Harold extracts structured data from every invoice automatically. Train it once on each supplier, and it handles the extraction, normalisation, and export forever — no manual data entry, no supplier behaviour change, no enterprise price tag.
Invoice processing automation is the use of software to extract, validate, and route invoice data without human data entry. Instead of someone reading a PDF and typing the supplier name, invoice number, date, and line items into a spreadsheet or accounting system, software does it automatically.
The challenge is that invoices are unstructured. Unlike a database record with defined fields, an invoice is a PDF designed for a human to read — and every supplier designs theirs differently. Effective invoice processing automation has to handle that variation reliably.
Harold solves this through a training model. Rather than applying a generic layout to every invoice, you teach Harold exactly how each supplier formats their invoices. Once trained, Harold recognises that supplier's layout and extracts the right data every time — even when the layout changes slightly between invoices.
The real problems
The average accounts payable team spends 8–14 hours per week manually keying invoice data into spreadsheets or accounting systems. A single transposition error can cause a supplier payment to fail, a VAT return to be wrong, or an audit to flag a discrepancy. The cost is not just time — it is trust.
Supplier A puts the PO number in the header. Supplier B puts it in the footer as 'Your Ref'. Supplier C doesn't include one at all. Generic OCR tools treat all of these the same way and get all of them wrong. You end up correcting more than you extract.
Solutions like Tungsten (Kofax), Yooz, and Basware are genuinely excellent — and genuinely expensive. Licences, implementation, consultancy, and training routinely land at £500–£2,000+ per month before you have processed a single invoice. For SMBs processing 100–2,000 documents a month, the economics simply do not work.
Most automation tools require your suppliers to email a new address, use a supplier portal, or change their invoicing format. None of those things happen reliably. The tool gets bypassed and someone ends up printing and rekeying anyway.
How Harold works
Upload 3–10 sample invoices from a supplier. Harold's DocuTrain wizard detects every field automatically — invoice number, date, supplier name, line items, totals, VAT, payment terms. You confirm or correct the mappings. That is the entire training process. It takes under 10 minutes per supplier template.
Once trained, Harold processes every future invoice from that supplier without any human input. Upload via the app, send to your Harold inbox by email, or pipe documents in automatically. Harold detects the supplier, selects the right template, and extracts every field in seconds — including line-level data like product codes, quantities, unit prices, and VAT amounts.
Extracted data is rarely ready to drop straight into an ERP. Supplier names need mapping to internal IDs. Dates need reformatting. VAT codes need translating. Harold's Rules Engine lets you define these transformations once. DAVIDBRENT LTD becomes DB-001. 30/01/2026 becomes 2026-01-30. The rules run silently on every extraction.
Reviewed and approved invoices export as CSV or Excel. With Zapier connected, Harold pushes clean structured JSON directly to your accounting system, ERP, or data warehouse — Xero, QuickBooks, Sage, FreeAgent, Oracle, SAP Business One, Airtable, or anything else in Zapier's 5,000+ app ecosystem. No custom integration required.
Harold Automations gives every document type its own email inbox. Suppliers CC your Harold address — nothing changes for them. Harold receives the document, detects the type, matches the DocuTrain template, applies your rules, and processes it. Your team sees clean structured data, ready to review and export.
Most AP automation is built for enterprise budgets. Harold is built for real businesses.
Under 10 minutes for most suppliers. You upload 3–10 sample invoices, Harold detects the fields automatically, and you confirm or adjust the mappings. The more sample invoices you provide, the more robust the template becomes — but most templates work well from just 3–5 examples.
Harold's confidence scoring will flag documents that don't match the expected template well. You review, make any corrections, and Harold updates the template automatically from your corrections. Over time the template improves with each correction.
Yes. All Harold plans include line item extraction. Supplier name, invoice number, date, payment terms, subtotal, VAT, and total are extracted by default. With DocuTrain training you can add line-level fields — product code, description, quantity, unit price, VAT code, and any custom fields your supplier uses.
Yes, via Zapier. Once Harold processes and approves a document, it can send clean structured JSON to a Zapier webhook. From there you can route it to Xero, QuickBooks, Sage, FreeAgent, or any of Zapier's 5,000+ connected apps — without writing a single line of code.
Harold accepts PDF (the most common invoice format), image files (JPG, PNG, TIFF), and email attachments forwarded to your Harold inbox. Scanned PDFs and image-based PDFs are also supported.
No. Harold is month-to-month. Cancel at any time from your billing settings. There is no minimum contract, no setup fee, and no cancellation penalty.
Sign up free. Train Harold on your first supplier in under 10 minutes. No contract, no card required to start.
From £29/month after trial. Cancel anytime.