ChatGPT is amazing. It really is.

It can write romantic poems, summarize complex text, provide engaging fun conversations, and answer sophisticated general knowledge questions. It can also produce a fake report for your boss of what you’ve (supposedly) been working on for the last few days — complete with perfectly (surface) plausible code snippets! It makes up stuff.

So could ChatGPT replace human agents in a contact center?

To answer that question let’s examine some key requirements for automated business-grade conversations. What is essential?

Predictable Responses — Enterprise requires that conversations are reliably predictable and can be signed off by legal, marketing, and CX departments. You really don’t want an automated system to just ad lib!

Real-Time Operation — Customer service unavoidably needs up-to-the-minute access to customer and product info (via APIs). Furthermore, this data needs to be seamlessly integrated with both conversation context and individually personalized customer history and preferences.

Custom Knowledge —Every company has unique product information and business rules. In addition to appropriate common-sense knowledge, these often dynamic ontologies must be an integral part of a dependable ‘knowledge graph’ referenced by conversation logic.

Complex Actions — Two crucial aspects of helping customers are a) to triage (ask a structured set of questions in order to understand the issue) and b) to execute series of carefully controlled actions, such as updating multiple related databases or APIs.

Rapid Learning — The system needs to be able to (permanently!) learn and fully integrate new facts and preferences on the fly. It also needs to be easily configurable — changing or adding new products, business rules, and skills.

Scrutable Operation — All knowledge and rules utilized in conversation, plus how decisions are made, needs to be full auditable. Black Box statistical models are inherently inscrutable.

In summary, ChatGPT is extremely impressive engaging in free-form conversations on a wide range of general knowledge topics. This makes it easy to imagine what it might do for the call center. However, its inherent limitations are that it produces unpredictable and frequently incorrect or meaningless responses, and its Black Box operation cannot be reliably enhanced or constrained.

Maybe as a final word, let’s get it straight from the horse’s mouth: