Author: Charles Gadalla, Advanced Analytics, SAP

42-47254170-compositedAfter a number of customer meetings on SAP Predictive Analysis this year, I realized that there are some strong misconceptions out there, like  “this predictive stuff is for statisticians,” and “I can see how it would be useful, but we don’t have those skills.”

This made me wonder:  Why should advanced analytics be the realm of a select few? And why are business users (who can clearly see a strong return on investment) willing to defer to the ‘smart guys’ who hold advanced degrees in mathematics and statistics?  After all, the business users are the ones who really understand their market, their competition, and their products.

Given that the data scientists know the algorithms and analytical techniques, I concluded that surely, at the very least, they should work together! It dawned on me that there is another way.  I’m proposing that a business analyst can, with the right tools, build a more business-relevant model than a statistician.

Consider the case of a sales manager who turns to his data scientist and asks for a sales forecast for the next year:

  • The data scientist might consider several different factors: seasonality, correlated product lines, algorithm selection.
  • The scientist tries to identify an objective, identify the required data, and then work with the best algorithms to see what they can do with the forecast.
  •  Let’s assume a Triple Exponential Smoothing algorithm (if there is strong seasonality) or an ARIMA model is selected.
  • Once the algorithm is run, the data scientist takes the output to the sales manager and reviews it with them.

Now let’s consider the situation if the sales manager went to his sales leads and asked them for a forecast too:

  • The sales leads, who know his customers, competitors, and company better than any data scientist, will get to work.
  • They may look at their pipeline, identify the key opportunities, consider the factors at play in each customer (“John is sick, so let me push this opportunity out by 2 weeks”, and so on).
  • Because they know more about the business, their model will be more relevant (even if it’s mathematically inferior).

For example, here’s a quick forecast analysis in two keystrokes when the data is loaded.

Step 1: Visualize Sales by Month

PA Pic 1

 

 

 

 

 

 

 

 

 

 

 

 

Step 2: Build a forecast

PA Pic 2

 

 

 

 

 

 

 

 

 

 

 

 

Together, Data Scientists and Business Analysts See the Complete Picture, Improve Forecasts

Surely it would make a lot of sense for both groups to be involved.  The business analysts could take on some of the aspects that they’re most comfortable with (like clustering their customers to find high/low value customers, or building a decision tree to identify the characteristics of the high value customers) to support the business.

The flip side is that this will free up the data scientists to focus on more sophisticated analysis that can really move the needle for the company.  For instance, they can build out an optimization equation using a Monte Carlo algorithm to simulate the potential outcomes of an acquisition that’s being contemplated by the board.

Now both groups are getting what they want: the analyst no longer has to rely on a data scientist to get an answer to a fairly straight forward analysis, and the data scientist gets to focus on higher- level decisions that will make a significant impact on the corporation.

SAP Predictive Analysis Serves Both Groups, Extends to Business Users Too

With SAP Predictive Analysis, both groups are served. The interface allows a ‘non-statistician’ to build a robust analysis using a handful of template algorithms by simple drag and drop.  But data scientists could use the Custom R scripting functionality to write their own R based algorithms that are tweaked to the uniqueness of that situation.

This flexibility is powerful, as both are hosted on the visualization platform of SAP Lumira, which is the underlying code base in SAP Predictive Analysis.  The output visualizations can be shared by e-mail,  the cloud, mobile devices, and so on.

Custom R Scripting Example:

PA Pic 3

 

 

 

 

 

 

 

 

 

 

 

 

 

And now a custom script could be coded in:

PA pic 4

 

 

 

 

 

 

 

 

 

 

 

 

 

The best news is that once the Custom R script is written, it can now be used by business users.  The data scientist will identify the variable names, and the business user will key in the data required.  And now the entire organization can benefit from the work of the data scientist who has optimized a script based on the business requirements!

 

 

 

VN:F [1.9.22_1171]
Rating: 4.5/5 (2 votes cast)
Predictive Analytics for Business Analyst and Data Scientist , 4.5 out of 5 based on 2 ratings