ChatGPT in Agent Experience

Today’s front and back-office agents are hindered by multiple internal problems of Enterprises that impact their ability to do their jobs effectively. This has fundamental impact on Enterprise operational and business performance. Impact of front-office agent productivity is felt on customer satisfaction and retention. Whereas back-office agent productivity impacts operational agility and responsiveness.

The key challenge is complex Enterprise processes and multiple disparate systems. Agents must navigate this complex landscape to get the necessary information and data to solve problems. Agents need real-time and contextual help to perform their jobs in the most effective manner. While some technologies exist to provide this help to some extent, it’s often either not real-time or not contextual.

ChatGPT, with its large Large Language Model (LLM) has the potential to transform customer and agent experience. It can remove the limitations of the traditional chatbots like irrelevant responses, non-contextual data and bad experience. ChatGPT even has the ability to redefine how chatbots are built and managed.

Focusing on Agent experience, as a first step, ChatGPT will significantly reduce the load on Agents as both customers and internal stakeholders will be able to get required information and help through ChatGPT. GPT-3 models can be trained on Enterprise data, like tickets history, Case history, Customer conversation transcripts. Once trained, the ChatGPT is likely to provide far better and personalized answers to customers and internal stakeholders that will reduce a significant volume from reaching agents.

Once the Agent picks up a ticket or case, a ChatGPT prompt will be automatically generated using identifier (like customer id, case ID or any other). Using this prompt, ChatGPT will create its view on why the customer is calling (based on previous call history) or what the case is about (based on case history). Thus, the agent, even before talking to customer or going through case, will have full context of the call/case.

If it’s a customer service call or chat, the front-office agent, with GPT’s multi-lingual interfaces, can easily converse with the customer through ChatGPT, even if the agent is not proficient in the other language. This will reduce the need to hire local agents for Enterprises who have operations in multiple countries but want to centralize call or service centers.

Most importantly, GPT-3 models like Curie and Da-Vinci have the ability to answer questions based on prompts. This will be used significantly by the agents to get real-time contextual answers to questions posed by customers (on the call) or by internal stakeholders. This will also be used by agents if they are unclear about the next step to take.

Prompts can also be auto generated through automations. Based on real-time customer conversation and speech-to-text and semantic analysis, prompts can be generated for ChatGPT. This will ensure that ChatGPT can keep feeding the agent real-time “next best Action”. This will significantly reduce agent training time and also ensure process compliance as Agent will be prompted the next step as per standard process.

GPT-3 models can also identify sentiment. It can be trained on historical customer interaction data and sentiment associated with it. Even without this training, GPT-3 can identify sentiment based on word context or emojis/emoticons. This will be very useful for the agent to warn about a potential negative sentiment and take appropriate action. A positive sentiment could mean an opportunity to cross/up sell.

The other use case will be cross and up-selling. Models can be built based on organization’s marketing content and customer profiles. This model with the right prompt (either manually by agent or automated based on customer context) can suggest the best offer that the agent can make to the customer.

GPT-3 models are auto-regressive models which are trained using reinforcement learning. This facet will be used to continuously make the models better by training them on every customer interaction or case solved by agents. The agent notes, customer feedback and transcript of conversations can be fed to the LLM models. These will make the contextual models better and future responses will be more accurate.

Overall, GPT-3 models and ChatGPT interface have the potential to transform agent experience and overall customer/stakeholder experience.?