How artificial intelligence (in particular LLMs) can assist agents in the contact center

Artificial Intelligence (AI) is becoming a normal part of operations in many industries. But despite that, there’s still an incredible amount of hype around it. Plenty of it justified, and plenty more that’s not.

Every day I see a lot of fear-mongering on LinkedIn about how AI will consume 80% of jobs. Equally, huge swathes of people embracing AI and offering tips on how to use it most effectively.

As we welcome services such as content generators, website builders, image creators, financial strategists, and code builders into our lives, it’s easy to get overwhelmed.

The pace of change is truly astounding.

But, crucially, as with pretty much anything in life, it’s how AI is harnessed that will determine its impact and effectiveness.

In this article, I explore:  

• How artificial intelligence is impacting the contact center

• How Large Language Models (LLMs) can assist agents and operations teams

• Use cases that generative AI can support

• What’s next for contact centers that embrace AI

How can AI help contact centers overcome operational challenges?

Contact centers have been under immense pressure for some time.

It’s now widely accepted that the experience you give customers is a differentiator for your brand and can provide a competitive advantage in a highly commoditized society.

And yet, contact centers are a large drain on resources – something senior management doesn’t like. And this means they’re often starved of investment.

When you then layer in other challenges faced by the contact center, including:

• Difficulty hiring and keeping agents
• Legacy infrastructure, systems and processes
• Lack of performance insight
• Poor mental health of agents and staff

It starts to paint an even more challenging picture.

To try and tackle some of these challenges, progressive contact centers are turning to artificial intelligence for support.

But how are they using AI in contact centers to drive operational efficiencies? Well, they’re automating repetitive and mundane tasks. Summarising large volumes of information in seconds. And they’re categorizing, filing and coding information that can be retrieved in the future.  

All these things that were once on a wish list for many Senior Contact Center leaders have now become a reality. Especially with the use of Large Language Models (LLMs).

Large Language Models and the contact center

By now, I’m sure you’ll have given an LLM, such as ChatGPT, a go. Venturing into the world of LLMs tends to start with feeding in prompts, usually in the hope of increasing efficiency.

Businesses too are getting to grips with LLMs to perform a wide variety of functions. Especially given the advancements made in where and how they’re hosted and tighter data retention policies.

There are two core areas of the contact center that can be supported by LLMs:

• Agents – automating repetitive or time-consuming tasks, such as summarizing a call, or summarizing previous customer interactions to help them solve the issue faster
• Operational teams – accelerating performance insights, for example, asking the LLM to read a transcript and summarize the actions the agent took

Just from this brief summary, you can quickly see that the potential is fairly far-reaching.

Any job that’s currently manual in nature, involves large amounts of data or is a time drain can likely be replaced or supported through the use of LLMs.

Let’s go into more detail about how LLM’s are changing the game in contact centers.

How can Large Language Models support contact center agents?

Here are some of the best use cases for LLMs in contact centers.

Call Summarization – Call transcripts are clunky things, especially if the agent has been on a long customer engagement. If a new agent picks up the conversation, then information is often lost, or there is a time delay while the new agent gets up to speed.

Summarizing call details, using support from LLMs has multiple benefits:

• Agents can instantly be brought up to speed without having to read the full historic interaction transcript
• Operational teams don’t have to listen to long recordings to find information
• Call summaries can be shared with customers as evidence their query has been taken seriously and the previous details are correct

Answering Questions About Interactions – You can prompt LLMs to review call transcripts and answer questions like, “What was the call reason?” or “What troubleshooting steps did the agent take?”

This use case is best for getting answers that require an understanding of the full context of an interaction. However, it’s important to keep humans in the loop to review for accuracy. Many contact centers are also combining machine-learning based conversation analytics with generative AI to get the best of both worlds: structured and contextual analysis.  

Call Classification – This is still a major challenge for contact centers and a resource drain to manually fix when calls are wrongly classified.

LLMs can be used to read the call transcript and decide, with a high degree of accuracy, how the call should be classified.

Access to Knowledge Base Articles – Agents waste a large amount of call time looking for information to help customers.

Through the use of LLMs the agent can ask a question and receive a real-time answer collating information from their (normally vast) knowledgebase articles. Saving time to reach the correct answer.

This powerful capability also means new agents can be on-boarded faster. There’s no longer any need to spend weeks on end in training, hoping that they’ll retain the right process knowledge to answer customer enquiries.

Things change quickly in the contact center as they’re at the mercy of wider company changes. With the right LLM solution, contact centers can make fast changes to process information without having to wait for an IT team to make the change.

Talk to the team to see how AI can help your contact center

What I’ve shared in this article is just the tip of the iceberg when it comes to the use of AI and LLMs in the contact center. Every week I see new use cases and opportunities to leverage this powerful technology.

If you’d like to know more about how AI can support your contact center operations, contact the team today and book your free demo.

FAQs

What's the best approach to get agent buy-in for AI tools?  

Frame AI as an assistant that eliminates tedious tasks rather than a replacement threat. Start with pilot programs involving your most tech-savvy agents and showcase early wins to the broader team. Provide comprehensive training that emphasizes how AI enhances their problem-solving capabilities and career development. Regularly collect feedback and adjust implementation based on agent input to ensure the tools genuinely improve their daily workflow.

Should we build AI capabilities in-house or partner with specialized vendors?  

For most contact centers, partnering with established AI vendors is more cost-effective and faster to implement than building in-house capabilities. Vendor solutions offer proven algorithms, ongoing updates, and specialized support. However, ensure vendors provide customization options to fit your specific processes and industry requirements.  

What are the best practices for combining traditional machine learning with generative AI in contact centers?  

The most effective approach uses machine learning for predictive analytics and pattern recognition, while generative AI handles content creation and conversational tasks. For example, machine learning algorithms can categorize and tag topics discussed on calls, predict probable customer satisfaction scores, and automatically score calls for objective QA criteria. Generative AI can generate call summaries, answer questions about transcripts, and recommend best practices or knowledge articles to agents in real time. This hybrid approach leverages ML's strength in data analysis with generative AI's natural language capabilities, creating a more comprehensive solution than either technology alone.

What are the best practices for keeping humans in the loop when using generative AI content in contact centers?

Maintain human oversight by requiring agent review and approval of AI-generated responses before they're sent to customers, especially for complex or sensitive issues. Implement confidence scoring systems where AI flags low-confidence responses for human verification. Train supervisors to regularly audit AI-generated content for accuracy, tone, and brand alignment. Create feedback loops where agents can quickly correct AI suggestions to improve future recommendations. For critical processes like escalations or refunds, always require human authorization even when AI provides the recommendation. Additionally, establish clear guidelines about when agents should override AI suggestions and ensure they feel empowered to do so without penalty.