1. Measure the whole manual path
Invoice data entry often looks like one task, but it is a chain: opening an email, downloading the attachment, checking whether all pages arrived, finding the supplier, typing header fields, entering lines, choosing item or GL codes, checking totals and finally posting. If you measure only keyboard time, you will underestimate the workload.
Sample at least 30 documents across repeat suppliers, unusual layouts, phone photos and multi-page PDFs. Record collection, key-in, checking and correction time separately. The invoice data-entry calculator can turn that baseline into a monthly estimate.
2. Standardise how invoices arrive
Automation is harder when documents live across personal inboxes, chat groups and desktop folders. Give suppliers or staff one controlled route. With beres, that can be a private document email address or a linked Telegram chat. The purpose is not to force suppliers into a new portal; it is to make forwarding the document enough.
A single email can carry several supported attachments. Each attachment becomes its own document. When separate photos appear to be pages of one multi-page document, the ingestion worker groups them automatically; merge and unmerge controls exist for the exceptions.
3. Separate transcription from judgement
Supplier name, invoice number, dates, quantities, descriptions and totals are transcription. Deciding which supplier code, stock item or GL account is correct is accounting context. Good automation handles the first and proposes the second, but it should not invent missing codes.
Import or sync master data before the pilot. A reviewer can then confirm uncertain matches. Those corrections are reused for later documents inside the same organisation.
4. Keep review focused on exceptions
Review is not a failure of automation. It is the control that prevents a plausible-looking misread from silently reaching the ledger. The useful question is whether reviewers can spend time on uncertain fields instead of retyping everything.
5. Deliver in the format the accounting system expects
For SQL Account, beres supports itemised eStream and summarised ARAP flows through an import file or API push. For AutoCount, direct delivery currently supports AutoCount Cloud only. A workflow is not complete merely because OCR produced text; it is complete when the approved entry arrives successfully and can be retried without duplication.
What to automate first
- Centralise document forwarding.
- Extract predictable invoice fields and line items.
- Match against real master data.
- Route uncertainty to a reviewer.
- Deliver through the accounting system’s supported method.
Start with representative invoices rather than the cleanest examples. A guided session can help map the process without accepting financial documents through the enquiry form.