Frontier AI Outperforms 12 CPAs on Month-End Close, but Replacement Fears Ove...

Attention: This is a machine-generated transcript. As such, there may be spelling, grammar, and accuracy errors throughout. Thank you for your understanding! Blake Oliver: [00:00:04] And there have been two recent studies that found that these AI assistants are actually only recommending a small group of prominent firms, meaning that most CPA firms are basically invisible to AI searches. David Leary: [00:00:19] Coming to you weekly from the OnPay Recording Studio. Blake Oliver: [00:00:25] Hello and welcome back to the Accounting Podcast, your weekly roundup of news in the profession. I'm Blake Oliver. David Leary: [00:00:31] And I'm David Leary. Blake Oliver: [00:00:33] And David, this week we're talking about a new study that found that frontier AI models perfectly completed month end close tasks, outperforming every CPA, 12 of them at less than one tenth the cost. We're also going to talk. David Leary: [00:00:49] So much in that that that study that I can't wait till we discuss it. It might be half the show. That's how much is in that study. It's amazing. Blake Oliver: [00:00:56] And then you've got a story about what clients are willing to pay for accounting services, which is still a decent amount. And it's interesting how it changes as they make more revenue, but it's not like linear. There's a there's a big shift that happens at like a million in revenue. So we're going to dig into that and as much accounting news as we can after we thank our sponsors. David Leary: [00:01:18] So our sponsors this week, we have Cloud Accountant staffing on Pei finance, accounting Technology Expo and Savant Labs. Blake Oliver: [00:01:26] And I was just going to read that ad for you, David, but I got signed out at Google. Oh, so I need you to read the first one. David Leary: [00:01:32] Yeah. Are you tired of the endless search for qualified accounting talent? You're not alone. Growing accounting firms are struggling to find available and affordable team members when they need them most. Cloud accounting. Staffing has a solution with a revolutionary candidate portal. Unlike traditional staffing agencies that waste your time with sales calls, paperwork deposits, the Cloud Accountant Staffing Candidate portal gives you instant access to highly vetted, qualified accounting professionals. No waiting, no hassle. Just top talent right now. What makes this difference? Speed and simplicity. What other firms make you wait weeks or months with cloud account and staffing? You could interview someone as soon as tomorrow. Their boutique support ensures you're actually getting quality talent that's both available and affordable. Exactly what a growing firm needs. The candidate portal puts you in control. Browse live candidates, make selections on your timeline, and build your offshore team without the traditional headaches. To find, review and book interviews with potential team members, all in less than ten minutes. Head over to The Accounting Podcast dot promo. That's cloud accounting. I'm sorry, that is accounting podcast.io/cas. Blake Oliver: [00:02:38] Thank you, Cloud Accountants Staffing and welcome to our live stream viewers. We've got Tino here in the chat saying three coffee emojis made a live event. Happy federal fiscal new year and happy end of 2026 extension season. Ashley says happy Friday. Edgar. What up? Accountants? Uh, Shelly. Happy day after deadline to those who celebrate. Great to have you with us. Uh, and if you haven't checked us out on YouTube or seen a live stream, go to YouTube, search for The Accounting Podcast, hit subscribe and that notification button icon and you will, uh, get to see us live. If you can catch a live stream. All right. Let's talk about this study. I feel like, well, I personally am not surprised by this because the recent, uh, um, new, the newest models like opus 5.5, fable Big Four that have just been incredible. And I'm using them to do all sorts of work that I never could have used before. And now we've got a study by a startup that specializes in AI for professional services that has demonstrated just how capable these models are. That company is called Mercor, MERCOR, and they are a startup founded in 2023 that specializes in using professionals, working with professionals like CPAs to train models, so teaching the models, evaluating them, giving them feedback so they can get better. And if you think about it, that's the only way that we're going to get models that can do the work that CPAs can do. And I think that's in big part why the recent models have been so good. So this company did a study and it's called Human Baselines for benchmarks. Ai now outperforms junior accountants. And basically the takeaway is that the latest models have perfectly completed month end close tasks. They can do them perfectly 100% of the time. And they outperformed every CPA in Mercury study at less than one tenth the cost. What did they do? David Leary: [00:04:45] And we see one tenth the cost. We're talking like $0.21 a task. Like crazy. Crazy cost reduction. Blake Oliver: [00:04:52] Saying yes. They tested 12 accountants on simplified tasks from their apex accounting benchmark, and they found that Frontier AI models were faster and more accurate than every participant. The accountants were described as junior. But they had 5.5 years of experience on average, and all of them were licensed CPAs. The participants completed four realistic month end close scenarios searching company working files, identifying the correct numbers, performing calculations, and delivering tables of results. The study was designed to measure AI augmentation, but the model scored perfectly on their own, leaving no room to measure improvement from pairing accountants with AI. So here is figure one, the first chart in the study and the headline here on this chart. The title is cloud is Faster and more Accurate than accountants on medium length, detail oriented accounting tasks. And we've got these black dots that represent accountants without AI. They had a 23 results here. And you can see that they go from zero up to it looks like maybe 90% on accuracy. And then clod alone is the purple diamonds 100% every single time. Minutes to finish the task. Uh, Claude, I mean, almost, you know, just a few minutes at most on every single task. Whereas the accountants took anywhere from 30 minutes to 180 minutes. Here is a chart from the study. This is figure two. The title is AI Models Past Accountants on these tasks in 2025 and now score at or near 100%. So we've got time going from left to right from July 2024 up to July of 2026, and in the spring of 2024, that's when we got GPT four. And that was the model that first started to be able to do work to the level where we really were paying attention to it. Blake Oliver: [00:07:00] We really started to think, oh, what can this do? And back then, GPT four was getting on this test like 5%, maybe a few percent. It wasn't able to complete hardly any of them. And then as time goes by, we get Gemini 2.5 Pro open AI oh three. And you can see that basically in July of 2025, the models, which have varying different scores of accuracy, began to surpass accountants. So the average accountant score without AI assistance on this test that they do is 37%. And it's a difficult test. So it makes sense that it's it's low and it was open AI's oh three that first began to beat them, getting closer to 60%. Gpt five getting around 70%, opus 4.8, which just was, you know, just, uh, this summer, uh, passes 80% and now opus five, 100% and all the other, you know, frontier models are close to or about that. They're all over 80% now. So I mean, that is a spectacular change. We're talking, uh, two years, we went from these models being able to complete month end close tasks like a few percent of the time to now 100% of the time on this test. Now, Merkur is careful to say that the results don't imply that accountants are replaceable. And there's two reasons why not. First, the tasks tested only a small set of skills that happen to be what the models are best at being detail oriented, searching through files, following instructions very closely. And second, most of an accountant's job was absent from these tests. There's no coworkers to ask questions of. The participants came to the task cold without context you'd normally have at a job, and that makes the task a lot harder. There's no. David Leary: [00:09:02] Client communication. There's things that you kind of have to do beyond what the eye is going to work on those tasks. Blake Oliver: [00:09:07] Yeah. And imagine, you know, you just sit down at your desk one day and you've got a new client that you have no idea who this is, and you've got to complete some sort of month end close task, right? That's why these are very difficult for humans to do. You don't have any history, so you don't measure skills like client communication on this test, asking the right questions, collecting, collecting context over time. And so Merkur says that these skills are going to become a larger part of accountants work as the models take on more of the detail oriented tasks. And you know what, what are we talking about with detail oriented tasks? I mean, putting together, um. Do you know doing depreciation schedules right. Or doing, uh, journal entries. Right. Like taking this data and doing the spreadsheet work that you'd normally have to do in accounting job. Basically, we're at the point where the models can do it 100% of the time for very cheap. So the work of accountants is it's just going to change so dramatically. David Leary: [00:10:03] I think they talked about that even if the models don't get any better, where they're at today is going to change everything. But the models probably will still get better. Blake Oliver: [00:10:13] Here's an item. Here's an example of one of the items. So there's a task that required, uh the response to identify 4000 of parking revenue had been mis