Schedule C from bank statements

Why doing it by hand takes a weekend

A modest sole proprietorship runs several hundred transactions a year across two or three accounts. Typing them costs hours, and every hour of typing introduces the errors that later refuse to reconcile — a transposed digit, a skipped row, a refund entered as a cost.

It is worse than it sounds because the work is not linear. You type for an hour, find a payment you cannot identify, stop to look it up, lose the thread, and come back to a spreadsheet where you are no longer sure what has been entered. Most abandoned tax-prep attempts die there rather than at the tax questions.

Extraction removes the typing entirely and, more importantly, removes the uncertainty about whether everything is in. The remaining work — deciding what each payment was for — is the part that actually needs you.

flowparse.io

Step 1 — Gather every account, every month

Start with a written list of accounts rather than the pile of PDFs you happen to have: the business checking account, every business card, any personal account or card that was used for business costs, PayPal, Stripe or another processor if money arrived that way, and anything closed during the year.

Download all twelve statements for each. Name them consistently — account, year, month — so a missing period is visible at a glance rather than discovered halfway through. This is ten minutes that reliably saves an afternoon.

Closed accounts are the classic gap. If the account no longer exists, the bank will usually still provide historical statements on request; ask early, because that request is the one thing in this process with a lead time you do not control.

Step 2 — Extract, do not retype

Upload the statements and let extraction produce the table: date, description, signed amount and running balance for every transaction, from any bank, with no template to set up. Scans and photographs go through OCR first, and anything uncertain is flagged rather than silently guessed.

Two details matter more than they sound. Debits and credits have to collapse into one signed amount, so a refund reduces the cost rather than inflating it. And the running balance must never be exported as the transaction amount — an error that puts a plausible wrong number on every row of the year.

See bank statement to Excel for the conversion itself and scanned statements if yours only exist as images.

Step 3 — Prove nothing is missing

This is the step that separates a defensible set of figures from a hopeful one. A statement extracted with three transactions missing looks perfect: every extracted row is right, nothing is flagged, and the totals are only wrong if you compute them.

The statement carries its own proof. Opening balance, plus every transaction, equals the closing balance the bank printed. FlowParse runs that check automatically on every statement, per account, and names the rows where the arithmetic breaks.

Then check the sequence: each month's closing balance should equal the next month's opening balance, with no gaps. Those two checks together are the difference between "I think this is my year" and "this is provably my year".

flowparse.io

Step 4 — Put the whole year in one place

Twelve statements across three accounts is thirty-six documents. Converting them one at a time and pasting the results together is where errors enter — mismatched columns, a file processed twice, one skipped entirely.

Smart Merge consolidates up to a hundred statements into one workbook with unified columns, duplicate detection across overlapping periods and a source-file reference on every row, so any figure can be traced back to the exact statement it came from.

That traceability is worth more than it seems. When your preparer asks what a $4,180 payment in March was, the answer takes ten seconds instead of a re-derivation.

flowparse.io

Step 5 — Group spending the way the form is grouped

With the year in one sheet, sort by payee rather than by date. The top twenty payees usually cover most of the volume, and each one is a single decision applied to dozens of rows — advertising, contract labour, insurance, office expense, rent, repairs, supplies, travel, utilities, and so on.

Keep a category column rather than splitting into tabs. One flat table with a category per row can be pivoted a dozen ways; a set of tabs cannot, and it hides the transactions nobody classified.

Leave anything you cannot identify in an explicit "unknown" bucket instead of guessing. A short list of genuinely unidentified payments is a normal, honest output of this process — and it is exactly what a fifteen-minute call with your preparer resolves.

A working category set

The point of this list is not to be a tax authority — it is to give the spreadsheet consistent buckets that map cleanly onto whatever lines your preparer uses. Adapt it to your business rather than forcing your business into it.

The transactions that are not what they look like

Transfers between your own accounts are neither income nor expense. If both accounts are in the dataset, the same money appears twice with opposite signs, and treating either as real inflates both revenue and costs. Pair them and mark them.

Owner draws are not a business expense, and money you put in is not revenue. In a sole proprietorship these are the most frequently miscoded transactions of all, because from the bank's point of view they look exactly like everything else.

Processor payouts are net. A deposit from a payment processor is gross sales minus fees minus refunds, so recording the deposit as revenue understates both your sales and your costs. The processor's own report is the other half of that picture — see processor statements.

Refunds and chargebacks reduce the original category rather than creating income. And a card payment that settles the card balance is a transfer, not a cost — the costs were the individual card transactions.

Do not forget the card

Business cards carry a large share of small expenses, and their statements are a separate document set with their own format. Convert them the same way and keep the account column populated so card spending stays distinguishable from checking-account spending.

The double-count trap is worth naming: the monthly payment from the checking account to the card is a transfer. The expenses are the individual transactions on the card statement. Counting both is one of the most common ways a reconstructed year overstates costs, sometimes dramatically.

Credit card statement conversion handles the layout differences, including the summary boxes that look like tables and are not.

What to do about missing documentation

You will have payments with no receipt. This is normal in a reconstructed year and it is better handled explicitly than quietly.

First, recover what exists: supplier portals, order-confirmation emails, card portals and online accounts often reproduce a purchase history that replaces the paper you lost. An hour of this recovers a surprising proportion.

Then list what remains — payee, date, amount, what you know — and give that list to your preparer. Whether a cost can be claimed without a receipt is a question of law and of your circumstances, and it is exactly the sort of thing a professional answers in a sentence. Software should not, and this one does not.

Gross receipts: the side people get wrong

Expenses get all the attention, but the income side is where reconstructed accounts most often go wrong, because money can arrive by several routes: direct client transfers, processor payouts, marketplace settlements, cash deposits.

Each route needs treating on its own terms. Direct transfers are usually gross. Processor payouts are net and need the processor's report to split out fees and refunds. Marketplace settlements can be net of a whole fee stack. Cash deposits are only visible if they were banked.

The practical test: does total income in your sheet agree with what your invoicing or platform records say you earned? If it does not, the difference is a question worth answering before the return is prepared, not after.

This is also the input for estimated tax

The same dataset feeds the payments you make during the year rather than after it. If you are making quarterly estimated payments, the profit figure they should be based on is exactly what this exercise produces.

Doing the extraction quarterly instead of annually turns a weekend into fifteen minutes four times a year, and gives you a running profit figure rather than a surprise. See quarterly estimated taxes from statements.

Getting it into accounting software or a preparer's system

If you keep books going forward, import the year rather than retyping it: QBO files for QuickBooks, a Xero statement CSV, or plain Excel and CSV for anything else.

If your preparer works from a spreadsheet — many do for sole proprietors — the Excel export with a category column is exactly what they want, and it is far easier for them to review than a shoebox.

Either way, keep the original statements. The extracted sheet is a working document; the bank's PDF is the record, and retention rules apply to it rather than to your spreadsheet.

flowparse.io

A worked example: one consultant, one year

A consultant with a business checking account, one card and a payment processor. Twenty-four statements, roughly 700 transactions, no books at all.

Conversion of all twenty-four documents takes minutes, and the balance check fails once — a statement whose mid-page summary box swallowed two rows, caught and re-processed immediately rather than discovered in April.

Sorting by payee, eighteen payees cover about two-thirds of the transactions and take under an hour to categorise. Nine transfers between own accounts are paired and excluded. The processor payouts are split against the processor report, which moves both revenue and fees by a four-figure amount.

What goes to the preparer: one workbook with a categorised year, a short list of eleven unidentified payments, and the note about the two recovered rows. The preparer's job starts where it should — at the tax questions.

Who this helps most

What a preparer actually wants from you

Ask any tax preparer what makes a sole-proprietor return slow and the answer is not the tax: it is chasing a client for numbers that do not exist yet, then trying to work out whether the numbers that do exist are complete.

Handing over a categorised, balance-checked spreadsheet with a source reference on every row changes the engagement. It moves the conversation from "can you find your bank statements" to "is this cost deductible", which is where their value actually is — and it usually reduces the bill.

For firms doing this at scale, the tax preparer workflow covers batching it across clients.

How this differs from our other tax pages

Bank statements for tax return is the general version of this job: organising a year of statements for any return, in any country. This page is narrower — it is about the US sole-proprietor case where the output has to fit the Schedule C shape.

Bank statements for self assessment is the UK equivalent, with the different form, deadline and vocabulary that go with it.

And how to prepare bank statements for taxes is the step-by-step guide behind all of them. If you are choosing where to start, start there and come back here for the Schedule C specifics.

How much you can trust the extraction

Field accuracy on modern extraction is high, and it is the wrong number to focus on. The failure that matters is a row that never came out at all, because it produces no error, no flag and no visible symptom.

That is why the balance check exists and why it runs on every statement rather than on request. It is arithmetic rather than an opinion: it needs no reference data, and it fails loudly. See extraction accuracy for what that proves and what it does not.

Everything opens in an editable preview before export, with uncertain figures highlighted, so the review effort goes where the uncertainty actually is.

Your statements, handled carefully

Bank statements are among the most sensitive documents anyone owns. Uploads run over TLS, processing happens in EU data centres, the original PDF is deleted immediately after extraction, extracted data is stored encrypted, and documents are never used to train models — the detail is on the security page.

Worth saying plainly: FlowParse is a hosted service and cannot be self-hosted. If your requirement is that client documents never leave your own machine, a desktop converter is the honest answer and we are not it.

A readiness check before you hand anything over

Run these six checks before the data goes anywhere. Each one takes under a minute and each catches an error that is expensive to find later.

The boundary, stated plainly

FlowParse turns documents into data. It does not compute your tax, decide what is deductible, apply a rate, issue a form or file anything with any tax authority. Those are decisions about your circumstances and the law where you live, and they belong to you and your accountant.

It is also worth being precise about what a bank statement proves. It proves a payment happened: this amount left this account on this date, to this counterparty. It does not prove what was bought, whether the cost was for the business, or how much tax was inside it. For anything where the tax authority wants to see the nature of the expense — a VAT or sales-tax claim above all — the invoice or receipt is the evidence and the statement is the corroboration.

What that leaves is still the slowest part of the job: getting a year of transactions out of PDFs, complete, correctly signed, in one sortable place. That is what this page is about.

Where to start today

Take the single messiest statement you have — the longest one, or the scanned one — and convert it. Check the closing balance against the extracted total. That one test tells you more about whether this will work for your year than any description.

Then do the rest of the account, then the other accounts, then categorise. The whole sequence for a typical sole proprietorship is an afternoon rather than a weekend, and the part that remains is the part where you actually know something the software does not.

Frequently asked questions

Related