19 Oct Phrase Based Models for Customer Service QA Automation and Agent Training

Phrase based models use natural language processing (NLP) and machine learning which allow AI to derive meaning from human language. This is a powerful customer service technology in multiple ways.

The Power of NLP

NLP is a branch of computer science focused on enabling AI to interpret and respond to human language at a human ability level. 

People are trained all of their lives to use language and automatically assess words in context. A computer program has to be taught to interpret which definition of a word is intended, what words are proper names of people or places (named entity recognition, NER), even to identify what was meant when words are misspelled, and all of the other nuanced analysis humans do automatically. AI, when trained on robust and applicable data sets, ‘learns’ and processes all of these details with amazing speed. 

Many everyday tools use NLP: GPS systems, email spam filtering, speech to text on phones, and assistants like Siri and Alexa. With every use an AI learns and improves. In the case of assistants, consumers with an iphone can opt in to improve Siri, dictation and translation. This allows Apple to store and review audio samples to continually train NLP in Siri. 

In customer service, NLP has been used alongside machine learning (and a multitude of other AI focused processes) to automate aspects of voice and text based service. Phrase based modeling is one such usage.

Phrase Based Models and Automating QA

QA and first contact resolution (FCR) for customer service is an ongoing challenge. Phrase based models are one way to automate QA monitoring, reduce the load on live agents and improve resolution speed. 

AI is trainable. A company can provide the AI with a custom dictionary of words and phrases that come up frequently in their business and train it to provide easy answers to a customer inquiry. This is the idea behind phrase based models.

The AI is taught to monitor for keywords and phrases and then apply a response or action.  

Use examples:

• Keywords for customer issues triggering alerts for agents to perform particular actions  
• Words and sentiment that indicate an upsell opportunity or a need for further information with AI prompting the agent to provide same or escalate the interaction to sales
• Phrases that trigger a warning to an agent or manager for customer harassment, abusive or litigious language
• Monitoring for spelling and grammar, including common misspelling of words often used in the company product line or industry, that warn the agent to correct the error prior to sending to the customer
• Language that indicates subjects to be avoided, causing a guiding prompt to agents (controversial subjects, politics, war)
• Identifying when compliance language is necessary, missing or incomplete (for example, regulatory issues such as GDPR & CCPA)
• Greetings and empathy that, when missing, cause the AI to alert the agent to use them (like “Hello”; “I understand it’s frustrating…”)

 The CSAT.AI example below shows how a new phrase can be added, with exact word match or matching vaguely and an automated action applied when the phrasing is detected. It also has a drop down menu for monitoring either agent or customer language for the chosen phrase model.

Sentiment Analysis

Word definitions are one thing. Deriving a person’s feeling or mood from language is even more nuanced. 

Sentiment analysis is parsing customer responses and actions to understand how they feel. Accurately assessing customer sentiment helps companies course correct providing customers better products and better service. 

AI is able to analyze mass data to derive customer sentiment over time and during interactions in real time. This kind of in the moment monitoring acts as both a training tool and an opportunity to shift a negative interaction to a five star one. 

It is much harder to win back a lost customer. That’s why tools like phrase based models paired with real time AI monitoring and sentiment analysis are so valuable. They are a powerful asset for customer retention, personalization, and satisfaction.

Conclusion 

The world of AI technologies is deep and complex, so this is not an exhaustive look. In the realm of customer service, these are some of the aspects of AI that are continuing to change the game and how it is won. Phrase based models dial in to a company’s industry, products, brand and customer base. They amplify customer service training, help automate aspects of QA and improve over time.