Every small-business owner knows the feeling. It is the week before the accountant needs everything, and the year’s record-keeping is a drawer of crumpled receipts, a camera roll full of bank screenshots, and a spreadsheet someone started in February and abandoned by March. The owner of a Montego Bay café gives up a Sunday typing receipts into Excel one line at a time. A Spanish Town hardware store pays a junior to do the same. The work is not hard. It is just long, dull, and easy to get wrong when you are tired.

This is exactly the kind of task artificial intelligence was built for — not the glamorous, strategic work, but the patient, repetitive tidying that stands between a business and a clean set of books. Over the past few articles this column has used AI chatbots one file at a time. This week we try something a step further: Claude’s new desktop app, working in a feature called Cowork, which does not wait for you to upload files one by one. You point it at a folder on your own computer and it works across everything inside at once.

Why “point it at a folder” changes the job

With an ordinary chatbot, you upload a file, get an answer, copy it out, and repeat. That is fine for one receipt. It is a slog for a full drawer.

The work is really two jobs — capture and cleanup — and Claude’s
desktop app can do both inside one folder. (Branded graphic by PGH Consulting, LLC)

Cowork is different because it runs as an app on your own machine and works directly in the folders you choose to share. You grant it access to a folder, describe what you want in plain English, and it reads the files where they sit and writes the finished result back into the same place — a real Excel file with working formulas, not text you have to paste somewhere and fix. And this is the part that matters for a backlog: you can drop hundreds of receipts and exports into that folder at once and it works across all of them, rather than making you feed them in one at a time. Nothing is uploaded sheet by sheet, and the output lands where your work already lives. (Cowork is currently a paid feature in research preview, available on the Claude desktop app for Mac and Windows; more on what that means below.)

It still helps to think of the work as two jobs: capture — turning physical receipts and PDFs into rows of data — and cleanup — taking data you already have and making it usable. Knowing which one you are doing tells you what to ask for.

Capture: get the pile into a folder

You do not need a scanner. Snap each receipt with your phone, or save the PDF invoices your suppliers email you, and drop them all into one folder on your computer — if that folder syncs through OneDrive or Google Drive, the phone photos land there on their own. Then open Claude’s desktop app, point it at the folder, and ask plainly: “Read every receipt in this folder and build me a spreadsheet with one row per receipt — vendor, date, total, and tax.”

The starting point: a folder of receipt photos, PDF invoices, and bank
exports, ready for Claude to read in one pass. (Branded graphic by PGH Consulting, LLC)

It works through the whole pile and writes the spreadsheet back into the folder. Where a normal chatbot makes you feed it images a handful at a time, here you can point it at a folder holding a few hundred receipts and let it grind through the lot — the batch is the whole point. If your business genuinely runs on paper — hundreds of receipts a month — a purpose-built capture tool such as Dext is still worth a look: it advertises better than 99 per cent extraction accuracy and syncs straight into accounting software like QuickBooks, Xero, or Sage. But for clearing a backlog, a folder and a plain-English request go a long way.

No formulas, no menus — you describe the job in plain English and
Claude works through the whole folder. (Branded graphic by PGH Consulting, LLC)

Cleanup: the year you already have

This is where working in your own folder really earns its keep. Drop a year of bank transactions — the file you exported and never touched again — into the folder, and describe the mess in words: “Open the transactions file, standardise the dates to one format, remove the duplicate rows, split this column into vendor and amount, then add a category column and group every transaction by type. Save it as a new Excel file so the original stays untouched.”

Picture a distributor with twelve months of cash sales to reconcile. She points Claude at the folder, asks it to strip out duplicates and fix the dates, then to flag any transaction over a set figure and anything that looks like personal spending mixed in with business. She asks for a category column and a sort by month. By mid-afternoon there is a clean, formatted Excel file sitting next to the original — built from data that this morning was unusable, and saved without overwriting what she started with. She has not been replaced by the tool; she made every call about what counts as business spending. The app simply spared her the four hours of mechanical sorting that used to surround those decisions.

The output: a clean, formatted Excel file written straight back into the
folder — a fast first draft to review, not the final word. (Branded graphic by PGH Consulting, LLC)

A note on security — and one real advantage

Receipts and bank exports are among the most sensitive data a business holds: account numbers, customer names, supplier terms. There is a genuine advantage to a tool that works on your own machine — the files stay in your folder rather than being uploaded one by one to a website, and any code the app runs is kept in an isolated space on your computer.

But “works on your machine” is not the same as “harmless,” and this is the part to take seriously. You are granting an app permission to read, change, and even delete files in a folder. So grant access narrowly — point it at the receipts folder, not your whole Documents drive. Keep it in the setting that asks permission before each step rather than letting it run unattended, at least until you trust it. It will ask before permanently deleting anything; let it. And the old discipline still applies: mask full account numbers and remove customer names the job does not need. If you would not hand the folder to a temp you just met, do not hand it over without watching.

What it will not do for you

Claude will miscategorise things. It can file a personal lunch as a business expense, misread a faded receipt, or land on a tidy-looking number where the real figure is ambiguous. For day-to-day cleanup this is easily caught. But the books that go to your accountant, the tax authority, or your bank are your responsibility, not the tool’s. Review the categories, reconcile the cleaned totals against your actual bank balance, and treat the output as a fast first draft — never the final word.

There is still a ceiling, though it is a high one. A folder of a few hundred receipts is comfortably within reach — that is exactly the kind of backlog it is built for — but it will not swallow several years of files in a single pass, and because it is doing real, multi-step work rather than answering one question, a very large batch can run through your plan’s usage allowance partway through and stop. If you are clearing a really deep backlog, feed it in sensible chunks — a quarter or a year at a time — rather than emptying the whole drawer in one go, and reserve Cowork for those bigger jobs while you use a normal chat for one-off questions. And remember it is a paid feature still in research preview: it needs a paid Claude plan, and the desktop app must stay open while it works — close the laptop and the job stops.

What to try this week
1. Put a folder of receipt photos or PDF invoices — a handful or a few hundred — on your computer and ask Claude’s desktop app to build a spreadsheet with vendor, date, total, and tax, one row each.
2. Drop one month of exported bank transactions into the folder and ask it to remove duplicates and standardise the dates — saved as a new file.
3. Ask it to split one messy column into separate vendor, date, and amount columns.
4. Ask it to add a category column and group your spending by type — then review every category yourself.
5. Before relying on anything, reconcile the cleaned totals against your real bank balance.

Keep Claude in the “ask before acting” setting while you learn it, and always verify the figures against your source data before acting on cleaned books.

Next week’s column moves from the books to the boardroom: using AI to write clearer board and management reports in half the time.

Peta-Gaye Hardy is the founder of PGH Consulting, LLC, where she helps finance and operations teams adopt AI in practical, low-risk ways. She writes the weekly AI in Finance & Business column and is based between Jamaica and the United States. Learn more at www.pghconsultinggroup.com. Follow on Instagram @pghconsultinggroup.

Disclosures: This article is informational and does not constitute investment, tax, legal, or accounting advice. AI tools can produce errors and figures should always be verified against source data. Some features described require paid subscriptions — Claude Cowork is a research-preview feature available only on paid Claude plans (Pro starts at about US$20/month), and dedicated capture tools such as Dext are separately priced. The author has no commercial relationship with Anthropic, Dext, Intuit (QuickBooks), Xero, or any product mentioned and was not compensated by them. Readers should consult a qualified professional before acting.