Receipt capture is the part of bookkeeping that looks solved and is not. The extraction is solved: both Dext and AutoEntry read a photographed or emailed receipt accurately enough that the remaining errors are caught by matching. The unsolved part is upstream, in getting the receipt to exist in a place the tool can see. We found that out by auditing our own supplier emails, and the result is at the end.
Dext and AutoEntry, side by side
| What | Dext | AutoEntry |
|---|---|---|
| Capture by email-in | Yes | Yes |
| Capture by mobile app | Yes | Yes |
| Fetching from supplier portals | Yes: Invoice Fetch connects to supplier accounts and rechecks them about every seven days | Not listed |
| Line-item extraction | Yes, as an add-on priced per document or per month | Yes, at two credits per document instead of one |
| Bank statement extraction | Yes, add-on | Yes, three credits per page |
| Approval workflows | Multi-level approvals | Multi-level approvals |
| Publishes to | Xero, QuickBooks, Sage and a long list of others | Sage, Xero, QuickBooks |
| Pricing model (practice) | Per client per month on an annual term, with client minimums per tier; tier prices are quote-based on the site | Credits per month, from a 50-credit tier to a 2,500-credit tier, 90-day rollover |
Sources: Dext practice pricing, Dext Fetch, AutoEntry pricing, checked October 2026. Prices change; check before you decide.
Practically: if your clients buy from suppliers with portals (telecoms, utilities, software, Amazon), Dext's fetch feature is the differentiator, because it goes and gets the invoice the client never forwarded. If your clients' receipts are mostly photographed on a phone or forwarded from email, the two are close and the pricing model decides it: per-client suits a practice with many low-volume clients, credits suit a practice with a few high-volume ones.
Where both stop
- They see only what reaches them. A receipt sent to the client's old email address, a "your bill is ready" notification that needs a login, a supplier that simply does not send one: none of these become documents.
- Matching is approximate. Both suggest matches; a person confirms. The suggestion is usually right, and "usually" is the problem at volume, because the wrong ones are indistinguishable from the right ones until someone looks.
- Chasing is generic. A reminder that a document is missing is not the same as a reminder that says which supplier, which month, which amount, and stops the moment it arrives.
- Nothing knows the client's answers. "That one's the van insurance" lives in an email thread. The tools do not read it.
What our own mailbox audit found
We measured it for our own business before building anything. Of twenty regular suppliers, one attached a PDF receipt to its email. Two sent a notification that an invoice was ready, with a portal login required. Eleven sent nothing at all, because their billing contact was an identity nobody read any more. The remaining handful were somewhere in between.
The lesson was not about software. The highest-leverage fix was configuration: repoint every supplier's billing email at one address a system reads, and most of the "needs a portal login" and "sent nothing" buckets become the "found it in the inbox" bucket. Only then is a capture tool working with a full deck.
When a practice needs its own AI inbox on top
An AI inbox is the layer we built after that audit. It reads the mailbox, decides what each message is (a receipt, a notification, a statement, nothing), matches receipts to bank lines on exact amount and date window, flags the ones that need a portal login with the URL, posts the confirmed ones to the capture tool or the ledger, and chases the specific missing ones with a reminder that stops when the document lands. It also reads the accountant's weekly query sheet and does the same for every row that says "Invoice?".
It does not replace Dext or AutoEntry; it feeds them, and it handles the part before capture that they cannot. A practice needs it when the team's hours are going on finding and chasing receipts rather than on reading them, which, after the audit above, we would guess is most practices.
How that fits into the wider bookkeeping workflow is in the practice guide to automating bookkeeping with AI; the accountancy practices page has pricing for a build like it.
Founder, Anco AI. Builds the systems this site talks about, and runs his own business on them.
