The Importance of Contact Center Data for AI Initiatives
June 2024
Contact centers are a great source of customer-centric data, as most of what is done for customers and prospects is captured and recorded by one or more of the department’s systems or applications. Structured inputs are saved in a system of record, such as a customer relationship management (CRM) solution or another customer tracking application. Unstructured inputs, including customer conversations (via voice or digital channels), freeform survey comments, etc., are collected in omnichannel recording systems and either transcribed or structured for analysis by conversation analytics applications. Information and insights contained in the structured and unstructured inputs provide actionable findings that can help improve contact center and agent performance and identify customer needs and wants.
Despite its high value, most customer data captured by contact centers is generally not shared outside the department, even though the information frequently surfaces business, procedural, operational, and systems issues throughout the organization. Although this data was overlooked in the past, this must change in the artificial intelligence (AI) era. Customer data collected in the contact center should be shared with other enterprise functions and business intelligence (BI) repositories to enhance the company’s bottom line and brand.
The fuel or enabler for AI initiatives is data. In many cases, the larger the repository the better, particularly when the information is verticalized and targeted for a specific use, like data from a company’s contact center or customer service organization. Another important source of information for AI initiatives is the enterprise’s knowledge base or knowledge management system. This typically contains detailed information about their products, services, policies, and procedures.
When it comes to generative AI-based solutions, the underlying training data leveraged by the applications is essential for their success. Generative AI technologies utilized in contact center solutions are primarily trained using large language models (LLMs), which can contain generic training data (public LLMs), be specific to an industry or company (private LLMs), or have a combination of both. Initially, public LLMs were leveraged almost exclusively, including various iterations of OpenAI’s Generative Pre-trained Transformers (GPT-3, GPT-3.5, GPT-4), Google’s PaLM, Anthropic’s Claude, Meta’s Llama 2, Cohere, Hugging Face, and others. Today, a growing number of contact center/customer service software vendors are building customized private models trained on contact center/customer service data because their clients are showing a strong preference for domain-specific and/or verticalized LLMs. Vendors are also: fine-tuning public models with their own data to improve accuracy; supporting the simultaneous use of multiple LLMs; or leveraging different ones based on use case. In addition, some enterprises are beginning to create proprietary LLMs for complete control of the data leveraged in their organizations by generative AI technologies. Since generative AI creates its original content based on the training data provided, it’s imperative that any LLM used, public or private, contains targeted, tagged, curated, and maintained information appropriate for the task at hand. It must also have effective guardrails to avoid hallucinations. The bottom line is that companies need to figure out how to share structured and unstructured customer data captured by the contact center, as well as information contained in their KM solution, with other departments to deliver additional benefits enterprise-wide.