7 Trip-Expense Tracking Problems Jettova's Receipt Scanner Kills

Trip-expense tracking is one of those tasks where the gap between 'I should do this' and 'I actually do this' is enormous. Everyone agrees it matters; almost nobody finishes a trip with a clean record of what they spent. The reason isn't laziness, it's friction. Each individual receipt-handling step is small. Stacked together over a week, they exceed the threshold where people will bother.

Jettova's receipt scanner is a one-tap workflow inside any saved trip. Open the trip, tap a category, point your phone — three seconds later the row lands in your record. Here are the seven specific failure modes the feature was designed to eliminate.

Problem 1: Typing. The biggest reason expense tracking dies is that typing a number into a form, with the keyboard covering the screen, on a phone, in a foreign restaurant, after wine, is the worst data-entry experience humans have invented. Jettova's fix is removing the keyboard from the path entirely. Claude Haiku reads the printed total off the receipt — no field to tap, no number to confirm, no comma to hand-place.

Problem 2: Categorization that the AI gets wrong. Every other receipt OCR product asks the model to classify the receipt as food / lodging / transport / activities. The miss rate is brutal — a hotel folio with three breakfast charges lands as 'food' a third of the time. Jettova's fix is moving that decision off the model and onto the user: you tap the right category before you scan. One tap, zero confusion. Haiku only reads what's printed; you provide the context.

Problem 3: PDFs. Hotel folios arrive as PDFs. Airline e-tickets arrive as PDFs. Rental car summaries arrive as PDFs. Almost every receipt scanner accepts only camera-roll images, which means you have to screenshot the PDF first, which means you give up. Jettova's fix is accepting application/pdf alongside JPEG, PNG, WebP, and GIF. The vision call rasterizes each page server-side. You drop the original PDF in directly.

Problem 4: Multi-currency. You upload three receipts in three currencies on the same trip. Most apps either pretend conversion at upload time produces a single accurate total (it doesn't — the rate the day you scan is not the rate your card actually moved) or they refuse the receipt entirely. Jettova's fix is honest: each row stays in its native currency, the panel warns you when the trip has multiple currencies, and the per-category rollups don't lie by pretending an exchange rate.

Problem 5: 'About $40 USD.' The classic OCR failure mode is the model getting cute and returning a string like 'around $40' or '<UNKNOWN>' when the total is blurry. Jettova's fix is a strict tool-use schema: total_amount has to be a real number or the call gets rejected. A separate validation layer strips currency symbols and commas before the row touches the database, so a '$1,234.56' parse still works.

Problem 6: Confidence theatre. Some scanners hide their uncertainty by surfacing perfectly-styled rows for everything, including the ones they barely read. Jettova's fix is a three-level confidence rating: high (clearly printed and unambiguous), medium (plausible inference), low (a guess from limited info). The low-confidence rows are still saved, so the rollup is complete, but the rating tells you which ones to glance at before you trust them.

Problem 7: Privacy doubt. Every cloud-OCR product stores the image — sometimes indefinitely. People notice, and rightfully stop uploading sensitive receipts. Jettova's fix is that we don't keep the image. The bytes go to Anthropic for the vision call, get extracted into structured fields, and the original is dropped. Nothing to leak, nothing to manage, no orphan-storage policy to misconfigure.

The throughline across all seven is the same: every step that requires the user to do something tedious is the step that breaks tracking. Removing the tedious step (typing, mis-categorization, format gating, currency math, OCR ambiguity, hidden confidence, privacy worry) gets the feature past the threshold where people will actually use it for a whole trip.

How fast is fast enough? In our internal testing, twelve receipts went from camera roll to typed rows in under a minute on a normal phone. Same week, same trip, same restaurants — but the difference between 'I'll do it later' and 'done' is what actually decides whether the budget gets honest feedback. The bet is that two seconds per receipt is below that threshold and four seconds is above it.

If you've abandoned expense tracking on past trips, the easiest test is one trip. Open any saved trip in Jettova, find the 'Track your actual spend' panel below the budget breakdown, and tap a category button on your next meal. If it doesn't fit into your workflow, we'd rather you say so than the alternative — but most people who try it on one trip end up using it on the rest.