The Adults in the AI Room – Episode 102: Why AI in the Contact Center Is Harder Than You Think with Michael Darwal

From Cost Per Minute to Cost Per Resolution: How Ibex Is Rewriting the Rules of Customer Experience with AI

Guest: Michael Darwal, Chief AI and Digital Officer at Ibex

Host: Seth Earley, CEO at Earley Information Science

Published on: November 5, 2026

 


Welcome to the Adults in the AI Room, brought to you by VKTR.

In this episode, Seth Earley speaks with Michael Darwal, Chief AI and Digital Officer at Ibex — a global customer experience management company with nearly 40,000 employees operating 31 contact centers across the US, Jamaica, Nicaragua, Honduras, the Philippines, Pakistan, and India. Michael shares why the promise of AI automating 70% of contact center work has not played out as predicted, what separates successful deployments from the 95% that fail, and how Ibex's WaveIX platform is delivering a 50%-plus reduction in cost per resolution by marrying best-in-class AI technology with two decades of customer journey data.

Key Takeaways:

  • Over 95% of AI contact center implementations fail to deliver on their initial ROI promise, often due to misaligned objectives and poor understanding of the customer journey.
  • The binary framing of "AI handles the call or a human does" is fundamentally wrong — success lives on a spectrum measured by CSAT, resolution rate, and cost efficiency.
  • Decades of historical call data are a competitive moat: mapping customer journeys before deploying AI agents dramatically improves targeting and outcome prediction.
  • The pricing model for the industry has shifted from cost per hour to cost per resolution, creating 50%-plus savings opportunities for clients deploying AI agents effectively.
  • AI-augmented agents achieve 40% faster time-to-proficiency, allowing Ibex to recruit from a wider talent pool and get agents to full effectiveness more quickly.
  • Agent satisfaction has risen alongside AI adoption — Ibex's employee NPS climbed from 77 to 82 as agents found their work more manageable and rewarding with AI tools.
  • AI is now helping build the knowledge bases it depends on — using tools like Sierra's Ghostwriter to compile and structure disparate client data into actionable agent reference guides.

Insightful Quotes:

"Over nine out of ten — 95% of AI implementations fail to deliver on the ROI or the promise of what they initially went out to achieve. And that's for a variety of reasons, but I'd highlight a misunderstanding from the get-go of what AI is trying to accomplish or what it's able to accomplish." - Michael Darwal

"Success doesn't just have to mean the AI agent handled 100% of it. I took what used to be a 20-minute handle time call, and that AI agent did the front end — gathering information, verifying, summarizing — and turned it into a seamless handoff. You could shave off 50% of the time." - Michael Darwal

"Search is the retrieval layer for AI. The agents are only as good as the knowledge behind them — and what it takes to build and maintain that knowledge base at scale is the foundational work you can't shorten, regardless of how capable the model is." - Seth Earley

Tune in to hear how Ibex is proving that AI in the contact center isn't a switch you flip — and what it actually takes to move from failed experiments to measurable, sustainable results.

And a final thank you to our sponsor, VKTR.


Links

LinkedIn: https://www.linkedin.com/in/michaeldarwal/ Website: https://www.ibex.co

Ways to Tune In:

Adults in the AI Room Podcast: https://www.earley.com/earley-ai-podcast-home Apple Podcast: https://podcasts.apple.com/podcast/id1586654770 Spotify: https://open.spotify.com/show/5nkcZvVYjHHj6wtBABqLbEiHeart Radio: https://www.iheart.com/podcast/269-earley-ai-podcast-87108370/ Stitcher: https://www.stitcher.com/show/earley-ai-podcast Amazon Music: https://music.amazon.com/podcasts/18524b67-09cf-433f-82db-07b6213ad3ba/earley-ai-podcast Buzzsprout: https://earleyai.buzzsprout.com/

🏆 Named one of the Top 97 AI in Business Podcasts by FeedSpot.

You can also find us on dLogos at dlogos.xyz/podcasts/earley-ai-podcast-271271ce — suggest guests, topics, and questions to help shape future episodes.


Podcast Transcript: Why AI in the Contact Center Is Harder Than You Think

Transcript introduction

This transcript captures a conversation between Seth Earley and Michael Darwal about why the majority of AI contact center deployments fail to deliver on their promise, how Ibex uses decades of customer journey data to design better AI agent deployments, and why the shift from cost-per-minute to cost-per-resolution is transforming the entire industry.

Transcript

Seth Earley: Welcome to what was formerly the Earley AI Podcast, now rebranded as the Adults in the AI Room. This is our first episode with the new branding. It's brought to you by VKTR. I'm your host, Seth Earley, and in each episode, we're going to be talking about how business and operations are being reshaped by artificial intelligence.

Today, we're going to be talking about AI in the contact center, a space where the stakes are really immediate, and the high volume and expectations from executives and customers are very different. When generative AI arrived, the dominant narrative was that it was going to automate away most of the traditional call center work. What's happened is definitely more nuanced and much more interesting. So joining me today is Michael Darwal, who is Chief AI Officer at Ibex, a global customer experience management company with nearly 40,000 employees, operating 31 call centers across the US, Jamaica, Nicaragua, Honduras, the Philippines, Pakistan, and India. Michael has spent 13 years at Ibex across finance, digital marketing, and now AI, building the company's WaveIX platform for seamlessly integrating AI agents into traditional customer experience workflows. Michael, welcome to the show.

Michael Darwal: Thank you so much, Seth. It's a pleasure to get to speak with you, and I appreciate getting the opportunity.

Seth Earley: Absolutely. So, let's start off with some of the common misconceptions about AI and the call center. When generative AI first arrived, the dominant narrative was that it's going to automate away 70% of traditional call center work. But what has actually happened, and what do executives still get consistently wrong about what AI can do in the space?

Michael Darwal: That's a great question. And I think if we take a step back — when generative AI really started coming to the forefront a few years ago, people were under the misconception that it would almost immediately go after 70-plus percent of contacts. The reality is that MIT has published a study, and there have been a number of studies, but over nine out of ten — 95% of AI implementations fail. When you say fail, it means fail to deliver on the ROI or the promise of what they initially went out to achieve. And that's for a variety of reasons.

I think one of the ones I'd highlight is just a misunderstanding from the get-go of what AI is trying to accomplish or what it's able to accomplish. Then how it really fits into the customer journey, and are we delivering a better result for the customer on the phone, as well as for the agents, and as well as for the company deploying this? And also, are you getting a true resolution through the deployment of AI, or are you simply measuring a single point in time and then getting a contact that recurs later on?

Seth Earley: Yeah. And as you're saying, many organizations are really looking at AI in the contact center as a binary choice — either AI handles the call, or a human does. But why is that framing incorrect? What does looking at this as a spectrum produce that binary thinking does not?

Michael Darwal: When we look at it as a binary, that's just a pass-fail: did an AI agent handle this all alone, or did that person eventually get transferred to a human agent, or did they have to call back because they were dissatisfied? And I think that's where a lot of the failure to produce initially estimated results comes from.

Looking at it on the whole spectrum means: are we trying to accomplish all three of our goals, or at least one? The three goals I would think about are: is the customer getting the answer they want or resolution faster and more effectively than without AI? Are we doing this in a way that produces positive ROI — saving money, serving more customers with the same number of agents? And are we resolving this the first time, or are they coming back? Those are the real measures. The binary framing misses all of that.

Seth Earley: So that makes a lot of sense. You're really facilitating the whole experience, and you're injecting AI assistance at various parts of that — gathering basic information, passing it to a call center rep, looking up knowledge and having answers in an anticipatory way, making the experience more effective by making the agent more effective, and of course, avoiding the need for a human agent if possible. Let's also talk about what's driving the 95% failure rate and what the successful deployments have in common.

Michael Darwal: Some of what's driving those failures is a disconnect between systems, a disconnect or misunderstanding of what the ideal customer journey looks like, and perhaps a failure to align what the AI agent's objective is with what the organization's objective is.

For example: if you had a subscription service and you automated canceling very easily — great, you could automate every single contact where somebody says they want to cancel. The AI agent just says, "Great, we'll cancel your service right now." But that's not a good business outcome, because the agent hasn't been trained to probe: why is this person canceling? What could be done to save them? How long have they been a customer? Is this someone worth offering a discount to?

Alternatively, for complex calls with 20- to 40-minute handle times, success doesn't have to mean the AI agent handled 100% of it. It could mean I took a 20-minute call and the AI agent did the front end — talking to the consumer, gathering their information, verifying them, accessing their account, clearing up the first four minutes — and then the human agent, because they already have everything summarized and queued up, resolves the remaining steps in eight to ten minutes. You could shave off 40 to 50% of the time. That's a better customer experience and a significant cost saving.

Seth Earley: And so that's what a successful implementation looks like. Tell us about the formula.

Michael Darwal: The key is using best-in-class technology and marrying it with unparalleled know-how, expertise, and analytics around what the customer journey should look like. On the Ibex side, we partner with what we believe is a leading conversational AI technology company in Sierra AI. Their technology is incredible — intuitive, empathetic, great voice. But what we bring is the historical understanding from having handled millions of calls over two decades: what does this customer journey look like, mapping it out with our business insights and analytics teams, looking at all the tools and technologies that have to be integrated, and mapping out the ideal situation, the middling situation, and what to do when it doesn't go well.

And we don't try to boil the entire ocean. We say: these 30% of your calls are going to be easier to automate. Let's start there. Maybe we replace the upfront IVR with an AI agent just to get routing done more effectively and efficiently. Then let's parse out that 30% of volume that's easiest to automate, send that to the AI agent, and start capturing information and transferring the rest. The key is using the best technology and marrying it with the expertise and insights as to what this customer journey should look like.

Seth Earley: So let's talk more about that. You have decades of customer journey data across all sorts of interactions and intents. How does that change what you can do with AI, and what are organizations missing when they try to deploy an agent without that?

Michael Darwal: By looking at all that data and insights — and then looking at the client-specific data — we can say: this percentage of your contacts look like this type, these have a shorter handle time, these are likelier to be easier to resolve. Let's listen to those, understand how they get QA'd, and identify which ones are most ready to automate. And then let's look at where there's high customer dissatisfaction — where customers have to get transferred through multiple agents, where there are long wait times while things get verified.

How do we start automating: tying into systems to authenticate users faster up front, allowing the AI agent to capture information and queue it up, query through a knowledge base, so when and if they do get transferred to a human agent, all of that is already there. All of that leads to a much better starting point: setting reasonable expectations, knowing what you should be targeting, and then knowing what comes next.

Seth Earley: And so that journey mapping is so important — understanding all the data sources at each point in the journey, what the pain points are, what customers need to accomplish, what they need next. Anticipating that, and making sure you're surfacing the right data to the customer, to the call center agent, or to the AI tool in that context.

One of the challenges here is that the traditional business model for call centers was really based on price per minute, price per hour. AI resolves issues faster. That's in tension with that model. How has the pricing model evolved, and how does the shift to cost per resolution change the industry?

Michael Darwal: Historically, the vast majority of traditional BPO or CX interactions were measured on a cost per hour, cost per minute, or cost per contact basis. Today, when we deliver an AI agent solution, it's cost per resolution. We look at: what did your typical cost per resolution look like based on average handle time and your usual bill rates across onshore, nearshore, and offshore deployment — and what does it look like using either standalone AI agents or AI agents integrated with human agents?

What we're seeing is 50%-plus savings on a cost-per-resolution basis for most of the clients we work with. And while there's definitely an impact to traditional BPO revenue on the agent side, we're not seeing that adversely impacting us, because we're able to absorb more calls or contacts than before with the same number of human agents. That's creating revenue opportunities — by being a leader in driving these AI agent solutions, we're capturing more volume from competitors.

Seth Earley: And then others have to compete with you on that basis. There's no turning back. You've seen AI increase agent throughput by 40%, enabling less experienced agents to handle more complex interactions. What does that mean for how you recruit, train, and develop people?

Michael Darwal: Through leveraging technology to make the agent more powerful faster, it allows us to recruit from a wider base, train much faster, and get agents to proficiency more quickly — which also leads to happier agents. We went from an employee NPS of 77 — which we already believed was best-in-class for a BPO — to 82 in the most recent data published at the end of our fiscal year. That's a combination of how we treat our employees and create opportunities for them, but a lot of it is also that the job is becoming more enjoyable because of these tools and technologies. Better throughput from recruiting to floor, more effective agents, better retention, lower churn. An employee NPS of 77 for a BPO is off the charts — 82 is uncharted territory.

Seth Earley: Coming back to entry-level employees: contact centers have been where people get a lot of their first professional experience, develop foundational skills. As AI handles more of the work, where does the next generation of experienced operators come from?

Michael Darwal: I still think it comes from people who have had that experience on the floor, because we have a great track record of taking people from agent to supervisor to floor manager to director. We have country managers who started as agents. What we do is leverage the understanding from agents who move on to QA, business insights and analytics, and workforce management — people who understand how AI technology is benefiting agents firsthand, whether from their own time on the floor or from managing those teams. A lot of the tools and technologies we develop internally for our AI agent lifecycle have come from people who started as agents. Those are the people who best understand how the technology is actually impacting them.

Seth Earley: One of the things we talk about a lot — and it's very important for any of these things — is that AI agents are only as good as the knowledge behind them, and search is the retrieval layer for AI. What does building and maintaining a knowledge base look like at scale? What is the foundational work you can't shorten, regardless of how capable the model is?

Michael Darwal: What's becoming exciting is that with the strength of AI models being developed now, we don't have to wait for a knowledge base to be perfect before deploying. We can actually help compile and build knowledge bases for AI agents faster by saying: let's look at all the historical data you have on calls and contacts, the entire customer journey. Let's look at whatever segments of knowledge bases you have, and access them using tools like Ghostwriter through Sierra, or tools we've developed internally, and say: how does this actually map out for good customer experiences? Let's recompile and rebuild a quicker reference guide for those knowledge bases.

AI is helping us utilize whatever disparate systems clients have to build a more complete knowledge base faster, rather than having to go into each individual system and build it the legacy way.

Seth Earley: I think this has been a great discussion. Thank you very much for joining me and helping our audience understand why AI and customer experience isn't just a switch you flip — and what it takes to get this right. To our listeners, thanks for tuning in to the Adults in the AI Room. Be sure to subscribe for more conversations about how AI is shaping the future of business. And thank you to our sponsor, VKTR. You can also find us on dLogos — the link is in the show notes. Michael, where can people find you?

Michael Darwal: They can find us at ibex.co. I really appreciate the time and the opportunity today, Seth. Very thoughtful questions, and happy to reconnect anytime.

Seth Earley: Sounds great. Thank you again. Bye now.

Michael Darwal: Thank you, sir. Bye bye.

 

 
Meet the Author
Earley Information Science Team

We're passionate about managing data, content, and organizational knowledge. For 30 years, we've supported business outcomes by making information findable, usable, and valuable.