Receipt Apps and What They Do With Your Data, Ranked

When you upload a receipt to a "receipt app," what actually happens to it?

Most people don't ask. The receipt goes in, a row in a dashboard comes out, and we trust the rest.

But the receipt is not nothing. It has a vendor name, an amount, a date. Often a customer email. Sometimes a home address. A line-item breakdown of what you bought. For B2C businesses, a buyer's identity. For B2B, a corporate spend pattern that's worth real money to anyone who can aggregate enough of them.

So: what do these apps do with that data?

I read the privacy policies. Here's the ranking, scored on four axes that actually matter.

Table of Contents

• The four axes
• Hubdoc
• Dext (formerly Receipt Bank)
• Expensify
• Wave
• FreshBooks
• Bench
• QuickBooks Receipt Capture (Intuit)
• Shoeboxed
• AutoFileEmail (us)
• What this means in practice
• How to actually pick
• The bottom line
• Sources

The four axes

The grading system, applied evenly to every app in this list:

1. Does it OCR or extract content? If yes, the app has a database of line items, vendor names, and amounts that's separate from the original PDF.
2. Does it use your data to train models? Buried in some privacy policies, called out in others. Material whether you agreed or not.
3. Where is it hosted? US, EU, or other. Affects who can subpoena it.
4. Can you fully delete? Some apps "soft-delete." Some retain copies for "fraud prevention" or "legal" purposes for years after you cancel.

I'm working from the apps' own published privacy policies and DPAs as of May 2026. Links at the bottom. If they change theirs, this article will go stale - read the source if accuracy matters to you for compliance reasons.

Hubdoc

OCR? Yes, by design. The product is OCR-into-your-accounting-software.

Training? Hubdoc's privacy policy doesn't explicitly grant training rights, but it allows "use to improve the service." Vague. Owned by Xero (acquired 2018), so any practice flows into Xero's data handling.

Hosted? AWS US-East and EU regions, depending on your account.

Fully delete? Yes, on cancellation, with the standard "we retain backups for X days" caveat. Reasonable.

Score: 4/10. The OCR-by-default architecture means even if you delete, copies of your receipt content lived in their indexing pipeline. Not a privacy disaster, but not the lightest footprint either. (Longer breakdown of the OCR-vs-filing trade in our vs Hubdoc page.)

Dext (formerly Receipt Bank)

OCR? Yes. Their pitch is line-item extraction.

Training? Their privacy policy mentions using "anonymized" data to improve their extraction models. That's training, just phrased politely.

Hosted? UK + EU regions primarily, US for North American customers. ISO 27001 certified.

Fully delete? Yes, but they retain "summary metadata" for 7 years for tax-record integrity. Reasonable, but worth knowing.

Score: 5/10. The model-training disclosure is more honest than most. The trade is that your receipt content has, in some abstracted form, helped train their OCR. If your buyer-side details show up in those receipts, that detail traveled.

Expensify

OCR? Yes, their "SmartScan" feature.

Training? Their policy permits use of customer data to improve services, including ML. They've been public about training their OCR on customer receipts.

Hosted? AWS US-East primarily.

Fully delete? Account deletion exists. Receipt data retention policies are less clear-cut. Their public policies have been revised multiple times in the last 24 months.

Score: 4/10. SmartScan is genuinely good at what it does. The trade is the same: your receipt content fed the model.

Wave

OCR? Yes, for receipt scanning. They added it after H&R Block acquired Wave in 2019.

Training? Privacy policy permits service improvement use. Not explicit about ML training.

Hosted? US (AWS).

Fully delete? Yes, but parent-company data sharing with H&R Block is permitted in their policy. So if you use Wave and an H&R Block service, your data may flow between them.

Score: 5/10. The H&R Block parent-company link is the fingerprint here. If you're indifferent to that flow, it's fine. If you're a consumer who already files with someone else, the parent-company sharing is non-zero.

FreshBooks

OCR? Optional via add-ons. Their core product is invoicing, not receipt OCR.

Training? Policy is conservative. No explicit ML training language.

Hosted? US and Canada.

Fully delete? Yes. Reasonable retention for tax-record purposes.

Score: 7/10. Lighter footprint because OCR isn't the core product. If you use FreshBooks for invoicing and just upload receipts as attachments, the footprint stays small.

Bench

OCR? Yes, but the bigger story is Bench is a bookkeeping service - actual humans plus automation. So your receipts are read by both software and people.

Training? Their policy permits aggregated and anonymized data use.

Hosted? US.

Fully delete? On cancellation, yes. Retention follows the bookkeeping-records standard (7 years for IRS purposes).

Score: 4/10. Not because Bench is sketchy - they're a legit firm - but because the data lives in more places (their software, their bookkeepers' workflows, their accounting backend). More copies, more surface area.

QuickBooks Receipt Capture (Intuit)

OCR? Yes.

Training? Intuit's privacy policy is unusually frank: they use customer data, including financial data, across their product family (QuickBooks, TurboTax, Mint legacy, Credit Karma). In 2024 they were called out by the FTC over disclosure practices in marketing.

Hosted? US (their own data centers and AWS).

Fully delete? Account deletion exists, but the data-sharing across the Intuit family is the part that catches people off-guard.

Score: 3/10. Powerful product, big company, broad data-sharing across an ecosystem most people don't realize they joined. Not the worst, but the largest single-vendor data footprint on this list.

Shoeboxed

OCR? Yes. Plus a physical mail-in option (you mail them shoeboxes, they scan).

Training? Permitted for service improvement. Not explicit about ML.

Hosted? US.

Fully delete? Yes. Physical receipts are shredded on a schedule.

Score: 5/10. The physical mail option adds a chain-of-custody question. The digital service is comparable to the others in this list.

AutoFileEmail (us)

I'll be specific about what we do, because we want this article to age honestly.

OCR? No. We never read the contents of your attachments. The PDF that lands in your Drive folder is bit-for-bit the PDF Stripe (or whoever) sent you. There is no extraction pipeline. You can verify this: open any filed PDF, compare its hash to the hash of the original email attachment. They match.

Training? No. We can't train on something we can't read. We don't run inference on your file contents. The only metadata we look at is the email headers we need to file: sender, date, message ID, subject (for the filename). Those live in our database for one reason - dedup, so we don't double-file the same email if it's seen twice.

Hosted? Cloudflare Workers (compute) plus Supabase (Postgres on AWS US-East). Your filed receipts live in your Google Drive, not on our servers.

Fully delete? Yes, and the deletion is small because there isn't much to delete. Email metadata, OAuth tokens (encrypted), and your billing record. The receipts themselves stay in your Drive, where they always were. Cancel us tomorrow and your folder structure is intact - we just stop adding to it.

Score: 9/10. Not 10, because we're a startup and you should never give anyone full marks. We could screw up tomorrow. The architecture is the cleanest on this list because we deliberately built it to have nothing to leak. More on the why we don't OCR reasoning if you want the long version.

What this means in practice

If you're a B2C operator and your customer emails appear on receipts, the OCR-heavy apps in this list have, at minimum, indexed your customers' personal details. That may or may not bother you, but you should know.

If you're a solo founder filing your own receipts and they're mostly SaaS subscriptions (no PII beyond your own), the privacy gap between the apps is smaller. It still exists, but the stakes are lower.

If you're in the EU or you have EU customers, GDPR makes some of this an actual compliance question rather than a vibes question. We have a GDPR-for-one-person-businesses primer if that's where you're heading.

How to actually pick

Three questions, in order:

1. Do I need OCR? If yes, accept the privacy footprint and pick the OCR app whose policy you can live with. Dext and FreshBooks are the more conservative options.
2. Does my CPA need a folder, or a dashboard? If a folder, you don't need OCR at all. A filer is enough.
3. Am I prepared to delete and verify deletion? If yes, the OCR apps are fine. If you want "no copies of receipt content anywhere but in my own Drive," you need a non-OCR option.

For most solo founders, the answer to #2 is "a folder," and the rest of the questions become moot.

The bottom line

OCR is a feature most receipt apps were forced to add because their bookkeeper-firm customers demanded it. It is not a feature most solo founders need.

The privacy cost of OCR is real but manageable. The dollar cost (covered in the Hubdoc/Dext piece) is the bigger annoyance for most one-person businesses.

If you want a filer that does the boring job and nothing else, AutoFileEmail is free for one inbox. Files land in your Drive. We never read them. Cancel and there's nothing to delete because we never had your receipts in the first place.

Or pick a different one. But pick deliberately. The privacy policy for the receipt app you use is worth the 10 minutes it takes to read.

Sources

• Hubdoc privacy policy: hubdoc.com/privacy
• Dext data processing addendum: dext.com/legal
• Expensify privacy policy: use.expensify.com/privacy
• Wave terms and privacy: waveapps.com/legal
• FreshBooks privacy: freshbooks.com/policies
• Bench privacy: bench.co/privacy
• Intuit privacy: intuit.com/privacy
• Shoeboxed privacy: shoeboxed.com/privacy

(All accessed May 2026. Privacy policies change. Verify before relying on this article for compliance work.)