… you could identify which employees are likely to turnover in the next 6 months?
… you could understand which transformers are likely to fail during our next major storm?
… your call center agents could delight customer with the best next-step recommendations?
The question is no longer if, but when. With the latest analytic solutions from SAP, customers can find the answer to these and many more important business questions… and you don’t need a PhD!
Predictive Analysis for Dummies
In November 2012, SAP released a new member to the SAP BusinessObjects Business Intelligence product line – Predictive Analysis. With this solution, utilities have access to solution that can help them move from a reactive “sense & respond” to a proactive “predict & act”.
Predictive Analysis allows customers to intuitively design complex predictive models. Once built, users can then visualize, discover and share hidden insights.
Firstly, Predictive Analysis will leverage PAL (Predictive Analytic Library) within the HANA platform. This means that common algorithms like K-means, Decision Tree, Linear Regression can be run natively within the database.
This is a game changer. To put this into context, an SAP customer wanted to cluster analysis on 10 million rows of data, analyzing 20 columns of data to understand financial risk and exposure. Using traditional predictive tools, this process took 48 hrs. With HANA and the PAL library it took 90 seconds!
Secondly, Predictive Analysis supports R integration (as does HANA). This means that you can leverage over 3500 add-on packages and even add your own functions. This is more algorithms than SAS, SPSS and Statistica combined.
Applying Predictive to the Utilities Business Case
If you are intimidated by math and talk of algorithms, then let’s get something straight right from the start. Predictive Analysis is designed for anyone to be able to use. Understanding the type of algorithm you need is 50% of the battle, therefore I’ve created the following guide to show you when to use which algorithms in a Utilities context.
Classification algorithms/Decision Trees:
- Propensity for equipment failure
- Propensity for incomplete repair, e.g. revisit
- Propensity to pay bill
- Clustering customers for marketing/upsell purpose
- Clustering customers based on credit risk or collection priority
- Clustering based on consumption patterns (potential fraud, unregistered meter)
- Identify energy theft or fraud
- identify employee time reporting anomalies
- Asset co-occurrence – help docs (have you checked does this equipment also have symptoms of X?) diagnose, preventive maintenance
- Repair/PM co-occurrence – help engineers prescribe additional maintenance; help identify new repair/PM processes
(NOTE: This algorithm is most notably known for market basket analysis for retailers. When the customers buys X, they also by Y.)
- Seasonality of outages, equipment maintenance
- Forecast customer revenue, energy consumption, maintenance costs, outages, reliability, energy procurement costs
Here you can see the list of algorithms within the Predictive Analysis Interface.
Hopefully this list of algorithms and use cases gives you some idea of the best algorithm to use in your use case.
What to Know More?
In you are interested in more details about Predictive Analysis, I have a couple of recommendations.
If you are just starting, then I recommend your have a look at the Predictive Analysis resources at SAP:
I also recommend this recent presentation by Charles Gadalla:
Finally, I put together a demo walk-through of building an analytic model to predict transformer failure:
There’s nothing stopping you from taking your utility to the next step in the path to greater BI maturity.