This Call Will Be Recorded for Quality and #AI Training Purposes…

TechCrunch reports today that Amazon expands Bedrock with conversational agents and new third-party models.

This seems a direct pitch to attract companies interested in exploring how Amazon’s marketplace for #largelanguagemodels could be used to build an #aiworkforce for their customer experience or customer service business.

Take a step back and consider how many hours have you spent on the phone trying to get satisfaction with a Customer Service Representative of a large company? #CSR operations are huge, costly businesses to run well. Two cases come to mind:

  • A national telecommunications company I worked with measured quality in terms of call times (shorter was better) and calls per representative (more was better). This compelled their 15,000 CSR employees to race from one customer angry about their cable bill to the next so quickly, they quit at predictable intervals. The company accepted that half of new employees would ghost on their first day, and annual turnover was in the triple digits. That churn costs.
  • A global platform-as-a-service leader I worked with measured quality using #machinelearning to do sentiment analysis on call recordings and messages. They then correlated the call content to call outcomes. How many calls, emails, and chats did the issue take? Did the human agent ‘connect’ appropriately with the customer across all channels? Did our employee apply our expensive training to communicate, de-escalate, or resolve things effectively? How well is this employee skilling up or performing in this role over time? These were the higher-level questions they could answer with the #businessintelligence they had built for their 40,000 person operation. That insight costs.

For companies exploring how to build an #aiworkforce-based CSR operation, questions and implications might include:

  1. Do we want to help our human agents with agent-facing #AI or replace them with customer-facing #AI
  2. How easily could we train a large language model using the years of recorded calls and messages we have?
  3. If we built a customer-facing #aiworkforce would our customers accept them?
  4. Which business lines, regions, or call types might we experiment with first?

Leaders and managers might consider “how do I skill up to manage both a human and #ai workforce?”

For us as customers and consumers, the implication might be that companies charge, or charge more, to get a human on the line.