Re: AMA: Predictive Analytics in Connected Planning
What are the alternative and preferred methods for "fuzzy matching" disjointed sets of data?
As I understand it, Predictive Analytics involves a blending of internal data -- such as financial sales / price / margin, customer satisfaction, CRM prospects and pipeline -- with third party data -- public financial statements, descriptive data sets, demographics, etc. This requires some means of determining when two items in separate systems are the same thing. That requires some degree of inference ... similar name, same postal code, same industry = 90% chance of match. How is that best accomplished?
thanks for your question - good point! One of the most important topics (where btw data scientists also spend most of their time on) is cleansing data. Once you start merging data from different data sources, you see this issue more than when you stay in one data set. It depends a bit on the magnitude of the data - and how complex your model is, but part of the script of the algorithm is exactly about this process. There also exist packages in R and Python which can help you doing such data cleaning. I always tell to my team: that data format is not aligned between data sources is not an issue, it's a different story when the format is not consistent. Data scientists will write rules to merge, transform, split, ... data to be able to match formats. Every time that this algorithm is triggered, the computer will go through this recipe and perform the same cleaning rules. it would be too nice when an algorithm is only about data mining and pattern recognition.
Would you be able to better articulate a few examples of predictive analytics applied to a commercial use case ?
If possible could you also confirm the recommended approach/options for implementing it to an existing large and complex project?
I am a big fan of small POCs, would this sort of approach work? Is there a free pre-built template that I could refer to?
My understanding is that it would be a challenge for someone with no previous experience/skills in this area to be able to explore/start delivering something. Particularly if the time to invest is very limited.
In commercial use, predictive analytics can be applied for different use cases. E.g. territory and quota management - by having more accurate predictions, based on data - you can set better quota's and assign more insightful to different territories. But of course, also in sales forecasting, pricing use cases are algorithms helpful. I did a few projects where predictive analytics was used to understand what the dynamic price elasticity of products in different regions for different customer segments was. By knowing this, you can drive more effective promotions.
There is not 'approach' i can recommend. Every project is different, every client infrastructure is different, so hard to come with a recommended approach. But starting with a PoC is definitely a good start. This is what I also do to give people an idea of how well such an approach can work. There are many courses online which you can follow to get started with is. What is sure is that it's not because you're a master anaplaner you know anything from predictive analytics. It's a different skill set you need, a different toolbox you need to understand and a different way of thinking. I believe that the role of us - master anaplanners - is to collaborate very closely with the data scientists in our organization and that we need to help them making their insights (via predictive analytics) more digestible for our planners, our end-users.