We use cookies and other similar technologies (Cookies) to enhance your experience and to provide you with relevant content and ads. By using our website, you are agreeing to the use of Cookies. You can change your settings at any time. Cookie Policy.

Straightforward Analytics

Originally published February 24, 2011

I am fascinated by the promise of incorporating the results of business analytics into the operational environment of an organization, yet I also think that people looking to hook into this promise are often not aware of some common-sense steps that need to be taken to make sure it pays off. So in this month’s article, I’ll provide my opinion of straightforward steps that could be taken to put a firm foundation in place for successfully applying analytic results.

Process Preparation

Even before you integrate your analytics, there are some considerations involving organizational preparation, including:

  • Define the Objective—In order to improve a business situation it is important to have a good understanding of the value driver and corresponding business objectives to be improved. High-level objectives (“become a smarter organization”) are usually too fuzzy to be used effectively. Instead, select a specific value driver (“increase sales revenue”) and a specific method (“by increasing the number of products purchased by each customer”).

  • Establish Measures—If the desire is to improve the business in a specific way, there must be a measure that reflects the current baselined value so that a target value can be established. Continuing our example, if you want to increase sales revenue by increasing the number of products purchased by each customer, you must have a means for determining the current number of products each customer has purchased, the average number of products purchased per customer, the average revenue associated with each customer’s sales, and the total revenue associated with the cumulative number of products purchased by each customer.

  • Define Justifiable Target(s) for Success—The measure of success is not just a target value. Instead, it needs to be defined in terms of how it addresses the corporate need. For example, “increasing sales revenue by increasing the number of products purchased by each customer” needs to be refined to specify the number of products per customer. In addition, that target number has to be justified in terms of business value. In this example, that might suggest evaluating the projected increase in revenue when the average number of products per customer is increased to different target levels. As with our previous step, this requires more than just speculation; instead, you should have some means of calculating average cost per product and what each incremental increase in product sales means in terms of increased revenue. In other words, our success objective should be adjusted to “increase sales revenue by increasing the number of products purchased by each customer from 2.6 to 2.9.”

  • Select Business Processes—The analysis alone will not achieve the intended increase in value. Rather, the analysis should help you figure out which business processes can be improved using the analytical results. Our example would potentially look at any business process scenario in which a company representative can engage a customer to purchase another product. This might span different functions, such as front-line sales or the inbound call center, and may involve different business processes with different sets of triggers and actions.

  •  Engage, Engage, Engage—If the business processes need to be adjusted to achieve the desired benefits of the analytics, you must ensure that the people engaged in and the people managing those business processes are willing to make changes to improve their business processes.

Data Collection and Preparation

As we have just discussed, we need data before we even get started doing our analysis. On the other hand, you might say that reviewing the current state to establish the business objectives qualifies as part of the analysis. Either way, before you can change your business processes in any way, you will need to get high quality data in the right format as a prelude to the actual analysis. This includes these steps:

  • Data Selection—Select the data sets that capture the operational/transactional aspects of the business process. In fact, some of this should already have been done to establish baseline measures. However, at this point there may be an interest in incorporating other data sets to acquire additional demographic and characteristic information to augment existing data. This may draw from internal data sources as well as external data sources. This selection process may be an iterative one, since the results of the analysis may need additional data values to provide greater precision in results.

  • Qualify the Data—Apply data quality techniques to ensure that the data meets specified quality objectives for key data quality dimensions such as completeness, consistency, etc. This might also include a statistical review of data values to ensure the suitability of variation in data to determine if there are any issues that might affect or bias the analysis results.

  • Data Preparation—Since the data might need to be joined and denormalized for the analytical applications and algorithms to work, the data sets must be prepared accordingly. This might also mean additional transformations or data reductions. For example, data attributes with values drawn from a continuous value range (such as a customer’s age) might be mapped into a smaller number of “age buckets” such as {under 18, 18-34, 35-50, 51-65, over 65}.

Data Analysis

At this point we are ready to start the analysis cycle. The tasks in the phase are likely to be interleaved with those in the deployment stage, as the models are tested and tweaked. During this stage, these steps are taken:

  • Identify Success Models—Review the existing data to select the situations that reflect the most successful scenarios. To continue our example, we’d like to see if there are any characteristics that are similar for the desired customer profile. Therefore, we might select  those customers that already have purchased more than the target number of products as input to our analysis.

  • Analyze—Depending on existing data mining and analytics algorithms, there are a number of different methods of analysis. Employ one or more of these methods to analyze the data to determine if there are any dependent variables associated with these successful scenarios and to capture the corresponding variable values that are indicative of success. 

  • Review and Validate—Knowing that correlation does not necessarily imply causality, it will be worth further investigation to review and validate the analysis. This can be done in two ways. The first involves reviewing the results with subject matter experts to help determine if the correlation makes sense altogether. The second is to apply the results to a test set of additional transactional data not used in the original analysis to see if that data set’s success scenarios can be identified. If so, the results are almost ready for a prime-time test.

Incorporate Results

Given an analytical model that has been reviewed and validated, we can adjust our business processes to take advantage of this newly discovered knowledge. Some typical steps include:

  • Scenario Evaluation—This step looks at the dependent variables and considers business rules to be used in evaluating scenarios as well as describing potential actions to take within each scenario. This is intended to answer the question: Given the dependent variables, what will we do with them?

  • Modify Processes—This step goes back to the selected business processes and modifies them accordingly. This can be as simple as changing a process script if the values in the dependent variables in a customer’s record  suggest an additional product purchase. Of course, it is best to work this step with the owner of the business process to do additional research and analysis and then adjust the process accordingly.

  • Deploy Measures—One last thing before taking the next step is to make sure there are some “probes” in place to monitor success. A good example would be to track the number of times the customer’s characteristics map to the model and then how often the additional product has been purchased.


Specify a time frame over which these process changes are put in place and continuously measure success. Clear successes can be more fully integrated and deployed, but less-than-stellar results might suggest reviewing assumptions and repeating the exercise to seek modifications to the models that might yield better results. In that case, you might jump back to the start of the analysis activity and make some changes to the analysis, or perhaps even bail on the entire model and start from scratch. Either way, hopefully this article has provided a straightforward view of the analytics process.

Recent articles by David Loshin



Want to post a comment? Login or become a member today!

Posted March 4, 2011 by Bansi Patel

Excellent post.

Is this comment inappropriate? Click here to flag this comment.