Skavio Flow: reviewing handwriting and low-quality scans before export
Skavio Flow includes source evidence and review warnings alongside structured document data. For teams handling handwritten notes or poor-quality scans, those outputs support an essential distinction: an extracted value is a candidate for review, not proof that the original was readable. The goal is not to fill every spreadsheet cell—it is to identify what the source supports before the data moves downstream.
Inspect the source before trusting the table
Flow turns invoices, orders, scans and text into structured outputs using custom fields and saved company workflows. Embedded PDF text is read locally; scans pass through OCR before structured extraction. That processing path does not, by itself, establish reliable recognition of handwriting. Start by examining the original document. Is the full page present? Are amounts, decimal marks and annotations legible? Does a handwritten addition change a printed quantity or total? Microsoft’s Document Intelligence transparency guidance identifies resolution, contrast, lighting, rotation and text characteristics as factors affecting OCR accuracy. AWS likewise recommends preserving suitable original documents rather than downsampling them unnecessarily, and keeping table text upright. These are general preparation principles—not claims that Flow automatically repairs poor inputs.
Use warnings to direct attention—not to certify everything else
Flow includes source evidence and review warnings. Use them to guide comparison with the original, but do not treat them as an exhaustive error detector. A result without warnings still needs review of consequential fields. Look for omissions as well as incorrect values. Microsoft notes that handwritten or difficult-to-read content can be absent from OCR output. A tidy table can therefore conceal a missing annotation or line, not just a misread character.
- Dates and references: compare the extracted invoice date, due date and document number with the source. Do not resolve an ambiguous digit by choosing the most plausible value.
- Amounts: check totals, tax figures, currency labels and decimal separators. One incorrect character can make an otherwise convincing amount wrong.
- Line items: compare quantities, units, descriptions and row alignment, including handwritten changes and any source rows missing from the result.
Resolve ambiguity with evidence, not arithmetic alone
Consider an illustrative purchase order where a handwritten quantity could be either 1 or 7. A plausible line total does not establish which digit was written. Nor does a recurring supplier format prove that this document contains the same value as last month’s. Correct an extracted field only when the source supports the correction. If the mark remains unreadable, request a clearer scan, the original digital document or clarification from the document issuer. Record the unresolved issue in your team’s review notes and withhold acceptance of the affected data. These are recommended review practices, not descriptions of automatic Flow controls. Do not assume Flow detects handwriting, flags every uncertain field, exposes numerical confidence scores or blocks exports automatically.
Export reviewed data—and keep unresolved issues visible
Flow provides XLSX document and line-item worksheets, CSV and JSON. Choose the format for the receiving workflow after checking dates, totals and line items against the source. Before handing over a file, distinguish confirmed values from unresolved questions. If your process permits an incomplete handoff, accompany it with a separate, explicit exception note. Do not assume every export format automatically preserves warnings or review context. Keep the relevant source available to the next reviewer. Reusing custom fields and saved workflows supports consistent extraction requirements across documents. Acceptance still depends on the evidence in each new document. A structured export is not, on its own, an accounting approval.