In the age of convergence, customer churn is a concern for service providers, challenging most retention techniques...
While concluding a recent project for a fortune 500 company my colleague came across a very peculiar problem. The model was delivering very good results in the Quality environment but failed miserably when it moved to the production environment.
After a lot of investigation and discussions with the database team we realised the database in the quality environment and the database in the production environment were not synched on a regular basis so the model was built on old data and tested on new data which caused so many of heart burns for everyone. The point is not about the fact this is something that can be resolved using very simple processes and a little bit of discipline but about the complexities that are involved in getting good results from a set of data.
The dots actually got connected when one of my friends was sharing with me the problems they were facing at his company in implementing a product that was supposed to help their sales team deliver better margins by using some of their proprietary algorithms. He said that they were more than 12 months into the implementation and the product was still not able to deliver results.
Now let’s get down to the connection. Every organisation has its own philosophy of collecting data. Most of the time the data an organisation collects is a process of evolution either driven by management needs or by the investments made in acquiring products or building in-house solutions. The larger the organisation the more complicated is the data it collects and larger the volume of data not only in terms of size but also in term of the number of variables. More importantly is the fact that the process never stops. So we are typically faced with an environment where the nature and type of data being collected by an organisation continuously changes.
Advanced analytics based products even though customisable are built on a set of fixed input criteria which is normally data. If an organisation does not collect the data that the product has been designed for we will typically notice one of the two things.
- The product will not be able to deliver the desired results
- The data transformation process during the implementation will be another large project in itself.
It is my firm opinion that data based solutions like the ones normally used in advanced analytics should be designed around the data collected and not vice a versa because it is next to impossible to convince an organisation to change its data collection strategy.
One more in favour of customised solutions I guess.