Home > Columns > Executive Interviews

CallMiner Executive Interview

How Agentic AI Is Reshaping Customer Service: A Conversation with CallMiner 

The customer experience industry is in a moment of rapid change. Artificial intelligence promises faster answers, smarter workflows, and more efficient operations, yet many organizations still struggle to turn insight into action. In this interview with Sheri Greenhaus, Managing Partner of CrmXchange, MJ Johnson, AVP at CallMiner, explains how real time guidance and agentic AI can close long standing gaps and reshape the role of the customer service agent.

Sheri Greenhaus: I’m going to assume you want to talk about the newest products and the overall vision. Would that be correct?

MJ Johnson: Absolutely. I’m happy to talk about that. I have opinions about the moment we’re in—from a customer service perspective and from CallMiner’s history, and I’m glad to discuss those themes.

Sheri Greenhaus: CCW was a big show. Everyone had AI all over their booths. I asked each one, “How are you different?” and every answer was the same. Thinking about CallMiner’s AI guidance: where did you see a pain point that truly set CallMiner apart? Was there a gap in the market you felt needed to be solved?

MJ Johnson: Great question. Our decisions come from multiple places: the market, prospective customers, existing customers, and our own experience delivering product.

We’ve had real-time capabilities for some time, and customers have used them successfully. Our approach to delivering high-quality customer experience starts with listening to as many signals as possible and using those signals for insights and decision-making. That’s foundational.

A lot of our recent efforts, though we also focus on customers and executives, are best understood from the agent perspective.

There’s a first phase: Getting agents trained and ready. Products like Coach, analysis capabilities, and coaching signals all support that readiness.

Then there’s the second phase: Agents are on the front line. Ideally, they are prepared, but in the moment the question becomes how to actually help them. If you throw prompts or alerts at them when they are already confused, it does not solve the problem.

After the interaction, there’s summarization, action items, and automation. And then there’s the broader view: looking across all conversations to understand effectiveness. Historically, the loop between post interaction insights and real time support required significant upkeep. The loop back to readiness through training and coaching has always been easier to manage.

But that second loop, between real-time and post-interaction, is unique to this moment. That’s where AI is changing things.

Sheri Greenhaus: So, this is guidance for human agents, not for no-agent environments.

MJ Johnson: Correct. And that ties back to your question about the pain point AI needed to solve.

We’re using AI in two major ways:

1. Analysis and comprehension. Understanding when agents “raise their hands,” what circumstances lead to that, what responses are effective, and what’s not working. AI classifiers give us more flexibility than traditional language-based extraction.

2. Real-time context. This is what we announced at CCW. In the moment, two things can happen:

• If we know agents almost always need help with a certain issue, we proactively present that information based on historical patterns.
• If that doesn’t solve it, and the agent still doesn’t know what to do, they can escalate. Supervisors have always had command-post visibility.

Now, conversational AI enters the moment armed with:

• the company’s knowledge base, and
• the context of the conversation.

With those two inputs, it can generate a response grounded in both the customer’s need and the company’s approved information.

Sheri Greenhaus: Step one is making sure the knowledge base is accurate and consistent. As consumers, we’ve all called customer service, gotten one answer, called back, and gotten a different one.

MJ Johnson: Exactly. We’ve always been able to identify inconsistencies and address them through coaching. Now, instead of waiting for coaching after the fact, we can reinforce correct information during the conversation through conversational AI.

Sheri Greenhaus: If the underlying data is inconsistent or inaccurate, could the guidance generate an uncertain response, for example “I am not sure, here are two potential options”?

MJ Johnson: We never want AI conclusions to be a black box. We provide reference ability and source traceability so humans can analyze outputs.

Our philosophy is that AI must be practical and useful, but also human-centric. It augments accountability; it doesn’t replace it. If there’s any concern about hallucination, the system exposes the source so it can be addressed.

Sheri Greenhaus: At CCW, everyone was talking about agentic AI. I heard different definitions. What’s yours?

MJ Johnson: Agentic AI facilitates connected workflows. Requests originate in human need such as an agent needing insight in the moment, or an executive reviewing conversations afterward. Some reasoning steps and data inputs cannot answer the question by themselves, but they can when combined. Agentic AI connects those pieces so the user gets a helpful starting point.

Effective agentic AI is multi-layered: optimizing connected AI workflows based on your data, your interactions, and the business outcomes that matter most.

Sheri Greenhaus: Is this working? Do you have customers using it now?

MJ Johnson: Yes. The agentic capability is built into our platform. Any customer using Analyze, Coach, and our AI Assist UI is accessing the agentic architecture.

The real-time capability is in production for our real-time customers, and we work with them to ensure the knowledge base and other components are in place.

Sheri Greenhaus: What feedback are you hearing from customers and agents?

MJ Johnson: Agents appreciate the difference between push and pull. Push is when the system alerts them based on conditions in the conversation. Pull is when they ask for help because they need more than what they have.

Less-experienced agents benefit most; it levels out uneven experience. But trust is the big factor. As AI becomes more present, agents need confidence that the information is reliable and grounded in what they’re expected to know, not just a bot that might hallucinate.

Verified information and accountability matter for any generative capability.

Sheri Greenhaus: With turnover, I’d assume training time decreases. Agents still need basics, but guidance helps when they’re stuck. Are you seeing that?

MJ Johnson: We always encourage investing in agents. Coaching is proven and valuable, but it won’t meet 100% of needs. Real-time guidance complements it. If a business chooses one or the other, they’ll end up paying for the gap somewhere.

Our goal is to use automation to give agents more meaningful work—engaging work aligned with supporting customer needs. Automation carves out certain use cases, and humans handle the more complex, brand-defining interactions.

Sheri Greenhaus: But isn’t it essentially the same work with easier and better information?

MJ Johnson: Businesses are grappling with which use cases to automate first. Once those are carved out, they need to ensure automation remains effective.

CallMiner’s approach starts with intelligence: gather insights, identify automation-ready use cases, and continually improve what agents receive. When agents are on calls, we fill gaps in the moment. AI makes it easier to act on insights. That’s a major enabling outcome of generative AI.

Then:

• Identify automation-ready use cases.
• For the remaining human-handled interactions, use insights to continually improve what agents receive.
• When agents are on calls, set them up for success with the information they need.
• Acknowledge they won’t have everything—and fill those gaps in the moment.

AI makes it easier for customers to act on insights. That’s one of the key enabling outcomes of generative AI.

Sheri Greenhaus: Looking ahead a few years, how do you see the agent’s role changing?

MJ Johnson: The role will still look familiar, but how it works will change a lot. Some agents will function as humans in the loop across many complicated processes.

Automation will become more effective, but humans will remain essential to ensure processes behave correctly, especially for compliance and customer success. And customers will still be able to “hit zero” and reach a human.

Agents won’t look like traditional agents. Their tools, resources, and what agentic AI empowers them to do will be very different.

Sheri Greenhaus: What’s involved in that loop?

MJ Johnson: Agentic AI farms out tasks to specialists across different domains. Loops can involve authentication, trust relationships, and interactions between systems across businesses.

You’ll have loops created for one business that can interact with loops in another. These loops coordinate reasoning, data, and insights to deliver answers to customer needs.

Some loops will be initiated proactively by humans. Others will run automatically in the background, presenting answers before the customer even asks.

Sheri Greenhaus: Where do you see all this going?

MJ Johnson: Customers believe in the promise of automation. They want simplicity. Many people don’t want to call a human unless absolutely necessary. They want quick resolution.

Customers want reduced complexity. Agentic architectures can be complex behind the scenes, but the experience should feel effortless.

Businesses already have insight into what customers care about and what they’re likely to contact them about. Use that data to design thoughtful agentic architectures so customers don’t have to think about the process.

Whether it is an airline, a dentist, a doctor, a chiropractor, or an insurance company, every one of these situations offers a chance to make things easier and more efficient for the customer.

Generative technologies and agentic AI finally unlock interconnected data relationships that weren’t possible before. You no longer need one massive data repository; you can orchestrate across many.

Sheri Greenhaus: So simply put, it’s getting the customer what they want when they want it.

MJ Johnson: Yes, keep it simple, easy, and effortless. Instead of “Press 3 for this, 4 for that,” the agent or the system should say, “How can I help you? What do you need?” And ideally, “I think I already know what you are going to ask, so go ahead.”

Sheri Greenhaus: Is there anything I didn’t ask that the audience should know?

MJ Johnson: Two things.

First, real-time capabilities have existed for years, but adoption and documented success haven’t been as broad as they could be. Real-time alerts require continual calibration because agents learn. What they needed the first time may not be what they need the third time.

Our new capabilities make the loop between insights and real-time guidance fresher, more relevant, and more impactful.

Second, differentiation. Anyone can claim to use generative AI for a specific use case. That’s not unique. CallMiner builds on a long-standing commitment to high-quality customer experience programs. That informs how we apply AI. Our chief product officer has been successful in augmenting existing technologies with generative models, making analysts’ lives easier and improving discovery.

The enthusiasm around generative AI is justified. But CallMiner’s strength is applying it on top of a proven foundation focused on customer experience outcomes, not AI for AI’s sake.