{"course":{"slug":"ai-for-small-business","title":"AI for a Small Business","url":"https://addaly.com/learn/ai-for-small-business","level":"beginner","track":"work"},"lesson":{"slug":"books-and-invoices","title":"Bookkeeping Without Handing Over Your Accounts","url":"https://addaly.com/learn/ai-for-small-business/books-and-invoices","position":48,"of":88,"minutes":8,"answer":"Use it to sort and to write; never to add up, and never as the place your records live.","keyIdea":"Use it to sort and to write; never to add up, and never as the place your records live."},"body":{"format":"markdown","text":"## What it is actually good at here\n\nFour things, reliably:\n\n- **Sorting messy descriptions.** Turning \"PAYPAL *SHPFY 4471\", \"UPI-SWIGGY-3382\" and \"TFR RE INV MAR\" into categories. This is language work, and it is the strongest use in your books.\n- **Reading receipts.** A photo of a drawer of paper turned into rows you check.\n- **Writing the awkward email.** Chasing an overdue invoice without burning the relationship.\n- **Translating your accountant.** \"What does she mean by accrual basis and does it change what I send her?\"\n\n## What it is not\n\n**Not a calculator.** A language model predicts the next piece of text. Ask it to total a column of 300 numbers and it will give you a number that looks right and often is not, with no signal that anything went wrong. Some tools now write and run real code to compute, which is different and much better, but you still check the top-level total against your bank balance. Sums live in a spreadsheet.\n\n**Not your records.** A chat window is not a ledger. It is not backed up, not auditable, not something your accountant or a tax inspector can work from. The books live in accounting software, or a spreadsheet you own. The AI is a helper beside them.\n\n**Not a tax adviser for your country.** Rules are jurisdictional, they change every year, and the answer will be fluent regardless. Use it to understand a concept. Never to decide a filing.\n\n## Strip before you paste\n\nMost of the sensitive data is not needed for the job.\n\nTo categorise a transaction you need the description and the amount. You do not need the customer's full name, their card number, their account number, their address or their phone. Replace them: `CUST-114`, `CUST-115`. Keep the mapping in your own file, on your own machine.\n\nNever paste, under any circumstances: full card numbers, CVVs, bank login details, national ID numbers, or anything from a payment processor's raw export you have not read.\n\nA five-minute find-and-replace before pasting removes almost all of the risk in this lesson.\n\n## Prefer the tool that already holds the data\n\nIf your accounting software has AI features (Zoho Books, QuickBooks, Xero, Wave, Tally, whatever you use) check those first. Not because they are cleverer. Because that vendor already holds your data under a contract you already signed. Pasting the same rows into a second company's system creates a second copy in a second place that can be breached, and a second set of terms you have not read.\n\nFewer copies, fewer problems.\n\n## Chasing money\n\nThis is where it pays for itself in a single afternoon.\n\nA design studio in Nairobi had KSh 480,000 outstanding across 14 invoices, none of them chased, because writing fourteen individually awkward emails is the sort of job that slides to next week for a month.\n\nFeed it the list: client reference, amount, invoice date, days overdue, and one line of relationship context (\"good client, always pays, just slow\" versus \"third reminder, going quiet\"). Ask for fourteen short emails, matched in firmness to the context, each one naming the amount, the invoice number and a specific date.\n\nTwenty minutes, and you read every one before it goes. The tone ladder (polite nudge, firmer, formal notice) is exactly what most owners get wrong under stress, either too apologetic to work or too sharp to survive.\n\n## The reconciliation rule\n\nOne habit makes all of this safe. Whatever the AI touched, the bottom line still has to match your bank statement. If it does not match, the AI's work is wrong until proven otherwise, not the bank.\n\nThat check takes two minutes and it is the whole safety net."},"check":{"question":"A shop owner pastes 300 bank lines and asks for categories plus a total per category. What should she trust?","options":[{"text":"The categories as a first pass to review, with the totals computed in a spreadsheet","correct":true},{"text":"The totals, since arithmetic is the easy part; the categorising is the hard judgment call","correct":false,"tempting":"Arithmetic feels like the easy half, but a text model produces a plausible-looking total rather than actually computing one."},{"text":"Both, because it showed its working line by line and the steps all check out","correct":false,"tempting":"Visible working reads like proof, but the steps are generated the same way the answer is, and a wrong step reads as smoothly as a right one."},{"text":"Neither: this is her financial record and no AI belongs anywhere near it","correct":false,"tempting":"The caution is reasonable, but it throws away the one part the model does well: reading messy transaction descriptions."}]},"prev":{"slug":"pricing-conversations","title":"Price: the one thing that never comes out of the model","url":"https://addaly.com/learn/ai-for-small-business/pricing-conversations"},"next":{"slug":"spreadsheets-without-fear","title":"Ask for the formula, never for the total","url":"https://addaly.com/learn/ai-for-small-business/spreadsheets-without-fear"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Bookkeeping Without Handing Over Your Accounts', in AI for a Small Business. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-small-business/books-and-invoices (Accessed: 31 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Bookkeeping Without Handing Over Your Accounts","courseTitle":"AI for a Small Business","module":{"index":6,"slug":"money-and-the-back-office","title":"Money, paperwork and the back office"},"answer":null,"keyIdea":"Use it to sort and to write; never to add up, and never as the place your records live.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"what-it-is-actually-good-at-here","runs":[{"text":"What it is actually good at here"}]},{"t":"p","runs":[{"text":"Four things, reliably:"}]},{"t":"ul","items":[[{"text":"Sorting messy descriptions.","b":true},{"text":" Turning “PAYPAL *SHPFY 4471”, “UPI-SWIGGY-3382” and “TFR RE INV MAR” into categories. This is language work, and it is the strongest use in your books."}],[{"text":"Reading receipts.","b":true},{"text":" A photo of a drawer of paper turned into rows you check."}],[{"text":"Writing the awkward email.","b":true},{"text":" Chasing an overdue invoice without burning the relationship."}],[{"text":"Translating your accountant.","b":true},{"text":" “What does she mean by accrual basis and does it change what I send her?”"}]]},{"t":"h2","id":"what-it-is-not","runs":[{"text":"What it is not"}]},{"t":"p","runs":[{"text":"Not a calculator.","b":true},{"text":" A language model predicts the next piece of text. Ask it to total a column of 300 numbers and it will give you a number that looks right and often is not, with no signal that anything went wrong. Some tools now write and run real code to compute, which is different and much better, but you still check the top-level total against your bank balance. Sums live in a spreadsheet."}]},{"t":"p","runs":[{"text":"Not your records.","b":true},{"text":" A chat window is not a ledger. It is not backed up, not auditable, not something your accountant or a tax inspector can work from. The books live in accounting software, or a spreadsheet you own. The AI is a helper beside them."}]},{"t":"p","runs":[{"text":"Not a tax adviser for your country.","b":true},{"text":" Rules are jurisdictional, they change every year, and the answer will be fluent regardless. Use it to understand a concept. Never to decide a filing."}]},{"t":"h2","id":"strip-before-you-paste","runs":[{"text":"Strip before you paste"}]},{"t":"p","runs":[{"text":"Most of the sensitive data is not needed for the job."}]},{"t":"p","runs":[{"text":"To categorise a transaction you need the description and the amount. You do not need the customer’s full name, their card number, their account number, their address or their phone. Replace them: "},{"text":"CUST-114","code":true},{"text":", "},{"text":"CUST-115","code":true},{"text":". Keep the mapping in your own file, on your own machine."}]},{"t":"p","runs":[{"text":"Never paste, under any circumstances: full card numbers, CVVs, bank login details, national ID numbers, or anything from a payment processor’s raw export you have not read."}]},{"t":"p","runs":[{"text":"A five-minute find-and-replace before pasting removes almost all of the risk in this lesson."}]},{"t":"h2","id":"prefer-the-tool-that-already-holds-the-data","runs":[{"text":"Prefer the tool that already holds the data"}]},{"t":"p","runs":[{"text":"If your accounting software has AI features (Zoho Books, QuickBooks, Xero, Wave, Tally, whatever you use) check those first. Not because they are cleverer. Because that vendor already holds your data under a contract you already signed. Pasting the same rows into a second company’s system creates a second copy in a second place that can be breached, and a second set of terms you have not read."}]},{"t":"p","runs":[{"text":"Fewer copies, fewer problems."}]},{"t":"h2","id":"chasing-money","runs":[{"text":"Chasing money"}]},{"t":"p","runs":[{"text":"This is where it pays for itself in a single afternoon."}]},{"t":"p","runs":[{"text":"A design studio in Nairobi had KSh 480,000 outstanding across 14 invoices, none of them chased, because writing fourteen individually awkward emails is the sort of job that slides to next week for a month."}]},{"t":"p","runs":[{"text":"Feed it the list: client reference, amount, invoice date, days overdue, and one line of relationship context (“good client, always pays, just slow” versus “third reminder, going quiet”). Ask for fourteen short emails, matched in firmness to the context, each one naming the amount, the invoice number and a specific date."}]},{"t":"p","runs":[{"text":"Twenty minutes, and you read every one before it goes. The tone ladder (polite nudge, firmer, formal notice) is exactly what most owners get wrong under stress, either too apologetic to work or too sharp to survive."}]},{"t":"h2","id":"the-reconciliation-rule","runs":[{"text":"The reconciliation rule"}]},{"t":"p","runs":[{"text":"One habit makes all of this safe. Whatever the AI touched, the bottom line still has to match your bank statement. If it does not match, the AI’s work is wrong until proven otherwise, not the bank."}]},{"t":"p","runs":[{"text":"That check takes two minutes and it is the whole safety net."}]}],"check":{"question":[{"text":"A shop owner pastes 300 bank lines and asks for categories plus a total per category. What should she trust?"}],"options":[{"text":[{"text":"The categories as a first pass to review, with the totals computed in a spreadsheet"}],"correct":true,"tempting":null},{"text":[{"text":"The totals, since arithmetic is the easy part; the categorising is the hard judgment call"}],"correct":false,"tempting":[{"text":"Arithmetic feels like the easy half, but a text model produces a plausible-looking total rather than actually computing one."}]},{"text":[{"text":"Both, because it showed its working line by line and the steps all check out"}],"correct":false,"tempting":[{"text":"Visible working reads like proof, but the steps are generated the same way the answer is, and a wrong step reads as smoothly as a right one."}]},{"text":[{"text":"Neither: this is her financial record and no AI belongs anywhere near it"}],"correct":false,"tempting":[{"text":"The caution is reasonable, but it throws away the one part the model does well: reading messy transaction descriptions."}]}],"order":[1,2,0,3],"why":null},"prev":{"slug":"pricing-conversations","title":"Price: the one thing that never comes out of the model","minutes":9},"next":{"slug":"spreadsheets-without-fear","title":"Ask for the formula, never for the total","minutes":9}}}