AI extraction in OpenText VIM needs a named owner
Pilots fail quietly when nobody owns the feedback loop
Capture products in the OpenText VIM family (whether you are on Information Extraction Services, Core Capture, or another supported path after older BCC/ICC options drop off) can apply machine learning and enrichment to invoice fields. OpenText’s public materials also describe modular extensions and AI scenarios on-stack or on BTP for enrichment and rules. None of that removes the operating need for a human owner.
The operating mistake is treating extraction quality as a project deliverable that ends at go-live. Models and templates drift when vendors change layouts, when a new channel appears, or when AP stops sending corrections back into the learning loop.
What “named owner” means
Write one name (and a backup) for:
Learning data and template governance. Who may promote a change. Who may not.
Exception feedback. How AP corrections become training signal instead of tribal knowledge in a queue.
Quality bar. Field-level expectations AP will accept for touchless candidates—without inventing a vanity “touchless %” for a slide.
Channel scope. Which inbound paths are allowed to use the AI extraction profile, and which stay rules-only until proven.
Incident path. What happens when extraction confidence collapses for a vendor family on a Monday.
Who carries it? A capture owner inside the business or AMS structure, paired with an AP lead who can reject bad “improvements.” The SI can build. The SI should not remain the forever owner by accident.
After a landscape or RISE move
If you just re-proved channels and ArchiveLink, re-prove extraction on the same page. New hosts, new mail relays, new scan vendors, and new Fiori validation workplaces all change what “good extraction” looks like in production. Re-baseline a small vendor set. Do not assume the old learning store survived the move unchanged until someone checks.
Clean core and no hype
Keep AI inside the enrichment and capture layers you already standardized. Prefer published extension patterns over hidden custom code that only one consultant can tune. Do not promise that AI replaces the Chart of Authority or exception ownership. Extraction proposes. Your process still decides.
A useful weekly rhythm: review the top failing fields, the top vendor layouts, and whether corrections were fed back. Thirty minutes. Named owner. Dated notes. That beats a quarterly “AI steering committee” that never touches a DP document.
What to put in the runbook
Capture product name and version/CE as installed.
Owner and backup.
Promotion path for template or model changes.
AP rejection criteria for releasing a vendor to a higher automation path.
Link to the channel list from the RISE re-prove page so extraction and inbound stay honest with each other.
If you cannot name the owner, you do not have an AI extraction capability. You have a hopeful setting.