How about making use of the 80% of customer data you have on hand but haven’t tapped into yet? And how about if that data can help you reduce churn by 50%? Text mining is one of the latest trends in data mining today, with many companies already benefiting significantly from their efforts around this practice.
You can download PDF version of this whitepaper here.
About 80% of customer data captured within enterprises is believed to be in unstructured format, in the form of call center notes, customer / employee e-mails and survey responses. Especially in industries (particularly telecommunications and finance) where there is a high level of interaction with customers through various channels (such as through a call center or a customer service office), significant knowledge about the customer base resides in this unstructured but captured data. Although business intelligence and customer analytics have become an essential part of corporate life in most leading companies today, very few look into opportunities residing in this 80% of customer data, keeping their analytics focus on structured data only.
Text mining is the means to actually being able to analyze this unstructured customer data so that valuable customer insight can be extracted from it. This method of analytics identifies the keywords, phrases, and patterns hidden in data kept in free text format, assisting companies make sense of each piece of communication.
A recent study conducted at a major mobile telecommunications provider found that there was a 40% increase in the performance of the telecom’s churn prediction models after the inclusion of customer complaint details via text mining. A financial services provider case demonstrated a 20% improvement in credit-risk decision models after the inclusion of call center conversations via text mining. Text mining is providing a means for companies to utilize extremely valuable customer insight in their decision making processes, and clearly the results justify the effort.
Focusing on the use of text mining in analyzing contact center records, there exist various opportunities, such as:
1. Converting qualitative information into quantitative data, providing ability to categorize reasons, risk, and opportunity levels of contacts, which enables enrichment of almost all data mining models (e.g. churn, cross-sales, etc.)
2. Identifying repeating complaint and inquiry reasons automatically, which can be used to resolve root causes and decrease contact center load, while increasing customer satisfaction
3. Identifying recently begun categories of complaints or inquiries, by automatically categorizing them, which can be used as early indicators of operational problems
4. Extracting competitor intelligence from customer interactions, by automatically identifying calls related to competitor prices, campaigns, and products from contact center records
The benefits are not only limited to for-profit companies. The Hong Kong government, for example, text mines its call center records so that is can quickly discover new complaints – “For instance, we can spot districts with frequent complaints received concerning public health issues such as dead birds found in residential areas. We can then inform relevant government departments and property management companies, so that they can allocate adequate resources to step up cleaning work to avoid spread of potential pandemics.”
Six key steps exist towards making full utilization of text mining on unstructured customer data:
1. Channel Analysis Identification: The first step is to analyze all the channels that customers interact with your company through, to determine where the text mining practice can be applied to. The most traditional channel for text mining is call centers, by analyzing the input made by customer service representatives into customers’ profiles in their CRM systems. This can apply to customer service center / dealer / branch interactions as well, if and only if the interaction is recorded by the customer service representative. A separate but emerging channel of interaction that also needs to be text minded are electronic channels, in the form of customer e-mails or web form entries. As long as the interaction is recorded via a company employee or via the customer into some form of accessible system, text mining can and should be conducted.
2. Lexicon Customization: Most tools for text mining have a built-in dictionary of common words and their categorization (e.g. locations, numbers). Although highly effective in identifying generic words and patterns within a document, their effectiveness can be multiplied by introducing business and mission specific words. For example, a text mining tool would not be able to categorize a company’s product names or the names of its competitors by default. It would simply treat them as yet another word. Or it would not necessarily search for ‘value added service’ as a phrase when analyzing a telecommunications company’s contact center data. All such business-specific words and phrases need to be defined and categorized for a more intelligent analysis of the text, through preliminary analysis of the records and establishment of a business dictionary.
3. Text Mining: Once the custom text mining dictionary is defined for the business problem, the next step is performing the actual analysis, using a text mining tool, which is readily available in most statistical analysis and data mining solutions. The outputs from the text mining activity, which is the list and frequency of keywords and phrases, needs to be reviewed by business users in order to identify those which have business meaning, as well as any requirements for further fine-tuning the business dictionary. Like data mining models, text mining models need to be run periodically to categorize new records and continuously identify new patterns emerging from them.
4. Analytics Integration: Text mining, on its own, is a valuable tool for understanding what customers are most interested in or concerned about. However, its value add to other customer analytics models is also immerse, and hence needs to be used as an input for most data mining models (such as in churn prediction and cross-sales analyses). By categorizing details of complaints, requests and inquiries, text mining provides the ability to flag customers for critical interactions and needs, which then becomes yet another input parameter for quantitative data mining techniques.
5. Reporting Integration: As text mining can categorize complaints and inquiries, it provides an ability to track the changes in different reasons for customer contacts, which needs to be integrated into the enterprise reporting platform. Monthly reports on ‘new keywords/phrases’ or ‘keyword/phrase trends’ would enrich standard contact center reports significantly, enabling identification of increasing customer interest or concern in specific areas or early alerts on certain services or offerings. These reports can also provide an understanding on the repeated customer call reasons and focus on their root cause or moving their resolution to alternative channels.
6. Operational Integration: Integrating text mining findings into operational systems in a real-time manner can help companies increase the efficiency and effectiveness of their customer interactions – such an effort can provide guidance to the contact center agents on the right department to forward requests to or warn them when the call details they enter present a high-opportunity or high-risk customer interaction.
Similar to traditional data mining models, text mining requires ongoing maintenance, as what the customers talk about changes over time, with changing products, services and propositions. As a result, companies need not see text mining as a one-time activity, but rather, need to assign ownership and resources continuously.
As in all types of customer analysis, text mining is as accurate as its inputs are. Companies which start investing in text mining need to make sure that their customer touch-points enter enough details about each interaction with clear phrasing. ‘Cst complains on price’ is very different than entering ‘Customer compared our international prices to ABC Telecoms for India and is complaining about our off-peak prices. He mentioned that he would churn unless we offer a good package soon’ and would not result in any in-depth insights even with the best text mining techniques. This requires training and customizations in performance management for such personnel. A more advanced alternative solution to this problem is making use of speech recognition, to automatically translate customer interactions into detailed scripts for analysis.
Text mining customer data is only the beginning for companies seeking to strengthen their analytical practices and understanding of their customers – the next steps can be an analysis of web chatter, press releases, blogs, etc., using a similar approach, for identification of what the market is talking about and what the key competitors are up to.
To learn more about making most out of your unstructured customer data, please contact firstname.lastname@example.org.