AI and Automation in the Contact Center – Part 1
This blog was written by a human …
Are you sick of blogs about Artificial Intelligence (AI) in the contact center yet?
I’m not surprised. AI is the topic of conversation – news reports, vendor marketing and trade-shows where AI is the only topic, prompting a ‘fear of missing out’. And there are a deluge of new AI companies/ apps springing up online seemingly every day.
But instead of adding to the hype and losing our heads, let’s breathe, stand back a bit, and consider where AI fits in the spectrum of automation in the contact center.
A Spectrum of Sophistication
Automation in the contact center exists on a spectrum of sophistication, usually taking input, doing some processing and producing a result:
| Input | Processing | Result | |
|---|---|---|---|
| Hard-coded response | e.g. DTMF button press | None | e.g. join 'support' queue |
| Robotic process automation | e.g. CLI | e.g. database lookup, analyse total spend | e.g. join 'valued client' queue |
| Artificial Intelligence (AI)/ Machine Learning (ML) | e.g. conversation transcript | e.g. content/ sentiment analysis | e.g. agent prompts, next best action AND the result is fed back in as input (so the machine ‘learns’ and gains accuracy over time) |
Generative AI and ChatGPT
Generative AI, particularly ChatGPT, is the latest headline-grabbing iteration, currently spawning hundreds of new apps/ websites every week. There is no point in listing what’s currently available; by the time you have read this, it will be out of date.
We are currently just scratching the surface of the possibilities of prompt engineering. But as this area grows, use cases will continue to emerge.
And ChatGPT still has problems to be wrestled with, which leaves it requiring a human in the loop, acting as a human guardrail. For instance:
- Bad actors – developing ways to use tech developments for villainous ends
- Hallucinations – untruth presented as truth (and fed back into the system)
- Data security and copyright infringement – private/ protected customer data being fed back into the system
- Successful training on domain-specific data.
The result? We think ChatGPT will not replace agents any time soon, but it may replace some agent tasks, e.g. after-call summaries. This will allow agents to spend more time on what they are good at: empathetic human interaction.
