Home > Columns > Executive Interviews

Cyara Executive Interview

Click on image below to view the ebook

With AI agents emerging as the next major shift in customer experience, Cyara’s Gayathri “G3” Krishnamurthy,  brings a grounded perspective on how enterprises can navigate the complexity—testing smarter, assuring trust, and preparing for the hybrid CX future.

Sheri Greenhaus: Let’s talk about Cyara. You work across so many different technologies—chat, IVR, AI agents. How does Cyara handle testing across all those channels?

G3: Cyara is a category leader in Agentic CX assurance. Contact centers today have complex technology stacks—CCaaS platforms, legacy systems, new AI agent suites, and customers interact across multiple channels: IVR, web, chat, voice, and now AI bots.

What makes Cyara different is that we test every part of that ecosystem, not just one component in isolation. We test AI agents, which are rapidly being added to contact centers, as well as other channels:

• IVRs
• Web and chat
• AI agents
• Telco networks
• Human agent connectivity and experience

There are many players who do parts of this, but we have the depth and breadth in all aspects of contact center testing like no one else. We have about 50 connectors to AI agent platforms and deep partnerships with leading contact center vendors. Our strength is full end‑to‑end assurance, from the initial dial into the IVR, through any bot interactions, all the way to the human agent. Contact centers often have a mix of old and new technologies, and we test the entire path across all those layers.

Sheri Greenhaus: If you’re testing the entire journey, where do you most often see breakdowns? Or does it vary by customer?

G3: Right now, the biggest focus in the market is AI trust, making sure bots don’t hallucinate, misuse information, or behave unpredictably. A recent Forrester report listed 650 vendors in the AI space, so companies are investing heavily in bots and need assurance that they’re reliable.

But AI is only one part of the equation. The real challenge is the end‑to‑end journey. A customer might hit an old IVR, then a new AI agent, then another legacy system. You also have telco variability, background noise, routing logic, and geographic differences.

Cyara can test in‑country dialing across 425 carriers in 140 countries, which is a huge advantage. If a customer in Munich can’t reach your contact center, you can’t replicate that easily unless you have someone in Munich call again. We can simulate that instantly.

Other major areas where customers struggle:

• Load testing for AI agents – AI consumes tokens and processing power, so load behaves differently than traditional IVR flows.
• AI Production observability – Everything may work perfectly in testing, but production is a different beast. Companies want a closed‑loop system that detects failures in production and feeds them back into testing.

Sheri Greenhaus: When you’re doing this testing, does the client have access to the results? Is there a dashboard? How do they see issues?

G3: Absolutely. And yes, there’s no shortage of dashboards in the contact center world.

Our approach to testing in the AI era has two layers:

1. Brain-level testing We test the bot’s core intelligence—misuse, hallucination, drift, regression when models update, and API-level behavior. This ensures the bot’s “brain” is functioning correctly before running full journeys.
2. End‑to‑end testing Once the bot’s intelligence is validated, we test the full customer path. In the past, you might run 100 deterministic IVR tests. Now, you might run 50 at the API level and 50 end‑to‑end.

Customers get full visibility into:

• Hallucinations and the reasons behind them
• Trends and changes over time
• Intent-level performance
• Outcome metrics: resolution rate, escalation rate, handoff behavior
• Contact center KPIs: CSAT, latency, guardrails, transcription accuracy
• Human‑agent‑style metrics applied to AI: empathy, rubric‑based scoring

We categorize metrics into business outcomes, traditional contact center KPIs, and AI‑specific performance indicators.

Sheri Greenhaus: Across industries, some things are easy to test, like handoffs or routing. But how does Cyara know when a bot is hallucinating? For example, if the bot says, “You need four in blue,” but the correct answer is “four in aqua,” how does Cyara detect that?

G3: Bots are built with guardrails, knowledge bases, workflows, rules, and system boundaries. They’re trained on what they can access, what they can retrieve, and what actions they can take.

Cyara uses the same boundaries during testing. That allows us to determine whether the bot stayed within its knowledge or invented something.

We also test using different personas and environments such as an angry techie, a young customer in a noisy coffee shop, etc., because behavior can vary based on context.

But hallucination detection fundamentally comes down to grounding:

• Grounding to the knowledge base
• Grounding to rules and workflows
• Grounding to allowed system access

If the bot produces something outside those boundaries, we flag it as hallucination.

Sheri Greenhaus: With all the information Cyara collects, I’m trying to picture how it actually gets into your platform. Some things are straightforward—like whether the IVR hung up or how many times an agent said, “I can’t hear you.” But when it comes to knowledge‑base issues, how does Cyara know something is outside the bot’s scope? Does the system ingest the knowledge base? Does it check against it? How does that work, and how is it reported back?

G3: When you create an AI agent, you start by giving it goals. For example, in a telco use case like mobile returns, you define the main goal as handle mobile returns, and then sub‑goals like authenticate the customer, check eligibility, and so on.

In the past, everything was deterministic: if this, then that. Today’s AI agents work differently. You give them loose, natural‑language instructions, and the model figures out the steps based on the knowledge grounding you’ve provided. If page 17 of a knowledge article says “authenticate first,” the agent knows to do that.

The challenge is that once you give the bot those boundaries, you let it operate autonomously. So the testing agent must be smart enough to probe those boundaries. We intentionally design edge‑case tests, not the happy path. For example, we’ll instruct the test agent: “Trigger a failure during authentication. See how the bot responds when the CRM doesn’t connect.”

So while the bot uses loose guidance to solve problems, our testing uses the same flexibility to push it to its limits and expose failures. It’s a fascinating shift: you build the bot with loose edges, and you test it by pushing those edges.

Sheri Greenhaus: Do customers give you the questions to ask, or do you already know the areas to probe? I imagine each company has specific issues they want covered.

G3: It’s a mix. Through professional services and training, we teach customers how to test in the AI agent paradigm, using goals, objectives, and sub‑objectives written in plain English. Instead of coding “if/then” logic, you write natural‑language test objectives, and our system translates that into testing behavior.

We guide customers, but we also support special cases where our team helps design more complex tests. It’s still early days for many organizations, and everyone is learning. Frontier models are improving rapidly, some say a thousand‑fold in the last couple of years. As AI becomes more capable, testing becomes even more critical.

In the IVR era, testing was often secondary. In the AI era, assurance and trust are board‑level conversations.

Sheri Greenhaus: Across everything you test—IVR, voice, bots—where are you seeing the biggest failure rates?

G3: On the legacy side, voice quality and connectivity are major issues. People often focus on IVR flow logic, but the “plumbing”—audio quality, telco stability, routing—is critical and often overlooked.

On the AI side, the biggest issue is hallucination and model drift. Bots are asked to do something specific, but the model may wander or interpret instructions differently over time. That’s the number‑one challenge we see.

Sheri Greenhaus: Are companies still coming to you mainly for voice testing, or is AI becoming the dominant area?

G3: AI conversations are absolutely increasing. Many Fortune 50 customers still rely heavily on voice and IVR testing, but they’re actively expanding into AI.

Most organizations are in pilot mode with AI agents. They see the value, but enterprise readiness as testing, assurance, governance, is the gap. It’s similar to the on‑prem‑to‑cloud transition: it won’t be an overnight switch, but it’s inevitable.

Sheri Greenhaus: Let’s talk about self‑service IVR. We’ve all had the experience of pressing 1, 2, 3, and none of the options apply. You try to zero out to reach a human. Do you test how often the IVR fails to meet customer needs or how many loops it forces?

G3: Absolutely. We test loop counts, dead ends, and how often customers are forced to escalate. We’re the market leader in that area.

And we don’t stop at IVR. We test the entire flow from IVR, voice quality, routing, agent landing, and even the agent desktop experience. For example, we partner with NiCE on observability, so we can monitor what happens at the Voice PoP and agent desktop in terms of connectivity and experience.

Sheri Greenhaus: If your testing shows agents can’t hear customers, is there an alert to management in real time?

G3: Yes. Near‑real‑time alerting is a major focus for us. You don’t want the same issue affecting hundreds of customers before anyone notices. We’re building toward immediate detection and remediation.

Sheri Greenhaus: Where do you see testing going in the next few years? What’s next?

G3: Two major areas:

1. Production assurance Being present at the moment issues occur, feeding that data back into testing, and remediating immediately.
2. End‑to‑end, non‑deterministic journeys AI agents will be interlaced across channels. A customer might start in voice, move to an AI agent, transfer to a human in chat, then receive an authentication code on their phone. Journeys will be complex, fluid, and AI‑driven.

Cyara’s advantage is that we already test telco, WebRTC, IVR, chat, and AI. As AI becomes embedded everywhere, our ability to test the entire journey becomes even more valuable.

Sheri Greenhaus: IVRs now include speech recognition— “Tell me what you need”—and sometimes they misunderstand repeatedly. Do you test that too?

G3: Yes. Speech recognition has evolved from DTMF menus to mobile IVR to NLU intent‑based systems. Now we’re in the GenAI and agentic era. We test all of it; speech recognition accuracy, intent detection, and how well the system understands and responds.

Sheri Greenhaus: Bots are getting smarter and doing more for consumers.

G3: Absolutely. People are already chaining bots together for complex tasks, even planning a 10‑day trip to Iceland.  An eager enthusiast can now connect agents – not just deep nerdy developers.

Sheri Greenhaus: Fascinating as well as a little scary.

G3: Exactly. That’s why Cyara’s role is so important. Bots are powerful, but reliability matters. Testing is becoming strategic about trust, governance, and risk mitigation.

Sheri Greenhaus: You seem genuinely excited about this space.

G3: I am. There are many bot companies, but very few thinking horizontally about assurance across all bots and all channels. It’s a fascinating challenge.

Cyara works with leading AI platforms, OpenAI, Google, Watson, Meta Llama, Cognigy, NiCE, and more. We test them independently, which is important because you don’t want an LLM testing itself.

We also partner with major CCaaS vendors, Zoom, Amazon Connect, Genesys, and support in‑country dialing across 140 countries and 425 carriers.

Cyara can test:

• AI agents at the component level
• End‑to‑end journeys
• Load testing
• Production observability
• Telco and connectivity
• Human agent experience
• Business and KPI outcomes

Sheri Greenhaus: Ultimately, Cyara brings order to the chaos of today’s evolving contact centers, giving companies the confidence to innovate without compromising trust.