What is IDP in insurance?

Intelligent document processing (IDP) classifies files and pulls fields from known templates. It is useful for high-volume, stable layouts: invoices, ACORD-like forms, a coverholder bordereau that truly never changes columns. It is not a synonym for insurance AI, and it is not a synonym for reinsurance operations software.

Classic IDP scores well when column A stays column A. Reinsurance packs are not stable. Slips vary by broker and year. SOVs are workbooks with extra rows and merged headers. Treaty wordings are manuscript plus endorsements. Bordereaux insert a column without warning. Template IDP then does one of two things: it fails the document, or it silently maps the wrong column because "Pol Ref" moved. Empty is better than wrong. Silent wrong is how you book a total you cannot unwind.

What template extraction is for

Keep IDP where layouts are still. Accounts payable is not this product. Do not rip out a working template extractor to make a category slide simpler. A missing field in classic IDP is often an empty cell, which is closer to a gap than a chat invention. The weakness is the map. When the template does not fit, you do not get a chase list that names hours clause. You get a low extract rate and a queue of unclassified PDFs.

Conflicts between two documents in one zip are out of scope unless you built a second product on top. IDP licences for stable forms do not need to be replaced because facultative zips exist. They also should not be forced to ingest those zips.

Why reinsurance packs break the template

A facultative pack is a zip: slip, SOV, loss run, covering email, maybe a scan of an endorsement. The TIV on the slip and the TIV on the schedule often disagree. The hours clause is often absent. The named insured on the MRC is a legal name; the schedule uses a trading style. None of that is a template problem. It is a pack problem.

Source-grounded extraction is the contract for that mess: field, document, page or cell, span. Gaps instead of guesses. Conflicts left as two numbers. The comparison of generic models, template IDP, and ops software is AI versus IDP versus reinsurance ops. Insurance AI is the wider parent category. IDP sits inside that category as the template job, not as the messy-pack job.

Worked example: ACME Construction Ltd

Use the same fictional file this site uses everywhere. ACME Construction Ltd, acme.example. Slip TIV USD 42,000,000 on Slip.pdf. SOV.xlsx totals USD 47,100,000 because a warehouse was added after the slip was typed. Hours clause absent. Limit USD 10,000,000 on slip page 2.

Template IDP will often classify one PDF, miss the workbook, or map a "limit" field from a header that is not the occurrence limit. If the slip template was trained on a different broker layout, the extract is empty or shifted. You do not get a chase item that says hours clause. You get an unclassified file or a thin record. You do not get two TIV spans. You may get one TIV from whichever document matched the template.

Operations software should emit a traced limit, a TIV conflict with both sources, and a hours-clause gap. That is the behaviour to buy for placing and treaty admin. It is not IDP with a nicer theme. Mystery-shop with this pack, not with a stack of invoices. The invoice is IDP's home pitch. The zip is the insurance operations pitch. Humans still map exceptions. Software should not silently follow the wrong column and call it complete.