The term "Smart Manufacturing" is today a niche phrase that only a few experts in manufacturing are conversant with. However this is likely to change very soon. Particularly with more and more people becoming curious about something called "3D Printing", which The Economist first blew the lid on by calling it the Third Industrial Revolution. According to Google Trends, the interest in 3d printing has sky rocketed in the last one year!

For many small and medium manufacturing businesses, overhead costs can be a significant contributor to the overall profitability and health of the company. We explained in detail how manufacturing overhead is typically accounted for and how to use analytics to manage this. Setting up useful analytics is a simple 3-step process as explained in that article.

Sophisticated predictive analytics models have to hit the road and become usable at some point in time. Let us take an example, a typical application would be something like a lead scoring problem. You have hundreds of prospects visiting your real or (more likely) virtual store and you need to predict which of these are "sales ready" so that you can make them the right offers. In other words you need to execute a lead scoring program and target only those prospects who attain a certain score.

We have written about the importance of calculating customer lifetime value (CLV) as a means to quantify the benefit from carefully segmenting and marketing to individual consumers. A general customer lifetime value formula will include the following components:

Business Intelligence and data warehousing experts have long known how free form text can be a challenge for assuring data quality. Data entry operators can and will frequently "mess up" input data with typographic errors (e.g. misspelling salesperson names in CRM), using inconsistent terminology ( e.g. "Corp" and "Corporation" while referring to the same company in different entries), mixing up numeric and non-numeric codes and so on. 

The objective of time series forecasting is to take historical data and build a model that would allow us to forecast what is likely to happen in the future. The key ingredients of the model output could be a level, a trend and a seasonal index. The model may include one or more of these indicators. A principal assumption in these model is of course, that there would be no major external or internal changes to the business. Clearly that is a very broad assumption and one that frequently fails!

At a recent data mining training, one of the participants posed an interesting problem. How can an online store, let us say for example a shoe store, sort their incoming merchandize in a rapid and efficient way? The products have to be automatically organized into categories that make sense not only from a consumers point of view but also from an inventory perspective.

Customer segmentation, also referred to as market segmentation, is defined as the process of finding highly similar sub-groups within a heterogeneous larger market. This approach was originally used in direct marketing to target and focus on well-defined market subsets.   

When you run a market basket analysis, the deliverable of the analysis is to generate practical rules that a store can apply in order to maximize their cross-selling opportunities. The objective of the analysis is to establish statistical confidences around the applicability of the rules. For example, when a rule states "IF X is purchased, THEN Y is also likely to be purchased", we would also need to know what is the chance that this rule is not a random occurrence. The "X" is called an antecedent and "Y" is called the consequent.

Applying predictive analytics or machine learning techniques to extract meaningful business value from the flood of social media data requires some necessary amount of drudgery at first. This is the part where the analyst needs to establish labels on the different inputs. For example, we need to teach a model how to identify good reviews and bad reviews based on the text which the reviews provide. The key phrase is "learning" - the model must learn what constitutes a good sentiment and how it differs from a bad sentiment. Once we prepare the model in this way, deploying it to score new unstructured data is a well established process.