Is AI Actually Creating Value?

AI is becoming a bigger part of how companies operate, but measuring its actual value is proving more complicated. Nasdaq’s Michael Normyle joins the IBKR Podcast with Jeff Praissman to discuss how traditional software metrics are evolving in the age of agentic AI, what new measures could emerge, and whether productivity could become a key way investors evaluate AI-driven value.

AUDIO PLACEHOLDER

Summary – IBKR Podcasts Ep. 422

The following is a summary of a live audio recording and may contain errors in spelling or grammar. Although IBKR has edited for clarity no material changes have been made.

Jeff Praissman

Hi, everyone. This is Jeff Praissman with Interactive Brokers. It’s my pleasure to welcome back to the IBKR Studio Michael Normyle from Nasdaq. Hey, Michael. How are you? 

Michael Normyle

Doing well, thanks. Glad to be back. 

Jeff Praissman

I have the pleasure today of going over another one of the great articles that’s on your website, “The Search for New Measures of Value in the AI Era.” As listeners remember, last month we did a podcast on a similar subject, but this is sort of a further take on it. So, Michael, historically, investors have focused on metrics like seat growth, ARR, and net revenue retention. 

But this article brought out some pretty interesting points that these measures might become less meaningful in the agentic AI world. 

Michael Normyle

Yeah. Just to give everyone a quick overview here, this is our second entry in our AI series as part of the launch of the Nasdaq Economic Institute. In that first one that you referenced, we were talking about AI and how it’s impacting entrepreneurship, and now we’re looking at how it’s impacting the way that companies are explaining their value during earnings calls. 

And this has been a big shift, especially for software companies, because historically, those software-as-a-service companies, or SaaS companies, defined their value by metrics like the number of seats. That’s basically the number of users, since that was directly linked to revenue, right? The more seats they had, the more revenue they had. 

They also reported annual recurring revenue, the ARR that you mentioned, which measured the recurring revenue over a calendar year, and then net revenue retention, which measures the percentage of recurring revenue retained and grown from existing customers over some period. All these help capture churn and upsells and stuff like that. 

In the agentic AI era, though, these measures become less relevant because the value provided isn’t necessarily tied to the number of users. It’s more about the outcomes and the productivity that the product creates. So it could actually be unrelated, or perhaps even inversely related, to the number of seats. Maybe now you have one finance professional who can run an AI agent to automate profit and loss statements, cash flow analysis, bookkeeping tasks, things that used to occupy an entire team. And so that’s precisely why we need new metrics for these companies to define the value that they’re creating. 

Jeff Praissman

Yeah, I should mention that the name of the article is “The Search for New Measures of Value in the AI Era,” and that will be linked to this podcast as well. As always, you can go to nasdaq.com and go into their Insights area, and it’ll be there as well. 

But yeah, you kind of alluded to it a little bit here. The article describes the evolution from on-premise software to SaaS and now AI. Looking back, though, you sort of described how they were measured, but are there lessons that investors can learn from previous technology transitions when trying to evaluate AI businesses today? 

Michael Normyle

Yeah, and I think really the biggest one that we’ve seen in the past, and we’re already seeing now, is that it takes time to settle on metrics that are useful for investors. In the piece, we talk about the internet era, for example, where some metrics persisted because they directly linked user activity to economic outcomes, like customer acquisition cost, customer lifetime value, conversion rates, and unique visitors. Others, though, kind of fell by the wayside because they proved to be “attention dressed up as value,” as we put it in the article, with things like eyeballs. So, loosely defined things that aren’t actually that helpful. And right now, we’re still in the very, very early stages of this agentic era, and there’s a lot of different metrics being used, and a number are very unique to the company reporting them. So it’s possible that some of these new measures maybe won’t prove enduring. The risk also is that if the value proposition related to AI becomes really unique to customers, then it’s going to be hard to land on universal metrics. So that’s something that we’ll see in time. 

Jeff Praissman

One overwhelming theme in the article is that AI creates value through work performed rather than users added, right? So it’s the overall results versus some measurable number of users. But how should, or how could, investors think about measuring, quote-unquote, “work performed” in a way that is both meaningful and comparable across different companies? 

Michael Normyle

Yeah. So this is definitely a challenge of this era, at least so far. I think there will be a degree to which some companies can use universal metrics and be compared, but perhaps not as broadly as we saw in the SaaS era. 

So for now, I think you can think about things like tasks completed without human intervention, for example, or the percent of time saved with the help of AI agents. So, relatively standardizable metrics that are probably applicable to a broad swath of agentic AI tasks. 

Jeff Praissman

And AI is obviously a household word at this point. But one stat from the article that kind of surprised me is that nearly four out of five Nasdaq-100 companies are discussing AI on earnings calls. So even companies that have nothing to do with AI on the forefront, like Nvidia or chips or anything like that, it’s involved with everything now. How can investors separate genuine AI-driven value creation from companies simply participating in the AI narrative? 

Michael Normyle

Yeah, and I think for the full context of that stat that we’re providing, before ChatGPT was released at the end of 2022, only about one in five Nasdaq-100 companies were talking about AI on earnings calls. Now it’s four in five, like you said. 

And you see this growth across any index, whether it’s the S&P 500, Nasdaq-100, Russell 3000, whatever you want to look at. 

And I think, unfortunately, there’s not necessarily a shortcut to recognizing which companies are merely talking about AI in passing during earnings and which are discussing genuine AI-driven value creation beyond looking at it company by company. So, yeah, I don’t think there’s an easy shortcut, unfortunately. I think it’s more representative of how pervasive AI has become as companies are trying to implement it either in their business or on their business. So it’s obviously grown in leaps and bounds in just a few years. 

Jeff Praissman

You know, one thing the article did a really nice job of, too, is highlighting the wide range of emerging AI metrics, from token usage and API calls to hours saved on a job or duties, and probably most importantly, tasks completed autonomously. 

Which, if any, of these categories do you believe are most likely to become standard metrics over time? 

Michael Normyle

Yeah, and with the metrics we were highlighting, we were selecting companies that we think are really at the forefront of measuring the value of agentic AI. So I think there’s a good chance that any and all of these that you mentioned do become reasonably standard metrics here. And the way that we also thought about them in the article is that they fall broadly into two categories: inputs and outputs. On the input side, you have token usage and API calls, plus LLM requests, which we also mention. And then on the output side, you have hours saved and tasks completed autonomously. So knowing both sides has value to investors because with inputs, you’re getting a sense of activity and potential costs to the company. And with outputs, you’re seeing the value created. So, of course, both sides are relevant. You might be concerned if you see that the inputs are far surpassing the outputs, at least in a relative sense. So ultimately, I think you probably do need a mix of these input and output measures to get a sense of efficiency and value created by these AI products. 

Jeff Praissman

Yeah, and that kind of leads me right into my next question. All these companies are taking different approaches to reporting the value of AI, right? So is there a risk that investors just won’t be able to compare these companies if everyone’s developing their own AI performance metrics? 

Will sector analysis be able to kind of bucket them and say, “Okay, we’re going to compare the best of token usage versus the best of output versus the best of time savings,” and so forth? 

Michael Normyle

Yeah, I think that’s definitely a challenge, and one that we highlight in the research, right? So if agentic AI does not lend itself to universal metrics, since each company’s value proposition is relatively unique, then that does create a real challenge in comparing companies and can create some uncertainty about what the true value of a company should be. 

And that’s one of the ultimate functions of the stock market, to have this price discovery happening all the time. If this makes it harder to do so, that might create some, like I said, uncertainty about the true value of a company. 

Jeff Praissman

Yeah. And I think, really, out of the metrics we discussed and the article discussed, productivity gain seems like a key outcome, right? The whole point of AI at the end of the day is to be more productive. 

So do you think that could really be the next major valuation metric, compared to everything else, kind of like recurring revenue became the key SaaS metric? 

Michael Normyle

Yeah, I think that does make sense, since it’s the most obvious value creation piece here, right? But we still don’t necessarily have a universal measure even with that, because already we’re talking about some saying time saved, some saying tasks completed, and those are two different measures of productivity. So that’s kind of the market, the earnings element of it. And then from the macro piece, you have the customer who needs to take that time saved and do something else with it. Because if that time saved isn’t translating to increased output, then the total output remains unchanged. Of course, that’s incumbent on the customer to do something with it and is unrelated to the value that the company we’re talking about here is creating. But it is a bigger-picture AI question. And I think there’s interesting research recently from the Bank of Korea that showed that AI was saving people an hour and a half a week in Korea, but in general, they weren’t using that extra hour and a half to do anything productive, so total output was flat. 

So that’s really a macro question and not a market question, but it’s something that remains to be seen because ultimately, the hope is that AI is this driver of productivity. 

Jeff Praissman

Yeah, so it would be great if every company let the employees leave an hour and a half early, but I highly doubt that’s going to happen at most places, at least. 

Michael, consumption-based pricing models have become increasingly common in software. How does that trend help pave the way for the AI era of business models and valuation frameworks? 

Michael Normyle

Yeah. So I think this is a bit of a predecessor to charging for token usage, for example. Instead of charging for the number of users, you’re charging for the amount of use. And we noted in the article that this is something a handful of companies helped normalize in recent years, predating agentic AI. And with that groundwork laid, we then see a study that we highlight from Metronome that, as of 2025, 85% of SaaS companies surveyed were experimenting with this consumption-based pricing. So I think this helped create some understanding of this pricing method and how it worked before it was adopted by these agentic companies. 

And so, in turn, I think it’s helping analysts and investors understand its meaning for valuation purposes by having some track record for a few years before it’s been adopted more widely for these AI-related reasons. 

Jeff Praissman

Michael, this has been great. I kind of just want to wrap it up, summing up everything we talked about. Once the dust settles, what do you think success looks like for AI metrics? 

Do you think it’s going to be a handful of standard measures like we use now, or will it be more siloed, where each company has its own way of demonstrating AI-driven value creation, or a hybrid of both? 

Michael Normyle

Yeah, I think success would be metrics that provide real insight into the value a company’s product creates and that are also relatively straightforward for investors to understand. Something like hours saved or tasks completed, those are kind of intuitive metrics. 

And then I think beyond that, the understanding will be helped if companies can land on those standardized measures, because even if ARR or net recurring revenue weren’t the most intuitive things in the world, by virtue of becoming standardized measures, that helped people understand them. And perhaps within this kind of framework for success, companies can then supplement those standardized measures with more custom ones that are specific to their company. That could hopefully help with proper valuations of companies along the way, while enabling some comparability across companies as well. 

So that’s my hope for a win-win for everybody. 

Jeff Praissman

Thanks again, Michael, for stopping by the IBKR Podcast Studio. Again, the name of the article is “The Search for New Measures of Value in the AI Era.” It can be found linked to this podcast, as well as on nasdaq.com. 

And until next time, Michael, I look forward to continuing the discussion. 

Michael Normyle

Yeah, thanks.