Quantifidata

BRIEF

The fear of being left behind can drive businesses to scramble and invest in state-of-the-art technology and tools to manage their data pipeline, just so they can enter the world of AI and digital transformation, hoping to leverage their collected data to perform predictive analytics. However, they often overlook the importance of asking the right questions, which is actually where the digital transformation journey begins. In this brief, we explain why asking the right questions is crucial and how it can help small businesses make informed, data-driven decisions and get on the right track.

Types of Predictive Models

Predictive modeling attempts to predict an output response variable as a function of input variables. ML algorithms are designed to perform a task of function approximation where they attempt to find the best mapping function, f(x), for x. The x here represents historical data and f(x) is a prediction of the new data. The function approximation task can be viewed as classification or regression. If the response variable being predicted is categorical, classification algorithms are used, with binary classification placing the output into one of two categories and multi-class classification predicting multiple categories. If the response variable is continuous, regression models are used, which predict a value rather than a category. The quality of classification algorithms can be evaluated using prediction accuracy, while a regression algorithm relies on the root mean square error (RMSE).

Questions facing Small Businesses

The literature is replete with explanations of predictive machine learning techniques. But how do small businesses considering entering the digital transformation space know which approach to use? The best way to approach this choice is to start with the question that needs to be answered. This in turn depends on the business imperative at hand. Is it a problem of reducing operating cost? Or increasing revenue ? Or ideating a new product that could generate a new revenue stream? This business problem should drive the question being asked. Suppose the questions are of the following nature- Will the new campaign be of interest to young adults 18-30 or adults 31 and older? Should an oil change business open or close its doors on weekends? Is a customer expected to churn or not in the near future? Is the retained inventory of a product expected to be higher in summer or winter? On the other hand, suppose questions are of the following nature? What is the number of older adults who signed up for the campaign in the first week? How many customers is the oil change business expected to see on weekends? How much of the product's inventory is expected to be retained in summer vs winter? What is the minimum price a customer is expected to be willing to pay for a new service product?

Matching Predictive Model Types to the Questions

While the former group of questions are best formulated as classification problems, the latter are regression problems. Both require predictive modeling, approximating a function f(x) given input variables x. Both rely on past data to make new predictions. But the choice of the model/algorithm is driven by the nature of the question. The specific problem at hand should drive the questions being asked, whether they are classification or regression problems. The classification problems try to place their output into two or some discrete number of categories. The regression problems on the other hand concentrate on predicting a value for the output function f(x). Each have their own suite of algorithms which come with tradeoffs (hyperlink) . Small businesses should focus on formulating their questions first before attempting to implement digital transformation strategies.