analytics deployment

You know that an organization is serious about analytics when they are keenly focused on first understanding how to deploy the solution. Increasingly more and more small and medium businesses are recognizing the value analytics brings to their organization. Knowing this, their attention has been, correctly focused on how to make the insights consumable. Analytics deployment is a critical factor in fanning adoption.

The analytics consumers are employees who are as far removed from a typical data scientist as sales representatives and assembly line operators.  Let me talk about two examples, companies that we have been helping with analytics deployment. 

The first example is a sales executive for a retail product that deals regularly with buyers from big box retailers. Retailers source products from a variety of manufacturers and importers. Their buyers are overwhelmed with the endless choices to make from a large number of suppliers – especially for commodity products.  When asked to choose between commodity supplier A and supplier B, the retailer has very little to differentiate other than price (which is a slippery slope for the suppliers). A unique differentiator for the producer of goods would be the knowledge they possess about the sales and marketability of their products. This is where analytics can play a key role and if the insights are intuitively displayed, that can be the difference between closing or losing a deal. For example, if supplier B could not only show which products did well last year in different geographies (descriptive analytics) but can also estimate (with the help of predictive models in the background) which ones are *likely* to do well next year, this would clinch the deal for B!

Savvy sales execs are starting to appreciate the full potential of data analytics as both a sales tool and a way to generate better insights.  However they also understand that the key is in improving how the analytics are presented and made accessible to the end user.

The second example is from a “traditional” manufacturing company. This company extracts minerals from ores and a key challenge is adjusting the process parameters to increase the amount of mineral “recovery” from the ore. The parameters can be different for ore samples coming from different mines and if they establish classes or categories for the samples as they are being milled, the operators can adjust the process parameters to maximize recovery. They recognize that using predictive analytics they can classify or cluster the samples. However to dynamically adjust the process parameters, the operators must be able to “consume” this analytics. For example, they must be an intuitive way to show them which cluster or class a given sample may belong to and instruct them on the correct manufacturing settings.

This requires that the predictive model must be applied on the real time data and the results of the prediction must be nicely presented to a lay person. Simply building a model that classifies historical data is only step one in the adoption of analytics. The key step – the money making one – is to be able to translate the model predictions into operational settings. While the display of results in this case may be trivial, the complexity comes in aggregating real time data which need to be scored (the “unseen” data) using the machine learning algorithm built in step one.

To conclude, I want to make 2 key points. In order to “institutionalize” the solution process, we need effective analytics deployment. The first point is that deployment requires an intuitive analytics consumer interface. The second point is that we need to think about the unseen data that needs to be scored before expending the effort building predictive models. Other wise the entire exercise will become purely academic.

Originally posted on Fri, Feb 13, 2015 @ 08:00 AM

No responses yet

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

simafore.ai