OpenText added a quiet feature to Core Capture for SAP Solutions in CE 25.2. You can start with a higher recognition rate for header fields on invoices or sales orders by pre-learning from documents you posted in the past. A Fiori app lets a user pick the training documents by archive document type, posting date range, company code and financial document type.
It's a sensible idea. Your SAP system already holds years of invoices where someone checked the vendor, the invoice number, the date and the amount. Why make the extraction engine start from nothing?
Because history isn't neutral. Those posted invoices carry every workaround your AP team ever used. The selection screen is where you decide which habits the engine inherits.
Four filters, four governance questions.
- Archive document type. Which images count as an invoice for training. If credit memos, freight bills and utility invoices share one document type in your ArchiveLink setup, they train together. Settle document types before you train on them.
- Posting date range. The period that reflects how you work now. If a vendor master cleanup, a new company code or a big supplier's layout change happened last year, a range that starts earlier teaches the old pattern.
- Company code. Whether one entity's conventions should shape another's. A shared service center may want one set. Entities with different languages or tax formats may not.
- Financial document type. Which postings are clean examples. Reversals, corrections and manual postings made to free a stuck invoice are poor teachers. Decide whether they belong.
Posted isn't always printed. The training draws on documents you posted, so its reference is what AP posted. Usually that's what you want. Sometimes AP posted something other than what the supplier printed, by policy or by habit: a normalized invoice number, a corrected date, a reference moved to another field. Check in the release notes for your tenant how the training pairs the image with posted data. Then decide whether those conventions are ones you want repeated on every new invoice.
Where it sits next to 26.2. In 26.2, OpenText describes an intelligent voter that combines knowledgebase, self-adaptive learning and LLM-based extraction. Historic training feeds the learning side before the first new invoice arrives. We've written about who sets the voter policy and turning on GenAI extraction without a VIM upgrade. If you plan both, keep them apart. Train, measure against a fixed supplier set, then switch on GenAI. Do both in the same week and nobody can say which one moved the correction rate.
A training brief that fits on one page:
- Owner: the capture owner, with the AP lead signing off the selection.
- Scope: document types, company codes and date range, with one line on why each.
- Exclusions: the financial document types left out, and the reason.
- Baseline: field-level correction counts on a fixed supplier set before training.
- Review: the same counts a few weeks after, reported field by field.
Version note. Core Capture for SAP Solutions is a SaaS add-on, and OpenText has shipped 26.2 and 26.3 since this feature first appeared in CE 25.2. On OpenText's lifecycle page, read on 14 November 2026, current maintenance for the Core Capture for SAP Solutions 24.2 row ends 30 April 2029. Confirm the historic training app is available in your tenant and how it behaves in your release before you plan around it.
Training on your own history is a good shortcut. Choose the history on purpose.
Let's work together.
Where does your landscape stand?
Take the free Assessment. It's about 10 questions, and you get a one-pager with your estimated level of effort and suggested next steps.