Training Call Center Agents vs Training AI Agents

Coming off some recent client implementations with Talkdesk & Five9 , before we set final dates on the next round with NobelBiz? , Genesys & 亚马逊 has me pondering the future of contact centers. Again. Just as I was 2-3 years ago with a NICE + RingCentral combo purchase we guided, right as 微软 Teams was catching fire.

Recent news that 谷歌 has officially entered the CCaaS space and seeing a sneak peak of their tool in Silicon Valley last week, right on the heels of T-Mobile news that they're building their own CX platform with OpenAI has me a bit shook. It's constant pondering of what's coming next, but I believe I can see the path to 2030 forming in the near distance.

Which leads me to the topic for this month. Call Center Agents vs AI Agents. Specifically the differences in choosing & training each. A decision that many of you will have soon.

In 2024, in order to remain at the top of my AI game & to keep my Top 50 CX Thought Leader title, I've learned to build and train AI conversational intelligence & GenAI agents via Efficiencies.AI toolset, including Google and OpenAI cloud & environments. Having trained contact center leaders and agents in my past (outside of AI), I can now compare the two.

First, where do you use a contact center agent or AI agent, or both in tandem? That depends on what you're attempting to do. Some companies are using AI agents to reset passwords without human augmentation. With a high success rate. They feel it eliminates the human aspects of being pushed or persuaded to give out sensitive password information to a bad actor. For tandem use cases, it's mostly business development for AI sales qualification, simple question/answering or AI appointment setting. Via phone or chat, many times after hours, to begin. While AI handles the majority of routine parts of the transaction, it still hands it off to a human for the sale or physical appointment. But, what if there are customer questions during that leg of the customer journey?! That's where you either have to train AI to recognize this & hand it off to a human or give it enough information to handle multiple intents. Much like transfers back to HQ in the BPO world.

Second, which is easier to train? If that's not the most loaded question of all questions. I'll shut that down quickly with this response....you shouldn't compare them in this way, if you don't have to. We're seeing more and more comparisons & BIG headlines come from some who say that you can move 75% to eventually 99% of transactions to AI in the future.

I'd rather be on the knowledgeable side of that bold claim. If you've made it this far in the newsletter, you're likely interested in understanding it too. I've learned an entirely new set of AI for CX terms and practices that I will share now to help you in training your AI agents. Whether you're in I.T., Training or Operations, it's beyond time to put your student hat back on to remain relevant in getting the best performance results & CX for your business:

Training Dataset

Call Center Agent: New hire, foundational learning and curriculum for how to handle customer interactions.

AI Agent: A collection of data used to teach a machine learning model by showing examples of inputs and the desired corresponding outputs.

Fine-Tuning

Call Center Agent: Ongoing learning, coaching and development once productive.

AI Agent: Training a bot on a smaller dataset or language model to improve its performance for specific tasks. Ongoing learning & development for the technology to narrow its focus.

Intent Recognition

Call Center Agent: Using active listening or data from a NLP/NLU IVR to understand what a customer is attempting to do during the transaction.

AI Agent: Ability to detect the purpose or goal behind a users input, or understand if multiple intents will likely be needed for the transaction.

RAG (Retrieval Augmented Generation)

Call Center Agent: No definition exists. Closest comparison is using Agent Assist technology to auto-pop relevant documents to better answer customer inquries.

AI Agent: Different types of RAG development and datasets can improve response accuracy and lead to higher satisfaction for GenAI self-service.

Agentic AI

Call Center Agent: No definition exists. Closest comparison is a tenured, cross-trained call center employee who is highly knowledgeable and skilled to handle a variety of transactions with little to no coaching or assistance.

AI Agent: A type of AI agent that can take autonomous actions or make decisions to achieve specific goals. It has the dynamic ability to complete a workflow with little to no human intervention.

Prompt Filtering

Call Center Agent: No definition exists. Closest comparison is training your agents to not give out misinformation or not answer questions they don't have concrete answers on. Or to transfer to someone who does or provide a generic response.

AI Agent: These are the same guard rails, it's how to avoid risky or harmful inputs from customers that could allow output risk.

Output Moderation

Call Center Agent: Essentially equal to Quality Assurance. Applying post-transaction checks to audit the answers given to the customer.

AI Agent: Ways to find, limit or re-frame AI generated responses that could be inappropriate, biased, incorrect or dangerous. Training for safety & ethics.

You can hopefully see & appreciate the consistencies and differences. While 'agent' is spelled the same in both uses, each operates and is trained quite differently.

While others are spending their time creating fun pictures or meme videos with AI, I am afraid to say that I have dove head first into learning how to train GenAI to do what I want and what customers will eventually want. But when I need a breather, there's always this.

If you're looking to quickly deploy a sales development bot to test after hours info gathering for sales reps to receive and call back to close the deal, let me know. Or if you're looking to test an appointment setting bot that can schedule appointments by interacting with your calendars and adding an event with customer info, let me know. Or if it's as simple as wanting access to your own sandbox to upload all of your FAQ's and then deploy a limited bot to your website to answer questions related to those, let me know. I'd like to see how well I've trained this new contact center AI team.

Or if you have interest in AI Certification, I am working with ICMI & HDI on a new option!

Josh Streets is a renowned customer experience technology advisor with over 20 years of experience in the scene. Currently serving as CEO of Scoreboard Group Consulting, he has a proven track record of identifying, consulting & investing in disruptive contact center innovations & best practices. He is an ICMI Top 50 thought leader, frequently speaks at industry conferences and is a regular contributor to CIO publications or media engagements on emerging AI trends. www.scoreboardgroup.com