How to Extract Handwritten Goods Receipt and Dispatch Data to Excel for Warehouse Reconciliation

The paper keeps coming to the dock

Warehouse receiving is one of the last fully paper-dependent workflows in the modern supply chain. A WMS might track every pallet movement inside the four walls. The ERP might generate purchase orders with digital precision. The transport management system might optimize routes down to the minute. But at the moment goods physically cross the dock threshold — the handoff between the supplier's delivery driver and the warehouse receiver — the data transfer mechanism is still a piece of paper and a pen.

The reasons are structural, not cultural. Delivery notes come from suppliers who each use different systems — or no system at all beyond a Word template and a dot-matrix printer. The driver who handed over the paperwork is already halfway to the next stop and can't answer questions about ambiguous quantities. The warehouse receiver works standing up, often in a refrigerated bay or a dusty loading area, wearing gloves, with a forklift horn blaring every two minutes. A tablet-based receiving app in this environment is aspirational. A clipboard is operational.

According to the Finale Inventory guide to the warehouse receiving process, the standard workflow involves pre-receiving preparation, unloading, quality inspection and counting, variance documentation, system updates, and organized putaway. In a paper-based receiving operation, the "system updates" step is where the bottleneck lives. Every handwritten note — every circled quantity, every scribbled batch number, every "3 damaged" annotation — must be manually transcribed into the WMS or inventory system before the goods can be put away. If the receiver is fast and the handwriting is clear, that step takes 15 minutes per delivery. If the receiver was rushed and the handwriting is rough, it takes 30. Across eight deliveries per day, that's two to four hours of typing — performed by someone who could instead be on the dock inspecting the next shipment.

Two layers of data on every receiving note

A warehouse receiving document is structurally different from an invoice or a standard form. It's a turnaround document: it leaves the supplier's warehouse printed with shipment data, travels with the goods, and returns covered in handwritten receiving confirmations., travels with the goods, and returns covered in handwritten receiving confirmations. These two layers — the printed layer that says what was sent and the handwritten layer that says what was actually received — have fundamentally different roles in the warehouse workflow. Traditional OCR that treats the document as a single text stream collapses this distinction, and the receiving workflow breaks as a result.

The printed layer contains supplier-generated data: supplier name and address, purchase order number, delivery note number, date, line item descriptions, ordered quantities, and sometimes batch numbers. This data already exists in the supplier's system and your purchase order. It's structured, consistently positioned, and — because it's printed — reads at near-perfect accuracy. The value of extracting it isn't the data itself (most of it is already in your PO) but the cross-reference: matching the printed PO number and line items to your open orders to confirm the delivery matches the order.

The handwritten layer contains receiver-generated data: actual received quantities (which may differ from the printed quantities), condition notes ("3 damaged," "short 2 cartons"), batch and serial numbers, pallet or location IDs, the receiver's signature or initials, and the timestamp of receipt. This is the operational ground truth — the data that determines whether inventory counts are accurate, whether suppliers get paid for the right quantities, and whether quality issues are traced to the right batch. As we explain on our handwritten delivery note to Excel page, "the printed layer says what was sent. The handwritten layer says what was actually received." The distinction is the foundation of the entire goods-in process.

Step-by-step: from paper goods receipt to structured Excel

The extraction workflow replaces the manual transcription bottleneck while preserving the dock-level workflow that receivers rely on. The receiver still uses a clipboard. The driver still hands over paper. What changes is what happens after the paper reaches the office.

Step 1: Capture the receiving documents

There are two practical approaches depending on your dock setup. If you have a document scanner at the receiving office, run delivery notes through the ADF (automatic document feeder) as a batch — a typical delivery with 3-5 pages takes under a minute to scan. If your receivers work at docks without nearby scanning equipment, a smartphone photo of each page works. Modern phone cameras produce images with sufficient resolution for both printed and handwritten text extraction. The key is to capture the full page — including margins where handwritten annotations, signatures, and dock stamps often live.

For operations where receiving happens at multiple docks or even multiple locations, Collection Link provides an alternative: generate a shareable upload page, and receivers at each dock submit their completed goods receipt forms directly to a central processing queue. No accounts, no login — just a link and a verification code. The documents land in your batch ready for extraction.

Step 2: Upload and define extraction columns

Upload all receiving documents for the day or shift in a single batch. Then define the columns you want extracted. This is where the two-layer approach becomes operational: you define columns that capture both the printed supplier data and the handwritten receiver data, keeping them separate in the output.

For a standard warehouse goods receipt workflow, a column set might look like this:

The critical column pair is "Ordered Quantity" and "Received Quantity." The ordered quantity is printed on the delivery note. The received quantity is handwritten — often written next to, on top of, or circling the printed number. The AI extracts both independently because they come from different data layers on the same page. The difference between them — the variance — is the first number any warehouse manager looks at to assess receiving accuracy.

Step 3: Let the AI process the batch

The AI reads each document, locates the data for each defined column, and populates a row in the output table. Printed fields (PO number, supplier name, item descriptions) extract at near-perfect accuracy. Handwritten fields — received quantities, condition notes, signatures — extract with accuracy proportional to handwriting legibility and form condition.

This is the mechanism that sets AI extraction apart from template-based OCR. Instead of defining zones on the page where each field should be (which breaks the moment a different supplier uses a different delivery note layout), you define what each field means. The AI finds "Received Quantity" by understanding that it's a number written near the printed line item, often circled or annotated — not by looking at pixel coordinates (350, 842). Different suppliers' delivery notes, different layouts, different positioning — same column definitions produce consistent output because the AI is reading for meaning, not position. We've detailed this mechanism in our guide to how AI handwriting recognition works.

Step 4: Review flagged fields and export

The output table shows one row per document, with extracted values for every column you defined. Fields where the AI was uncertain — ambiguous handwriting, poor contrast, partially obscured values — get flagged for review. The warehouse clerk scans the flagged fields (typically 2-4 per document), corrects any errors, and exports.

The export format matters for downstream integration. Excel (XLSX) is the most common target, and it feeds directly into most inventory management workflows — QuickBooks import, WMS data upload, ERP integration via CSV. The spreadsheet's column structure matches the WMS field names you used in your column definitions, so the data maps directly without reformatting.

Time comparison: manual transcription of a 3-page delivery note with 12 line items and handwritten annotations takes roughly 8-12 minutes. AI extraction plus flagged-field review takes 1-2 minutes per document. Across 8 daily deliveries, that's a reduction from roughly 80 minutes of typing to 16 minutes of review — a full hour reclaimed per day for dock-level work that actually requires a human.

Column strategies that capture both printed and handwritten data

The column definitions you choose determine the quality of your output. Here are strategies for the most common warehouse receiving scenarios:

Variance tracking. Define both "Ordered Quantity" and "Received Quantity" as separate columns. The output automatically includes both, and the variance (Received - Ordered) is one formula away in Excel. This replaces the manual process of flipping between the delivery note and the PO, typing two numbers, and calculating the difference — a 30-second mental operation on every line item that accumulates to real time across a full delivery.

Batch traceability. For warehouses handling lot-tracked inventory (pharmaceuticals, food, electronics), define "Lot Number," "Batch Number," and "Expiry Date" as separate columns. The receiver writes these on the goods receipt form — sometimes pre-printed by the supplier, sometimes handwritten from the product label. The AI extracts whichever format they appear in.

Condition reporting. Define a "Condition" column with inference support — e.g. "Condition (options: Good, Damaged, Short, Over, Wrong Item)." The AI reads the receiver's handwritten notes ("3 boxes crushed," "missing 2 units") and infers the appropriate condition category. This is inferred extraction: the AI classifies the document based on what it reads, even though the receiver didn't write a standardized condition code. If you also define a free-text "Notes" column, you get both the structured classification and the original comment.

Handling common warehouse document challenges

Warehouse documents face physical challenges that office documents don't. Here's what to expect and how to work around it:

Carbon copies. Multi-part NCR forms produce progressively fainter copies. The top (white) copy extracts normally. The second (yellow) copy is 15-20% fainter and will produce more flagged fields. The third (pink) copy is often too faint for reliable machine reading — at this point, human review of the entire form may be faster than correcting a majority of flagged fields. Best practice: always process the top copy when available. If you only have the third copy, scan it at higher resolution (300 DPI minimum) and expect to spend more time on review.

Oil, water, and dust. Dock-level documents collect environmental contamination. A delivery note that sat on a forklift seat for an hour will have smudges. A goods receipt form handled with warehouse gloves will have dirt marks. The extraction accuracy drop from contamination ranges from minimal (light dust) to severe (water damage that has smeared ink). The AI flags the affected fields. The pre-extraction step you can control: keep a clean clipboard and a document sleeve at each receiving station. A $3 plastic sleeve protects the form from direct handling and pays for itself in reduced review time on the first contaminated form.

Multi-part shipments. A single PO might arrive across three trucks on three different days — three delivery notes, three sets of handwritten annotations, one purchase order. Process each delivery note as it arrives, and consolidate the Excel outputs when all deliveries are complete. The column structure (PO number as the key field) makes this a straightforward VLOOKUP or Power Query merge.

FAQ

Can the AI read handwriting on top of printed text?

Yes. The AI processes the printed and handwritten layers separately, understanding that the handwritten "197" written next to the printed "200" is a correction, not noise. This two-layer reading is the foundation of the delivery note extraction workflow. However, when handwriting directly overlaps printed text — the receiver writes directly on top of a printed number — accuracy drops. Most receivers write next to or below the printed quantity, which preserves both layers.

Does this work with delivery notes from different suppliers?

Yes, without per-supplier configuration. Because the AI finds data by understanding what it means rather than where it sits on the page, the same column definitions work across different suppliers' delivery note layouts. A "PO Number" is a PO number whether it's in the top-right corner on Supplier A's form or the bottom-left on Supplier B's.

Can I import the output Excel directly into my WMS?

Most WMS platforms — including Manhattan, Oracle WMS Cloud, Fishbowl, and SAP EWM — support CSV or Excel import for goods receipt transactions. The extracted data, structured in columns you defined, maps directly to the WMS import template. The only additional step may be adding warehouse-specific internal fields (warehouse code, bin location default) that aren't on the delivery note — these can be added as fixed-value columns in Excel before import.

What about dispatch forms — does the same workflow work?

Yes. Dispatch forms (picking lists, packing slips, outbound delivery notes) have the same two-layer structure in reverse: the warehouse prints what should be picked, and the picker handwrites actual quantities picked, substitutions, and any issues. The column definitions change — you'd define "Picked Quantity," "Location Picked From," "Picker ID" — but the extraction mechanism is identical.

How do I handle the receiver's signature?

Signatures are extracted as an image reference — the AI recognizes them as signatures and doesn't attempt character recognition on them. For audit trails and compliance requirements (ISO 9001 Clause 7.5 documented information), the original scanned document is retained alongside the extracted data. The signature lives in the scan. The structured data lives in the Excel. Both are preserved.