Document intake and exception routing
Extraction accuracy is only one part of document automation. The operating question is what happens when a document and a record disagree.
Reference build 02
Keelix’s own build — not a client result · in progressWhat this exists to fix.
Paperwork arrives continuously and is checked by somebody who knows what it should say. When volume rises, checking is what gets skipped, and the errors surface later as a payment dispute or a failed audit.
Extraction alone does not solve this. A system that reads a document perfectly and files it without reconciling it against the purchase order has automated the filing and left the actual work untouched.
How it is put together.
Plain enough for an owner to follow, specific enough for whoever maintains it afterwards.
-
Classification
Incoming documents are sorted by type before anything is extracted, because an invoice and a certificate of insurance need different schemas, different checks, and different owners.
-
Extraction
Per-type schemas with required and optional fields. Each field carries its source location so a reviewer can jump to where the value came from.
-
Reconciliation
Extracted values are compared against the record the business already holds — purchase order, contract terms, expected expiry date. This comparison is deterministic code and is where the actual value of the build sits.
-
Routing
Agreement files automatically. Disagreement routes to a named owner with both values shown and the difference stated plainly. Unclassifiable documents route to a person rather than to a default type.
-
Expiry tracking
Documents with a validity window are tracked, and the responsible person is notified ahead of expiry rather than after a certificate has lapsed.
-
Fallback
Every document remains retrievable in its original form, filed by arrival date, whether or not any automated step succeeded.
How it will be evaluated.
The procedure is published before a run exists, so the evaluation cannot be rewritten around a preferred result.
-
Freeze a versioned document set for each configured type, together with the expected business record, source locations, ownership rules, and sample-selection note.
-
Run classification and extraction against the per-type schemas. Preserve the proposed values and source locations so each field can be checked against the original document.
-
Run the deterministic reconciliation rules against the expected purchase order, contract term, or expiry date. Record agreement, disagreement, and missing-reference states separately.
-
Exercise unknown document types, mismatched values, expiry alerts, and dependency failure. Confirm that each exception reaches its named owner and that every original remains retrievable.
-
Compare handling time and cost with the manual baseline using the same document set. Retain the fixtures, rule version, run record, calculation method, and exception log before publishing a figure.
What it has to do to count.
Agreed before construction, published whether or not they are met.
- Classification errors route to a person, never to a default document type.
- Every automated filing decision records the comparison that justified it.
- A disagreement always reaches a named owner, never a shared inbox.
- Expiry notifications fire ahead of the date, with the lead time configured per document type.
- Reconciliation logic is covered by tests using real document variants.
- Cost and handling time per document are recorded against the manual baseline.
What has been run and retained.
No completed evaluation run has been published.
Where it stops, refuses, or hands back.
A build with no stated limits has not been examined closely enough.
- Reconciliation is only as good as the record it compares against. Where the business does not hold a reliable expected value, this build cannot manufacture one.
- Document types outside the configured schemas are routed to a person rather than handled generically.
- No figures are published above because this has not yet run against a real document population.
Operational guidance
This pattern, fitted to your operation.
An Audit works out whether a build like this is the right first one for your business, and what it would take.