When the books can't be trusted: a rebuild from the bank up

The company itself was doing fine. Revenue was real, customers were real, and operations ran on a solid ERP system that the ops and IT teams kept in good shape. The problem was that the books, in any meaningful sense, didn't exist.

Financial reports were produced once a year for the tax preparer, and they never balanced. Nobody had looked at the trial balance in years. Most balance sheet accounts carried unexplained items that had been sitting there for four or five years, and the cash accounts were off by close to half a million dollars because they had never been reconciled. The accounting had been kept without any real double-entry discipline, so there was nothing underneath the reports to stand on.

The executive team already knew something was wrong. What they needed was to find out what was actually true, and they needed it quickly, because it's very hard to manage a business on numbers you don't believe.

Start with the record that can't lie

When books are this far gone, you don't really fix them, you rebuild them, and the place to rebuild from is the most objective record available, which is the bank. Every dollar that ever moved in or out of the company is sitting in the bank transaction detail regardless of how it was recorded, or whether it was recorded at all.

So that's where we started, with years of transaction-level data across every account. This used to be the part that made a rebuild a six-month engagement, because someone had to read and classify all of it by hand. Working in parallel with Claude, that part compressed dramatically. The AI went through years of transactions and sorted them into the structure we defined, we reconstructed the P&L from the transaction level up, and we rebuilt the balance sheet account by account, using the ERP's operational data as a cross-check along the way.

The old balance sheet residue didn't get carried forward like it had been every year before. Each item was traced back to what it actually was, explained, and cleared, and for the first time in the company's history the cash accounts were reconciled down to the penny.

The same rule as always

We ran this rebuild on the same rule we run every engagement: AI does the mechanical work, and a CPA controls the file from A to Z. The classification standards were ours, and every judgment call about whether something was an asset or an expense, or owner activity, or just noise, was made by someone who has been closing books for twenty years. That's what makes the result defensible. These aren't approximated financials, they're a clean set of books that a tax preparer or a lender or an auditor can walk through line by line and follow.

What the company has now

The company now has a completely new set of books built on proper double-entry accounting, reconciled every month, with a close that takes two days. The executive team sees real margins, trends, and where cash actually stands every month, and for most of that information it's the first time they've ever had it.

That last part matters more than the cleanup itself. Books that don't exist are expensive in ways that never show up as a line item: decisions made on bad numbers, prices set blind, months of running the business on instinct because nobody trusted the reports. The rebuild didn't just clean up the past, it gave the executive team something they could actually steer with.

The quiet signs this is you

Nobody thinks their books are fiction until they go looking. In practice the tells are pretty consistent. Reports only get produced when the tax preparer asks for them. Nobody reviews the trial balance. Balance sheet accounts carry old balances that nobody can explain but that roll forward year after year anyway. Bank reconciliations don't happen, or the cash number in the books is one that nobody in the building would bet on.

None of this usually means anyone was dishonest. In most cases the person keeping the books was simply in over their head and nobody was in a position to check their work. It happens far more often than most executives would guess, including at companies doing very real revenue.

If some of this sounds familiar, the encouraging news is that the fix is much faster than it used to be. What was once a six-month forensic project can now be done in weeks, without giving up any of the rigor.