View through data
Data science recognizes two basic types of data. Structured data and non-structured data. The majority of models and analytical tools use structured data. Most often they are transactional data (purchase data, utilization data, etc.) or data from website traffic, etc. Analyzing such structured data, be it customer, financial or other, is a common practice today.
Unstructured data, often ignored, bring new and vital information that usually remain hidden.
Non-structured data are especially useful in the following situations:
- You do not have access to structured data or you do not have them at all.
- You need to know answers to specific questions you cannot get from structured data, such as customers’ feelings, your reputation, your products, or competitors on social networks and forums.
- You need to complement your structured data, (e.g., to finetune a customer churn forecast model with non-structured data, such as records from your call center).
Unstructured data sources, i.e., data for text analytics, include, for example:
- Email communication with customers.
- Customers’ posts to different social networks and discussion forums.
- Basically any other text data from both internal and external sources.
Unstructured data analysis is more demanding than structured data analysis. However, the benefits can be fundamental.
A real-life example:
A customer files in a complaint. The system analyzes the email via text analytics to understand the severity of the issue and combines it with data from the CRM system about the customer’s importance. Instantly, the complaint is assigned priority based on the severity of the issue and the given customer’s importance.Customers file different types of queries or complaints to their service or product providers. Through text analytics, we are able to identify the content and sentiment of such communications and determine their significance. By analyzing customer data (structured), we are able to determine the importance of individual customers.Once these two inputs are linked, it is possible to identify the priority of individual queries with regard to both the importance of the customer and the importance of his query. Thus we can prevent a significant customer from churning.
