Where does it go?
Only to FinStat, and only the documents you choose to send. No access to your machine, your firm’s folders, or anyone’s accounts.
For transaction advisory, deal diligence and SBA 7(a) lenders
The seller’s export in. A ledger you can defend out.
A quality of earnings report asks whether reported earnings are real. Before anyone can answer that, someone has to reconcile a messy ledger by hand — and on a live deal that is weeks, on the critical path. FinStat is the accounting tool your AI needs to do that part. Every transaction, not a sample. It never connects to the seller’s accounting system, because a seller in diligence will not grant access to one.
1,000,000 free tokens · no credit card · unlimited workspaces, no per-deal fee
01 / The job, start to finish
From the data room to a ledger you can defend.
The EBITDA bridge cannot start until the ledger is trustworthy. Today that means an analyst rebuilding a year or three of history before anyone can test a single add-back. With FinStat, that reconstruction takes hours, not the first weeks of the engagement.
Hand your AI what the seller actually sent.
The accounting export, bank and card statements, tax returns, internal financials. PDF, image, CSV or the export zip. No live connection to their system, and none required.
FinStat normalizes, reconciles and seals it.
Every statement proven against the bank’s own printed balances. Every transaction tied to its page. Statements for one account recognized as the same account across periods, even when the institution is spelled two different ways. Anomalies surfaced rather than smoothed over. Tax returns and internal financials your AI reads beside FinStat’s reconciled books.
You build the bridge.
Add-backs, non-recurring items, related-party costs, the revenue quality opinion. Every figure traces to a sealed document, so a number you defend to a lender has a page behind it. The judgment stays yours.
02 / What FinStat does on a deal
Accurate books, fast, from whatever the seller sends.
FinStat builds highly accurate books and converts existing ones. It analyzes the seller’s existing books directly from their export, and it builds new books from the statements when the existing ones cannot be trusted — so you can put the two side by side.
- Statements, digital or scanned — checking, savings, wallet, credit card, line of credit, loan and brokerage statements, read fast from digital and image-based PDFs.
- Claims and detail — invoices, bills, credit and debit notes, receipts, processor settlements and payroll statements.
- Checked and sealed — every document is checked against the accounting law of its form, then sealed. The seal makes tampering detectable, so the file behind every number in your report is the one FinStat checked, and your AI’s analysis runs on data known to be exactly what FinStat checked.
FinStat sits ready for your AI to hand it work. Your AI decides what to do itself and what to give FinStat — by design — so the analyst’s time and the AI’s tokens go to the judgment, not the reconstruction. How FinStat compares →
03 / What the engine catches first
A shorter list of places to look.
Real findings from one completed engagement, anonymized. None of this is an answer. It is a list of places to look, each with its evidence attached — the first-week checks a good analyst does by hand, done for you. What any of it means stays with you.
- The account that is not what it saysA “payroll processing fees” line holding the wages themselves.What it means
Over 40% of revenue, in an account named for a fee. Read at face value, the business appears to have almost no labor cost and a crushing administrative burden. The opposite of the truth.
- The same account, twiceOne institution spelled two ways becomes two accounts.What it means
Five apparent accounts in that engagement were duplicates. Separately, one last-four was shared by two unrelated institutions, so matching on last-four alone merges accounts that have nothing to do with each other.
- The impossible balanceAn expense account carrying a credit balance.What it means
Funded, on inspection, by receipts that are not refunds. Whatever it is, it is not a supply expense, and left alone it nets against cost and overstates earnings.
- The label that misleadsAn account called life insurance, mostly holding retirement plan contributions.What it means
The two have opposite add-back treatment. An owner’s personal policy comes back; a staff retirement match does not. Read the label and you get the bridge wrong.
Every one of those is machine-findable. Every line of the bridge is still yours.
October 1, 2026SOP 50 10 8.1 takes effect, for applications issued an SBA loan number on or after that date
$3M purchase priceand above, on 7(a) acquisitions and business expansions, where a quality of earnings report becomes mandatory — measured before buyer equity or seller financing
TTM + two yearsthe Cash Proof must reconcile bank statement data to the income statement and tax return for each period, on a trailing twelve-month basis and the last two fiscal years
The lenderis who the report is for. It needs an “independent, experienced financial professional” and “may not be prepared by or for the borrower or seller” — no credential is named
Source: SBA SOP 50 10 8.1, Appendix 15 · effective October 1, 2026
04 / Proof of cash
Every close is the next open.
The mandated cash test is a continuity chain: each period’s closing balance has to be the next period’s opening balance, on the same account, from the bank’s own printed figures — across the trailing twelve months and the two most recent fiscal years, on every account the business actually used. This is how the engine runs it.
For the cash proof, statements are enough.
No login to the target’s bank and no connection to their accounting file. The statements are both the input and the evidence, which is why a seller in diligence has nothing to refuse.
Accounts are identified, not assumed.
Nothing tells the engine which statements belong together. It binds each period to an account from the printed facts — owner, institution, account number. That matters in the field: on a live engagement the ledger presented twenty-three institution-and-account pairs and only eighteen were real accounts. Getting that wrong by hand is how a cash test quietly reconciles to the wrong thing.
The chain holds, or it comes back as a question.
Every period reconciles against its own printed controls and its closing balance has to equal the next opening balance to the cent. Nothing the documents cannot settle is guessed at — it returns as a review item with the source page attached, for a person to decide.
The last period costs what the first one did. That is what makes the lookback the SBA now requires a routine run rather than a project.
05 / Lender questions
“What does the AI do with the data?”
Written so you can hand it over as it is.
Does it touch the target’s systems?
No. It never logs in to a bank, never connects to their accounting file, holds no funds and cannot move money.
Does it decide the add-backs?
No. It reconstructs the books from the documents and surfaces what looks wrong. The add-back schedule and the earnings opinion are the analyst’s work, and they should be.
Can we show where a number came from?
Every figure traces to a sealed source document. Show the lender the page, not a summary. A sealed document cannot be edited afterwards without it being detectable.
The privacy policy and the subprocessor list are public, written to go in front of a lender or an engagement letter as they are.
06 / Questions deal teams ask
Straight answers.
- Does this replace analysts?
- No. It replaces the weeks before the analysis: rebuilding a ledger nobody can yet trust. The bridge, the add-back judgment and the revenue quality opinion stay with the people whose names go on the report.
- The seller will not give us access to their accounting system.
- They do not need to. FinStat works from what a seller will actually hand over — an export, statements, returns, PDFs.
- We look back three to five years. Does it all fit?
- Export year by year rather than all dates at once. Very large QuickBooks exports can be cut off without warning, and the general ledger is the file that truncates first — a QuickBooks limit, not an SBA one. A silently truncated ledger is exactly the kind of quiet data loss diligence exists to catch.
- What about IRS transcripts? The rule names them.
- It does: the analysis must reconcile accountant-prepared statements, tax returns, internal financials and IRS transcript data. Your AI reads the transcript beside FinStat’s reconciled books and ties it out, the same way it tied the rebuilt ledger to the filed return.
- What does it cost per deal?
- Nothing per deal. Every engagement gets its own workspace and workspaces are unlimited: no seats, no per-client fee. After the free tokens, metered work is $30 / 1M tokens. Metered work is the model-backed part: reading a new document into books, drafting a chart, and rebuilding or auditing an existing set of books. Reports, review items, verification, exports and re-reading a sealed document are not metered.
- What happens when something does not add up?
- A statement that does not reconcile is refused, with the reason named at the line. Ambiguity becomes a review item with its evidence attached. The engine does not quietly pick an answer to keep the file moving.
- Can the target’s own accountant do the first pass?
- Yes, with their own account, and hand you sealed, source-tied documents. No shared credentials. You keep the review, and you can still see every source page behind every figure.
Before you scope the engagement
Start on one deal.
Take the target whose books you least want to rebuild. Hand the export to the AI you already use, with FinStat connected. Ask what does not tie. How FinStat compares →