Intelligent POD Workflow — Logistics Operator
Client
A courier and logistics operator processing tens of thousands of Proof of Delivery (POD) documents every month across multiple hubs.
Problem
After the operator scaled its barcode-based renaming (see the Advance File Renamer case study), the next bottleneck appeared: not all PODs arrived with readable barcodes. Handwritten signatures, stamped consignment notes, and multi-page manifests came through daily, and staff had to identify the document type, locate the consignment number, and file each one manually. Claims disputes were decided by whether a POD could be found — and found quickly.
Solution
Gravitas India built an intelligent document workflow on top of the existing renaming engine. A computer-vision model classified every incoming document — delivery note, signed POD, manifest, or returns form — extracted the consignment number and delivery details from printed text, handwriting, and stamps, and routed each file into the searchable archive automatically. The human only stepped in when the model flagged low confidence.
Implementation / Technical Approach
- Document classification model — computer-vision classifier distinguishing POD types, trained on the operator's own archive of scanned documents.
- Hybrid extraction — printed text via OCR, handwritten and stamped fields via a vision model, with confidence scoring per field.
- Human-in-the-loop review queue — documents below confidence threshold routed to a triage screen, so the archive never contained mis-filed PODs.
- Rename-and-file pipeline — integration with the existing batch renamer meant extraction results flowed straight into the consignment-numbered archive structure.
- Search layer — every POD searchable by consignment number, mobile number, date, or hub, cutting lookup time from minutes to seconds.
Technology
Computer Vision OCR Handwriting Recognition Document Classification Human-in-the-loop
Outcomes
Within a month of deployment, 93% of incoming PODs flowed through the archive without human touch. Lookup time for a claims dispute fell from several minutes to under ten seconds — an 11× improvement — and the triage queue freed two staff-hours per day across each hub. The operator now handles claims disputes with an archive that is not just searchable but trusted.