Call Us
805-225-4959
805-225-4959
AI-assisted financial reconstruction
A five-year, 13,000-transaction cleanup is the kind of job most firms lose money on, or quote so high the client walks away. We ran this one with Claude doing the heavy compile work, and it became one of the most profitable engagements we have taken.
Five years brought current in about eight weeks, by one bookkeeper, with every statement closed to a $0.00 difference. Here is how the math changed.
What we were handed
Before any of it was rebuilt.
The problem with cleanups
Our client, a Ventura County entertainer running a solo performance business and a small side venture, arrived five years behind on taxes. The money had moved through eight bank and card accounts and four payment processors, feeding an accounting file that had quietly stopped telling the truth. Nearly everything had been swept into a single catch-all bucket that was later deleted, and three financial lives were blended into one. The old file was past saving, so we scrapped it and built a brand-new QuickBooks company from the statements up.
A backlog like this means opening more than 2,000 files: over 500 bank, card and processor statements running to 3,400+ pages, plus roughly 600 exported transaction files. Then rebuilding 13,000+ transactions by hand, one line at a time, and reconciling every account to the penny. Do the arithmetic and it is 600+ hours of mechanical work.
The first bookkeeper who took this on walked away from it. Not for lack of skill, but because the volume was simply more than one person could work through by hand. That is the honest problem with deep cleanup: quote it at what the hours are worth and the client cannot afford it, quote a fixed fee to win the job and you absorb the overage yourself. For most small firms, it is the work you turn away.
The popular workaround is to ship the data entry offshore, and it is where a lot of firms now turn. It lowers the hourly rate, but it does not fix the economics. You are still paying a team by the hour, still coordinating across time zones, and still reviewing every line yourself, because the responsibility never leaves your desk. The margin stays thin and the oversight stays heavy. Offshoring changes who does the typing. It does not make the job worth taking.
We wanted to say yes. So we changed how the work gets done, not just who does it.
The approach
Instead of hand-keying five years of backlog, we used Claude, Anthropic's AI, to do the mechanical heavy lifting: read every statement, extract every line, reconcile each period to zero, flag whatever did not add up, and hand back a clean QuickBooks Online company file for the client's CPA to file from. Our bookkeeper set the rules and reviewed the results. The AI did the compiling.
What the AI actually did
This is the part of a cleanup that is pure mechanics: enormous, unglamorous, and exactly what an AI is suited to. Claude carried all of it.
Bank of America, Chase and Square each print their statements differently. Claude read all of them, including scanned and reformatted PDFs, without a person retyping a single page.
Every statement was reconciled to a $0.00 difference. When a period did not tie out, that was the signal to look again. The reconciliation was the proof the compile was correct.
Across the full dataset, Claude flagged roughly $140,000 of "phantom income" that was really currency conversion, family support that had been booked as sales, and transfers that had been counted twice.
With the old QuickBooks company scrapped, Claude coded every transaction to a fresh chart of accounts and produced import files one account per statement period, so the new company loaded in order and reconciled cleanly from the first statement forward.
Illustrative figures. Every statement period was closed to a $0.00 difference before a single row was allowed into the books.
The AI did the compiling. The judgment stayed with us.
The division of labor that made it work
Where the human stayed essential
Claude never made the calls that matter. A person decided what counted as income versus mechanics, where one entity ended and the next began, and how each account should be coded. Every output was reviewed by our bookkeeper before the file went to the client's CPA.
That is the whole model. The technology removed the keystrokes, not the responsibility. It let our expertise land on the judgment, the anomalies and the sign-off, instead of being buried under data entry. The client is not paying for typing. They are paying for books a professional will stand behind, and a CPA can file without second-guessing.
Nothing reached the client's CPA that a person had not looked at and stood behind.
The economics, rebuilt
The change is not that we worked harder. It is that the largest, most mechanical part of the job stopped consuming billable time. Here is the same engagement, two ways.
The by-hand figure is a conservative labor estimate, 600+ hours at standard bookkeeping rates, to open the documents, key 13,000+ transactions and reconcile every account. Built from the actual file: 2,000+ source documents and 3,400+ pages of statements.
By hand, 600+ hours is months of full-time work, and it defeated the first bookkeeper who tried. One of ours, working with Claude, delivered the entire reconstruction in about eight weeks, while still managing a full roster of other clients.
By the numbers
The takeaway
The backlog that used to be a money loser is now work we can take on, price fairly and deliver at a healthy margin. We handed the client's CPA five years of clean, fileable books, each account traceable to the statement it came from and every dollar accounted for exactly once. The client got unstuck, the CPA got a file they could trust, and we walked away with an engagement that pencils.
The difference was putting AI on the part of the job that was always mechanical, and keeping our judgment on the part that never was. That combination is what lets a firm our size take on a reconstruction this large, and still make it worth doing.
How we would do yours
However many years back it goes, and however many accounts it runs through, the process is the same three steps.
Bank, card and payment processor PDFs, as far back as the backlog goes. No spreadsheets to prepare first.
Every line extracted, every statement reconciled to zero, every anomaly flagged for a person to rule on.
Clean, properly categorized, and ready to hand to your CPA or drop straight into QuickBooks.
Send us the statements. You'll have a fixed quote within a day, and no hourly meter running while we look.