Made possible by

Rokt helps marketers reach high-intent customers in the Transaction Moment™—when they’re actively completing a purchase online. Powered by AI and first-party data, Rokt Ads connects your brand with over 400 million global shoppers and delivers outcomes you can count on. Learn more at rokt.com/emarketer to get started today.

Marcus Johnson (00:00):

In today's economy, every ad dollar counts. That's why performance marketers are turning to Rokt ads to reach 1.1 billion unique customers globally in the transaction moment. When they're completing a purchase online, you only pay when customers engage. Yes, please. Learn more at rokt.com/emarketer. That's R-O-K-T.com/emarketer.

(00:32):

Hey gang, it's Friday, March 13th. Grace, Jacob, and listeners, welcome to Behind the Numbers, an EMARKETER podcast made possible by Rokt. I'm Marcus. Joined me for today's conversation. We have two West Coast folks. We start with our tech and AI analyst. Grace Harmon.

Grace Harmon (00:49):

Hi, Marcus. Thanks for having me.

Marcus Johnson (00:50):

Of course, of course. We're also joined by one of our tech analysts, Jacob Bourne.

Jacob Bourne (00:55):

Hi, Marcus. Glad to be here today.

Marcus Johnson (00:57):

Yes, indeed. Glad to have you. Today's fact.

(01:04):

So we're talking about the share of new car sales that are electric. So in California, which is where Jacob... Ah, Grace, you too. You're just not from that state. So, yeah, where both of you live right now, California, that state has, I think it's a third of all registered electric vehicles. So it's got most in the country. But what share of new cars bought by people in America do you think are electric?

Jacob Bourne (01:32):

I guess small, like 5%.

Marcus Johnson (01:37):

Okay. Grace?

Grace Harmon (01:38):

6%.

Marcus Johnson (01:39):

Okay. 10.

Grace Harmon (01:40):

10? Oh, 10.

Marcus Johnson (01:41):

Close.

Grace Harmon (01:42):

Okay.

Marcus Johnson (01:42):

Yeah. 10% of new cars bought by people are electric. That's up from 2% right before the pandemic in 2019. In Norway, the highest country of them all, 97%.

(01:55):

So almost every single new car bought in Norway is electric. And that's up from 56% in 2019. So a huge jump even in that country as they're basically close to everyone buying one. Nepal is second. 73% of new cars being electric, which is fascinating. I went to look into why. Significantly lower import duties are in electric vehicles, huge fuel savings in the country. Lots of locally produced hydroelectric power as well. Government policy is aiming for 90% of EV sales by 2030 because of the high pollution levels in cities and also access to affordable Chinese EV brands.

Grace Harmon (02:36):

Okay.

Jacob Bourne (02:36):

Yeah.

(02:36):

I think that last one is probably one of the biggest ones. And really one of the biggest barriers in the US is that the models aren't really particularly affordable.

Marcus Johnson (02:43):

Yes. Yeah. Anyway, today's real topic. How AI is changing the workplace? Efficiency gains or higher demands?

(02:56):

Well, I thought was interesting as well though. Now we've stopped talking about it. But I was like, why isn't America higher? And people be like, "Oh, well, you can't really go as far as you can with the gas." But Americans only drive 37 miles a day, according to the Federal Highway Administration.

Jacob Bourne (03:14):

Yeah, but that's the average though. There's a lot of people that like to take road trips, and if you're taking a road trip on a route that doesn't have an EV charger or a fast EV charger or one that's compatible with your particular EV, then...

Marcus Johnson (03:29):

That's true. But what's also true is Americans have about eight cars each family. And so you could have one electric one for your runarounds just going to work and back because people don't... And then one for those more further trips. But I was surprised at how low I thought this was. And it saves so much as well. It saves like two grand, up to two grand to get the Level 2 electric installer in your house. But then you save like a thousand to $2,000 a year on gas or on electricity. And also oil changes because it doesn't have those. Anyway, this has got nothing to do with the episode. Let's crack on. Just learning about it.

Jacob Bourne (04:07):

It's related to technology very early.

Marcus Johnson (04:10):

Good save. Thanks, mate. All right. Burger King is rolling out AI headsets that track employee friendliness, writes Danielle K of the BBC. An AI chatbot called Patty, very clever, in the headsets, answers questions from employees about how to prepare menu items and flags when a product needs restocking, as well as tracking how often they're saying please and thank you to customers and then figuring out a score and how nice they're being friendliness score. Turns out AI is trying to help employees in all kinds of ways, but is it making us more productive? Justin Lahart of the Journal writes that, "Investment in AI ignites a fire under US economy, but technology hasn't yet fulfilled promise to make humans work more efficiently." He roughly defines productivity as the amount that the average worker can produce in any given hour. Grace, based on the research you've been looking at, how much is AI really helping with productivity?

Grace Harmon (05:07):

Yeah, I mean, some of the data is conflicted. I would say that the impact tends to be incremental and task-specific rather than more broadly transformational across the company level, at least at this point. I think the thing about AI is it's such a huge technology and it's advancing so rapidly, it feels like the economic and work benefits and effect should match that, but we aren't quite there yet in mass.

Marcus Johnson (05:31):

Yeah. Yeah. I like that line you said because the economy said something similar. They said most usage consists of discreet tasks rather than wholesale automation, Jacob.

Jacob Bourne (05:40):

I think the answer is really an unsatisfying one, and that's that we don't really fully know yet. I think productivity has been notoriously difficult to track across a wide variety of industries. You have a use case like coding, which is one of the earliest use cases for AI, and it's also a very well-defined use case and maybe one of the easiest ones to really determine whether or not productivity is happening. And even there, you see, like Grace mentioned, it's conflicted. I've seen studies that say, "Wow, yeah, AI is really boosting coding productivity, making coders, software developers faster." And then other studies that say the exact opposite.

(06:19):

And so I think what's happening here is that the speed of AI output, I mean, it produces output at lightning speed. It creates this illusion of productivity, but then not all studies factor in all of the work that you have to do after the output is produced, such as debugging code, checking for errors, which AI is not infallible, it still produces errors, checking to make sure it's really the context is great, is the right context for your project. All these things, these extra steps that humans have to do, the human review process can really eat into those productivity gains.

(06:55):

So I think that there's long-term possibility for really hard quantifiable productivity gains, but it's not quite there yet. And it's just also very hard to measure.

Marcus Johnson (07:09):

Yeah.

Grace Harmon (07:10):

I think that you brought up one good point, which is how are you measuring productivity? Is it how fast you do the work? Is it how much work there is to do? One thing I've been thinking about a lot related to productivity, it's from a recent Hard Work episode, but they were talking about the lump of labor fallacy, which is the incorrect belief that there's a fixed amount of labor to be done in an economy. And I think that it's really up to debate at this point if AI automation is going to cause permanent mass employment or if it's going to add to productivity and open up new industries and new jobs like the Industrial Revolution did.

Marcus Johnson (07:42):

Yeah.

Jacob Bourne (07:42):

Yeah, that's a great point, Grace. And actually, it's something else I've been thinking about along those lines is if we just look at productivity as output per hour, I think that misses other aspects of the economy such as what is the actual value of the output that's produced in that hour. And I think it's ultimately people that determine the value. And so actually, to me, there's still going to continue to be a premium on human judgment. And so even if you get faster at certain tasks, that doesn't really give you the full picture of the work that's being done. I mean, there's so much of qualitative output that really matters as well.

Marcus Johnson (08:28):

It might bring up the conversation of how are we measuring things again because GDP for a while now, people have started to say, "Why are we still using this measure?" Because if that goes up, it doesn't necessarily mean people's lives are getting better. We made this a measure to track output, to track how much we were building things and making things during the industrial revolution. Is it still relevant today? And Jacob, something you were saying about, so what if you're freeing up all this time over here, you're maybe having to reallocate it over there, some data to back that up. AI, it can help workers save time to data points there. In 2023, MIT found ChatGPT reduces completion times for writing tasks by nearly 40%. And then the second one here in a study of consultants of BCG, Harvard Business School found AI-driven productivity improvements of 12 to 25%.

(09:17):

Unrealistic professional tasks, University College London found 15 to 30% in real world settings. The question is, where is that freed uptime going? And the economist had an article saying these productivity gains assumed that all that time saved is redeployed productively and that workers neither shirk nor produce lower value output. Early evidence points to a messier reality. Some studies suggest workers spend more total time working when using AI, others that the technology is sometimes used to generate low quality slot that requires editing or verification.

Jacob Bourne (09:46):

So it also raises the question, well, what is the impact of that faster produced written content? Is it having the same impact on the people that are reading it? And I think that's also an open question.

Marcus Johnson (09:58):

Part of this as well is just the technology's new, right? We're talking about why isn't it shaken up the world? And it's because it's a couple of years old. I mean, AI a long time, but gen AI and how widely accessible that's been to the general public is only quite new. Disruption hasn't happened yet, probably because not enough people and businesses are using it. And some of those numbers to point out our Wall Street Journal looking at Census Bureau survey saying just 10% of businesses reported using AI in some way, and that's up from 6% a year ago. Another one about people, that's businesses, according to our estimates, 60%, 60% of people in America don't use GenAI. By 2029, over half still won't use GenAI. And the share who don't use it for work is even higher. 75% of folks saying they don't use it at work.

(10:48):

Zooming in a bit, only about 13% of working age adults use AI every day. Our track is every month. They're saying every day, just 13% of working age adults going to Federal Reserve Bank of St. Louis. And then lastly, McKinsey, finding two-thirds of companies are just at the piloting stage and just one in 20, 5% high performers who have deeply integrated AI are seeing it drive over 5% of earnings, which again is a fraction. So it's part of the problem is just too new to have really made a kind of disruption. What do you guys make of this? The other part here is this, the AI investment is kind of created a productivity illusion, so to speak. So Jason Firman of Harvard estimating some 90% of GDP growth in the first half of 2025 came from spending on data centers and related CapEx. Research from the Federal Reserve Bank of San Francisco found that underlying productivity gains once the effect of such investments is excluded close to zero.

(11:45):

So part of this productivity argument is a ton of money being invested, but not really anything else.

Jacob Bourne (11:51):

Yeah. And I think part of the reason why all that money is being invested is because there's this narrative that AI is going to reduce the need for human labor. But I think the evidence is there that it's not quite there yet. And I'm not so sure it will ever be there for a variety of reasons, but I think what's happening is that the massive spending on data centers, AI chips, AI cloud contracts is then creating this sort of justification for squeezing payroll budgets. And again, is part of this narrative that, well, AI reduces the new need for human labor. So it's a bit of a circular reality that's happening. And I think that's why also there's this laser focus on this question about productivity and whether AI enhances productivity, because it all kind of hinges on that question, all this money that's being spent.

Grace Harmon (12:44):

I think the money being spent also hinges on monetization because at the baseline, if you're going to get returns on all this CapEx, how good is your revenue model? How good is your business model? Do you have AI that's good enough to sell or is it just going to be limited to enterprise cases? Sell to consumers, I mean.

Marcus Johnson (13:01):

It seems as though, based on what we said, productivity's not really had as much of an impact as expected. Bureau of Labor statistics showing US non-farm business productivity measure, five-year average, last year, 2025 was a bit lower than all the years from 2020 to 2024. So we've been getting more productive in other ways. AI is not the only thing that makes us more productive.

(13:23):

What's strange though, is must be some cognitive dissonance because you can see this chart on the screen. 77% of US full-time desk workers said AI tools make them more productive according to EisnerAmper. So people feel like it's making them more productive, but what does that really mean?

Grace Harmon (13:37):

I think the data's pretty split on that. I was looking at different data earlier today that it's more like 35%. I think it really differs from source to source.

Marcus Johnson (13:46):

Okay.

Jacob Bourne (13:46):

Yeah. I think also these models are really general purpose too. And so they can be used for a variety of things and you're shifting between tasks. So I mean, I think that the measurement around this is just very, very difficult. And so, yeah, you might be getting productivity gain for one task and maybe losing it in another, and then also depends on how good the model is performing at a given time as well.

Grace Harmon (14:14):

Yeah. And sometimes it's as simple as just needing to be able to pay for team ChatGPT access so that employees can just at the baseline be able to use the tools at all. Having clear communication about how workflows are going to change, about where governance lies when you have questions for IT about or for another team member about data privacy, who do you turn to, how job roles are going to be impacted. I think that that's part of what makes things so slow and uneven.

Marcus Johnson (14:38):

Yeah. I'll jump ahead here because that's one of the things I wanted to talk about when we look at what can companies do about this. And Grace, you hit the nail on the head with offer training and access to gen AI tools. As research from McKinsey found 48% of US workers said training would boost their AI usage. 41% said access to gen AI tools. And training is really important because you can see from this chart on the screen, lack of knowledge and skills was the number one barrier to AI adoption for marketers ahead of lack of budget, uncertainty on ROI, concerns over data privacy or security. People are saying, "I don't know how to do this." And so training is going to be the best thing you can do and what Grace is saying, also give them access to the things so that they can use the services.

(15:24):

That's probably not a bad idea. I want to circle back to actually an article that our colleague Gaja Savilla was writing. It talks about this idea that we were sold on this bill that AI was going to make things, it was going to help us with our work, make things a lot faster, a lot more efficient. And he was basically a senior analyst guardian severely writing, quote, gen AI was pitched as relief from busy work, faster drafts, cleaner code, fewer routine tasks, an eight month Harvard business review field study of 200 people in the US tech firm revealed a more complex outcome showing that AI adoption was mostly voluntary and employees leaned in even when AI use wasn't mandated, the result work sped up, responsibilities multiplied and AI assisted tasks spilled into more hours. Grace, what do you make of this and some of the trade-offs of using AI?

Grace Harmon (16:13):

Yeah, I think when it comes to some of the trade-offs, and Jacob was speaking about this earlier, is that it can speed up output, but it can also introduce errors and over-reliance and quality control issues that require more human oversight. It can also widen performance gaps. So there's more benefits for highly skilled workers, there's more coordination challenges for teams. And then if you're talking about training, there's also less opportunities for women and older generations. So I think that that creates some more issues just in terms of access and ability to maximize the technology.

Marcus Johnson (16:46):

Jacob, anything to add?

Jacob Bourne (16:48):

Well, I think it's also... I mean, I agree with everything that you and Grace both said. I think that the speed of AI advancement also is maybe paradoxically slowing adoption in a way, because I think even though it sounds counterintuitive, when you have these AI tools that are changing on a daily, weekly, monthly basis, I think it creates anxiety, it creates information overload, it creates this uncertainty and maybe a lack of confidence over, am I staying up with the AI curve fast enough? And I feel like a lot of people feel that way. And it's just hard to feel confident in using these tools when the ground is constantly shifting between our feet every day.

Marcus Johnson (17:27):

Yeah. That's a great take.

Jacob Bourne (17:30):

And the other thing about it is this AI evolution has become very politicized. I think we have stoked fears about job loss, about mass surveillance, about security breaches, loss of data control by companies. And I think this entire environment is making people and companies too more cautious and maybe even in some cases more resistant. So I think in some ways this frenzy around AI is maybe counterproductive to adoption that's really going to show these kind of productivity gains that everyone is hoping to see.

Marcus Johnson (18:06):

Yeah. Grace, as asked by Christopher Mims, the Wall Street Journal titled, or in the title, it says, "Adoption is slow and uneven," talking about AI adoption. Is this because of the things... I mean, Jacob's listed a few reasons why that might be the case. Are there any other reasons why it might be slow and uneven? And also, how can companies overcome this?

Grace Harmon (18:27):

I think one other thing is clear use cases. To follow up on one thing that Jacob said with how fast models are moving and developing, I can imagine that it would be really difficult to always have a concrete idea of what you should be using and how you should be using it. For example, Anthropic is known to create really great coding tools, but depending on your use case with so many models coming out constantly, I think it's hard to keep track of the best applications.

Marcus Johnson (18:51):

Good point.

Grace Harmon (18:51):

Yeah. I think leadership buy-in is another big deal, like training, as we've said, access is what you said.

Marcus Johnson (18:56):

I thought this was interesting. That's a great point. I thought this was an interesting one from Workhelix Chief Executive James Milin. He said, "Willingness to experiment, but not the business role is the most important factor determining worker's eagerness to adopt AI." So how willing are they to experiment? Not necessarily would this be more useful for that other person. So if person A is more eager, but actually a technology is more suited for person B, person A is going to be the person who's adopting it the most. Within companies, people who could get the most out of AI might need the most encouragement, education, and guidance to get there. An age who's saying often an imperfect proxy since some young workers are anti-AI, circling back Jacob to one of the things you were talking about, and actively work to thwart companies' AI adoption efforts. So it's not just the company saying, "Here's the thing, whoever's most eager to use it are the people who should be using it."

(19:52):

It's doing a full audit and saying, "Where could this technology be best applied?" And helping that person in that role get up to speed with it and work it into their role. Gajo is saying teams redesigning workflows around AI will benefit over those who simply layer it on without adjusting output expectations. I thought was a really good way of putting it as well. Let's end with this, Grace. And this question was inspired by piece that you wrote. I wanted to talk about how this infusion of AI is affecting how workers view their roles. You'd written a piece, this quote here, "AI's perceived impact on jobs is already reshaping worker behavior even before its effects are fully felt." You're saying anticipation of future changes is driving behavior more than lived change, setting up tension between expectations and reality. Could you elaborate a bit on this for us and talk a bit about what you found in your research for this piece?

Grace Harmon (20:45):

Sure. Well, obviously there's a lot of anxieties from job security. I think that's pushing employees to rethink what skills make them uniquely valuable and whether they need to prepare more for an AI-driven future, either to be good candidates for future jobs or to maintain security in their current role. It's also pushing some employees, especially gen Z-ers, to be sabotaging team efforts to adopt AI. I think a lot of workers do at this point see AI as a collaborator. They can take on repetitive tasks and first drafts. And as we've talked about, free them up for higher value work, how higher value work is defined is still yet to be seen. But I think that what I said earlier in my article is that there's this drive to prepare for a future that we have not yet lived.

Marcus Johnson (21:29):

And you said sabotaging companies, AI efforts. It's a big share of... It's not like a couple of people over here. I think there was some research. Is it 30, 40%?

Grace Harmon (21:39):

I think something like that. And it's things like under-reporting the efficacy, intentionally getting bad outputs, things like that.

Marcus Johnson (21:46):

Yeah. Yeah, because they're nervous about if I show that this thing did really well, do I become obsolete?

Grace Harmon (21:53):

Yeah.

Marcus Johnson (21:53):

Yeah. Jacob.

Jacob Bourne (21:56):

I think it's that concern, but I think it's also this pressure around productivity itself. This using AI is supposed to make me faster, so the expectation is to be faster. And a very recent Anthropic study, which analyzed about 10,000 AI conversations, found that only 9% showed users fact checking the AI. And then only 16% showed users questioning the AI's reasoning, and only 20% showed users noticing when context was missing, like when AI just wasn't getting the context of the output right. And so I think to me, that signals that there's over-reliance on AI, there's inappropriate use of AI. And I think it really comes down to this faster, better metric that is really fueling the AI race more than are we really getting the full value out of AI? I think the incentives around this are maybe the problem more than the technology itself.

Marcus Johnson (22:49):

Yeah, that's a good take. I love this quote from Charlie Munger. It's Warren Buffett's right hand man. "Show me the incentive and I'll show you the outcome." And so if you're saying that we want you to be faster, the incentive is to show that you're being faster, then that's going to be the result or the outcome, whether it's good or bad. Really quickly, anything else you guys want to talk about before I close out? Is there anything we didn't touch on?

Jacob Bourne (23:18):

I mean, I guess my long-term belief in this is that a functioning economy is still going to require people, and that means people that still retain their full breadth of skills and human judgment. And I think that one concern I have around this is just the erosion of human skills because of over-reliance on AI, which I think it really comes down to AI can really enhance our existing skills or it can erode them. And it really depends on how we adopt the technology.

Marcus Johnson (23:46):

Yeah. What does working with this technology look like?

Jacob Bourne (23:49):

Yeah, working well, right?

Marcus Johnson (23:51):

Yes. Yes. We'll end there. Thank you guys so much for hanging out with me today. Thank you first to Grace.

Grace Harmon (23:57):

Thanks, Marcus. Have a good one.

Marcus Johnson (23:58):

Of course. Thank you. Same to you. Thank you so much to Jacob.

Jacob Bourne (24:01):

It was a pleasure to be here. Thanks, Marcus.

Marcus Johnson (24:03):

Thank you, sir. Pleasure was all ours. Thank you so much to the whole production crew. Helping out with this one. We've got Lance on this one, Luigi and Danny as well. Thanks to everyone for listening in to Behind the Numbers, an EMARKETER podcast made possible by Rokt will be back Monday. Of course we will. Happiest of weekends.