Measuring the True Cost of AI With Front
Tech Talks DailyNovember 17, 2026
3791
22:4917.92 MB

Measuring the True Cost of AI With Front

What does AI really cost when a cheap model fails, retries the task, hands the work to a person, and leaves the customer returning with the same problem?

In this episode of Tech Talks Daily, I welcome back Dan O'Connell, CEO of Front, to discuss AI tokenomics and why token consumption is a poor substitute for business value. Our previous conversation focused on Bring Your Own Agent and the need for companies to choose the agents that fit their customer experience. This time, we turn to the economics behind those decisions.

Dan describes tokens as the new cloud bill. As businesses deploy agents across a growing number of customer interactions, usage-based charges can rise quickly. That number may show activity, but it does not explain whether the work created a useful result. A high token count can represent productive automation, repeated attempts, unnecessary complexity, or a task that eventually required human intervention.

His preferred measure is the total cost per successful outcome. In customer support, that means looking beyond the model's headline price and accounting for every part of the journey. Model retries, agentic retries, human review, handoffs, coordination, response time, and the final customer result all affect the true cost to serve.

The idea matters because lower model prices can be misleading. A cheaper model may save money on the first attempt while increasing the number of retries or transfers. A larger model may cost extra per request while completing a difficult task with fewer steps. Leaders therefore need to compare operating results rather than price lists.

Dan explains how model routing can help. Simpler tasks may be directed to smaller models, while difficult work can be sent to models with greater capability. The choice depends on the task and the tradeoffs the business is willing to accept. Dan frames those tradeoffs through three dimensions: quality, speed, and cost. In many cases, a company can optimize two while accepting a compromise on the third.

That does not mean every basic task should automatically use the cheapest option. A slower response may be acceptable for internal analysis but frustrating during a live customer conversation. A small reduction in accuracy may carry little risk in one workflow and create expensive repeat contacts in another. The orchestration layer needs rules informed by the customer experience, the complexity of the request, and the cost of failure.

Customer service gives companies a useful environment for measuring those results because many of the existing measurements already connect work with outcomes. Cost to serve, time to resolution, deflection, accuracy, customer satisfaction, and repeat contact rates can all reveal whether AI is helping. Dan argues that agent performance should be evaluated alongside human performance using comparable standards.

Resolution is particularly important. If an AI system provides an answer and the customer returns an hour, a day, or a week later with the same issue, the task was not resolved. The first interaction may look successful in an activity report while creating extra work, frustration, and cost later. Measuring only initial response or deflection can therefore reward the wrong behavior.

The conversation also considers what automation means for the people working around customers. Front's research, as discussed in the interview, found that 54 percent of surveyed leaders see issue resolution as support's main purpose today, while 40 percent believe it should remain the main purpose. Dan interprets that difference as an opportunity to automate lower-level work while allowing employees to spend additional time building customer relationships.

The research also reports that 75 percent expect support, customer success, and account management to become increasingly unified. Dan believes this will affect team structures and the platforms companies choose. Customers do not think in internal departmental boundaries. They expect one company to remember their history and coordinate the response, whether the work passes between an agent, a support specialist, customer success, or account management.

Shared context and coordination therefore become part of the AI cost equation. A handoff without the customer's history can force someone to repeat the problem, reducing satisfaction and extending resolution time. The technical cost of the interaction may appear low while the operational and customer costs continue to grow.

Dan recommends treating AI autonomy as a dial. Companies can begin with a person reviewing recommendations, maintain audit logs showing why the system acted, and increase autonomy as evidence builds. The same outcome measurements applied to employees should also apply to agents, including whether the customer received an accurate answer, achieved a lasting resolution, and had a good experience.

The larger message is that AI economics cannot be separated from service design. A model, router, agent, or platform earns its place by contributing to an outcome the business and customer value. Usage may help explain the bill, but it cannot prove the result.

Should companies continue celebrating rising AI usage, or start asking what each successful outcome cost and whether the customer would agree it was successful? Listen to the episode and share your thoughts.

Useful Links

[00:00:03] What does an AI system really cost once retries, human intervention, handoffs, latency and failed resolutions all enter the calculation? Well in today's episode I welcome back Dan O'Connell, CEO of Front onto the podcast. But this time we're going to have a conversation about AI tokenomics and why token consumption tells us very little about business value.

[00:00:32] Because he's going to explain why companies should be measuring the total cost of successful customer outcome, how model routing can match simpler work with lower cost models, and also where quality, speed and price all force difficult trade-offs. Because we're going to discuss why an accurate answer is just not enough if that customer returns with the same problem.

[00:00:58] And also why AI efficiency should create stronger customer relationships rather than just another impressive usage chart. Yeah we've got a lot to talk about today so enough from me. Please join me in welcoming Dan back to the podcast now. So thank you for joining me on the show today.

[00:01:20] For anyone that missed our previous conversation, can you tell them a little about who you are, what you do and how your work at Front is connecting AI with customer service? Yeah it's great to be here Neil. I'm Dan O'Connell the CEO of Front. And then for those that may not be familiar with Front, we're a customer support platform. We focus on businesses that have complex needs. And so we can automate requests that come in from customers entirely.

[00:01:45] We can also be the platform for where the humans are getting involved for escalations and high value relationships. And so really it's an end-to-end solution. I'm excited to be here and share a little bit more about our story. Well it's a pleasure to have you back on the podcast. Last time you joined me we were talking about bringing your own AI agent. So much gold in our urge. People listening to go back and check that out if you've not listened to it. But one of the reasons I was excited to get you back on was to discuss tokenomics.

[00:02:13] Because I read that you've described tokens as the new cloud bill. And I think that's something that many people listening will resonate with. So why can token consumption become expensive so quickly? And why is volume alone a poor measure of value? We've all seen those new stories a few months ago of measuring the value that way. I think we all agree that that's not the way. But tell me more about that. Yeah I think to no surprise look like a lot of businesses are deploying more and more agents today.

[00:02:40] And so a lot of those businesses that provide agents are billing on usage. And so one is the more places that you're appointing an agent. Obviously it's going to rub up the usage bill. And so then that naturally goes into like understanding the tokens that are powering that usage. And the reason I think that's not the right way to measure things is ultimately it's not talking about the impact. And I don't think that's terribly surprising. We see a lot of articles around this of how do we understand the true value of AI as it's deployed.

[00:03:08] And for us in the customer experience space, the customer support space, like I always go down to like can you go and measure and understand the outcome and the cost to an outcome? Because ultimately that's what you want to solve for. You know today in a customer support organization your cost to serve, you know your time to resolution, you know your deflection rate, like everything is pretty well measured. And so fortunately I think we play in a space where we can clearly define and provide that ROI.

[00:03:37] But again, it's really pushing people to think about the ROI that they're getting and the value that they're getting from these tools as opposed to being focused on just the usage and the spend that comes out at the top line. I must admit I found it incredibly refreshing when I was reading how you argue that companies should be measuring total cost per successful outcome. I mean, it's so obvious. I think it's what everyone was thinking. It was great to hear you saying this in the public arena.

[00:04:02] But I suppose the other question is what should that calculation include beyond just the model's headline price? Yeah. So a couple of things. I think it's got to take into account, you know, are there model retries or agentic retries that are happening on that? It's got to obviously take into account if there's any human intervention that goes into it. So again, I don't think it's just being based on like, hey, here's the token spend for this one action. There's going to be multiple action. There's going to be a human handoff.

[00:04:30] Perhaps there's human intervention that goes back to the agent. And so you've got to look at that total cost to serve piece in that and understand some of the coordination that's happening behind the scenes. And as my granddad used to say, you buy cheap, you buy twice. So there is a phrase that, you know, a cheaper model, it can become costly when it fails, retries a task or needs human connection.

[00:04:53] So how should companies compare models using real operating costs rather than just a price list that has been maybe gamifying the system somewhat? Yeah, I think what's really unique and important at this moment in time is you have flexibility in models. And different models can excel at different tasks. And so you hear a lot of people talk about routers these days. And routers, again, is how do I go pick a cheaper, simpler model to go do a more basic task?

[00:05:19] And how can I make sure that I'm using a complex model for something that's more difficult? And so, again, I don't think it's like rocket science. Like, these are very practical applications. I think we all got so excited about the AI transformation two and a half years ago that we kind of rushed at how do we go implement and not necessarily think through all of these things. And so I think naturally every business is now thinking about optimization. So what is the task at hand? What's the complexity of that task? If it's more simple, I can go use a mini model. Right.

[00:05:49] Perhaps I'm willing to give up some speed or some latency because it's going to be save me on some costs and some dollars. And then, again, having more flexibility to say, hey, here's this really complex task. I want to go point out at a larger model. Right. That's more complex. That's going to have a higher cost associated with it. And so, again, I think you want to have experts within a business that are understanding some of this optimization. And really understanding what are the tasks and what's really needed for those tasks.

[00:06:17] And, of course, customer service does give companies a direct way of connecting AI spending with real results. But for many people listening, it's knowing what to measure is the sticking point. So which measurements or metrics reveal whether that customer is actually receiving the right outcome? What are you seeing now? Yeah, I think, look, there's internal metrics that you want to measure before you deploy AI.

[00:06:41] So, you know, if you think about building an agentic first experience for customer support, you want to make sure and run your own tests and evaluations to say, hey, if these types of questions come in, am I happy with the agentic response? So there's obviously accuracy is number one. You want to make sure that people, when they ask a question, you're going to get a great experience that's accurate. Ultimately, at the end of the day, I always go back to the customer experience. And so that ultimately means their customer satisfaction.

[00:07:10] And ultimately, did you resolve the task for them? Because if you've resolved it, but they come back a week later or a day later or an hour later, you haven't resolved anything. You've probably just introduced more friction into the process. And so I think a lot of times it comes down to, hey, if you're measuring your team's output as a human, right? We talk about customer support. We measure in customer satisfaction. You want to have that same measurement applied to all of the agentic experiences the same.

[00:07:37] And so you want to be able to go and look across the team to say, hey, when I have a human, here's what the engagement and satisfaction looks like. When I have an agentic first experience, here's what the agent is doing and how they are also stacking up against the humans, right? But ultimately, delivering hopefully a comprehensive experience to the customer. And you mentioned a few moments ago how companies are beginning to route simple work to faster models, complex work to more capable ones.

[00:08:03] Any tips on what decisions the orchestration layer would make on their behalf or what should the orchestration layer make on their behalf? Yeah, again, I think it comes back to the business making sure that they are taking the time to think through these cost optimizations. I think one of the best ways to control cost for AI is really getting into the weeds of these things. And that means that you clearly understand the tasks at hand.

[00:08:28] And you also understand the technology to say, hey, here is a task that's relatively simple that might require a mini model. And when you pick mini models, there's dimensions that you give up. I naturally kind of think there's three dimensions that you get from models. You get quality so that it can be measured in terms of accuracy. We talk about speed in terms of latency. So how quickly can a model get your response back? And then that last dimension is cost. I usually think you get two out of the three when you pick models.

[00:08:58] And so there's always going to be tradeoffs. If you're picking a model, it's not a top end model, I would say like, but you're usually getting two of the three. And so there's always going to be tradeoffs when you think about the task. Hey, this is a basic task. I'm going to save some costs by picking a mini model. That might mean it's a little bit more slower to get a response. Perhaps there's like some tradeoff in terms of accuracy as well. And I think like everyone needs to be really well adept in understanding those tradeoffs and thinking about the evaluation of it.

[00:09:24] And before you join me today, I also saw how you surveyed, I think it was CEOs, founders and customer leaders at companies, including OpenAI, OpenRouter, ObserveAI, Dialpad and so many others about what they're seeing out there and how they see their own role evolving as a CEO yourself. I'm interested in your thoughts here. But one of the things that I did pick up was that 54% of those leaders see issue resolution as support's main purpose today.

[00:09:53] But on the flip side of that, only 40% believe it should remain that way. So what is it that is replacing that traditional support mandate? Yeah, I think what was really interesting on that survey, and as I said, we did that across, I think, some of the most cutting edge companies that are out there when they think about support is like, they wholeheartedly naturally think that like, look, over the next few years, we can automate a lot of the low end tasks that come in.

[00:10:19] And so that means that they don't immediately then say, look, I need fewer people around the customer. I think it's really around they want to transform them the customer experience. Managing our own business today, I constantly think about this. We have 9,000 customers globally. How do we have stronger relationships with the right customers? And so naturally, I think there's like this evolution that is happening. And you can see it from folks that are perhaps on the cutting edge or leaning into more of AI automation.

[00:10:47] But it is, look, they want to go and see support transform. And I think you also said that means spending more time with building relationships. But I think the other piece that's playing out is you start to see a blurring between the lines of support and customer success and account management. And I think that blurring of the lines of roles and responsibilities is not just happening around teams that service the customer.

[00:11:10] You can see that playing out in engineering product and design as well, where how we build software and the teams also blurring on that. And so I think that's also a dynamic that is starting to play out. Another standout stat there in the research was 75% expect support, customer success, and account management to all become more unified. So what does that mean for team structure incentives and accountability?

[00:11:36] I'd imagine that's what many customer service managers are thinking about when they hear stats like this. Yeah, I think one of the biggest things is I think it brings a lot of questions around the platforms that people use. And you think about like the traditional way that we've built companies is we pretty much align functions around specific tasks or units of work.

[00:11:54] And I think that gets completely flipped on its head when we think about agentic experiences that no customer ever thinks about, hey, here's how you have the support team organized and here's how you have the customer success team organized. But we've always done it that way. And so I think the whole way that we think about providing a single customer experience is one company changes. I think that means that those teams are going to consolidate to fewer platforms.

[00:12:19] It's why we talk a lot about our belief at front is we want to unify your customer facing teams on a single platform. We want it to be a place where humans and agents can get this customer work done and for that customer experience to feel like it's coming from one company. And I think that wholeheartedly is one of the dynamics like we believe is playing out in the market is people are going to unify around fewer platforms. And that's partly because you start to see these blending of these teams together.

[00:12:47] And obviously, you're someone that is living and breathing this space at the moment. And I'm curious when you look through that research, did anything stand out and surprise you at all? Or are you so used to seeing everything that nothing surprises you anymore? Or were there a few surprises in there? Yeah, I think there's always going to be surprises.

[00:13:05] I think some of the things that you can see different dynamics within different verticals, you know, tech B2B has some really interesting, as I say tech B2B, but technology vertical as a whole, I think tends to be like the canary in the coal mine in terms of potentially what the future can look like. And that's very different than perhaps some more traditional industries that are out there. And so naturally, anytime we do some research, you can find our research at research.front.com. I think there's very much those interesting trends that play out.

[00:13:34] Different dynamics across these different industries. How much they believe in an AI first touchpoint, first perhaps want to take a slightly slower approach to the rollout and application of AI. I was talking about roles converging a few moments ago. And I would imagine as these roles converge, customer platforms will need to give teams and AI agents the shared context. And we mentioned that word heavily in our last interview.

[00:14:00] And also workflows that work effectively across those traditional boundaries. What are you seeing here? Yeah, I think context is, you know, there's two things that I think have become increasingly important in an agentic world. And again, this is a world where more agents are going to show up and be proliferate, you know, have some proliferation. And so I think, look, the shared context and then coordination of that ultimately is going to have the biggest impact on a customer experience.

[00:14:26] The most frustrating thing for any customer comes down to if they have to start over and repeat themselves. And the reason they have to do that, right, if I answer the first call and I pass it and I pass it to Neil, perhaps I didn't provide you with notes and context. That customer is now starting over. And so, again, I think this becomes like really the critical part of a customer experience is how do you make sure that agents and humans all have the shared history? They all have the shared context and then you're able to coordinate that handoff in a seamless way.

[00:14:55] So the customer experience feels like one experience from one company for them. And so we get really excited. Our DNA has been around coordination when we first started front that we thought customer work is better when more people can be around it. And that stays true today, that mission, but it's slightly evolved in terms of it's now people and agents and they want to have a seamless coordinated experience across that customer experience.

[00:15:19] And if we do have anyone listening that are concerned about a risk that an efficiency program could lower costs whilst damaging service quality and trust, are there any controls that you could share that would help leaders recognize a problem like that before customers start leaving? Yeah, I think number one is anytime you're thinking about implementing AI is making sure you understand clearly the governments and the control that you have for that.

[00:15:44] We naturally think and encourage us even in our own AI deployments, we think of AI in terms of a dial. Can I start off and make sure that a human is always in the loop? Do I have full audit logs and understanding of like why AI made a recommendation and not? But then I think it's really important to make sure that you're measuring the outcomes that are happening from AI. We talk a lot about the customer experience, customer support place where traditionally we will send out surveys. We can now do some pretty magical things like inferring customer satisfaction from an experience.

[00:16:13] And so, again, if you're doing that for humans, you want to make sure that applies to agents. And so you understand at the end of the day, when an agent gets involved, are they are they powering or providing a great experience? It's not just about the accuracy of the answer. It's ultimately did they resolve the issue for the customer and did they have a good a good experience with that? Yeah.

[00:16:35] But at the very beginning of our conversation, we're talking about how you once described tokens as the new cloud bill and AI is redefining the role of customer support. And also the world of support is evolving so rapidly at the moment. I'm curious from all the conversations you're having, everything you're seeing and hearing, what excites you about this evolution and the art of the possible maybe? Yeah, I think the biggest art of the possible gets into, I think, openness wins in terms of opportunities.

[00:17:03] We can see that more agents are going to show up. And so I think people are consolidating to fewer platforms. They want to be able to bring their own agents to those platforms. And so we're super excited about that. We've had some interesting announcements that are coming up and have made around the ability to choose our native agents, use an agent from a third party or build your own to go empower customer experiences. And so I'm excited about that trend.

[00:17:27] I'm also excited about the opportunities of agents to provide a seamless experience across every channel. And again, I go back to telling these stories of customers get frustrated when they have to restart their problem and start the same story to a different person. And I think there's opportunities for that all to go away and to really unify humans and agents together to provide a seamless single experience from a company.

[00:17:51] And again, if I was to ask you to put all those talking points, all the conversations you're having with clients and customers and inquiries, when they're asking questions around connecting AI costs to business value and so many other areas around AI, are there any trends in the kind of things that people are coming to you and asking for your help with or speaking to you on the show for all of a busy tech conference? Are many people asking the same things or are there any big trends there? Yeah, I think it really is all over the place.

[00:18:21] And I think that's why it's such an interesting time. And I think everyone comes in and will ask you questions from, hey, how should I be thinking about implementing AI in my customer journey? Should it always be an agentic first touch point? Should I always have an AI handoff? Should I have a human first touch point? And so I think these are those moments and times where businesses, one, believe in the opportunity that agents can provide a seamless experience to their customers and do a good job.

[00:18:49] And so I think the vast majority of businesses recognize that's a very real opportunity and experience. I think that means that they can transform and elevate the teams they have around the customer and they're excited about that. It allows them to go and build deeper in-person customer relationships. And so, again, I think the challenge is businesses don't always know where to start. And so they naturally will come to us with some guy asking for some guidance and help.

[00:19:14] And again, I go back to it depends, depends on the type of business, depends on the type, the complexity of the questions that they're getting, depends on the type of experience they want to provide. And so we want to be seen as a trusted partner to them to help think through those questions in a really thoughtful way and tailor it to the experience that they ultimately want to provide as a brand. And I think, you know, that's ultimately for the customer in the company to go and figure out, like, what is that type of experience they want to deliver?

[00:19:40] For everybody listening, I'll include a link to our previous conversation, Bring Your Own Agent, a fascinating discussion. I'd urge people to check that out. But also for people listening today interested in exploring AI tokenomics, the changing role of customer support, exploring the art of the possible. We've explored so much today. But where would you like me to point everyone listening? Yeah, you can find us at www.front.com if you're interested in some of the research that we've done,

[00:20:07] which I think is super interesting across the customer support space and customer experience space. You can find that at www.research.front.com. I'm fairly active on LinkedIn. You can always reach out to me and say hello. And just want to say thanks for having me, Neil. This has been super fun and great. No, it's a huge pleasure for me to have you back on the podcast today. We covered a lot in a short amount of time. And I think we'll need to get you back on next year to complete your hat trick of appearances and see how things are evolving. Maybe Southampton will secure a promotion. Maybe.

[00:20:37] We're going to do some real work. Look forward to speaking again. Thanks again. Thanks. Take care. I think Dan's argument gives a timely reminder for leaders to question the AI bill. Token spend is an input, while resolve customer problem. That is the result the business actually needs.

[00:20:57] So measuring the distance between those two points means accounting for retries, model choice, human handoffs, response times and customer satisfaction. And whether that issue remains truly resolved. And that means resisting the temptation to send every task to the largest model, when sometimes a smaller one may deliver the required result at a much lower cost.

[00:21:24] So thank you to Dan for returning to the podcast and connecting AI economics with the daily realities of customer service. So head over to techtalksnetwork.com. There'll be a blog post associated with this episode. And you'll find all the links and more to everything that we talked about today. But before I go, a question for you to take away and ponder.

[00:21:48] If your AI usage doubled tomorrow, could you realistically show whether customer outcomes had improved with that increase in cost? Let me know. techtalksnetwork.com. So a huge thanks to Denodo for supporting the Tech Talks Network, helping us produce more than 60 interviews a month. And when it comes to trusted data products, it all starts with the right foundation. And trusted data products start with Danodo.

[00:22:16] Because they can help you create, manage and deliver business-ready data products faster with secure real-time access across all of your data sources. And you can learn more by simply visiting danodo.com. Time for me to go now. I'll be back again real soon with another episode. But thanks for listening as always. Bye for now.