Today we had a fantastic session on Predictive Analytic @ BI Conference Organized by Silicon India , Hyderabad chapter. I will post the discussion we had on a separate blog article and in this article I want to provide some precursor about this topic.
Why Predictive Analytics ?
Prediction stands for Classification and estimation of a value or behavior that is yet to occur.
Renowned business management guru Peter Drucker once said that what cannot be measured cannot be managed. In light of this theory, reporting, scorecarding, key performance indicators (KPIs), and other strategies emerged as management tools, helping to ensure that performance is consistent, and does not deviate from operational and profitability objectives. Perhaps the greatest benefit of those tools is that they give managers the ability to monitor processes after they have been executed, and react accordingly by fixing them when things go wrong.In addition to what happened yesterday or last week, managers are anxious to know what will happen tomorrow. For example, what will be the demand for a product or service? What revenue can be expected and from what channels (direct sales, partners, etc.)? Which customers are likely
to buy a product? If accurate expectations can be formed about the future, managers can proactively structure workflows and allocate resources to maximize productivity and profits. The more precise the expectations, the greater the profits that can be realized – more efficiently and with fewer resources
Modern information systems collect tremendous volumes of data. A customer profile can include more than 200 descriptive variables, such as income, age group, gender, occupation, and education level. A crime incident may also have many attributes – weather conditions, location indicators, and victim characteristics. How can it be decided which factors matter and which ones don’t? And how does one determine the relative importance of each factor. For example, how much more important is income than education when performing credit application approvals? With the aid of predictive modeling technology, managers can do what the human mind alone cannot. They can assess which factors are relevant or assign weights to those factors. Predictive analytics eliminates guesswork, helping managers make better decisions.
Predictive Analytics can serve only as a guide to the future based on its understanding of the past. Changing circumstances, such as the entrance of a new competitor into the market, may disrupt the accuracy of these predictions. Second, data mining can only analyze the data it is presented with. Information about the customer's motivation or emotions is not typically captured in enterprise data systems, so data mining is better at analyzing what the customer did than at understanding why the customer may have done it
The purpose of any CRM activity is to turn an action the customer would have taken into one that delivers the most value to the enterprise. This may mean accelerating or bundling a planned purchase or delaying or dissuading a planned switch to a competitor. Most enterprise decisions about how to behave are based on what the customer has already done (collected from operational systems and stored in the data warehouse). The problem is that these actions are a lagging indicator of the customer relationship; they tell you what the customer has already done and leave the enterprise to extrapolate the implications: Why did the customer do it? Will the customer do it again? What else might the customer do? Although not a bad place to start,enterprises are increasingly looking to identify leading indicators about the state of the relationship. These leading indicators can be the words and thoughts of the customers, directly expressed with the intention that the enterprise should react to them. This idea of capturing the voice of the customer will grow more prevalent as technologies mature, and as best practices and business cases enter the public domain. Now is the time for forward-looking, analytical organizations to prepare for the coming questions and initiate pilot programs and proofs of concept to understand the real potential behind these questions..
What is the future of the predictive analytics market?
Interest in unstructured data is growing rapidly in today's social business environment, although the variety of unstructured data that exists for organizations to tap into is much more diverse. Unlike pattern discovery, which relies on technology (unless you are starring in "A Beautiful Mind," identifying patterns from numerical displays is almost impossible), content analytics seek to duplicate something that most people would agree is better-performed by people. The drawback to the manual approach is that it almost always fails in at least one respect. The sheer volume of information now available would defeat most organizations' ability to comprehensively analyze it all, forcing a potentially misleading reliance on sampling. In the event a large enough team can be assembled, the issue becomes whether enough consistency in the categorization can be created. Even simple distinctions between good and bad can be difficult to define ("Not as bad as I expected" is a very ambiguous statement, for example). Applying technology to the problem allows both of these pitfalls to be avoided, albeit at the risk of having a completely consistently inaccurate interpretation if the deployment is not correctly managed. Content comes in many forms and with many different characteristics. In most cases, a text-mining solution can be employed to analyze content, but this approach is not always the best. In particular, a decision between using speech analytics to analyze an actual conversation (for example, between a customer service agent and the customer) or using a text mining system to analyze the call enter agent's notes may result in very different analyses. In most cases, we would recommend using speech analytics to analyze the call recording.Analyzing Web content, particularly from social media sites, can also be performed using either text-mining applications or dedicated media-monitoring service providers. In general, we recommend the use of mediamonitoring service providers as the better long-term solution (text-mining applications are best-suited for tactical or ad hoc analysis).
Combining traditional data with the positioning information from mobile devices and the ability to append information to the identified location allows significant improvements in predictive power and the ability to deliver context relevant messages to customers.Incorporating location into predictive models can drive a variety of applications in multiple industries.
This is quiet a fascinating topic and I look forward to explore more and welcome you to share your views /predictions about predictive analytics.
Your Loving P&C