Ask your receipts: natural language vs the spreadsheet
A spreadsheet is a grid you fill. Asking your receipts is a question you already wanted to type: “How much did I spend at Checkers this month?” The difference is not a prettier chart. It is whether the total still opens the original till slip. That is the product behind AI Insights.
What spreadsheets are good at
Excel or Google Sheets are honest tools. You can sum a column, pivot by month, and share a file with an accountant. If you enjoy typing merchant names, they will never go out of fashion. The failure mode is the same every year: the grid is only as good as the last evening you had the energy to copy totals from fading paper. Miss a week of Checkers trips and the “budget” is fiction.
Bank CSV imports look like automation. They are not receipt archives. A single SuperSpar or “POS PURCHASE” line does not tell you VAT, line items, or which bag of dog food blew the grocery number. You cannot tap the bank row and see the slip. For South African till-slip VAT fields, see what to capture.
What “ask your receipts” means
Natural-language insights start after capture. You scan with the camera, pick a gallery photo, or upload a PDF. Merchant, date, VAT, line items, and total are stored with the image. Then you ask a question in ordinary language. The system has to pick a date range, a merchant or category, and the receipts that belong — then show the work.
Examples that match how people actually talk:
- How much did I spend at Checkers this month?
- Show grocery spending by month.
- What did dining cost this month versus last month?
- Which categories took the most money this year?
If the answer cannot open the source slips, it is a chatbot paraphrasing a spreadsheet. AI Insights is built to keep that drill-down. We do not claim the model never mis-hears a category name; you can see which receipts were used and correct the underlying extract.
Why the original slip is the point
A number without proof is an argument. A number with the Food Lover’s Market image attached is a record. Returns, warranties, and (if it ever matters) SARS supporting documents all need the picture, not a cell in row 482. Spreadsheets throw the picture away at the first copy-paste. Scanning and then asking keeps both layers: the query for speed, the image for evidence. That is also why we talk about closed testing honestly — testers should try the question on their own slips, not a marketing table.
Compared with “expense report” software
Expensify-class tools exist to get a trip approved. Genius Scan exists to make a PDF. Smart Receipts exists to file and export. Those are real jobs; we named them on the 2026 comparison. Asking questions about your own receipts is a different job. You are not submitting to a finance team. You are trying to remember whether the last month of Pick n Pay was an outlier.
A spreadsheet can answer that if you typed everything. Natural language is for the people who will not type everything, but will photograph a slip if the payoff is a later question. The app still lets you edit a wrong total. Automation that you cannot correct is worse than a slow grid.
A worked example: Checkers this month
Suppose you shop at Checkers twice a week. A spreadsheet version of the month is fourteen rows you might skip. A bank version is fourteen “CHECKERS” authorisations with no VAT split and no bananas-versus-batteries. The ask-your-receipts version is: scan each slip (or the emailed PDF), then type “How much did I spend at Checkers this month?” The report should list those slips. If one total looks high, open the image. If a slip was never scanned, the answer is incomplete — which is the truth, not a bug dressed as intelligence.
That is also why we lead with AI Insights on the homepage rather than OCR theatre. Plenty of apps photograph paper. The useful product is retrieval plus proof. Read the walkthrough on AI Insights, including the Checkers example prompt. If you only remember one habit: scan the slip while it is still black, then ask later.
Android, web, and what closed testing is for
Capture is Android: camera, gallery, PDF. Review can be the same phone or the web portal with Google Sign-In. Encrypted cloud storage is how the image follows you. Closed testing means you join a Google Group and a Play opt-in with the same account — not that the public listing is live. We will not put a Play badge on this site until that listing returns 200. Testers should try real questions on real slips and tell us when a total cannot open its source image. That bug would miss the whole point of this article.
Limitations, said plainly
Natural language will mis-parse a nickname you never used as a merchant. A category suggestion can be wrong. A crumpled slip can extract a bad total. None of that is a reason to go back to an empty grid; it is a reason to keep edit tools and the original photo. We also will not run a “vs Expensify” scorecard. Different jobs. If you need company approvals, use the company tool. If you need to ask your own till slips a question, that is the experiment running in closed testing.
A realistic workflow
- At the till or the same evening: capture the full slip (Android camera, gallery, or PDF).
- Glance at merchant, date, VAT, total. Fix obvious misses.
- Let categories land; change the ones that are wrong.
- When you are curious — Sunday night or month-end — ask the question on phone or the web portal (Google Sign-In).
- Open any total that looks wrong and read the image. That is the loop.
Pinning a useful insight on the dashboard is for questions you will ask again. It is not a substitute for scanning. Empty archive, empty answers.
What we will not say
We will not say this beats Expensify for companies, or Genius Scan for contracts, or that OCR is magically accurate on every crumpled thermal. We will not put a Play purchase badge on the site while we are in closed testing. Play prices for Basic are on the pricing page. The honest line is: if the job is asking your own receipts a question, with the slip as proof, that is what we are testing on Android.
South Africa-specific questions
Local merchants (Checkers, PnP, Woolworths, garages) are how people actually query spend. VAT on the slip is a field, not a vibe — capture it. Digital copies and SARS retention are a separate article: can you throw the paper away. Insights do not change those rules; they only help you find the file.
Spreadsheet or questions?
Keep a spreadsheet if you like typing and you already never miss a week. Use ask-your-receipts if the archive is photos-plus-fields and you want to query it in the same words you would text yourself. Either way, keep the original image. Go to AI Insights for the product walkthrough.