I Watched AI Burnout Finance Teams for a Year. Here’s What Finally Started Working

Full transcript

Lightly edited for readability. The audio is the original.

0:00The 40% who got handed broken AI work

Last month, 40% of finance pros who use AI got handed work that looked finished, came out of the tool, and was quietly broken. The kind of thing that reads great on the first pass, and then falls apart the second you lean on it. The person downstream spent almost 2 hours fixing each piece. And if you work in finance, a lot of the time that person downstream is you.

There's a name for this now. Researchers call it workslop, and it's showing up all over finance teams that were promised the opposite. AI was supposed to hand you your evenings back. Fewer late closes, fewer Sunday board decks. For a lot of teams, it's doing the reverse.

The hours didn't leave, they moved. Today, I want to show you the line. The exact line between the AI that pays you back and the AI that just makes more work with a nicer font. Because both are real, and the difference is almost entirely in your hands.

Welcome to Excel at AI in Finance, the show for finance pros who want to use AI to get home sooner and be the one their company can't afford to lose. I'm Mike, I run F9 Finance, and I'm a senior finance leader working in the trenches. Each week, I share the AI tools, workflows, and conversations that get you where you want to be. All right, let's get into it.

1:20Two finance jobs, same week, opposite results

Let me make this real with something off my own desk. I had two jobs land in the same week, and I pointed AI at both of them. Same tool, same me. The results could not have been more different, and that one week taught me pretty much everything I now believe about this.

The first job was the monthly variance commentary. You know the drill. Revenue's off plan, margin moved, and somebody on the leadership team wants a clear paragraph explaining why by the end of the day. So I fed the numbers in and asked AI to draft the narrative. 30 seconds later, I had a confident, polished write-up.

Full sentences. Sounded like me on a good day. And it was wrong. Not obviously wrong, quietly wrong. It had pinned a big chunk of the unfavorable variance on volume when I knew, because I'd lived the month, that it was a timing issue on a couple of large accruals.

The AI couldn't have known that. It just wrote the most plausible-sounding story the numbers could support. If I'd forwarded that to my CFO, I'd have been defending it in a meeting I really didn't want to be in.

So I fixed it. And here's the part that stuck with me. Fixing it, going back through, unwinding the wrong logic, rewriting the explanation, took me longer than if I'd just written the thing myself from scratch. The AI didn't save me an afternoon, it cost me one. Plus the 30 seconds.

And the worst part? For about a day, I almost trusted it. It was that convincing.

Now, the second job that same week. I had a messy data export. 47 cost centers, inconsistent naming, a few thousand rows that needed cleaning and reformatting before I could even use them. Normally that's a mind-numbing afternoon of find and replace and manual sorting.

I handed that to the same AI tool. It cleaned it, standardized the names, flagged the duplicates, and handed it back to me in about 2 minutes. And I could tell in one glance that it was right. Because I could see it. Clean columns, consistent labels, row counts that tied out.

10 seconds to check, a whole afternoon back. I remember closing the laptop early that day. The same tool that burned me on Tuesday just handed me my Thursday afternoon. Same week, same tool. One job made it worse, one job made it effortless.

And for a while, I told myself the difference was luck, or the prompt, or the model having an off day. It was none of that. The difference was where I pointed it. And once I saw the pattern, I couldn't unsee it.

So let me back up and show you what's going on underneath. Because this right here is the entire AI game for finance.

4:48Every finance team got AI, and the work changed shape

So here's what's happening across finance right now, and I'll keep this in plain language the whole way. You don't need to be a machine learning person to get any of it. In about a 2-year window, every finance team on the planet got access to AI. Overnight, seemingly.

And the promise was simple. The grunt work shrinks, you get your time back, you go home earlier. For some teams, that's exactly what happened.

For a lot of others, the work didn't shrink at all. It changed shape. It went from doing the work to checking the work.

5:18Workslop, defined

And there's now a name for the worst version of that. A team out of Stanford and a research group called BetterUp coined it this year. They call it workslop. Work that looks polished, came out of an AI, and doesn't hold up. So it gets passed along, and someone downstream has to catch it and redo it.

Now let me show you how big this is, then I'll show you the other side, where AI genuinely earns its seat. Both of those are true at the same time. And let me be clear up front, I'm not down on AI here. I use this stuff every single day, and some weeks it genuinely gives me my life back. This is a where-and-how conversation, because the same tool in the same week can be the best hire on your team or the source of your next fire drill.

5:59The real cost: 2 hours and $186 per person, per month

Start with the cost, because the numbers are wild. That same study surveyed about 1,100 full-time workers here in the United States. 40% of them said they'd gotten workslop in the previous month. Not last year, last month.

And each incident, each piece of AI slop somebody had to untangle, took just under 2 hours to sort out. Almost 2 hours per piece to fix work that was supposed to save time.

They ran the math on it. It comes out to roughly $186 per employee per month in lost productivity. For a company of 10,000 people, that's around $9 million a year. Gone. On work that looked done.

6:40What workslop looks like inside a finance function

And picture what that looks like inside a finance function. It's the AI-drafted reconciliation that balances to the penny, but tied out the wrong two accounts. It's the board commentary that reads beautifully and quietly overstates a margin. It's the formula an AI wrote that returns a clear and confident number that happens to be off by a sign.

None of those announce themselves. They all look finished, and every one of them lands on someone's desk to catch.

And there's a second cost stacked on top of the hours. More than 60% of the people who got handed workslop said they trusted the sender less afterward, rated them less capable. So there's a career cost sitting right underneath the time cost. And I'll come back to that, because for a finance pro, it might move more than the dollars.

7:28MIT: 95% of enterprise AI pilots returned nothing

Now, zoom out from individuals to whole companies. MIT put out a report this year on enterprise AI. They found that 95% of company AI pilots returned nothing. 95%. Tens of billions of dollars spent, and almost none of it showed up in the P&L.

And here's the detail that matters for us. The projects that did work were mostly the boring back office ones, not the flashy front office stuff that got all the budget and all the headlines.

7:57AI makes easy tasks easier and hard tasks harder

One more finding, because it explains the whole thing. The human factors researchers, the people who've studied how humans work with automation for decades, have a line about AI I think about constantly. AI makes easy tasks easier. And it makes hard tasks harder.

I want to say that again. AI makes easy tasks easier. And it makes hard tasks harder.

It's great at the stuff that was already simple. And on the genuinely hard stuff, it quietly hands you a brand-new job you didn't have before. The job of checking its work.

So here's the tell in all of it. And once you see it, you'll see it everywhere. Almost every one of these failures, the workslop, the dead pilots, the wasted hours, happened in the same place. They happened when AI got pointed at the judgment, the call, the assumption, the number you'd have to defend to your CFO.

8:50Where AI genuinely pays off in finance

Now the other side, because if I stopped there, you'd think the answer is to keep AI away from finance, and that's the one conclusion that will actually cost you. There's a study that came out this year in the Journal of Accounting Research. Real field data, not a survey. They tracked 277 accountants and more than 200,000 real transactions across dozens of small and mid-sized companies. Actual work.

And where AI got used well, the results were genuinely good. Productivity went up. The month-end close got faster. But here's the part to hold on to. It wasn't that the accountants did the same work faster.

Their work changed. The AI pulled them off the data entry, the routine coding, the manual grind, and pushed them up toward the communication and the review, the higher-value stuff. And the ledgers actually got cleaner and more detailed, not messier.

So the value was real. But notice where it showed up. It showed up on the high-volume checkable work, the stuff you can verify at a glance. And it showed up as an assistant to the accountant, with a human still making the call.

Think about what that frees up. The 2 hours you'd normally spend keying journal entries or fixing formatting turns into 2 hours on the why: explaining the numbers to the business, pressure-testing the forecast, the work that gets you noticed. AI took the part of the job nobody wanted anyway, the keying and the cleanup, and handed back the part that makes you valuable.

And the same study found the flip side, which nails the whole point. When the accountants leaned on the AI for the low-confidence calls, the judgment calls, and just trusted it, the error rate went up. The exact place people wanted AI to save them the most thinking is the exact place it hurt them the most.

So put the two piles side by side. The place where AI created work and the place where AI created value.

And the line between them comes down to one thing. And it's the least technical thing in the room. It's where you pointed it. Same tool, same data, opposite result depending on the job you gave it.

10:59AI is a multiplier, and the verification tax

So here's my take, and it's simpler than most of the headlines you're reading. AI is a multiplier. That's it. It multiplies whatever you point it at.

Point it at high-volume work you can check in a glance, and it multiplies your output. 10 times the reconciliations, the cleanups, the first drafts, in a fraction of the time. Point it at judgment you can't easily check, and it multiplies your rework instead. You eat a beautiful, confident wrong answer, and now you've got to go find the wrong part.

The mechanism underneath this has a name I use with my own team. The verification tax. Every single thing AI hands you has to be checked before you can trust it. That's the tax.

And on some work, the tax is almost nothing. My data cleanup? I glanced at it, the columns were clean, the counts tied, 10 seconds, done. I kept basically all the time I saved.

But on the variance commentary, the tax was brutal. To verify it, I had to rebuild the reasoning. I had to go re-derive why the number moved, which is the entire job I was trying to skip. So the time savings went negative. I paid the tax and got nothing back for it.

And here's the trap, the one that creates all that workslop. When the verification tax is high, it's tempting to just not pay it. Skip the check, send it along, cross your fingers and hope. And the second you do that, the mistake doesn't disappear. It travels.

It goes downstream, to the next person, to your manager, to your CFO with your name on it. The time you thought you saved just moved to someone else's desk as rework. And now, you're the workslop sender in someone else's story.

Flip it around, and the good version is just as clean. When I hand AI the checkable work, the reconciliations, the cleanup, the first ugly draft, the verification tax is tiny. A glance tells me whether it's right. So I keep almost every minute it saved me, and I spend those minutes on the judgment calls that actually need my brain.

Low verification tax, and you get real time back. High verification tax, and it's a trap dressed up as a shortcut.

13:02The two camps everyone gets wrong

Now, let me tell you what almost everybody gets wrong about this, because the popular takes will walk you straight into that trap. Most of what you read about AI and finance lands in one of two camps. Camp one says AI is coming for your job. Learn it or get replaced. Panic now.

Camp two says AI is overhyped. It hallucinates. It's a toy. Wait for it to grow up. And both camps miss the question you actually have, sitting at your desk on a Tuesday with real work due.

13:31The CFO gap: 88% want it, 36% trust it

And the money is already moving on this. Heading into this year, something like 88% of CFOs put efficiency and productivity in their top three priorities. And about 60% are raising their AI spend. But only about a third of them, 36%, feel confident they can turn that spend into real impact.

Now sit with that gap for a second. The budgets are flying, and 2 out of 3 finance chiefs aren't sure it's working. That's the exact value-versus-work problem we're talking about, just sitting at the top of the house.

Here's the read that matters. The finance work that was always checkable, the reconciling, the formatting, the first-pass drafting, that's the part AI takes first. And what lands back on your desk is more of the judgment, more of the part only you can do: the assumptions, the story behind the number, the call you'll defend in the boardroom.

So your job is shifting. It's moving toward the harder, more valuable half that AI can't touch without you. And that's genuinely good news if you're ready for it. It's a problem only if you keep pointing the tool at the wrong half and wondering why it keeps making a mess.

The finance pros who figure this out end up amplified by AI. They feed it the work it's good at, and they keep the work that makes them worth keeping.

I've watched a version of this play out in real time. The analyst who really gets where AI helps and where it hurts becomes the one the director pulls into every new project, because they've become the person who makes the tool safe to use. That's a real durable edge, and almost nobody has it yet.

15:063 things to do before your next month-end

Okay, so all that said, what do you do about this? Well, it's three things, and you can start right away.

One. Take your recurring finance work and split it into two lists. First list, everything you can check in a glance. Reconciliations, data cleanup, first draft formatting. A rough variance draft you're going to rewrite anyway.

Second list, everything you'd have to rebuild to trust. The estimates, the key assumptions, the board narrative, the stuff you would defend in a meeting. Point AI at the first list this month. Leave the second list on your desk for now. You'll get there, but start where it pays.

Two. Put a check step on anything AI touches before it leaves your hands. Simple rule. If it's going out with your name on it, you read every line and you don't forward it. That one habit is the whole difference between using AI and becoming the person who sends workslop.

Three, measure it. For one month, just notice. Did AI hand you time back, or did it move the work into checking? If you're spending more time reviewing than you save, that's your signal.

You pointed it at the wrong things. Move that task back to your desk, try a different one, no guilt, no stress.

That right there is the whole skill, knowing which list a job belongs on. And this matters way past this one month. The finance pros who learn to draw this line, who know exactly what to hand a machine and what to keep, become the person their whole team trusts with AI. When the tools change, and they will change again and again, that person is the one everyone else turns to.

AI earns its seat in finance when it takes the work you can check in a glance and hands back the work only you can judge. Point it at the judgment and all you've done is mail the busy work downstream.

16:52The 3 questions to ask before you point AI at anything

So before your next month-end, I want to give you something you can hold in your head. Three questions. Ask them before you point AI at anything.

Question one: can I check this in a glance, or would I have to rebuild it to trust it? That's the whole test. If a quick look tells you whether it's right, hand it over. If the only way to trust it is to redo the thinking yourself, AI is about to add work, not save it.

Question two: if this goes out with my name on it, am I reading it or forwarding it? Be honest with yourself. Forwarding AI work you haven't read is the single move that turns a good analyst into the person nobody trusts. Read it, every time. Your name is on it.

And question three: am I pointing this at the busy work or the judgment? The busy work is where AI pays you back with real hours. The judgment is where it costs you, unless you're the one still making the call and just using AI to get there faster.

Now the hard version of it. If the honest answer is that you're using AI to skip the judgment, that time doesn't vanish. It lands on someone else's desk as rework, and they know exactly whose desk it came from.

And look, this is bigger than one close, one deck, one month. The finance pros who learn to draw this line don't just get their nights back. They become the ones their company can't afford to lose.