Operators, utilities and logistics businesses run on paper that never sees a scanner: creased delivery notes photographed at a gate, handwritten inspection sheets, stamped bills of lading, contractor invoices. DocxIntel reads them on your own infrastructure, including at sites with no connectivity at all.
Accuracy is measured field-by-field against a human-adjudicated ground truth, not character-by-character. The published benchmark mix includes photographed and degraded pages, and your own threshold is agreed in writing on your own field documents before the Proof of Value begins.
A delivery note that has been folded into a pocket, rained on and photographed at dusk is not an edge case in this sector. It is Tuesday.
Purchase orders, goods receipts and contractor invoices are read and compared line by line, with disagreements raised as named exceptions rather than a match percentage.
Delivery notes and proof-of-delivery photographs captured by drivers and gate staff are quality-corrected and read, including signatures, stamps and handwritten quantities.
Bills of lading, manifests, certificates of origin and customs paperwork are extracted with container, seal, weight and consignee details intact.
Existing field forms are read as they are: handwritten sections, ticked checkboxes, annotated sketches, meter values and attached photographs.
Incident reports, permits to work and manual meter readings are captured into structured records so trends and overdue actions become queryable.
Title blocks, revision histories, tag numbers and datasheet parameters are extracted and cross-referenced, so drawings and specifications can be reconciled against each other.
This is the reconciliation that holds up payment runs and creates contractor disputes. Every step runs inside your own environment, including at sites where the connection is unreliable.
At an offshore or remote site the same pipeline runs locally, so a lost satellite link delays synchronisation rather than stopping recognition.
One document flow, one site or terminal, measured on the images your field teams actually capture.
Usually contractor invoice matching, proof-of-delivery capture or inspection report digitisation. We ask specifically for the photographs your current tool rejects, because a sample of clean scans produces an accuracy number that will not survive the first week in the field.
Centrally in your data centre, or locally at terminals and remote sites that cannot depend on a link. Both are supported by the same bundle, and the choice is about connectivity and latency rather than licensing tier.
Containers and open-weight models are deployed on your hardware — air-gapped, private cloud or managed single-tenant. Your team holds root, the keys and the change window, and no site document is visible to us.
Document types, field schemas, tolerance rules for quantity and rate variance and per-contractor form variants are configured directly. Your field forms do not have to be redesigned to be readable.
Your own accounts payable and inspection teams adjudicate field-by-field, and the report shows per-field accuracy, confidence distribution and the exception rate you would staff for. Miss the agreed threshold and you keep the report and owe nothing further.
Adding sites, contractors or a year of historical delivery notes for an audit is a configuration change and some GPU time. The licence does not move because the shipping schedule got busy.
An indicative schema. The final field list is agreed against your own forms, contractor variants and tolerance rules during the Proof of Value.
Format and image-quality coverage in a specific deployment is confirmed in writing during the Proof of Value against your actual field captures, not against a generic list.
Cloud IDP platforms perform well on clean scans. The difficulty in this sector is that clean scans are the minority, and the pages that matter most are the ones captured on a phone at a gate.
Where documents are read
Works at a site with no connectivity
Creased, angled, glare-affected phone photos
Existing field forms usable as they are
Cost of a busy shipping quarter
Illustrative cost at 100,000 pages a month
Re-running a year of delivery notes for a dispute
Engineering drawings and contract rates exposed
Cost figures are our own arithmetic on publicly published list rates as of 2026 — roughly $0.0125 per page for agentic cloud parsing and around $0.30 per page on a usage-based IDP growth plan — not vendor quotations. Verify against current vendor documentation before making a decision.
The objections that come up once someone has actually seen the photographs the field sends in, answered without hedging.
It is the input DocxIntel was built to survive. Deskew, dewarp, perspective correction, glare suppression and contrast normalisation run before recognition, so a creased delivery note photographed on a bonnet is treated as a normal document rather than an exception. This is the failure mode that sends most extraction tools to manual review, and it is where the accuracy difference actually shows up.
Thirty to forty-five days, fixed fee, deployed in your environment on your own field captures. The accuracy threshold and the conversion price are both agreed in writing before we begin.
