Summary
NLP can help digital platform organizations extract hidden insights and automate language tasks to understand user behavior and act on predictions.
Take Our Content to Go
As our world becomes more and more digitized, organizations need to leverage behavior data to provide a better customer experience and detect potential problems before they arise. This can be achieved by analyzing behavior patterns from digital platforms customers operate.
Businesses that offer access to these platforms as a service need an efficient way to quickly identify groups of similar users to address their problems at system-wide levels, rather than spending time trying to fix a million individual issues. Natural language processing (NLP) can help organizations synthesize user feedback from emails surveys, social media form fill-outs, and other unstructured text sources, deriving insights on customer sentiment and anticipating certain behaviors so businesses can act proactively.
Using NLP techniques, extracting metadata from text such as entities, keywords, categories, emotion, relations, and syntax is easy. Deep learning can encode the meanings of individual words and sentences in context or even of entire documents and use this information to categorize documents, extract relevant facts, or infer characteristics of the authors.
Once sentiment is identified, Mosaic can create metrics to understand just how much a negative experience costs to quantify the negative interaction with a customer. Natural language processing will analyze social interactions to inform important customer metrics like churn and customer experience measures.
There are three fundamental objectives to help achieve this: a pipeline to bring structure to unstructured data and extract essential insights from records, followed by three analysis phases for these extracted data. The phases of analysis would consist of:
- Phase 1: Identification of key inflection points using time-series or graph analytics.
- Phase 2: Grouping or clustering of patterns for these inflection points to identify and recommend common actions/interventions to improve the customer experience.
- Phase 3: Visualizing the customer journey by creating an application that integrates with existing organization technologies to make the results easily accessible to decision-makers and staff.
Phase I | NLP-Driven Data Pipelining
The first and most important objectives are to develop the necessary NLP and analytics pipelines to continuously extract critical insights from semi-structured customer surveys, emails, form submissions, and other miscellaneous sources. Typical information Mosaic has seen in past project experience comes from data sources like CRM records and emails to customer interactions with learning management software. Of crucial importance is identifying two contrasting experiences: friction points that determine when a customer is dissatisfied or needs troubleshooting, and resolutions where a single interaction, or series of exchanges, with staff and software help a customer overcome their friction points.
Identifying and extracting friction points and resolutions represents the primary analytical challenge of NLP. While the best approach may yet be discovered as new insights are found in the data, Mosaic presents an initial approach. Specific to the unstructured text problem, we would first begin by using manually labeled definitions and preceding events to identify related friction points and resolutions.
With word embeddings, we can identify sections of text which are semantically like labeled examples. This would provide us with a set of initial candidates for friction points and resolutions. It may also be essential to utilize more classical NLP approaches such as N-Grams and sentiment analysis to identify key terms and pairings for each case and any associated meta-data such as the time of the event, and the events preceding it, the mode of communication, etc.
These new data transformations and either the graph structure or time series can be used as features to train a statistical model to automatically classify and distinguish events as resolutions, friction points, or superfluous. Once an NLP and analytics pipeline is established, events and text defining these critical milestones can be automatically and continuously identified and extracted. This information can then be indexed and stored for further analysis.
Phase II | Trend Identification
Once friction points, resolutions, and important steps along a customer’s pathway are reliably identified, the following analysis phase can begin, identifying known and emergent inflection points in the customer experience. By looking for trends in either the time-series structure of events or the graph’s structure, which defines the relationships between the events leading to friction points and resolutions, it will be possible to look retrospectively at what events typically lead to positive outcomes.
This will enable Root Cause Analysis by surfacing key sequences that lead to these outcomes—identifying these patterns will also inform how best to visualize the customer’s journey and make the most impactful recommendations to improve the student experience. The information extracted and analyzed up to this stage will provide a rigorous statistical picture of the series of events that lead to resolutions, including understanding the impact different paths may have on a customer’s experience.
Phase III | Visualizing the User Journey & Recommender Engine Development
The next task will be to understand and visualize the user journey at the individual and operational levels through aggregations that can be filtered depending on the relevant metrics or demographics of interest.
By visualizing the user journey concerning friction points and resolutions, it will be possible for decision-makers and analysts at digital platform organizations to identify common patterns of friction points that stagnate progress towards resolutions. These groupings and aggregations will provide a deep statistical insight into the user experience.
More importantly, these aggregations will allow businesses to extract and understand shared experiences on how customer service can best respond where needed. This can be done retrospectively and in real-time to identify emergent groups to be proactive in ensuring these customers have the best chance of success.
Concurrent with visualizing the customer journey, novel insights will be used to inform recommendations that enable customer service representatives to proactively assist customers in finding their friction points and recommending appropriate action or resolution to overcome this friction and progress towards resolution.
This recommendation system will ingest data from a particular customer experience. Using any commonalities across the user base and based on a history of successful customer service resolutions, recommended actions will bring each customer closer to resolution while minimizing pain points. Recommendations that effectively and reliably guide customers away from their friction points down the path of success will enhance retention and improve the user experience.
Final State Solution
The final objective of this project is to develop a user-friendly software application that showcases all the above and integrates seamlessly with existing products used by the digital platform business. This “full-stack” development project includes the necessary infrastructure to run and execute the NLP and analytics pipelines, the recommendation engine, and the database(s) to store and index extracted insights for later analysis.
The resultant software must make it simple and intuitive for customer service to create and visualize customer journeys for individual customer groups and operational aggregations (e.g., using demographics). This will allow organizations to monitor emerging trends in real-time for individual customers and groups. The recommendation engine will need to be integrated into this software to seamlessly provide customer service reps with important recommendations and historize both data, recommendations, and resulting actions to quantify the long-term effectiveness of the approach.
Finally, this software must integrate with existing systems, such as Salesforce and other CRMs, and make it straightforward for the business to fully own, maintain and sustain each piece of the software from the data ingestion to the NLP and analytics pipeline, various models, databases, and the final application.
Mini Case Studies from Mosaic’s Portfolio
The above process can be applied to a range of industries, wielding the power of NLP and advanced analytics to uncover powerful insights to increase productivity, customer satisfaction levels, and sales processes.
Logistics | Real-Time Recommendations for Brokers
In the hyper-competitive third-party logistics industry, every minute counts. Third party logistics (3PL) brokers operate in the trucking spot market, where agents match one-off shipments with truckers (carriers) willing to transport them. A leading 3PL firm approached Mosaic with the goal of using machine learning and NLP analytics to prioritize carriers for brokers to call, thereby helping the agents to source low-cost carriers with the fewest possible calls. We achieved this by creating an analytics toolbox with a set of models and data processing scripts that provide AI-based insights and recommendations based on internal and external data sources.
Healthcare | Health Forum Insights
Mosaic was asked to create a proof-of-concept web-based healthcare insight platform to understand if first-person social media posts could offer unique insights into the experience of Americans living with chronic disease, and specifically if it was possible to infer demographic information based on the linguistic content of a user’s forum posts. Mosaic undertook the task of scraping two health-related online forums to access unstructured text data and available structured data.
The unstructured text was analyzed using NLP techniques and a web-based analytics platform to develop a successful classification model which inferred crucial demographic information about users based on the language they used in their posts. This enabled the development of demographic profiles with shared attributes, which provided critical first-person insights into their experience written by people living with chronic diseases. The web application built by Mosaic was designed to have a simple and intuitive user experience for analyzing these social media posts, which helped understand and communicate the experience of living with chronic disease, ultimately informing better decisions around public health.
Conclusion
Natural Language Processing presents ample opportunity to derive insights from unstructured data and support enterprise-wide customer-facing communications and escalation processes. Given the complexity of today’s digital platforms and users’ heavy reliance on them, customers will have questions or issues that must be resolved promptly. Being able to anticipate, identify, and successfully resolve such pain points creates a positive experience for the customer and contributes to the overall user experience of the digital platform. This can trickle down into higher user engagement and sales, with organizations touting quick customer service support and resolutions.
Mosaic has extensive experience in this space. We have worked across many business domains developing ground-up data science solutions, including NLP, recommender systems, time-series, graph-based, and statistical/machine learning solutions. Not only that, but Mosaic also has the expertise and experience in software application development and the infrastructure needed to support user-facing applications. In short, Mosaic’s skill set promotes ease of use when communicating critical results and insights to users and analysts to make better decisions. Get in touch with our team today to learn how we can tailor an NLP solution to meet your organization’s needs.