GenAI + Customer Service/Experience

GenAI for Customer Service is an obvious opportunity in 2024. In the Predictive vs Generative AI, the leapfrog has been the machine’s ability to understand commonly used human language. And for any enterprise one of the biggest source of incoming natural language data is customer communications.

I have been tinkering with OpenAI APIs to give some of these solutions a try.

I learn by doing and get to understand the double-click on challenges, nuances etc better when implementing these quick hacks.

For the next experiment, I wanted to build a solution which would do a sentiment analysis and draft a possible response for any inbound customer communication.

Why sentiment analysis for Customer Service?

As someone who has been on both sides of a customer-care channel, I do understand the need to triaging. Not every customer communication is the same. Some are high priority, and some may present an opportunity to convert a customer and so on.

Imagine all incoming messages are auto-mapped by

  • Emotion
  • Urgency
  • Key Concerns
  • Relevant Product/Service

Would this help the team-leader to smartly allocate cases (or one could even automate case-allocation basis rules built on these fields).

Customer Experience guiding Product Development

I am a big fan of marrying the aggregated analytics information (funnel metrics) with the granular feedback from customers for guiding the product roadmap.

As the cases grow, imagine a tag-cloud of issues that visually guide the product management and customer experience teams to the opportunities – my bet is there would be clear insights around bugs, possible new features and broken customer journeys.

Sample tag cloud for a hotel basis demo queries recieved

GenAI powered responses to Customer Queries

Just for fun, also added a draft response. This was super interesting, because

  • in most cases the prompt I used was over-promising. Almost like an instant resolution to the customer’s issue.
  • While the language was great, in some cases it assumed a negative experience when there was none (refer to the first one)
  • Identified cases where it needs to ask for more information for the resolution.

Imagine no longer having standardized email templates or holding responses. AI drafts generated for each case as it lands in the agent’s queue.

While this is not even scratching the surface of AI driven customer support, it was a good to see it in action rather than imagining the possibilities.

AI Experiments done so far