By: Mark Cummings, Ph.D.

Most everybody has known a frustrating experience engaging with an automated help system. With Generative AI (GenAI) this experience can be greatly improved while reducing the costs of its delivery. The key challenge is a well implemented GenAI model. To deliver a better experience, organizations need to move up the AI learning curve and develop optimal implementations. One of the best ways is to join a domain focused user group.

Background

Early attempts to reduce costs of providing help desk and customer service took one of two directions: off shoring or end user automated data entry. 

Offshoring involves staffing in countries with lower labor costs. Unfortunately, it was often prone to poor user experiences because of accents, unfamiliar vocabulary, or general cultural gaps.

Automated data entry, whether voice or data driven, uses a fixed script to prompt a customer to input each piece of data serially. While this can be cost effective it too can deliver a poor experience. And such systems can result in a dead-end if an end user doesn’t have the requested data or doesn’t fully understand the request.

Similar automated approaches were also used in an attempt to drive down the cost of sales. They intended to reduce time per session by automating the front-end collection of basic customer data to optimize the interaction time with live staff.

Early AI Help Desk Implementations

With the introduction of GenAI new opportunities for reducing costs are emerging. The challenge is that we are early on the learning curve with this new technology.

Earliest implementations sought to lower training costs by having staff use an AI chat system to get key information to end users. This has two notable downsides: possible delays in answering the questions and delivering the wrong information.

The delay comes from the time it takes to manage the multi-step request loop: First, a staff member receives the prompt and engages the AI; then AI processes the prompt and communicates the inference to the staff member; then the staff member reads and processes the inference and finally, communicates the inference to the customer.

Wrong information can be more subtle. Often, this originates from not accurately structuring the prompt, i.e., asking the wrong question. For example, the author recently engaged with a help desk staff member about pricing, who then responded with the pricing of a specific plan. However, while it was an accurate description of a plan, it was not the desired plan that was eligible for a discount that the author qualified for. As a result, a sale was lost.

One of the easiest ways to implement an AI-based help desk is to use an AI intelligent agent to perform the same end user data entry used in earlier generations of help desk technology. This has the advantage of leveraging support staff with a lot of experience with these scripts and thus able to focus on the infrastructure changes involved in replacing deterministic code with an AI intelligent agent. The downside is that such implementations have the same tendency to produce a poor customer experience. That is to still get stuck if the end user doesn’t fully understand what is being requested or doesn’t have the requested information readily available.

Optimal AI Implementation

In prior generations of technology, the best customer experience routinely came from a properly trained staff member with a friendly demeanor, who was well informed, and had ready access to the necessary information.

GenAI presents the opportunity of creating an effective agent that has these characteristics while being much less expensive. That is, systems that fully leverage AI’s powerful capabilities intended to maximize its performance with an implementation focused on replicating this more human-like behavior. This can be achieved by an AI application having an appealing persona that operates against specified objectives, algorithms, and constraints.