

Ocado[5] found that this approach improved their precision of detecting fraud by a factor of 15.
 

Deployment

How you make use of your trained model depends on whether your ERP provides ML functionality; even if it didn’t originally, it may have been included in an upgrade. See also Sean Jackson’s blog post on  "Getting More Value from your ERP System”.

If you don’t have access to ML from within the ERP, then it is possible to have a developer create a standalone solution (which could be deployed to a public cloud and accessed via an API) but this is much more challenging technically and in terms of business process integration.
 

Final Thoughts

This blog has used B2C examples to highlight some ways that ML could be used, but there are likely to be opportunities wherever there is abundant data.  These could include:

  • Vendor Payment History: Probability of Vendor breaching contractual terms
  • Manufacturing Execution Systems (MES): Predicting machine breakdown
  • Image capture: Image classification/recognition

The deployment of ML models, especially outside of the native ERP environment, remains challenging. However, given the forecasted growth of ML and AI, it is worth considering if these techniques could be successfully applied to your business.
 

This blog was written by Jim Goodison, Principal Consultant at Lumenia. For further information please send an email to Jim Goodison.

 

[1] https://gtnr.it/2DxYMat

[2] https://bit.ly/2Kthsd3

[3] https://mck.co/2yXp5kA

[4] https://youtu.be/nYMBHwf4owY

[5] https://bit.ly/2HxhD7t