8 Oct 2026
My AI Agent Found $4,200 I Didn't Know I Was Losing
My AI agent just found me $4,200 a year in payroll fees that I didn't know I could negotiate. Now, mind you, that $4,200 is from one vendor. And honestly, I wasn't even looking for that. I was looking at something completely different. So if I'm here to tell you that I'm surprised about something, I'm happy that I just found $4,200. Let me explain what's going on.
Welcome to Fix My Business, the show that helps you fix your business. I'm your host, Scott Todd. I have built multiple seven-figure businesses after leaving my Fortune 300 VP job. And today I want to walk you through two of the experiments that I've been running on this channel. And I'll kind of give you some updates because I can't stand when people run experiments or they tell you something they never get back to you. So here we go. Experiment number one.
I talked about it in last week's episode, and it was using AI to do my bookkeeping. And I told you at the time that I was running parallel systems, my CPA versus AI. Which one was gonna battle it out? And essentially I'm happy to report to you that the September run, if you will, was smooth for well, both of them. I mean, the CPA did what the CPA does.
And the AI does did what the AI did. Now, here's the difference. My C EPA has years and years of experience with knowing how to code my transactions. And so the rules are built in. Whereas AI needed help more frequently with coding the transactions. But here's the thing. Every day, Monday through Friday at 8 a.m. or thereabout, I would get an a message from my
AI bookkeeper who would say, Hey, here's yesterday's transactions. How should I code them? And in the beginning, I needed to give a little bit more, but by the end of the month, it was really doing a great job of suggesting what the coding should be. So it became less and less of like, hey, code it over here versus just scanning it and saying you're good to go. And then it started to pick up on some.
Scott Todd (02:29.698)
well-established rules like, hey, when we get revenue into the bank account, well, then that's just revenue. when we have certain expenses like payroll, boom, it just hit the payroll. When we paid our VAs, it knew their names and it hit them as well. And it just started even not even recommending anything to me. It just did it. And so at the end of the month, when I compared the revenue and expenses from the CPA and the AI, what happened was
The revenue and the expenses aligned. It matched as you would expect that it would. Okay. Now, there were some transactions that were coded differently. The CPH coded them one way, the AI coded the other. And I don't really believe that that is a deal breaker. And the reason it's not a deal breaker is because it's not necessarily that the coding has to be exact and precise. So when I say coding, what I mean is.
That if I spent the money on software, that it's coded software, or if I spent the money on travel, that it's coded travel. It helps a great deal, especially if you're comparing year over year, or you're comparing different transactions, or you're looking at different categories. That's important. But at the end of the day, the most important part is that the revenue and the expenses total the exact amount, right, of profit, right? That's what we're looking for. Now, here are some of the differences though.
First of all, the AI agent got me my financial statements much faster than the CPA did. In fact, I had the financial statements on the first day of the month from the AI. And it was well after that that I got them from the CPA. So the human involvement of looking at it faster to me was through the AI. Clearly, the lower cost option is the AI. The AI, I don't really know what it cost me.
Because I'm just on a subscription and it's using like my grok bot subscription for 200 bucks a month. And clearly that was just some fraction of it, but way less than what I spend with my CPA. But the one clear difference that I found was the thing that kind of surprised me the most. And that was the fact that my AI agent began suggesting to me cost savings.
Scott Todd (04:54.796)
Now, some of them I just dismissed. Like, yeah, no, it's good. I'm I'm okay with this subscription, but it began to look at my subscriptions. And then it brought to my attention something that I was really surprised about. And here's what it was. I pay my payroll company. I have multiple businesses, so they each have their, they each have a payroll account with this company. And I pay them a fee every time we run payroll.
So I think on one company it was like $140 a month. And the other was a hundred, I'm sorry, $140 per pay period times 26 pay periods. And on the other company, it's like 106 times 26. That was what we were paying. And the AI looked at that and said, Hey, I think you have a cost savings here. And I was like, What do you mean? And they said, it said, Hey, there is a lower tier plan.
that you're avail that you qualify for if you're not using these other services. And quite frankly, I didn't even know I had access to those other services. So in almost three years, I had never used those services once. They were like background checks and other things that just really aren't relevant to my business. And in fact, I called up the payroll company and I said, is this true? Like, is
Is this true? Because like I'm paying over $100 per pay period when in fact I think I qualify for a lower plan that's like less than $40 per pay period. And they looked at my account and they said, yeah, that's true. So with that one suggestion, my AI agent literally found for me $4,200 in annual savings. And it wasn't, again, it wasn't even something that I was looking for.
I didn't hire it to go out there and do it, do that. I hired it to record transactions. And, you know, I think that the thing that surprises me the most is that at the end of the day, as a business owner, I cannot examine every single transaction and know that I'm on the very best plan. But and I'm not saying this to dog on a CPA at all, but the CPAs in the world.
Scott Todd (07:23.982)
They're not necessarily looking at those transactions either, saying, I think you have the opportunity to save money. And I just was really taken aback by the fact that my AI agent just delivered to me this gift that will continue to keep on giving. So that was $4,200 for the next 12 months. It's $4,200 a month, a year, sorry, $4,200 a year for multiple years. Again, I've been paying this for three years. So right there alone.
I just paid for a bowload of AI usage on that savings alone. And it was all proactive. And if your AI agent, if you're using them, if they're not suggesting ways that you can save money, well, then you should definitely be proactive and ask it: hey, look at these bank transactions that you now have access to. Where's my cost savings opportunities? Question everything. Now, earlier this year.
I did that with another service. I used it with Cloudflare. And I think my Cloudflare bill was almost, I don't know, $350. And AI suggested that, hey, that seems like it's kind of high. Give me the bill. And I gave it the bill and it said you should cut off all of these other services. You don't need them. And it also took my Cloudflare bill in January down from $350 a month down to $30 a
So again, AI is very, very good at that. And it's not something that I've seen a proactive effort from the CPAs and the bookkeepers of the world. And again, I'm not here to dog on them. I think that they do a fantastic job. It's just an opportunity with AI that we can do something that's different. And you see this across the board. Lots of people are talking about AI saving on money. Well, this is how it helped me. Now, word of caution.
Do not fire your accountant off of my podcast. Do not do it. Don't do it. But see, here's the thing: that experiment that I just ran gave me an answer to a question that I didn't ask. And what I want it to do is I want the AI to bring to me ideas that we can test.
Scott Todd (09:48.43)
Constantly, because that's the most important part of growing the business, is constant testing. And I mentioned this in episode 79 of this podcast, which was a few months ago, where I think it was in July, where I talked about this experiment that we were going to run on churn. And when I talked about the testing, I said, Hey, whenever we run something, we want to create a hypothesis.
Whenever we run a test, we want to create a hypothesis and we want it to have a verdict of either true or false. Okay. It has to be true or false. And at the time, at the time, I said that the test that we were running in one of my companies is that in a SaaS product that we have, we did not allow certain people to accept reviews. And we turned that on, thinking that sellers with
Reviews actually churn less. And that is still remains true that they churn less. And in fact, we did see a decrease in our churn from where we were. It was, it was let's see, it was 2.84%. And since we talked, it went down to 1.97%. Now that's a that's an improvement of 30%. That's pretty good. Okay. And
I could have just come back on here and said, Hey, we yeah, we wanted to test to see if churn got reduced and it did. What yeah, we won. But as we dug into the numbers, we realized that the thing that we did, the thing that we did to reduce churn while churn was down.
The thing that we did didn't have an impact on it. So again, what we did was we turned on reviews for certain sellers that didn't previously have it. And we saw that churn rate of that membership go down. But here's how we know it didn't have an impact. Because when we looked at the numbers, we realized that ultimately the P score, which is the correlation between that thing.
Scott Todd (12:00.756)
And something...
