Insight of the Week: AI Powered Agent Training

Insight of the Week: AI Powered Agent Training

By Kerry Robinson

Looking back, it’s obvious, but it took me a while to realize it: AI is an amazing agent training tool!


I’ve talked before about the challenges of ensuring your Gen AI agents - voicebots and chatbots - do the right thing, and can’t get led astray.


Traditional QA won’t cut it. You can’t write a test script, because Gen AI isn’t scripted. There is an infinite number of conversations your AI agents can have. So to make sure they do the right thing and don’t do the wrong thing, I’ve advocated augmenting explorative testing and evaluation with automated simulations, and evaluations.


Then someone asked if we could use Gen AI to help with their agent training program. Yes… of course we can! We can just point our Generative AI simulation and evaluation framework at a human agent, instead of an AI agent.


Obvious, right!


The cool thing is that this provides a nice, step-by-step path towards an AI-first contact center.


Here’s how it could go:


  1. Pick a small number of contact types that your junior agents handle
  2. Grab the training materials you use to train agents on those contact types
  3. Build a set of scenarios that describe the situation of a customer that would lead them to get in contact about those things
  4. Add some dummy accounts to the CRM that match the scenarios
  5. Use those scenarios to feed the simulation pipeline. The scenario and personal details should tally with a CRM entry
  6. Make calls to junior agents - using a Simulation AI agent - to help them practice.
  7. After each practice call, run an automated evaluation - using an Evaluation AI agent - against the conversation to give instant feedback.


There are 2 key pieces of AI you need for this:


  1. Simulation AI – an AI agent (voice or chatbot) that will call or chat via your CCaaS platform, to simulate a given scenario
  2. Evaluation AI – an AI agent that’s given a conversation transcript and provides an evaluation – probably a score on certain metrics, and some advice on how to improve


With these two AI agents and your existing CCaaS platform, you can be up and running, delivering classroom training, or even interleaving your training with real calls when your agents are not as busy. I went into more detail about AI agents in last week’s article.


The coolest thing about this approach is that it provides a safe, easy on-ramp to your AI first contact center journey:


  • Deliver immediate value, even if you start with just one use case.
  • No need to address the compliance issues of a customer-facing AI Agents.
  • It forces you to really understand the Objectives and Key Results (OKRs), Standard Operating Procedure (SOP), and the relevant Key Performance Indicators (KPIs) that you’ll need to evaluate human agents, and build automation in future.
  • If and when you decide to roll out customer-facing AI Agents, you’ll have your simulation and evaluation framework in place.


But, the biggest win of all is that you’ll know how well your agents handle specific tasks, and you can use that as a baseline for the performance AI agents need to achieve before putting them in front of customers.


What do you think? Might this be a way for you and your team to get started with AI agents?


Kerry Robinson is an Oxford physicist with a Master's in Artificial Intelligence. Kerry is a technologist, scientist, and lover of data with over 20 years of experience in conversational AI. He combines business, customer experience, and technical expertise to deliver IVR, voice, and chatbot strategy and keep Waterfield Tech buzzing.

Subscribe to Kerry's Weekly AI Insights

要查看或添加评论,请登录

Waterfield Tech的更多文章

社区洞察

其他会员也浏览了