A large proportion of customer interactions on your digital channels are text-based, and if leveraged in real-time these can form very valuable nuggets of information and customer insight . When it comes to exploiting these, most organizations have 2 problems:
- Capturing these text interactions and structuring the data in such a way that it can be easily assimilated and used is an enormous challenge;
- Projects that attempt to undertake real-time analytics to understand customer sentiment and intent tend to be high risk, and often fail or overrun due to the complexity of the task in hand.
So why are text capture and analytics so hard?
Starting with capture – most behavioral data capture methodologies and technologies are simply not well set up to capture text. They use the JavaScript tag to capture in session customer behavior and interactions, but it's extremely difficult to configure tags to capture text data from all of the likely sources of text input, across all channels. I'm referring to chatbot interactions, reviews and complaint feedback and also text from less obvious places such as forms that are completed and even text that is inputted but then deleted or changed. And even if an organization manages to capture text in sufficient detail, they will most likely struggle to structure and assimilate this complex, unstructured data in a format that is immediately usable to make meaningful real-time interventions that customers value and increasingly expect.
And when it comes to analytics, there is a vast array of text analytics solutions for organizations to choose from, offering varied capabilities, but each of these require significant configuration effort. Ultimately, if and when a successful configuration has been achieved, success of the overall project will still be dependent on the quality, structure and granularity of the data that is being fed into text analytics solutions. In addition, where real-time use cases are being strived for, naturally data latencies and the ability of the text analytics solutions to deliver sentiment analysis within milliseconds will be critical to the success of the project.
How does Celebrus solve these issues?
Thankfully Celebrus is a proven enterprise solution that provides an elegant answer to the problems outlined above.
The answer to the 'capture' challenge is really quite a simple one. We are the undisputed masters of data capture and our unique tagging free capture technology captures more data than any other solution. In fact, this patented technology simply captures everything in terms of visitor behavior and experience. Naturally that includes text inputs on any channel and device, which is something that we are confident that no tagging-based collection solution can achieve in a comprehensive way, regardless of their claims!
But in addition to capturing all of the text-based data that an organization could ever wish to possess, Celebrus is also able to structure this notoriously hard to manage data instantly. We achieve this with our published data model which enables the complete array of customer data captured by Celebrus to be processed within milliseconds and assimilated into a highly structured format. This format is highly compatible with the partner technologies to which we connect, enabling millisecond data latencies to achieve the genuine real-time use cases for which we are renowned.
However, when it comes to the subject of text sentiment analysis, Celebrus has a new string to our bow which will dramatically benefit our clients.
With our latest release, Celebrus has rolled out Natural Language Processing capabilities into the product. We have embedded the leading language library fastText into the product to enable out-of-the-box functionality for real-time text analytics. This enables organizations to run text analytics models in real-time scoring environments, leveraging our tried and tested connectors to Microsoft Azure Machine Learning Studio and Openscoring Server. The machine learning models running in these real-time environments can be easily and quickly connected to the fastText library to train the model with no additional configuration effort required.
This functionality provides enterprises with an easy way to enrich customer profiles with the results of real-time semantic analysis. These powerful enrichments based on the text a customer inputs not only improves the view an organization has of their customer but enables a wide range of new use cases. In a similar vein to our other exciting new feature, Automated Marketing Signals, fastText enables organizations to instantly determine sentiment and identify opportunities or threats for instant action. For example, fastText can identify sentiment signals that indicate a high likelihood of customer churn, and this data can be connected to a decisioning application to determine the most appropriate next best action, perhaps based on the status of the customer in question. Or other text interactions can be classified following scoring to ensure that they are routed to the most appropriate department as quickly as possible. Or perhaps your organization simply wants to score all text inputs to feedback forms to enhance customer profiles and make this data available for analytics – fastText provides a quick and easy means to deliver this.
