The first real warning showed up in 2020.
ScaleFactor, an Austin-based startup that raised over $100 million on the promise of AI-powered bookkeeping, shut down quietly. The official reason was the pandemic. But a deeper look told a different story.
The platform wasn’t actually running on AI. Behind the dashboard, there was a team of human bookkeepers doing the work manually, split between Austin and an outsourced office in the Philippines. At the same time, the company was selling the product as fully automated.
The cracks started to show. Customers reported errors. One business owner found a $17,000 mistake that had been sitting there for months before anyone caught it.
Most people moved on. The AI-in-accounting story was too attractive to slow down for one failure.
Then, the Bench collapsed in December 2024. Over 35,000 small businesses were suddenly locked out of their own financial records. No warning. No proper shutdown. Just gone. Many discovered their books were months behind.
Fourteen months later, in February 2026, Botkeeper shut down as well. Eleven years of work, close to $90 million raised, and the entire company closed in 72 hours.
ScaleFactor. Bench. Botkeeper. Three companies, five years, more than $300 million raised. All built on the same idea: hand over your books, and the AI will take care of everything.
All three are now gone.
Table of Contents
What actually went wrong?
ScaleFactor sold a vision. A machine learning system that would replace accountants.
In reality, it was humans doing manual work behind a layer of automation. Once customers and investors started asking tougher questions, that gap became impossible to hide. The pandemic didn’t cause the failure. It just gave the company something to point to.
Bench took a different route. It didn’t hide the human layer. It leaned into it, combining software with a bookkeeping team and scaling fast.
But scale came at a cost. By mid-2024, customers were already feeling it. Financial statements were delayed. Support slowed down. And many clients felt like the person handling their books didn’t really understand their business.
When Bench shut down, customers were told to move to a startup called Kick. A company that most of them had never heard of. They had to migrate their entire financial history, right in the middle of tax season, to a platform they hadn’t even chosen.
Botkeeper went deeper on the AI side than both. By the time it shut down, it claimed to categorize about 80 percent of transactions with 98 percent accuracy.
That sounds impressive until you look at what actually happened. The company went from operating to closed in three days.
Ninety million dollars. Three days.
None of these companies failed because something dramatic broke technically.
The real issue was simpler. The system was never built to handle the messy, unpredictable nature of real business finances.
When something unusual came up, there was no experienced person in the loop to catch it, understand it, and act on it. That’s not a limitation of AI. That’s a design choice. And the businesses using these platforms ended up paying for it.
And the industry still hasn’t learned.
You would expect all of this to slow things down.
In some cases, the opposite is happening.
There are platforms today pushing even harder into the AI-only model. One is marketing a fully autonomous system with no human review at all. Another is openly aiming to become a pure software subscription product, removing professionals from the process entirely.
From an investor’s perspective, it makes sense.
Software businesses scale better. Margins are higher. Recurring revenue is predictable. It’s a cleaner story.
But step back for a second.
Think about your bank balance, your taxes, payroll, and money coming in. Now imagine handing all of that over to software, with no one checking the output before it drives your decisions.
Not assisted. Not reviewed. Fully unsupervised.
Most business owners wouldn’t actually agree to that. Not because they don’t believe in technology, but because the risk is real, and they’re the ones who carry it.
A delayed payment. A payroll classification mistake. A tax rule that changed last quarter.
These are not edge cases. They are normal situations. And they require context, someone who understands your business.
That part doesn’t get automated.
The problem with “98% accuracy.”
This is where things start to get uncomfortable.
Let’s say your business processes 500 transactions a month. At 98 percent accuracy, that still means 10 errors every single month.
If no one reviews those errors, they don’t stay small. They flow into your reconciliations. Then, into your financial statements. Then, into your tax filings.
After three months, you’re not dealing with 10 mistakes. You’re dealing with a chain of errors that now shape how you run your business.
And that 98 percent assumes clean, simple inputs.
Things like multi-entity setups, deferred revenue, contractor payments, or state-specific tax rules are exactly where these systems start to slip. Quietly.
No alert. No warning. Just numbers that are slightly off, until something breaks later.
A miscategorized expense can distort margins for an entire quarter. A missed accrual can make your cash flow look healthier than it is.
When those numbers influence hiring or pricing decisions, the impact is not small.
Accuracy percentages don’t mean much unless someone is responsible for catching what’s wrong and fixing it quickly.
What actually works?
The setups that hold up in the real world look very different.
The best firms don’t try to choose between AI and human expertise. They separate responsibilities clearly.
Automation handles the heavy lifting. Categorization, matching, reconciliations, and flagging issues. The repetitive work.
This alone can save five to fifteen hours a week for most service businesses.
But the moment you move into decisions, things change.
Tax strategy. Revenue recognition. Compliance. Cash flow planning. Anything that doesn’t follow a predictable pattern.
These require context.
It’s not about whether a transaction is tagged correctly. It’s about what that transaction means for your business, right now.
That understanding builds over time. Through experience. Through continuity. Through someone actually knowing how your business works.
That layer doesn’t come from software.
Firms like Numetix are built around this split. AI handles the data. Professionals handle the decisions.
Remove either one, and the system breaks.
The bottom line
ScaleFactor shut down in 2020 after raising $100 million. Bench shut down in 2024 after raising $113 million. Botkeeper shut down in 2026 after raising $90 million.
More than $300 million has gone into a model that tries to replace professional oversight with automation.
It has now been tested multiple times at scale.
The result has been consistent.
For any business looking at bookkeeping solutions today, the question is not whether AI should be part of the process. It should. The efficiency gains are real.
The real question is simpler.
Is there someone qualified reviewing what the system produces before it shapes your decisions?
Because when something goes wrong, or when a deadline hits, software doesn’t take responsibility.
People do.
