A model is a terrible thing to waste
How often do you use your streaming provider’s recommended movie selections? How often would you use them if you had to open a report in a separate browser…Yeah, that’s what I thought.
(For a TL;DR Video see https://youtu.be/AEzp7I3B9zE)
We all know the adage about trees falling in forests –but if a model makes a prediction and nobody uses it, does it make a difference? The answer is clearly no. This challenge is one of the biggest sources of failure and irritation for predictive analytics and machine learning. Building the best mathematical model in the world is a waste if it fails to yield meaningful results for the business or its predictions are not accessible by the people who need to make decisions. It may as well not exist at all. As the hype around big data, machine learning, and digital transformation reaches a fevered pitch the fundamental question remains: how can I drive meaningful value for my organization from these technology trends?
To address this, SAP has a focus on designing and delivering the intelligent ERP. What makes it intelligent is the use of technologies like machine learning to uncover critical patterns and actionable insights in the data, and then delivering these to end users directly via the ERP system where they must make decisions. Instead of expecting users to look at worklists, reports, and data visualization to identify the right action to take, the system itself becomes prescriptive. This is accomplished through a set of machine learning libraries sitting in the underlying SAP HANA platform that can analyze large volumes of data at scale and continuously adapt to new data and new patterns as they emerge.
For example: by looking at all the historical movements of materials in your ERP system, these models can detect patterns and predict the time it takes for specific materials to move between different locations and can identify scenarios where there is a risk that a critical material will not make it to its destination in time. If you are manufacturing large volumes of products, a single material being out of stock can significantly impact the production line and result in major bottlenecks and lost revenue. By identifying these potential issues in advance, organizations will be able to rapidly address or avoid these challenges.
The enhanced Stock in Transit screen, which makes use of statistical regression against historical data, is an example of this. By proactively alerting users of potential delays based on historical delivery times it empowers them to take action and prevent interruptions. This allows companies to be more agile in sending and receiving goods, and helps them avoid other stock related problems. (Learn more about machine learning in S/4 HANA at http://machinelearning.saps4.me/)
Sounds great, but what if you need to build and use a model that isn’t already built into your SAP ERP system? Enter SAP Predictive Analytics, Application Edition –your personal tailor for the intelligent ERP. Utilizing a special integration framework, users can take any data in their SAP S/4 HANA system and build predictive models against them. From there, the models can be pushed directly into SAP S/4 HANA where they will self-manage and self-adapt to any new data or information that becomes available.
The results of these models can be embedded directly in business process screens so that end users can consume the results without needing to ever leave their standard workflows. The best analytics can fail if they are not consumed in the right place, in the right time, in a way that is useful for someone to make a critical decision in the moment –SAP Predictive Analytics combined with the Intelligent ERP is designed to ensure that this no longer happens.
This being said, when it comes to machine learning there are still risks. By making things this easy, we have to also ensure we prevent users from making mistakes in their models and then implementing them across an ERP system where they can wreak havoc. The key to ensuring continuous quality is transparency –experts need the ability to look under the hood of every model to assess and customize its fit for the situation or the unique needs of their organization. Just being able to add algorithms to an ERP screen is reckless unless you are able to ascertain how accurate and trustworthy they are. You must to be able to have an understanding as to why a prediction was made or a decision was suggested or taken. It is essential you be able to trust that it works, and will continue to work. In short you need an open approach and not a black box. Only then can you ensure that bad models are not providing erroneous predictions resulting in inappropriate actions. SAP Predictive Analytics, Application Edition allows for this level of transparency –but that is a blog for another time.